Sie sind auf Seite 1von 14

Solvent effects in organic chemistry - recent developments

MICHAEL H. ABRAHAM AND PRISCILLA L. GRELLIER


Department of Chemistry, Universiry of Surrey, Guildford, Surrey, GU2 5 X H , United Kingdom

JOSE-LUIS M. ABBOUD
Consejo Superior de Investigaciones Cientificas, Institute de Quinlica Fisica Rocasolano Serrano, Madrid 119-28006, Spain

RUTHM. DOHERTY
Naval Surface Weapons Center, White Oak Laboratory, Silver Spring, MD 20910, U.S.A.
AND

Can. J. Chem. Downloaded from www.nrcresearchpress.com by 189.4.76.40 on 11/28/13 For personal use only.

ROBERT W. TAFT
Department of Chemistry, Universio of California, Irvine, CA 92717, U.S.A. Received May 20, 1988

MICHAEL H. ABRAHAM, PRISCILLA L. GRELLIER, JOSE-LUIS M. ABBOUD, RUTHM. DOHERTY, and ROBERT W. TAFT. Can. J. Chern. 66,2673 (1988). Solvent effects on a number of different processes have been surveyed, and results of the application of multiple linear regression analysis are discussed. The processes examined include examples of solubility of gases or vapours, distribution coefficients of solutes between water and a series of solvents, and solvent effects on conformational equilibria, on keto-en01 tautornerism, and on reaction rates. It is shown that two particular equations, that due to Koppel and palm and extended by Makitra and Pirig, and that due to Abraham, Kamlet, and Taft, can cope quite satisfactorily with solvent effects on these various processes. It is pointed out that interpretation of parameters obtained from equations that involve macroscopic quantities such as AG* or AGO is not necessarily straightforward, and that some model is needed in order to interpret these macroscopic quantities in terms of qicroscopic quantities that can characterise, for example, solute-solvent interactions. MICHAEL H. ABRAHAM, PRISCILLA L. GRELLIER, JOSE-LUIS M. ABBOUD,RUTH M. DOHERTY et ROBERT W. TAFT. Can. J. Chem. 66, 2673 (1988). On a CvaluC les effets de solvant sur un certain nombre de processus et on discute des rCsultats de l'application de l'analyse de regression linCaire multiple. Les processus Ctudies comportent des exernples de solubilitt des gaz ou des vapeurs, des coefficients de distribution des solutCs entre l'eau et une sCrie de solvants ainsi que des effets de solvant sur 1'Cquilibre conformationnel, sur la tautornerie ctto-Cnolique et sur les vitesses de reaction. On dCmontre que deux Cquations particulikres, celle de Koppel et Palm qui a CtC Ctendue par Makitra et Pirig ainsi que celle d' Abraham, Karnlet et Taft, peuvent rendre compte d'une fason satisfaisante des effets de solvant sur ces divers processus. On note que llinterprCtation des parametres obtenus & l'aide de ces Cquations qui impliquent des quantitCs rnacroscopiques cornrne le AG* ou le AGO ne sont pas simples; des modkles sont ntcessaires pour interprkter ces quantitCs macroscopiques en fonction de quantitks microscopiques qui peuvent caracteriser, par exemple, les interactions solutC-solvant. [Traduit par la revue]

Introduction The first experimental observation to show that "inert" solvents could affect the rate of a chemical reaction was provided by Berthelot and Saint-Gilles ( I ) . However, it was the classic paper of Menschutkin ( 2 ) in 1890 that initiated the widespread investigations on solvent effects that have charactensed a large area of physical organic chemistry. Menschutkin ( 2 ) showed that rate constants for the reaction of triethylamine with ethyl iodide at 373 K were highly dependent on the solvent, there being factors of up to 700 between rate constants in hexane and acetophenone. Since the work of Menschutkin, there have been published a host of papers on solvent effects in organic chemistry, mainly dealing with reaction rates but also on equilibria. At the outset, it is useful to consider the objectives of studies of solvent effects on reaction rates and other processes. In principle, there are two main objectives, (a) the prediction
'TO whom correspondence should be addressed.
Rinled in Canada 1 lmprime au Canada

of rate constants in other solvents, and (b) to reach some understanding of the various influences that might affect reaction rates. The first objective is certainly important as regards testing theoretical models in the Popperian sense, but practically has been but little used. In the few recorded instances in which rate constants have been estimated for some particular purpose, rather simple procedures have been used. Thus log k values for the solvolysis of t-butyl bromide or t-butyl iodide in aqueous organic media can be estimated from corresponding values for t-butyl chloride, cf. the well-known GrunwaldWinstein mY equation. Again, rate constants for the Et3N/EtI reaction can be estimated from corresponding values for the related Pr3N/MeI reaction, and vice versa, if required. By far the greatest use of solvent effect studies has been in developing an understanding of the factors that influence reaction rates, that is to probe the reaction at a microscopic level in order to obtain information on the nature of any solute-solvent interactions, etc., that may be set up. Unfortunately, there is no direct logical connection between

2674

CAN. J . CHEM. VOL. 66, 1988

the macroscopic solvent parameters that are used in studies of solvent effects, and microscopic details of reaction processes. In order to establish such a connection, it is necessary to construct a "link", usually based on some theoretical model. As an example we can take the effect of solvent dielectric constant (E) on reaction rate constants. It is quite well known that for a restricted range of solvents, usually aliphatic non-halogenated, non-hydroxylic solvents, there is often a quite good correlation between rate constants (as log k or AG*) and functions such as 1 / or ~ (E - 1 ) / ( 2 ~ + 1). Steiner and Huisgen (3a) found for the addition of tetracyanoethylene (TCNE) to butylvinylether a correlation of the type?
- 1)/(2~ + 1) + constant [ l ] AG* = - 1 6 . 1 ( ~ Such a correlation, by itself, leads to no information whatsoever of a microscopic nature. The "link" or model, in this case, was provided by Kirkwood (4) who showed that the electrostatic part of the Gibbs energy of a species was given by eq. [2], if the species could be regarded as a dipole of moment p. in a sphere of radius r. In eq. [2], AGE is in kcal mol-', p in Debyes, and r in A.

Can. J. Chem. Downloaded from www.nrcresearchpress.com by 189.4.76.40 on 11/28/13 For personal use only.

considered by Steiner and Huisgen (3); and Abraham (7) has listed a number of transition-state dipole moments obtained via the Kirkwood equation [3]. More sophisticated reaction-field equations have been applied to the Et3N/EtI reaction and to the heterolysis of t-butyl chloride (8, 9) leading also to transitionstate dipole moments. volumes of activation can also be correlated with the pressure derivative of the Kirkwood function to yield the microscopic quantities p: / r and hence p. + , as shown by McCabe and Eckert (10). Activation entropies can similarly be correlated with the temperature derivative of the Kirkwood function to give the same microscopic quantities (11). Application of the Kirkwood function and its derivatives has mostly been to a series of pure organic solvents. Indeed, following Menschutkin, for many years studies on mixed solvents were rather.uncommon. The advent of the WinsteinGmnwald equation (12, 13) specifically designed to apply to solvolyses in aqueous organic media brought about a radical change. Since that time, there have been a very large number of studies using such mixed solvents, usually in terms of the original mY equation [5] or some variant.

Laidler and Eyring (5, 6) then combined eq. [2] with the transition state theory, to yield eq. [3] for the effect of solvent dielectric constant on rate constants, as AG*. [3] AG* - AG;

The total nonelectrostatic effect is denoted as A+, and A , B and the symbol =+ stand for reactants A and B and the transition state. If A+ is zero or constant, then a plot of AG* against (E - 1 ) / ( 2 ~ + 1) will yield a line from whose slope the term in the square parentheses can be evaluated. Then knowing pA, rA, k g , and r g , the term p g /r: can be deduced. If r + can be estimated, the microscopic quantity p+ can be obtained. Steiner and Huisgen (3) carried out these calculations and found a reasonable value of 10.3 D for p + . They also calculated a dipole moment of 17 D for the zwitterionic intermediate, so that the transition state seems reasonably close to the intermediate.
[ 4 ] (CN)2C=C(CN)2

[5] log k/ko = mY = m log (k/ko)r-BuC' Here k and ko denote rate constants in a solvent and in the standard solvent of 80% aqueous ethanol, and m is defined as unity for the solvolysis of t-butyl chloride. The difficulty with the Winstein-Gmnwald equation is that no link is provided between the macroscopic quantities in eq. [5] and microscopic parameters characteristic of the reaction or transition-state. Hence from eq. [5] it is impossible to deduce any such microscopic parameters. Quite recently, however, there has been an interesting new approach to the study of medium effects in highly aqueous solutions. Blandamer, Engberts, and co-workers (14) have shown that by assuming a specific solvation model for the transition state, it is possible to apply pairwise group interaction parameters to reactant/solvent and transition-state/solvent interactions and hence to explain the effect of additives to aqueous media. They have carried out such an analysis for the hydrolysis of 4-methoxyphenyl dichloroacetate in aqueous solutions of urea and alkyl-ureas, and extension to other systems is eagerly awaited.

+ H2C=CH-OBu

+ [Tr]

We have dealt with this in some detail, because it is a very simple example of how a model or "link" can be set up in order to connect microscopic details of reaction mechanisms with bulk solvent properties. Other cycloaddition reactions were
'we shall deal in this paper not only with rate constants but with Gibbs energies. In order that the constants in various regression equations are all compatible, we shall transform any equations in log k into equation in AG*/kcal mol-', using the factor - 1.36425 at 298 K .

Multiple linear regression analysis We now turn to processes in pure organic solvents. As far as rate constants are concerned, it is now quite clear that there is no single solvent parameter that will satisfactorily correlate log k values for a variety of widely differing reactions (15). Hence there has developed considerable interest into correlations by the method of multiple linear regression analysis (MLRA). The first general system of analysis by MLRA was devised by Koppel and Palm (16, 17), specifically for the correlation of rate constants. They argued that a number of solvent influences could be specified on general chemical grounds, and put forward their well-known equation [6]:
[6] l o g k = log ko

+ g . f ( ~ )+ p.f(q) + c . E + b.B

Here f ( ~ ) is a dielectric constant function, usually taken as (E - 1 ) / ( 2 ~+ l), f ( q ) is the refractive index function ( q 2 - 1)/(q2 + 2), and E and B are measures of solvent electrophilic solvation ability and nucleophilic solvation ability,
3 ~ oarrecent article, see ref. 3 b .

REVIEW,

Can. J. Chem. Downloaded from www.nrcresearchpress.com by 189.4.76.40 on 11/28/13 For personal use only.

respectively. Although Koppel and Palm did not explicitly set out a model to link the macroscopic quantities in eq. [6] with microscopic quantities, their approach is not too dissimilar to that of Abraham, Kamlet, and Taft and their co-workers (18-22) who have devised a more general equation based on a cavity theory of solution. In this approach, the solution of a solute in a solvent may be broken down conceptually into a number of stages, ( a ) the creation of a cavity of suitable size in the solvent, (b) reorganisation of solvent molecules around the cavity, and (c) introduction of the solute into the cavity, to set up various solvent-solute interactions. The endothermic or endoergic term (a) depends on the forces holding the solvent molecules together and on the size of the cavity. A measure of the solvent forces is the Hildebrand cohesive energy density, (6;)i, defined as the enthalpy of vaporisation of the solvent at 298 K per unit volume: [7] (6A), = (AH: - R T ) / V , To a first approximation, the cavity size can be taken as proportional to the solute volume V2; note that solvent properties are denoted by the subscript 1 and solute properties by the subscript 2. The reorganisation energy is probably not important, at least in terms of Gibbs energy, but the final term (c) will involve a number of exothermic or exoergic interactions, depending on the nature of the solvent and solute. Acid-base interactions will be set up if the solvent is a hydrogen-bond acid ( a and the solute a hydrogen-bond base (P 2), or if the solvent is the base ( P I ) and the solute the acid ( a 2 ) In addition there may be dipolar interactions of the dipole-dipole and dipole induced dipole type, proportional to the dipolarity ( . r r x ) of solvent and solute (.rrT~ 5 ) .A . full equation for the solution of a solute in a solvent will therefore be given by eq. [8]. [8] AGO = A G + ~ A.rrT..rr* + a l . p2 + CP1. a 2

+ ~(6;)i.V 2 If we consider only the case of a single solute in a series of solvents, all the solute parameters are constant to leave an equation for the effect of solvents only:4
Equation [9], the AKT equation, can be applied not only to the solubility of a single species, but to equilibrium constants, rate . ~ of the solvent parameters .rrf, constants (as AG'), e t ~Values a and p have been listed for a wide range of solvents (23). Some of these, together with a number of cohesive energy densities, are given in Table 1. Whenever possible, the latter have been calculated from calorimetrically-determined AH: values at 298 K. Because the original Koppel-Palm equation [5] contains no parameter that can deal with a cavity term, it is restricted to processes in which cavity effects largely cancel out, i.e. to correlations of rate constants where initial-state and transition-state cavity contributions do, indeed, in general cancel. Makitra and Pirig (24,25), however, have added the 6; cavity term to the Koppel-Palm equation giving a more general equation [lo] that covers the same range of applicability as the AKT equation [9] . 6 4 ~ hterm e d.6 in eq. [9] is a polarisability correction term, with 6 taken as 0.5 for polyhalogenated solvents, 1.0 for aromatic solvents, and zero for all others. 'Note that in the first papers of AKT on this subject, the parameter 6Hwas used (18, 19) rather than the better parameter 8;. 6~akitra and Pirig (24, 25) use a log K term rather than AGO in eq. [lo].

Equations [9] and [lo] have both been applied to numerous solubility-related phenomena, generally with rather good results. The equations clearly closely resemble each other in the identification of the various solvent properties that will influence solvent-solute interactions, viz. solvent dipolarity as .rrf or f (E), solvent hydrogen-bond acidity or electrophilicity as a , , or E, and solvent hydrogen-bond basicity or nucleophilicity as p i or B. Both equations have their disadvantages. The AKT equation [9] suffers in that there is no parameter directly related to solvent-solute dispersion interactions, and that for solvents with 6 not equal to zero, it is not so easy to interpret the d. 61s. .rr'f terms. On the other hand, the modified Koppel-Palm equation [lo], includes a solvent basicity term B that is directly derived from the change in the OH infrared stretching frequency of phenol with the solvent, when both phenol and solvent are present in dilute solution in C C 4 (26). Thus the B-scale is actually a solute scale and is not a solvent scale at all. Although a solute B-scale can probably be used as a solvent B-scale for nonassociated compounds, it is most unlikely that basicities of associated bulk solvents can be matched to those of monomeric solute species .7 There have been numerous applications of eqs. [6], [9], and [lo], and we shall review these by taking examples from a variety of processes. It is as well, however, to point out a number of conditions that should be obeyed in such applications. Firstly, the solvent parameters in any given data set must not be colinear. If any two parameters are colinear over the given data set, incorporation of both of the parameters into eqs. [6], [9], or [lo] may lead to meaningless coefficients. Secondly, parameters must be available for enough data points; usually five data points per each explanatory variable would be regarded as the minimum number. Thirdly, some statistical procedure, based on Student's t-test or the F-statistic, must be used to avoid the accumulation of extra unnecessary variables. In the equations that follow, we shall either exclude all parameters for which the confidence level (CL) is less than 95%, or else show these parameters specifically. Fourthly, and very importantly, results of application of equations based on some chemical model must make general chemical sense. Thus if solvent hydrogen-bond acidity ( a seems to be important in a process that does not involve any solute hydrogen-bond bases, the resulting equation cannot be taken seriously. Similarly, for a process in which there is no, or little, volume change with respect to the species involved, the cavity contributions should cancel out, and the term in 6; should be redundant. This must certainly be the case for spectroscopic transitions that obey the Frank-Condon principle. However, we shall not discuss these here, since they have already been examined in detail (23, 27).

Application of MLRA to equilibria There are a number of equilibria in which the cavity term (in 6 ; ) should cancel between reactants and products, especially
7The original Koppel-Palm equation [6] used v(Me0D) in pure solvents as the basis of the B-parameter, but Koppel and Paju (26) suggested the alternative v(Ph0H) solute scale because of practical difficulties over the v(Me0D) solvent scale. Koppel and Paju were well aware of the assumption involved in using a solute scale for a solvent scale, but concluded (26) that it was a satisfactory approximation.

CAN. 1. CHEM. VOL. 66, 1988

TABLE1. Some solvent parameters used in this worka No. Solvent Water Methanol Ethanol 1-Propano1 2-Propanol 1-Butan01 t-Butanol 1-Pentan01 3-Methylbutanol 1-Hexan01 1-Heptanol 1-Octanol Ethylene glycol Formamide Trifluoroethanol Hexafluoro-2-propanol Acetic acid Benzyl alcohol Benzonitrile Nitrobenzene Acetophenone Iodobenzene Bromobenzene Chlorobenzene Anisole Ethyl benzoate Phenetole Benzene Toluene o-Xylene m-Xylene p-Xylene Pyridine Propylene carbonate Dimethyl sulphoxide Nitromethane Dimethylformamide Acetonitrile N-Methylpyrrolidinone Nitroethane Dimethylacetarnide Propnonitrile Dioxan Cyclohexane Propanone Tetrahydrofuran Butanone Ethyl acetate Decalin Butyl acetate Pentyl acetate HMPA 1,2-Dimethoxyethane Cyclohexane Hexadecane Di-n-butylether Decane Octane Diethylether Di-isopropy lether Heptane Hexane Pentane Iso-octane

.~r?

aI

PI

Can. J. Chem. Downloaded from www.nrcresearchpress.com by 189.4.76.40 on 11/28/13 For personal use only.

TABLE 1 (concluded) No. Solvent 1,2-Dichloroethane 6


n ;
I

PI

(631

1,1,2,2-Tetrachloroethane
Dichloromethane Trichloromethane 1,1,2,2-Tetrachloroethylene 1,1,2-Trichloroethylene 1-Bromobutane I -Bromoheptane Tetrachloromethane I, 1,1-Trichloroethane cis- l,2-Dichloroethylene

Can. J. Chem. Downloaded from www.nrcresearchpress.com by 189.4.76.40 on 11/28/13 For personal use only.

'For an extensive list of a ; , cr , and calculated from AH: values at 298 K.

P I see ref. 23. Cohesive energy densities are in kcal dm-3 as

conformational equilibria where reactants and products will have almost exactly the same intrinsic volume. Although there have been numerous studies on solvent effects on conformational equilibria (28), only in very few cases have sufficient solvents of a wide variety been examined. In one study, Abraham (29) measured the nmr coupling constant, J, between the C1 and C2 protons in 1,1,2-trichloroethane in 22 aprotic solvents as 10% solutions. Solvent parameters are available for nineteen of these solvents, Table 2, and an analysis through eq. [9] leads to: [ll] J
= (6.27

TABLE 2. Parameters for the conformational equilibrium of 1 , l ,2-trichloroethanea Solvent Pentane Hexane Decalin Tetrachloromethane Benzene 1,1,2-Trichloroethylene Di-isopropylether Diethylether Trichloromethane 1-Bromoheptane Dichloromethane 1,2-Dichloroethane Butanone Propanone Acetonitrile . Nitromethane Dimethylformamide Dimethylacetamide Dimethylsulphoxide Carbon disulphide J(29) 6.25 6.21 6.17 6.17 5.94 6.13 6.04 6.05 6.03 AC0(28) 1.25 1.24 1.17 1.14 0.63 0.92 0.84 0.79 0.75 AC0(30)

t 0.05) + (0.25

* 0.10)6 - (0.70 + 0 . 0 9 ) ~ ;
- (0.45 t 0.12)PI

Abraham and Bretschneider (28), and also Chen et al. (30), calculated AGO values for equilibrium [12] from the original (29) J-values. The two sets of AGO values are not the same, due to different assumptions over the absolute values for the conformers (I) and (II), and so we have correlated both sets of results, with AGO in kcal mol-', eqs. [13], [14].

1,1,2- richl lo roe thane^


Mesityl oxide Sulpholane

6.13 5.92 5.64 5.31

1.03 0.58 0.32 -0.02

0.48 0.30 0.08 -0.15

"All values of J obtained as 10% solutions, except where shown. bBulk pure solvent.

As expected, the term in 6; is not significant, and the main factor influencing the equilibrium is solvent dipolarity. Since (11) will have a larger dipole moment than (I), any increase in T; should decrease AGO, exactly as observed in all three correlations. An unexpected observation is that solvent basicity also favours (11) over (I). Now in (11) the acidic C1 proton is trans to the C2 chlorine atom, and hence will be slightly more acidic than the C1 proton in (I), so that an increase in basicity should favour (II), as observed. Although the absolute values

of the coefficients in eqs. [13] and [14] differ, the general interpretation remains the same. However, the calculated AGO value in the gas phase is not the same; if we take 6 = 0, P I = 0 and, from Abboud et a l . (31), T : = - 1.1, then on eq. [13] AGO (gas phase) = 2.46 kcal mol- and on eq. [14] AGO (gas phase) = 1.24 kcal mol-'. Abraham and Bretschneider (28) gave a range of possible values from 1.7to 2.7 kcal mol-' with a6'best" value of 1.9 kcal mol-', whereas Chen et a l . (30) using their own values of AGO extrapolated a value of 0.81 kcalmol-'. Interestingly, Abraham and co-workers (32) suggested a much lower value of -0.44 for the gas phase value of T:, and this lower value yields AGO (gas phase) as 1.72 from eq. [13] and 0.86 from eq. [14], in much better agreement with the extrapolations given (28, 30). Another equilibrium that involves conformers which differ in

'

2678

CAN. I. CHEM. VOL. 66, 1988

dipole moment is that given as eq. [15]. Abraham and Griffiths (33) obtained AGO values in 13 non-hydroxylic solvents from which we have calculated equation [16]. The fit is not very good, but we have included all the solvents studied, including CC14 that the authors felt was anomalous. General chemical sense is maintained in that the 6; term is not significant, and that an increase in solvent dipolarity favours the more polar conformer (IV).

(111)
Can. J. Chem. Downloaded from www.nrcresearchpress.com by 189.4.76.40 on 11/28/13 For personal use only.

p. =

3.17 D

(IV) p.=4.32D

A gas phase value of 1.5 kcal mol-' was suggested (33), as compared to one of 1.57 kcal mol-' calculated from eq. [16] if .rrf = -0.44 for the gas phase (32), but no less than 2.60 if .rr.; is taken as - 1.1 for the gas phase.8 Another, much studied, equilibrium that might also be uninfluenced by solvent cohesive energy density is the wellknown keto en01 equilibrium. A recent study by Emsley and Freeman (35) has yielded AGO values from results at zero concentration in 21 protic and aprotic solvents for pentane-2,4dione, eq. [17].

'171

Me\

c II

Ff"\ /Me c
0

A,

Me,

Il

FH\
0.

pentane-2,5-dione in 12 solvents, but in neither case are there really enough solvents for MLRA to be carried out rigorously, although the dominant effect can still be seen to be that of solvent dipolarity . One of the first processes to be investigated by either the extended Koppel-Palm equation [lo] or the AKT equation was that of solubility. Various parameters can be used to express solubility, for example, the Ostwald solubility coefficient L, the Henry's law constant K H , or AGO, the standard Gibbs energy of solution based on some particular standard states. To avoid confusion over the sign of the coefficients in regression equations, we shall either set out equations in terms of AGO, or when equations have been cast in terms of log K values, we shall recast these as equations in AGO. Naturally, this does not affect any "goodness-of-fit" as denoted by the correlation coefficient. On the cavity theory of solution, solubility of a solute in a series of solvents (or transfer of a solute from one solvent to another) should be inhibited by increase in cohesive energy density so that in any regression on AGO, the 6; coefficient should be positive. Any solute-interaction that is set up will aid f ( ~ ( ) E, , solution and hence the coefficients of terms in f(~), and B in eq. [lo] or terms in IT?, a , , and P, in eq. [9] should correspondingly all be negative. These considerations must be kept in mind when assessing any particular application, see the fourth condition, above. Makitra and Pirig and their co-workers have applied eq. [lo] to the solubility of a number of gases in a series of solvents (24, 25,36-41). For example, the solubility of Freon-22 (CHClF2) in 15 solvents is given by eq. [20] where N is the mole fraction solubility at 293 K (40): log N(CHC1F2) = 0.838 - 0.416;/100 [20] C
n = 15,

/Me

I1

+ 0.00281B
- 0.053E

,o

sd

0.161,

r = 0.966

Of the 21 solvents, parameters were not available for formic acid or diglyme, but the remaining 19 solvents yielded eqs. [18] and [19]. The %CL for the a l term in eq. [18] is 96.7, so that this parameter is just about significant; AGO is in kcal mol-'.

To be consistent with our other equations, we can simply ~ the standard Gibbs transform eq. [20] into [21], where A G is energy of solution in kcal mol-' for the transfer from the gas phase to solution:

As suggested by Emsley and Freeman ( 3 3 , polar solvents markedly favour the more polar keto form. Hydroxylic solvents are not anomalous and so the en01 hydrogen-bond must remain intact, again as pointed out by Emsley and Freeman. There is a possibility, eq. [18], that solvents that are hydrogen-bond acids may stabilise the keto form preferentially. The keto form should be more basic than the en01 form, certainly, but the effect is quite small. Emsley and Freeman (35) also gave results for 3-ethylpentane-2,Cdione in 6 solvents and for 3-methyl1.1 has been obtained from gas phase spectrophotometric data (31) whereas that of -0.44 is an indirect value from the known gas phase AGO value for the 1,2-dibromo-4-tbutylcyclohexane conformational equilibrium (32). Suppan (34) has suggested that gas phase spectrophotometric shifts should not be compared with those in solution, and it is possible that for chemical processes the value of -0.44 is better.
-

Makitra and Pirig (40) point out that the two main factors influencing solution are solvent basicity which favours solution and solvent cohesive energy density which inhibits solution. The solvent electrophilicity, E , also seems to inhibit solution, but for the 15 solvents examined, the correlation coefficient between 6; and E is no less than 0.749, so that 6; and E are no longer independent parameters. With the exclusion of solvent THF, Makitra and Pirig (40) obtained the more sensible c~rrelation,~

he gas phase value of

The problem of parameter colinearity still exists if the AKT equation is applied to the solution of CHC1F2 since for the 14 solvents for which we have available values there is a marked ' ~ o t ethat Makitra and Pirig list ti& values as 8&/1000when 6& is in kcal dm-3 (Table 1). In addition, many of the 6; values of Makitra and Pirig differ substantially from those that we consider to be the "best" values.

TABLE 3. Correlations of AG?, kcal mol-' at 298 K on the mole fraction scale, from methanol of some ion-paus and pairs of dissociated ions, through the AKT equation [9Ia

Coefficients Solute Me4NI Me4NBr MeaCl Eta1 EtaBr Et4NCl

A G ~
10.9 15.1 18.3 10.9 15.4 18.2 9.5 28.2 32.8 36.2 26.6 31.2 34.7

s.7~;

a.al -6.2 -10.4 -13.3 -6.8 -11.4 -14.2 -8.0 -16.4 -20.7 -23.9 -16.3 -20.6 -23.7

h(6&/100)~ 2.2 2.8 2.9 2.4 3.1 3.1 3.3 6.3 6.7 6.8 6.3 6.7 6.8

n
18 18 17 18 16 15 11 14 13 13 14 13 13

sd
0.3 0.5 0.6 0.4 0.5 0.6 0.5 1.1 0.9 0.9 1.2 1.1 1.0

h a ! M e q + ICan. J. Chem. Downloaded from www.nrcresearchpress.com by 189.4.76.40 on 11/28/13 For personal use only.

M e q + Er M e p + ~i E t q + IEtq +B ? ~ t +a ~i

-15.6 -18.8 -20.0 -14.9 -18.0 -18.9 -14.8 -42.0 -44.5 -45.5 -39.3 -41.9 -42.8
=

0.997 0.996 0.997 0.997 0.997 0.997 0.990 0.988 0.993 0.994 0.981 0.988 0.992

"All solvents are aliphatic with 6 see ref. 44 and also Table 4.

0, and in all cases the term in P , is statistically not significant,

correlation between 8; and a l ( r = 0.722); we have had to exclude tributylphosphate for which we have no reliable 6; value. With the exclusion of any term in a l ,the best equation is, The coefficients in eq. [24] make general chemical sense. Thus the (8;) term shows that increase in cohesive energy density inhibits solvation, exactly as expected on a cavity theory of solution, whilst increase in solvent hydrogen-bond acidity considerably aids solution, again exactly as expected for a solute that is a strong base. On the other hand, the reported (41) coefficients in eq. [25] make little general chemical sense - there is no apparent effect of solvent cohesive energy density, increase in solvent polarity as f ( ~ retards ) solution, and an increase in solvent basicity B aids solution of the basic trimethylamine. This is an example of how application of the regression eqs. [9] and [lo] must be carried out in accord with general chemical principles as well as with statistical ones. The number of possible applications of eqs. [9] and [lo] to the solubility of solutes is almost limitless, but we give possibly the most complicated example studied to date, that of the solubility (as transfer Gibbs energies) of tetra-alkylammonium halides and ion-pairs (44) using results for up to 18 aliphatic aprotic and hydroxylic solvents for which 8 in eq. [8] is zero. or (%A X-) it would be For compounds of type expected that solvent dipolarity and hydrogen-bond acidity would greatly reduce the value of AG:, and that solvent cohesive energy density would considerably increase AG: through the endoergic cavity term. This is exactly as found, see ref. 44 and a summary of the regression equations in Table 3. What is somewhat unexpected is that solvents that are hydrogenbond bases, all of which have lone pairs of electrons, seem to have no effect at all on AG:. The coefficient in the b. P term is always statistically insignificant, see also Table 4, and hence there can be no interaction between these basic solvents and the quaternary ammonium positive centre. The unit positive charge, even in the dissociated ions, must be almost completely shielded from the solvent. Inspection of Table 3 shows that the coefficients of .rr; and a , very with solute in a chemically sensible way. Both the s..rrf and the a . a , term become more

Both eqs. [22] and [23] lead to a chemically sensible interpretation, namely that the endoergic cohesive energy density term is counterbalanced by an exoergic solute-solvent interaction term that arises through hydrogen bonding between the somewhat acidic CHClF2 molecule, and basic solvents. The above discussion shows the difficulty of MLRA when explanatory variables are not independent. Other difficulties may arise if general chemical sense is not maintained. Abraham, Kamlet, and Taft (42) used a truncated version of eq. [9] to account for the solution of a number of solutes into a set of 22 solvents that were all non-acidic ( a , = 0), using just the .rr; and (SH) terms. Makitra and Pirig (41) re-examined this work and showed that for two solutes, trimethylamine and 4-nitrobenzyl chloride, the truncated version of the AKT equation did not apply when the solvent set was extended to cover hydrogen-bond acidic solvents. They concluded, obviously erroneously, that the AKT equation was not satisfactory. Clearly, for solutes that are hydrogen-bond bases, a term in a must be included in the regression equation. If this is done, then for all 15 solvents mentioned by Makitra and Pirig (41) the best equation for transfer Gibbs energies (43) from a standard solvent, in kcal mol-' is given as:

%fix

This compares very favourably with the corresponding correlation using the five parameter extended Koppel-Palm equation for which r = 0.905 for the same 15 solvents. Makitra and Pirig (41) do not give the resulting equation for these 15 solvents, but after the exclusion of chloroform find eq. [25]:

2680

CAN. J. CHEM. VOL. 66, 1988

TABLE 4. Correlations of AG:, kcal mol-I at 298 K on the mole fraction scale from methanol of some ion-pairs through the AKT equation[9Ia Coefficients Solute
% CLb

d.6

s.7~;

a.al

b. P,

h(6;/100)~

sd

.srf

a1

PI

6;

Can. J. Chem. Downloaded from www.nrcresearchpress.com by 189.4.76.40 on 11/28/13 For personal use only.

"Solvents in the first set, n = 18, are aliphatic with 8 = 0. bPercentage CL. The entry "100" denotes CL > 99.999%.

negative (exoergic) along the series I-, Br-, and C1- both for the R4NX and (%Nf + X-) solutes, leading to increasing stabilisation of the chlorides by dipolar and by hydrogen-bond acid solvents. The h.6; term becomes more positive (endoergic) from Me4Nf to Et4N+ for the ion-pairs, but not so for the dissociated ions. Furthermore, the increase in h.6; is always in the order I- < Br- < C1-, the reverse of the order of ionic radius. It was suggested (44) that solvent electrostriction and possible solvent reorganisation around the ions or ion-pairs may play a significant role. One difficulty over the reported (44) regression is that for the 18 solvents examined (Table 3) there is a non-negligible correlation between IT: and 6; ( r = 0.585) and between a ,and 6; (r = 0.641), so that extension to a wider set of solvents would be of interest. Results are available (45) for five other solvents (nitrobenzene, benzonitrile, chlorobenzene, benzene, and tetrachloromethane) with several of the ion-pairs, and for an extra eleven solvents for Et4NI (the five given, plus ethyl benzoate, 1,1,2,2-tetrachloroethane,dichloromethane, 1,2-dichloroethane, anisole, and bromobenzene) and so we have applied eq. [9] to an extended solvent set. Unfortunately, this still does not resolve the problem of colinearity ( r = 0.435 and 0.658) for the two above cases over the full 29 solvents, but if the equations in Table 3 are artificial, then incorporation of a different set of solvents should alter the magnitude of the coefficients considerably.1 In Table 4 are details of regressions for ion-pairs, where now the d.6 term has had to be included. We give also the b. p term, just to show how insignificant it is. Results in Tables 3 and 4 are very similar, as are the results in the two sections of Table 4. Hence we can conclude that the original equations (44) in Table 3 are, indeed "robust". A process closely related to solubility is that of the distribution of a solute between solvent phases. Indeed, since the distribution, or partition, coefficient K , is defined by eq. [26] where A, and A , are the activities of the solute in solvent (s) and water (w), K p is simply the ratio of the solubilities in the two phases, provided that the ratio of the secondary medium activity coefficients is unity. Thus for distribution of a given solute between water and a number of solvents, log K , or AGO, might
- namely

5 . Carbonyl compounds used in TABLE extraction experiments (46)

Vanillin

Orthovanillin

M e & . CHO

Isovanillin

&cHO

Salicylaldehyde

CH3COCH2COCH3 PhCOCH2COCH3

Acetylacetone Benzoylacetone

'O~hisis quite a good test of the "robust" character of a regression that the coefficients should remain the same for various subsets of data.

be expected to depend on (6;), and on parameters that reflect the various solute-solvent interactions in the solvent phases. Makitra and Pirig (46) have applied the extended Koppel-Palm equation [lo] to the distribution of several carbonyl compounds between water and a number of solvents. Because the chosen solvents must be immiscible with water, the solvent variation is, perforce, limited. Furthermore, all the distributions refer to the water-saturated solvents and not the normal, dry, solvents, so that correlations using parameters for dry solvents would be expected to be not as good as usual. For a number of the carbonyl compounds, there are too few data points available, but for those shown in Table 5, we give the best regressions obtained using the AKT equation [9]; as usual we have transformed the given (46) K p values into AG; values in kcal mol-'. Results are in Table 6 for six carbonyl compounds. By and large, the regression coefficients make general chemical sense. Thus distribution is hindered by the endoergic (a;),

TABLE 6. Application of the AKT equation [9] to AG: values for extraction of some carbonyl compounds"
Solute Vanillin Orthovanillin Isovanillin Salicylaldehyde Acetylacetone Benzoylacetone

AG;
-2.18 -2.74 -1.61 -1.94 -1.30 -3.94 0.67 0.48 rt 0.65 + 0.08 -t 0.35 k 0.38
?

~ri
-5.29k 0.63 -3.97 + 0.45 -4.95 k 0.61 - 1.61 i 0.15 -2.41 rt 0.35 -2.81 ? 0.42

aI

PI
-1.71 -1.45
rt
-

(6;/100 0.15 5.06 rt 1.16 2.93 i 0.82 4.35 + 1.13 1.95 + 0.63 1.71 * 0.72

n
16 16 16 18 20' 20'

sd 0.29 0.20 0.27 0.19 0.19 0.15

-2.21 k 0.56 - 1.28 + 0.38 -1.75 k 0.55 -2.0Ok 0.38 -1.75 5 0.33

0.35 +0.93 2 0.63 +0.89 0.23

* *

0.975 0.974 0.974 0.936 0.943 0.981

"Results for solvents no. 1-26 in Table 2 of ref. 46 after conversion to AGO. bExcludingsolvent tri-n-butylphosphatefor which no reliable value of (ti;), was available.

TABLE 7. A summary of the "best" regressions reporteda for distribution of some carbonyl compounds, in terms of AG:
Can. J. Chem. Downloaded from www.nrcresearchpress.com by 189.4.76.40 on 11/28/13 For personal use only.
Carbonyl compound Vanillin Orthovanillin Isovanillin Salicylaldehyde Acetylacetone Benzoylacetone Constant 4.07 3.62 4.41 0.85 3.47 0.84
f(11)

f(~)
-

E 0.17 0.13 0.18 0.08 0.08 -0.09

B
-0.01 -0.01 -0.01' -

6A/lOO -1.70 -1.75 -3.54 0.64 -0.65

sd 0.31 0.22 0.31 0.18 0.14 0.14

-9.86 -19.19 - 10.44 -4.34 -16.20 -12.57

1.68 -2.03

13 12 12 16 15 11

0.971 0.966 0.964 0.958 0.960 0.988

"From Makitra and Pirig (46), after conversion by the factor - 1.364.

term, and favoured by the various other terms that reflect exoergic solute-solvent interactions. The only exceptions are the small, but statistically significant endoergic P I terms for acetylacetone and benzoylacetone and the lack of any term in for salicylaldehyde. All distributions are favoured by increase in solvent IT?, as expected and are usually favoured by increase in solvent cx value, the latter arising no doubt through solute (base) - solvent (acid) interactions. The one marked exception, salicylaldehyde, is internally hydrogen-bonded and hence unable to interact with external hydrogen-bond acids. That is, the intramolecular hydrogen-bond competes effectively with any intermolecular hydrogen-bond. It is rather difficult to compare results in Table 6 with those of Makitra and Pirig, since these workers give several "best" equations, all obtained after removing one or more solvents. We give in Table 7 the coefficients for some of these equations, after multiplication by - 1.364 in order to transform them into coefficientsrelated to AGO. There is very little chemical sense to be made out of these coefficients -thus increase in dispersion hinders partition, whilst increase in solvent cohesive energy density actually aids partition into the solvent. In truth, there are neither enough solvents nor a sufficiently varied collection of solvents to obtain "robust" MLRA eauations for the distributions. We have considered these processes, however, because Makitra and Pirig (46) have used them as examples of the extended Koppel-Palm equation [lo].

"best" equation obtained by Koppel and Palm when a variety of solvents, including aprotic and hydroxylic, was considered is,"

More recently, Dvorko and co-workers (48) applied the extended Koppel-Palm equation to 28 aprotic and hydroxylic solvents:

As Dvorko and co-workers point out, both eq. [27] and eq. [28] lead to the conclusion that AG* is reduced by solvent electrophilicity (E) and to a lesser extent by solvent polarity. Very significantly, solvent nucleophilicity (B) seems to play no part, in contrast to the conclusion of Fawcett and Krygowski (49). Abraham, Taft, and Kamlet (50) also examined the t-butyl chloride reaction using a truncated version of the AKT equation [9], without the (6;)1 variable. For a variety of aliphatic aprotic and hydroxylic solvents they obtained,

Application of MLRA to kinetics The first of MLRA lo kinetic phenomena was that K o ~ ~ and el (I7) who listed regression for the Ko~~el-Palm equation L6] forvari0us processes. Shorter and cO-workers(47) have correlation analysis to solvent effects on the reaction between diazodiphenylmethane and carboxylic acids, but one particular reaction that has been studied b y several workeis is the unimolecular decomposition of t-butyl chloride at 298 K. The
7

Although the explanatory variables in eq. [29] are different to those in eqs. [27] and [28], the conclusions are very similar, namely that solvent polarity and hydrogen-bond acidity (or electrophilicity) reduce AG* and hence increase log k . The conesponding t-butyl bromide and t-butyl iodide reactions have also been examined by Dvorko and co-workers (48) and Abraham, Kamlet, and Taft (50), but the most recent set of
" ~ that ~ in eq. t [27], ~ f(~) = (E - 1 ) / ( 2 ~ + l), and that the original equation was given in terms of log k.

2682

CAN. J. CHEM. VOL. 66, 1988

TABLE 8. Application of AKT equation [9] to the t-butyl halide reactions (51)"
Coefficients CL%b

"Converted from log k regressions to AG* in kcal mol-'. bWhere the value 100 is used, the CL% > 99.999. 'A variety of aliphatic aprotic and hydroxylic solvents. dReference 57; the coefficients of 6 are 1.30 and 1.77 for the Bu'CI and ButBr reactions, respectively.

Can. J. Chem. Downloaded from www.nrcresearchpress.com by 189.4.76.40 on 11/28/13 For personal use only.

correlations is due to Abraham and co-workers (5 1). Details are given in Table 8. The solvents studied are a number of aliphatic aprotic and hydroxylic solvents, so that the d.6 term is zero, cf. eq. [29]. Although the regression coefficients in Table 8 are numerically smaller than those in eq. [29], for T ; and a l , the general chemical sense is the same. As now seems clear from a number of different regressions, eqs. [27] to [29] and Table 8, the two main factors influencing the reaction are solvent dipolarity and solvent electrophilicity (or hydrogenbond acidity). This is chemically quite reasonable because the transition-state seems to behave as a dipolar species of dipole moment around 8 D, with a substantial negative charge on the leaving chlorine atom. Transition states (V) and (VI) have previously been suggested (52), both of which would be stabilised by solvents with large T ; and a , values.

(V) p = 8 . 8 0 D

(VI) p = 7.29 D

Results in Table 8 show that there are minor accelerating effects of solvent cohesive energy density (suggested to be due to electrostriction of solvent by the dipolar transition state) and solvent hydrogen-bond basicity, P I . Now since solvents that are hydrogen-bond bases are also nucleophiles, it follows that solvent nucleophilicity plays but a very minor r6le in stabilising (V) or (VI). Bentley (52) has quite correctly pointed out that the identification of solvent hydrogen-bond basicity, P I , with solvent nucleophilicity is only an assumption. However, it can be shown (53) that there is a reasonable connection between the nucleophilic solvent parameter Y + of Kevill and Anderson (54) and solvent p 1 values,

determine their individual contributions. The studies discussed here (5 1, 53) as well as other recent work (55, 56) has finally enabled the extent of nucleophilic and electrophilic assistance to be established. It might be thought somewhat odd that the positive centre in transition state (V) and (VI), carrying some 0.8 units of charge, is not stabilised by nucleophilic solvents. However, we already know that R4N+ centres are completely unaffected by solvent nucleophilicity (as P I ) , see Table 3 and especially Table 4. The very small nucleophilic assistance to the t-butyl chloride reaction is therefore quite consistent with effects on R4NC1and (R4N+ C1-) solutes. In Table 8 are given also regression coefficients for the t-butyl bromide and t-butyl iodide reactions (51). Nucleophilic assistance is not statistically significant in either case. The variation in the coefficients of ~i and Pl is as expected on general chemical grounds. Along the series RC1, RBr, and RI the transition states will probably become slightly more dipolar, but much less susceptible to electrophilic assistance, hence the order of IT; coefficients in AG' is numerically RC1 < RBr < RI but for a1 the order is RC1 > RBr > RI. More recently, Abraham and co-workers (57) have extended results to cover aromatic and polyhalogenated solvents as well as the aliphatic solvents examined before (5 1). Although the fits are not as good as for the 21 aliphatic solvents, the various regression coefficients agree very well with those for the aliphatic solvent set, and the general conclusions based on the 21 solvent set remain unchanged. One of the difficulties with MLRA is not only that quite a large number of solvents are needed, but also that there must be a wide variety of solvents. A number of Menschutkin reactions have been studied over a very wide range of solvents, and Koppel and Palm (17) have examined that between tri-npropylamine and methyl iodide,

For 28 aprotic solvents, eq. [32] was found, but the correlation coefficient decreased markedly when the solvent set was extended to include hydroxylic solvents, eq. [33].

There seems little doubt, therefore, that solvent dipolarity and electrophilicity are the dominant factors that determine the magnitude of AG* and hence log k for the unimolecular heterolysis of t-butyl chloride. Because solvent nucleophilicity and electrophilicity are colinear over many of the hydroxylic solvents used in solvolysis studies, it has been difficult to

Both equations indicate that solvent dipolarity and polarisability greatly reduce AG* and hence increase log k. However, on eq. [32] solvent electrophilicity is important but on eq. [33] it is statistically not significant (CL < 80%). The related reaction between triethylamine and ethyl iodide has recently been examined by Abraham and co-workers (58) who find for 33 aprotic and hydroxylic solvent^,'^

amino-4'-nitroazobenzene (cis-DENAB) in a number of aprotic and hydroxylic solvents and correlated AG* separately with T; and a l . We have applied the MLRA equation [9] to as many solvents as we had parameters for. The parameters P I and (8;)' are not statistically significant, and the best regression equation is,

Can. J. Chem. Downloaded from www.nrcresearchpress.com by 189.4.76.40 on 11/28/13 For personal use only.

According to eq. [34] the over-riding factor influencing the reaction rate is solvent dipolarity, with solvent hydrogen-bond acidity and basicity being statistically not significant. Presumably the stabilisation of the leaving iodide ion in the transition state by electrophilic assistance is exactly counterbalanced by stabilisation of triethylamine in the initial state. If the leaving halide is altered, but the incoming amine is kept constant, then on the above suggestion, the effect of solvent electrophilicity will be to stabilise the transition state in the order RI < RBr < RCI and hence to lower AG* in the same order, since a leaving chloride will be electrophilically solvated better than a leaving bromide or iodide. Auriel and Hoffmann (59) have studied the reaction of 1,4-diazabicyclo[2.2.2]octane, or Dabco, with 2-chloro-, 2-bromo-, and 2-iodoethylbenzene using a wide variety of aprotic and hydroxylic solvents. The best equations are given below:

with all three parameters being significant at 399.99% CL. Such an equation suggests that the transition state is stabilised by dipole-dipole interactions and by hydrogen-bond interactions between acidic solvents and the nitro group in the rotational transition state, which may be close to the valence bond structure (VII). The lack of any stabilisation of t h e = k ~ ,

centre (no p effect) is not surprising in view of our previous discussion on R4Nf compounds. Whitten and co-workers (60) also examined the T, T* transition of trans-DENAB for which we find,

In all eqs. [35]-[37] terms in P 1 and (66), were not significant. Exactly as found for the Et3N/EtI reaction, the main solvent influence is dipolarity, stabilising the transition state and increasing log k. For all three reactions, the initial-state effect of solvent hydrogen-bond acidity will be the same, and hence the variation in the a . a l term must be due to variation in transition-state property. As suggested above, the largest stabilising effect will be on the Dabco/Cl transition state, just counteracting the effect on the dibasic Dabco amine. But for the bromide and more so for the iodide, the stabilisation effect on the transition state is less, leading to more positive a . a l coefficients and hence to a retarding effect on the rate. The coefficients of T; for the various Menschutkin reactions are all large and negative in eqs. [34]-[37], reflecting stabilisation of the dipolar transition states, a lowering of AG*, and an increase in log k. Other studies (15) have yielded a dipole moment of 7.6 D for the Et3N/EtI transition state and a charge separation of around 0.4 units, so that there is considerable independent evidence for transition-state structure. This is not necessarily so for other reactions that have been somewhat less studied, and we give as a final example the thermal cis-trans isomerisation of azobenzenes. Whitten and co-workers (60) determined rate constants for the isomerisation of cis-4-diethyl''Note in eq. [34], AG* is calculated with respect to solvent DMF.

with %CL of >99.9999 (T;), 95 ( a , ) , and 99 ( P I ) . Consideration of the Frank-Condon principle suggests that transDENAB itself is but marginally hydrogen-bonded. Presumably this is also the case for cis-DENAB, so that the transition state for isomerisation is preferentially stabilised by hydrogenbonded acidic solvents. It follows from the coefficient of T; in eqs. [38] and [39] that the dipolarity of the Frank-Condon excited state and the isomerisation transition state must be very similar, as pointed out by Whitten and co-workers (60). The latter suggest dipole moments of 8.0 D for trans-DENAB, 25 D for the excited state and 20.5 D for the transition state. From comparisons of the s . ~ term ; in eqs. [38] and [39] with those in, for example, eqs. [34] to [37] and Table 8, it seems as though the dipolarity of the DENAB transition state is no more than that of the t-butyl chloride transition state ( p = 8.80 or 7.29 D). Possibly the degree of charge separation in the transition state is a determining factor, rather than the actual dipole moment. Asano and Okada (61) studied the isomerism of the related compound cis-4-dimethylamino-4'-nitroazobenzene (cis-DMNAB) in a different set of aprotic plus hydroxylic solvents, leading to

in good agreement with eq. [38]. Asano and Okada (61) showed also that for a number of other 4- or 4,4'-substituted azobenzenes there was very little solvent effect on the rate of isomerisation, e.g. for the ratio k(MeOH)/k(hexane) they found 2.30 (4-nitro), 1.44 (4-N,N-diethylamino), 1.88 (4-N,Ndimethylamino-4'-chloro), and 1.08 (2,4-dinitro) as against a ratio of 25 500 for cis-DMNAB itself. Clearly, a push-pull mechanism leading to a rotational transition state close to (VII)

2684

CAN. J . CHEM. VOL. 66, 1988

Can. J. Chem. Downloaded from www.nrcresearchpress.com by 189.4.76.40 on 11/28/13 For personal use only.

is favored only by the specific combination of a +R and -R substituent in the 4 and 4 ' positions. Although we have mostly dealt with solvent effects in terms of the AKT equation rather than the Koppel-Palm equation, both equations when properly used give results that are generally reasonable. Sometimes the AKT equation leads to statistically better fits, as we have seen in many of the examples discussed above, but sometimes a better fit is obtained using the Koppel-Palm equation. Zhong-yuan et al. (62) have applied several regression equations to the substitution of 2,4-dinitrofluorobenzene by the ethyl ester of tyrosine and find for the 11 solvents studied that the Koppel-Palm equation in B, E, and f(~) leads to the best fit of the log k values. For a general discussion of solvent effects, especially on kinetic processes, the excellent book of Reichardt (63) is highly recommended. The application of MLRA to chemical kinetics is clearly limited only by the number of reactions that have been studied in a wide variety of solvents and we have given but a selection of such processes. It must be kept in mind, however, that just as kinetic data relate only to differences between transition state and initial state, so do all the regressions based on log k or AG* values. Although we have analysed regression equations in terms, especially, of transition-state effects, this can only be done through "chemical intuition", there being no logical connection whatsoever between any log k, AG*, or regression constant and a transition-state effect. In order to establish such a connection, solvent effects on log k or AG* must first be broken down into initial-state and transition-state contributions; we deal briefly with this aspect of solvent effects in the next section.

viously investigated (51). Values of AG:(BurX) were obtained and combined with AG* values to yield the corresponding transition-state values via eq. [42]. The standard solvent was DMF. It is then possible to apply the AKT equation [9] to A G ~ ( T ~ ) , AG:(BU'X) and 6AG* = AGB - AGhMF and then to see exactly how each term in the equation originates. Details are in Table 9 for the case of Bu'C1 and Bu'Br. Inspection of the coefficients for the BurC1 reaction shows very clearly that the a : term in AG * arises almost exclusively through a transitionstate effect, as indeed does the a l term as well. The rather small pl term is again a transition-state effect, showing that a small amount of nucleophilic participation may take place. However, the negative 6; term in the AG equation can now be shown to arise through two positive endoergic effects on initial state and transition state that do not quite cancel out. Effects on the Bu'Br reaction for the same 21 solvents are very similar, but in this case the effect of nucleophilic participation is even smaller as judged by the term. Abraham and co-workers (57) also carried out an analysis using a further set of solvents that included aromatic and polyhalogenated ones as well as the original (53) 21 aliphatic solvents. Although a d.6 term is now required, the remaining coefficients are very close to those obtained for the 2 1 solvent set, compare Tables 9 and 10. However, the p l term in both SAG* and AG:(T~) is rather less significant in the. 30 solvent set, Table 10. The only other reaction for which there are sufficient results to apply this "unified" method is the Menschutkin reaction of triethylamine and ethyl iodide. Abraham (15) has given a preliminary account of work on these lines, but si6ce a more extensive and detailed account is in preparation (58) we refrain from an analysis until a later date. Although it is unlikely ever to be applied to a great many reactions, we feel that it is important to use the unified method for the analysis of typical organic reactions. In the way, not only can solvent effects on AG+ be broken down into individual interactions, but it is then possible to assign exactly these interactions to initial-state and transition-state effects.

'

Initial-state and transition-state effects and MLRA The separation of solvent effects on rate constants (as log k or AG* values) into initial-state and transition-state contributions is now a well-known technique in physical organic chemistry, see for example a number of reviews (7, 15, 64). Values of AG* are rather more convenient to use than log k values, a general equation for the separation being (7),
where AG: represents the Gibbs energy of transfer of a species from a standard reference solvent (A) to any other solvent (B). Values of AG:(reactants) can be found by conventional thermodynamic methods (7) and are combined with kinetic measurements of AG; and AG; to give the resulting effect on the transition state, AG:(T~). It is not our intention to discuss the various applications of eq. [41], or equivalent equations that can be set up using any thermodynamic parameter such as H, S, V, etc., but only to deal with one specific recent advance, as follows. Suppose that eq. [41] can be applied to a given reaction, using quite a large number of different solvents. It is then possible to apply the method of multiple linear regression analysis, not only to AG' values but also to values of AGy(Tr). In this way, the exact origin of all the terms in an equation such as the AKT equation [9] can be determined, without recourse to chemical intuition. Unfortunately, application of this method, known as the "unified method" (15), requires the availability of a very large quantity of experimental data, both kinetic and thermodynamic, and but few examples are known to date. We discuss now, the one complete published analysis on these lines -that of the t-butyl halide reactions carried out by Abraham and co-workers (57). These workers first studied the 21 aliphatic solvents pre-

Summarising discussion We have devoted much of this review to showing how the method of MLRA can be applied to various phenomena in organic chemistry and can lead to some understanding of solvent effects. As with all linear free energy relationships, equations such as [9] and [lo] are based on the assumption that individual effects act independently and can be summed to yield an overall effect. Considering the complicated nature of solvents and solution, especially those involving solvents such as water and alcohols, it is remarkable that equations based on such an assumption hold over a wide range of processes and solvents. No doubt there will be situations in which this basic assumption breaks down, but it certainly seems to lead to chemically reasonable results for the processes we have considered. There are, of course, other approaches to the study of solvent effects, notably factor analysis (FA) and principal component analysis (PCA). Factor analysis itself has been used in two quite different ways. Chastrette and Carretto (65) applied FA to several solvent properties and obtained four factors, or parameters, that were simply regarded as abstract factors, although it was shown, for example, that the first factor was strongly

TABLE 9. Application of the AKT equation [9] to initial-state and transition-state contributions in the t-butyl halide reactions
for 21 aliphatic solvents at 298 K (57) Coefficients AGO

CL%

v;

aI

PI

(6:)1/100

sd

v;

a,

PI

(6A)1/100

"For the Bu'C1 reaction. bFor the Bu'Br reaction.

Can. J. Chem. Downloaded from www.nrcresearchpress.com by 189.4.76.40 on 11/28/13 For personal use only.

TABLE 10. Application of the AKT equation [9] to initial-state and transition-state contributions in the t-butyl halide reactions for 30 solvents
at 298 K (57) Coefficients AGO AG?(T~)" AG?(BU'CI) SAG* 7.22 -0.73 7.95

CL %

vi
-7.34 -0.85 -6.49

s
1.19 -0.09 1.28

aI

PI
-1.21 0.38 -0.83

(S8)l/lOo 0.42 1.15 -0.73

sd 0.66 0.32 0.68

v
100 95 100

s
99 32 99

al

p,
97.5 87 88

(s$)~/~oo 96.5 100 99.9

-5.46 -0.01 -5.45

0.9805 0.9633 0.9863

100 8 100

"For the BurCl reaction. bFor the Bu'Br reaction.

correlated with solvent dipole moment and f(~), etc. Svoboda et al. (66) also analysed solvent properties (35 properties for 85 solvents or a restricted set of 20 properties for 51 solvents) and showed that a minimum number of four abstract factors were required. They then rotated these factors to yield four orthogonal factors to which were ascribed a physical significance: AP (electrophilic solvation), BP (nucleophilic solvation), EP (polar solvation), and PP (dispersion solvation), and then set out a multiple linear regression equation, where A is the dependent parameter (log k, AG*, etc.) and Ao, a , b, c , d are constants to be found by MLRA. Application of eq. [43] is no different in principle to the multiple linear regression equations we have already discussed. Whether eq. [431 is more general, or leads to better fits of the data, than the AKT equation [9] or the Koppel-Palm-Makitra-Pirig equation [lo] could be decided by the simple, although lengthy procedure of comparing the three equations as applied to as many sets of data as possible. This is quite beyond the scope of this review, but since Svoboda et al. (66) list regressions for a large number of kinetic and spectroscopic processes, some idea of the relative success of the various MLRA equations can be obtained. Thus for some SN1 reactions in hydroxylic and aprotic solvents, eq. [43] leads to good fits, with r = 0.972 and 0.990, compare the AKT eq. [9] where r ranges from 0.980 to 0.997, Table 8, and the simple Koppel-Palm equation [6] where values of r 0.982 and 0.985 are obtained (17). On the other hand, for SN2reactions of the Menschutkin type, the same "all-solvent" correlations by eq. [43] results in correlation coefficients between 0.764 and 0.846 as compared to values by the AKT equation [lo] of 0.927 to 0.980, eqs. [34] and [37], and avalue

of 0.798 using the Koppel-Palm equation, see eq. [33]. At the moment, there does not seem to be any advantage of eq. [43] over the previous eqs. [9] and [lo]. Wold and co-workers (67) take a rather different view. From a PC analysis of I3C NMR shifts of indenyl-lithium, they derived two principal components, 0 and 0 2 . Although 0 1 could be correlated with solvent parameters such as T;,the second component O2 seemed to be system-specific. It can then be argued (68) that the appearance of system-specific parameters precludes the use of MLRA, because the latter is based on the assumption that general solvent parameters can be used in the correlation of solvent effects on various processes. In principle, Wold and co-workers (67, 68) are no doubt correct. The complicated nature of solvents and solvent-solute interactions must give rise ultimately to system-specific factors. However, in practice, it appears that MLRA can be applied to a reasonably wide range of processes, and, as we have seen, can yield coefficients in equations such as the AKT equation [9] that make general chemical sense. Our conclusion is therefore that if MLRA is properly applied, with due consideration given to matters such as colinearity of explanatory variables, etc., it is possible to derive equations that can be interpreted in a chemically sensible way, and that can lead to further understanding of a rather large range of thermodynamic, kinetic, and spectroscopic processes.
1. M. BERTHELOT and L. PBANDE SAINT-GILLES. Ann. Chim. (3 me Ser), 65, 385 (1862); 66, 5 (1862); 68, 225 (1863). Z. Phys. Chem. 5,589 (1890). 2. N. A. MENSCHUTKIN. and R. HUISGEN. J. Am. Chem. Soc. 95,5056 3. (a) G. STEINER (1973); ( b ) M. ESSEFAR, M. EL MOUHTADI, D. LIOTARD, and J.-L. M. ABBOUD. J. Chem. Soc. Perkin Trans. 2, 143 (1988).

2686

CAN J . CHEM. VOL. 66. 1988

4. J. G. KIRKWOOD. J. Chem. Phys. 2, 351 (1934). and H. EYRING. Ann. N.Y. Acad. Sci. 39, 303 5. K. J. LAIDLER (1 940). K. J. LAIDLER, andH. EYRING. The theory of rate 6. S. GLASSTONE, processes. McGraw-Hill, New York. 1944. Prog. Phys. Org. Chem. 11, l(1974). 7. M. H. ABRAHAM. 8. M. H. ABRAHAM and R. J. ABRAHAM. J. Chem. Soc. Perkin Trans. 2 , 4 7 (1974). and R. J. ABRAHAM. J. Chem. Soc. Perkin 9. M. H. ABRAHAM Trans. 2 , 1677 (1975). 10. J. R. MCCABEand C. A. ECKERT.ACC. Chem. Res. 7, 251 (1974). 1 1 . H. KWART and T. H. LILLEY. J. Org. Chem. 43, 2374 (1978). and S. WINSTEIN. J. Am. Chem. Soc. 70, 846 12. E. GRUNWALD (1948). E. GRUNWALD, and H. W. JONES. J. Am. Chem. 13. S. WINSTEIN, SOC.73,2700 (1951). J. B. F. N. ENGBERTS, J. JAGER,and M. J. 14. W. BLOKZIJL, BLANDAMER. J. Phys. Chem. 91, 6022 (1987). Pure Appl. Chem. 57, 1155 (1985). 15. M. H. ABRAHAM. and V. A. PALM. Org. React. USSR, 8,291 (1971). 16. I. A. KOPPEL and V. A. PALM. In Advances in linear free energy 17. I. A. KOPPEL relationships. Edited by N. B. Chapman and J. Shorter. Plenum Press, London. 1972. P. W. CARR, R. W. TAFT,and M. H. ABRAHAM. 18. M. J. KAMLET, J. Am. Chem. Soc. 103, 6062 (1981). M. J. KAMLET, andR. W. TAFT.J. Chem. Soc. 19. M. H. ABRAHAM, Perkin Trans. 2, 923 (1982). R. M. DOHERTY, and M. J. 20. R. W. TAFT, M. H. ABRAHAM, KAMLET. J. Am. Chem. Soc. 107, 3105 (1985). R. M. DOHERTY, J.-L. M. ABBOUD, M. H. 21. M. J. KAMLET, ABRAHAM, and R. W. TAFT. Chemtech. 566 (1986). R. M. DOHERTY, M. J. KAMLET, and R. W. 22. M. H. ABRAHAM, TAFT. Chem. Brit. 22, 551 (1986). J.-L. M. ABBOUD, M. H. ABRAHAM, and R. W. 23. M. J. KAMLET, TAFT.J. Org. Chem. 48, 2877 (1983). YA. N. PIRIG,A. M. ZELIZNYI, M. A. DANIEL 24. R. G. MAKITRA, DE AGUAR, V. L. MIKOLAJEV, and V. A. RAMANOV. Org. React. USSR, 14, 421 (1977). and YA. N. PIRIG.Org. React. USSR, 16, 535 25. R. G. MAKITRA (1979). and A. I. PAJU. Org. React. USSR, 11,121 (1974). 26. I. A. KOPPEL M. J. KAMLET, and M. H. 27. R. W. TAFT, J.-L. M. ABBOUD, ABRAHAM. J. Soln. Chem. 14, 153 (1985). and E. BRETSCHNEIDER. In Internal rotation in 28. R. J. ABRAHAM molecules. Edited by W. J. Orville-Thomas. Wiley, London. 1974. J. Phys. Chem. 73, 1192 (1969). 29. R. J. ABRAHAM. and W.-C. LIN. J. Phys. Chem. 90, 30. J.-S. CHEN,R. B. SHIRKS, 4970 (1 986). and J.-L. M. ABBOUD. J. Am. 31. M. ESSFAR,G. GUIHENEUF, Chem. Soc. 104, 6786 (1982). L. E. XODO, M. J. COOK,and R. CRUZ. 32. M. H. ABRAHAM, J. Chem. Soc. Perkin Trans. 2, 1503 (1982). and L. GRIFFITHS. Tetrahedron, 37,575 (1981). 33. R. J. ABRAHAM Chem. Phys. Lett. 94, 272 (1983). 34. P. SUPPAN. and N. J. FREEMAN. J. Mo1. Struct. 161, 193 (1987). 35. J. EMSLEY and YA. N. PIRIG.Org. React. USSR, 15, 547 36. R. G. MAKITRA (1 978). YA. N. PIRIG, T. I. POLITANSKAYA, and F. B. 37. R. G. MAKITRA, MOIN.Org. React. USSR, 15, 561 (1978).

38. R. G. MAKITRA, YA. N. PIRIG,T. I. POLITANSKAYA, and F. B. MOIN.Org. React. USSR, 16, 549 (1979). YA. N. PIRIG,and T. I. POLITANSKAYA. Zhur. 39. R. G. MAKITRA, Priklad. Khim. 54, 54 (1981). YA. N. PIRIG,T. I. POLITANSKAYA, and F. B. 40. R. G. MAKITRA, MOIN.Zhur. Fiz. Khim. 55, 758 (1981). and YA. N. PIRIG.Zhur. Obshchei Khim. 56, 41. R. G. MAKITRA 2657 (1986). M. J. KAMLET, andR. W. TAFT.J. Chem. Soc. 42. M. H. ABRAHAM, Perkin Trans. 2, 923 (1982). J. Chem. Soc. B, 299 (1971). 43. M. H. ABRAHAM. 44. R. W. TAFT, M. H. ABRAHAM, R. M. DOHERTY, and M. J. KAMLET. J. Am. Chem. Soc. 107, 3105 (1985). J. Chem. Soc. Perkin Trans. 2, 1343 (1972). 45. M. H. ABRAHAM. and YA. N. PIRIG.Org. React. USSR, 16, 535 46. R. G. MAKITRA (1979). and J. SHORTER. J. Chem. Soc. Perkin Trans. 2, 47. D. MATHER 1179 (1983) and previous papers. 48. G. F. DVORKO, E. A. PONOMAREVA, and N. I. KULIK.Usp. Khim. 53, 948 (1984); Engl. Trans. p. 547. and T. M. KRYGOWSKI. Austral. J. Chem. 28, 49. W. F. FAWCETT 2115 (1975). R. W. TAFT, andM. J. KAMLET. J. Org. Chem. 50. M. H. ABRAHAM, 46, 3053 (1981). R. M. DOHERTY, M. J. KAMLET, J. M. HARRIS, 51. M. H. ABRAHAM, and R. W. TAFT. J. Chem. Soc. Perkin Trans. 2, 913 (1987). Am. Chem. Soc. Ser. 215, 255 (1987). 52. T. W. BENTLEY. R. M. DOHERTY, M. J. KAMLET, J. M. HARRIS, 53. M. H. ABRAHAM, and R. W. TAFT. J. Chem. Soc. Perkin Trans. 2, 1097 (1987). J. Am. Chem. Soc. 108, 54. D. N. KEVILLand S. W. ANDERSON. 1579 (1986). M. H. ABRAHAM, J. M. HARRIS, R. W. TAFT, 55. R. M. DOHERTY, and M. J. KAMLET. J. Org. Chem. 51, 4872 (1986). M. R. SEDAGHAT-HERATI, 56. J. M. HARRIS,S. P. MCMANUS, N. NEAMATI-MAZRAEY, R. M. DOHERTY, R. W. TAFT, and M. H. ABRAHAM. Am. Chem. Soc. Ser. 215, 247 (1987). P. L. GRELLIER, A. NASEHZADEH, and R. A. 57. M. H. ABRAHAM, C. WALKER. J. Chem. Soc. Perkin Trans. 2, 1717 (1988). P. L. GRELLIER, and A. NASEHZADEH. 58. M. H. ABRAHAM, Unpublished work. 59. M. AURIEL and E. DE HOFFMANN. J. Am. Chem. Soc. 97,7433 (1975). T. F. MATTOX, and D. G. WHITTEN. J. Org. 60. K. S. SCHANZE, Chem. 48, 2808 (1983). and T. OKADA. J. Org. Chem. 49,4387 (1984). 61. T. ASANO S. CHENG-E, and H. TE-KANG. J. Chem. Soc. 62. Z. ZHONG-YUAN, Perkin Trans. 2, 929 (1985). Solvent effects in organic chemistry. Verlag 63. C. REICHARDT. Chemie, Weinheim, New York. 1974. 64. E. BUNCEL and H. WILSON. Acc. Chem. Res. 12,42 (1979). and J. CARRETTO. Tetrahedron, 38, 1615 65. M. CHASTRETTE (1982). 0. PYTELA, and M. VECERA. Coll. Czech. Chem. 66. P. SVOBODA, Cornrnun. 48, 3287 (1983). D. JOHNELS, S. WOLD, and U. EDLUND. Acta. 67. B. ELIASSON, Chem. Scand. B36, 155 (1982). Acta. Chem. Scand. B40, 270 68. S. WOLD and M. SJOSTROM. (1986).

Can. J. Chem. Downloaded from www.nrcresearchpress.com by 189.4.76.40 on 11/28/13 For personal use only.

Das könnte Ihnen auch gefallen