Generalized Cumulant Expansions and

Spin-Relaxation Theory
Cite as: J. Chem. Phys. 49, 376 (1968); https://doi.org/10.1063/1.1669833
Submitted: 11 December 1967 . Published Online: 18 September 2003

Jack H. Freed

ARTICLES YOU MAY BE INTERESTED IN

Stochastic Liouville Equations

Journal of Mathematical Physics 4, 174 (1963); https://doi.org/10.1063/1.1703941

Critique of the generalized cumulant expansion method

Journal of Mathematical Physics 17, 1148 (1976); https://doi.org/10.1063/1.523041

A comparison of generalized cumulant and projection operator methods in spin-

relaxation theory
The Journal of Chemical Physics 62, 4687 (1975); https://doi.org/10.1063/1.430417

J. Chem. Phys. 49, 376 (1968); https://doi.org/10.1063/1.1669833 49, 376

© 1968 American Institute of Physics.

 THE JOURNAL OF CHEMICAL PHYSICS VOLUME 49, NUMBER 1 1 JULY 1968

Generalized Cumulant Expansions and Spin-Relaxation Theory*

JACK H. FREEDt

Department of Chemistry, Cornell University, Ithaca, New York

(Received 11 December 1967)

The generalized cumulant expansion method of Kubo is applied to an analysis of spin-relaxation theory
appropriate for NMR and ESR studies of molecular systems. It leads to a general, formal solution of the
equation of motion of a suitably averaged magnetization operator (and also the spin-density matrix).
This solution permits a convenient perturbation expansion in the region of motional narrowing; i.e., when
the random functional spin perturbation XI (t) obeys I XI (t) I T,< 1, where T, is a correlation time for the
random process. It is shown how this expansion valid for times t»Te generates the time-independent R or
relaxation matrix to all orders in I XI (t) I T" and detailed expressions are given through fourth order. While
the R matrix supplies the Iinewidth and dynamic frequency shift of the main Lorentzian resonance line,
it is found that by formulating the cumulant method without the restriction t»Te, weak subsidiary
Lorentzian lines are predicted. These subsidiary lines appear as second-order perturbation corrections in
I X, (t) I Te. They have widths given by Te- I and, in general, large frequency shifts. The problem of
(anisotropic) -rotational diffusion is discussed in detail, and it is shown that the spin-Hamiltonian approxima-
tion for liquids rests upon a Born-Oppenheimer-type approximation appropriate for random modulation
of the nuclear coordinates, i.e., Te-I«Wn.o, where hWn.o is the energy separation between the ground electronic
state and the nth excited state coupled by the angular-momentum operator L. The cumulant method is
then used to obtain higher-order corrections to the g-tensor line-broadening mechanism.

I. INTRODUCTION expansion method, and he develops several important

The earlier theories of spin relaxation of free radicals
in liquids as developed by Freed and Fraenkel1 and
Kivelson2 were based upon the general formulations of
Kubo and Tomita,3 Bloch,4 Redfield," and Abragam. 6
All these formulations are characterized by an approxi-
mate solution to the equation of motion of the magnet-
ization operator (or equivalently the spin-density
matrix) valid to second order in a spin-dependent
perturbation X1(t), which is a random function of time.
The approximations, which are not easy to justify, are
summarized by Abragam. 7 Furthermore, they may not
be conveniently generalized to obtain further correc-
tions, i.e., higher-order terms in an appropriate expan-
sion series for the equations of motion. In fact, straight-
forward attempts at carrying out such an expansion
series are found to lead to divergent terms. The problem
lies in properly grouping the higher-order terms.
Kub08 ,9 has more recently shown how one may obtain
a general and formally correct solution to the spin-
relaxation problem by means of a generalized cumulant
* Alfred P. Sloan Foundation Fellow.
t Supported in rart by PHS Research Grant No. GM 14123
from the Nationa Institutes of Health and by the Advanced
Research Projects Agency.
1 J. H. Freed and G. K. Fraenkel, J. Chern. Phys. 39, 326
(1963) .
2 D. Kivelson, J. Chern. Phys. 33, 1094 (1960).
3 R. Kubo and K. Tomita, J. Phys. Soc. Japan 9,888 (1954).
4 F. Bloch, Phys. Rev. 102, 104 (1956).
6 A. G. Redfield, IBM J. Res. Develop. 1, 19 (1957).
6 A. Abragam, The Principles of Nuclear Magnetism (Oxford
University Press, London, 1961).
7 Reference 6, p. 276. Note that assumption (b) of this reference
was not utilized in the Kivelson-Kubo-Tomita approach, which is
then not in a useful form to handle problems involving degenerate
spin states, d. Ref. 1.
8 (a) R. Kubo in Fluctuation, Relaxation, and Resonance in
Magnetic Systems, D. ter Haar, Ed. (Oliver and Boyd, London,
1962), p. 23. (b) R. Kubo, J. Math. Phys. 4,174 (1963).
9 R. Kubo, J. Phys. Soc. Japan 17,1100 (1962).
376
theorems on the utility of this method. In the present
work, we have developed the application of this method
to the spin-relaxation problem, so that some useful
results may be obtained from it.\O Thus, it becomes
possible to explicitly obtain the higher-order terms in
the time-independent relaxation or R matrix which
governs the relaxation of the spin-density matrix in
the motionally narrowed region. This result is applicable
to any well defined random process modulating Xl (t).
It is also possible to remove the "coarse-graining" in
time assumption (i.e., consideration of time intervals
t»Tc, where Tc is a correlation time for the random
process) which is inherent in the R-matrix treatment
even to higher orders. It is found that removal of this
assumption usually leads to the appearance of weak
Lorentzian subsidiary lines as well as the main Lorent-
zian resonance line (the widths and frequency shifts of
the latter being given by the real and imaginary parts,
respectively, of the R matrix). The inclusion of finite t,
hence the appearance of the subsidiary lines, may also
be regarded as a perturbation in X1(t). While the
cumulant method leads to a formally correct solution
to all orders in roughly Xl(t)T c, it leads to reasonably
convergent expansions only for I Xl(t)Tc I < I, and thus
it is not expected to be very useful in the near-rigid
solution or slow-motion region, except perhaps for
special cases, such as a Gaussian random modulation. s
The general features of the cumulant method as
applied to spin relaxation are developed in Sec. II.
Explicit expressions for the R matrix to fourth order
10 H. Sillescu and D. Kivelson O. Chern. Phys. 48,3493 (1968) ]
have just developed a general method for calculating the resonance
line shape which, however, requires at the outset that the molec-
ular motion be determined by a classical rotational-diffusion
equation. It appears to be more "tailor-made" for this particular
case than the more general cumulant method. We wish to thank
these authors for communication of their work prior to publi-
cation.
 SPIN-RELAXATION THEORY 377

in essentially Jel(t)rc are developed in Sec. III. In where we have employed a convention due to Kub08a
Sec. IV, the method is illustrated by applying it to a such that for any operators a and b
two-jump Markov modulation of the hyperfine-inter-
action term in ESR. Examples of rotational modulation axb=[a, b]. (2.6a)
of the spin-orbit operator and the g tensor are discussed This differential equation may be integrated formally to
in Secs. V and VI, respectively. The discussion in Sec. V
leads to a Bom-Oppenheimer-type approximation
appropriate for random modulation of nuclear co-
8 xt (t) = expo [if dt'Jelt(tl)x] S.,(O), (2.7)
ordinates. which becomes
II. THE CUMULANT METHOD
We wish to develop the Cumulant Expansion Method
Sx t (t) = <expo [if dt'Jelt(t')X J) SX (2.8)

so that it leads to a general and useful solution to the upon taking an ensemble average. The exponent in
spin-relaxation problem. One starts with the line-shape Eqs. (2.7) and (2.8) is an ordered operator (indicated
function, or spectrum lew), which is given byl,3,6 by subscript 0), which must be interpreted as8 ,9

lew) = ~
11"
I'"
0
G(t) coswtdt, (2.1) <expo [if dtIJelt(tl)XJ) =~Mn(t), (2.9)

where G(t) is the autocorrelation function of the spin where

angular momentum:
G(t) =Tr[S.(t) S.,]. (2.2)
Here S., is the x component of the angular momentum
and S.,(t) is given by =injl dtlll! dt 2 • . .
o 0
1
0
10 1
- dtn
S.,(t) = (8.,(t) )
X (Jelt(tl) xJe l t(t2)X' . 'Jelt (tn) x), for n~ 1. (2.10)
= (exp(iJet) Sx exp( -iJet) ), (2.3)
Mo(t) =1. (2.10a)
where the angular brackets imply an ensemble averaging
and the tilde indicates an operator for a single inde- The time ordering indicated in the first equality of
pendent spin system (e.g., a single molecule or free Eq. (2.10) may be shown to imply the second equalityY
radical). The trace in Eq. (2.2) is taken over the spin Time ordering is necessary because, in general, the
states of a single system. The Hamiltonian Je for the operator Jel(t i ) need not commute with Jel(tj) for
system is separated into a time-independent component ti~tj. In the second equality the time ordering is auto-
JCo and a perturbing Je l (t), which is a random function matically achieved such that
of time: t'2 tl'2 t22. ... tn. (2.1Ob)
(2.4)
The expansion of Eq. (2.9) may be regarded as a gen-
The ensemble average of Eq. (2.3) (which is to be eralized moment expansion, where the nth moment of
regarded as equivalent to a macroscopic sample con- the random function
taining many independent spins) is taken over the
ensemble defined by the probability distribution charac- mn(tl, t2•• ·tn) = (Je l t(tl)XJel t(t2)x •. ·Jelt(tn)x) (2.11)
teristic of Jel(t). It is also assumed that the random is itself an operator.
process is stationary, and ergodic,S where the station- The basic idea of the cumulant method is to find a
arity arises by assuming that the ensemble of Jel(t) solution to Eq. (2.9) of form
remains at equilibrium, and the ergodicity assures that
all the spin systems are able to experience the same
range of effects from Jel(t). <expo [if dt'Jelt(t')xJ) =expoK(t), (2.12)
It is convenient to define an interaction representa-
tion by where
S.,t(t) =exp( -iJCot) Sx(t) exp(iJCot) (2.Sa) (2.12a)
and n-l
Jelt(t) =exp( -iJCot) Jel (t) exp(iJCot). (2.Sb)
Each Kn is still an operator and is of nth order in the
The equation of motion of Sx t (t) is then given as operator Jelt(t)x. It may be regarded as a generalized
cumulant, which is known to be closely related to the
(djdt)8.,t(t) =i[Jelt(t) , 8",t]
11 P. Roman, Advanced Quantum Theory (Addison-Wesley
(2.6) Publishing Co., Inc., Reading, Mass., 1965), p. 311.
378 JACK H. FREED

moments. 9 The ordering prescription on the cumulants The nature of the ordering prescription is illustrated
as implied by Eq. (2.12) means that by [d. Eq. (2.10)]

expoLK,,(t) == E(P!)-IO[LKn(t) ]p, (2.13)

p=O n=1 0(M12) =0 [i{ dtlml(tl) T
where the cumulant expansion also follows the time-
ordering called for in the moment expansion. Thus, by
comparing terms of the same order in Eqs. (2.9) and
(2.13), while preserving operator and time ordering,
expressions may be found for the Kn in terms of the Thus,
Mn. In general, the Kn will be composed of sums of
products of Mi where i~n. Kub09 gives general K 2= - [dtl[! dt2[~(tl, t2 ) -ml(tl)ml(t2)], (2.1Sb)
theorems for the validity of solutions like Eq. (2.12) in o 0

the form of generalized cumulant functions, and points and it will be shown to be equivalent to the second-order
out they are just as valid as the moment expansion in term of the earlier theories. a- 6 Also,
Eqs. (2.9) and (2.10). When the two expansions are
compared, one obtains to fourth order
Kl =Ml, (2.14a)
K 2 =M2 -!0(M12) , (2.14b)
K a=M3 -![0(MIM 2 ) +0(M2M 1)]+to(Ml a), (2.14c) Often, it is possible to define Xlt(t)X so that ml(t) =0,
which simplifies the cumulants to
K 4 =M4 -!Q(M22) -![0(MIMa)+0(MaM1)]
+t[0(M12M2 ) +0 (M1M 2M 1) +0(M2M12)] Kl =M1, K 2=M2, Ka=M3, K 4 =M4 -!0(M22).
-to(MI4). (2.14d) Then K4 becomes [utilizing Eq. (2.16a)]

It follows from a general theorem of Kub09 that the Kn (t) may be written in the form

Kn(t) =in[ dtl[' dt 2.. .

o 0
["-1 dtn(Oxlt
0
(t1) xX l t(t2) x •• 'Xlt(tn)x)c, (2.17)

where the subscript c implies a "cumulant average," i.e., the appropriate collections of moments. Thus, for example,
Eq. (2.16b) may be rewritten in this form as

where the last two terms in Eq. (2.16c) are shown to
be equal to the last term in Eq. (2.16b) by permuting
the numbering of the ti in Eq. (2.16c) to bring the
~(ti, tJ into standard order (while preserving the
proper time ordering) and then interchanging the order
of integration. We shall have occasion to use the
general form Eq. (2.17) for Kn(t). Another important
property of the Kn(t) is that a cumulant is zero if any
one of the random variables contained in it is uncorre-
lated with the others.9 For example, if XI(tl ) and
XI (t2) are uncorrelated, then
m2(t1, t2) =ml(tl)ml(t2) (2.18a)
and
(2.18b)
In Eq. (2.16b), the second term under the first integral
subtracts out that component of m4(tl , t2, ia, i4) in which
Xl (t i ) for i = 1 and 2 are uncorrelated with the values
for i=3 and 4. It has the very important role of
subtracting out the component of M 4 (t) which diverges
for large t.
The utility of the cumulant method can perhaps be
better seen by transforming Eqs. (2.8) and 2.12) to
the differential equation [d. the form of Eqs. (2.6)
and (2.7)]
S}(t) =K(t)S}(t) (2.19)
with
 SPIN-RELAXATION THEORY 379

It is convenient to transform the time integrals in Eq. Equation (2.21) follows when one recognizes that
(2.19a) by letting - Re(R) gives the linewidth for the line· shape function
in Eq. (2.1), so that lew) is nonnegligible only for
so that I W-Wo I "-' I Re(R) I, (where Wo is the center fre-
quency), and finally that lew) is determined mainly
t;=t- tTi.
i=)
by the behavior of G(t) around the region
t"" I W-Wo I-I.s
Then
Note that Eq. (2.21) is a generalization of the more
common narrowing condition (I HI(t) I)Te«1. We note
also that the resonant frequency is shifted from Wo to

x <OX)t(t)xXlt(t-TI)x, °Xl t (t-

0 ~Tir>e'
[wo- rm(R)], so these arguments also require Eq.
(2.21a).
Utilizing the cumulant method, it is, however, not
(2.19b) necessary to pass immediately to the asymptotic limit
of a time-independent R matrix. The cumulants may
The connection with what may be regarded as a more be explicitly determined for finite values of t. This
conventional time-dependent perturbation theory ig results in corrections which will be seen to often have
obtained as follows. Solve for K (t) for times t»Tc , the characteristic of introducing subsidiary weak
where Te is a correlation time appropriate for the random Lorentzian components into the spectrum. The general
process modulating XI(t) .Sb Since the cumulant average procedure may be illustrated by the following simple
vanishes if any of the Xlt(tl)x in Eq. (2.19b) are example. Suppose we have
uncorrelated, the only nonvanishing contributions to
Eq. (2.19b) come from times Ti ;STc. Thus, a negligible [XI(t) , JCo]=O,
error is introduced into Eq. (2.19b) by letting all the so that
upper limits tend to infinity. Also, only those terms of xlt(t)X=XI(t)X
Eq. (2.19b) which are not exponentially decaying in t (i.e., only a secular perturbation); also let (XI(t) )=0.
(e.g., correlations depending only on the differences in Suppose, further, that
times such as are characteristic of the assumed sta-
tionary-random functions) will remain nonnegligible. (XI(tl)xXI(tl-T)X)=(1 XI(tl)X 12) exp( -I t lire).
Thus Kn(t»Te) achieves a steady-state valueS with, (2.22)
at most, a sinusoidal time variation. We can therefore
define Then consider the second-order cumulant K 2 (t)
Kn(t»Te) ""Kn(t---+ ao) =exp( -innt)R(n)* (2.20)
K2 (t) = - (! XI(tI) X12)[ exp( - IT I/Te)dT
where R(n)* is a time-independent operator and nn is o
also an operator. It is shown in Appendix A that nn =n =-(1 XI(tl)X 12)rc[1-exp(-1 tllre)]
independent of n provided n operates on the eigenstates
t».c
of JCo. Thus we set ~ - (I XI(tl)X 12)Te. (2.22a)
K(t---+ 00) =exp( -int)R* (2.20a)
where Now for secular perturbations, one has
[XI (tl) x]aa'tlW = [X) (tl ) aa-XI(t l ) a'a]5 atloa,w, (2.23)
so that
(2.24a)
The matrix R* is the complex conjugate of the familiar
relaxation matrix, and we now have a prescription for
where
obtaining it to higher orders.12 It is readily seen from
Eq. (2.19b) that R(n) is of order (I Xlt(t) In)Trl or that K2 (t)aa'W = - Ll2Tc[1-exp( - I t liTe) ]Oa~a'/l' (2.24b)
successive terms differ by a factor of order I XI t)t) I Te.
with
This may be expected to yield a convergent expansion
provided that I Xlt (t) I T e< 1. The validity of using this (2.24c)
Then:
asymptotic limit rests upon having
(2.25)
I Te Re(R) I «1 (2.21) where
and
I Te rm(R) 1«1. (2.21a)
12 The reason for introducing R* rather than R, at this point, is
discussed in Sec. III.
380 JACK H.
and the expansion
eXp[-.:l2Te2 exp( - I t liTe]
= L[( - )nln!].:l2nTe2n exp( -n I t liTe) (2.27)
" -.:l2Te2(1+2i.:lTe)
has been utilized in order to make Eq. (2.26) easy to
invert according to Eq. (2.1). Equation (2.26) is seen
to be essentially an expansion in powers of .:l2Te2. By
making use of Eqs. (2.1) and (2.5), we obtain as the
spectrum implied by Eqs. (2.25) and (2.26)
[(2)(W) = eXp(.:l2Te )
2
t
(_)n (.:lTe)2n
1I'.:l n-o n!
X (nl.:lTe)+.:lTe
[(nl.:lTe) +.:lTe]2+[(W-wo2) 1.:l2]
"" (.:l2TeI1l') {[ (.:l2Te) 2+ (W-Wo)2jl
-[Te- 2+(W-WO)2]-1}, (2.28)
where WO=Waa' and the approximate result in Eq. (2.28)
is valid if .:l2Te2«1. We note at resonance, the amplitude
of the subsidiary line is .:l4Te4 times that of the main
line (or very small in this approximation). The ratio
of the integrated intensities is, however, .:l2Te2 empha-
sizing that the subsidiary line is, in fact, a second-order
correction to the main line. It is a weakness of the
cumulant expansion method that K 2 (t) evaluated for
finite t includes corrections higher than second order in
.:lTe which have been dropped in the approximate
equality of Eq. (2.28). In general, if one wishes to
carry out a solution to mth order in .:lTc, it is necessary
to evaluate all Ki(t) for i= 1, 2· . 'm but then to retain
only the terms up to mth order in each of the Ki(t).
For asymptotically large t, there is no such problem.
Equation (2.28) is not a useful result yet, because
there are often large frequency shifts associated with
the subsidiary lines, which may be obtained from the
K 3 (t) term. In the present simple case, assuming
(2.29)
we have
K 3 (t)aa'aa' = -(i.:l3Te3 )[(1 t liTe) -2+2 exp( -I t liTe)
+ (I t liTe) exp( - I t liTe)]. (2.29a)
Then
[exp(K2+Ka)]aa'aa' =exp- {.:l2Te(1 +i.:lre) 1t I
+.:l2d(1+2i.:lTe+i.:l1 t I) exp( - I t liTe)
-.:l2Te2(1+2i.:lTe) I. (2.30)
The second term in the exponent in Eq. (2.30) becomes
negligible for t»Te • Then, under the approximation
.:lTe«1, we have .:It<<1 for times of the order that make
a nonnegligible contribution to this term. Thus, we can
let
(1 +2i.:lTc+ i.:l1 t 1):::exp[2i.:lTe+i.:l1 t IJ,
FREED
so that
[exp(K2 + K 3) Jaa'aa,""exp- (.:l2Te( 1+i.:lTe) 1t 1
+.:l2Te2 exp(2i.:lTe) exp[ - (Te-1-i.:l) It IJ
I. (2.30a)
When the second term in the braces of Eq. (2.30a) is
expanded in the manner of Eq. (2.27) and only terms
of lowest order are kept, then it is seen that (W-WO)2
in the denominator of the subsidiary line in Eq. (2.28)
must be replaced by (w-wo-.:l)2, which gives an
appreciable frequency shift .:l, when compared to the
width of the main line of .:l2Te for .:lTe«1. Similar
techniques should be applicable to obtain still higher-
order corrections to the subsidiary lines.
The above discussion is readily generalized for any
secular perturbation [where Eq. (2.23) is validJ,
characterized by a single exponential decay of its
correlation [e.g., Eqs. (2.22) and (2.29)]' Then the
subsidiary line is approximated by
- (C2/1r) {Te-2+[W-wo- (C3/C2) Te-1J2 1-1, (2.31)
where
(2.31a)
and
(2.31b)
and it has a maximum amplitude of 1C2Te 12 times that
of the main line. The fact that the subsidiary line
appears opposite in sign to the main line may be
expected on simple physical grounds. It is well known
that the Lorentzian line shape is not a very good
approximation to the true line in the far wings, because
its intensity does not decrease fast enough to yield con-
vergent moments. 8 •6 •8 The very broad subsidiary line
will, however, be most effective in subtracting out some
intensity from the wings of the main Lorentzian compo-
nent, so that the resulting composite line does decay
more rapidlyY
In this simple example, we have avoided specifying
the exact nature of the stochastic process. It is impor-
tant to analyze the process carefully in a particular
problem, because it is possible that divergences could
be incorrectly introduced into the higher-order terms of
R.14 It is also necessary to specify the details of the
13 One may show that for Ll1".< 1, the approximate expression of
Eq. (2.28) does have a finite second moment unlike a pure
Lorentzian.
14 Thus, in our simple example, if we were to let
m4=Ll4 exp[ - (1"1 +T,+1"3) IT.],
then
while
m,(ll, I,) m2(ta, (4) = Ll4 exp[ - (Tl+1"3) IT.],
which diverges when integrated over 1"2 as the upper limit on T2 is
allowed to approach infinity. In the examples of Markov processes
of Sees. IV-VI, however, this latter term cancels an equivalent
term in the proper expression for m4.
 SPIN-RELAXATION THEORY 381

stochastic process in order to obtain explicit formal lent expression for S",(t) is then
expressions for the subsidiary lines following the method
S.,(t)aa' =iwaa' S"' ....' + LR(n) *""'1313' S.,f3f3', (3.3a)
indicated above, since Eq. (2.31) has limited applica- n=1
bility.
where R(n)*a"'f3f3' is the complex conjugate of R(n) .... ,f3f3'.
III. THE RELAXATION MATRIX In the interaction representation [d. Eqs. (2.S), (2.6) J
Eq. (3.3a) becomes
Before explicitly developing equations for the R or S.,: .... , = 2: exp[ -i(w",,'-Wf3w)tJ2:R(n)*""'wS"':f3/3',
relaxation matrix, we note that R is usually obtained f3f31 n-1
as a solution to the equation of motion of the spin- (3.3b)
density matrix 'U 1.4 ••• 6: where only terms for which
(3.1) I w....' -Wf3{J' I ;s I R .... ,{J{J, I
need be retained.1 •s- s
To avoid confusion, we use Eq. (3.1) as the basis for
We now rewrite JC1(t) in the form
generating the R matrix, utilizing the procedures of
Sec. II. We define an ensemble averaged spin-density JC1(t) = 2:Fq(t)Aq, (3.4)
matrix u(t) = ('U(t) ) where the defining characteristics q

of the ensemble are the same as discussed after Eq.
(2.4). Our solution is again valid for times t»Te as
discussed with respect to Eq. (2.20). The final result
may be expressed as
(3.2)
where a, a', (3, and (3' are eigenstates of Xo. The equiva-
where F q(t) is a function of spatial variables and is thus
a randomly varying classical function of the time, and
A q contains only the spin operators. The subscript q
labels the different components contributing to JC 1 (t).
We can then write down the R matrix to fourth (or
higher) order using the results of Sec. II. Since it is
usually possible to define JC1(t) so that (Fq(t) )=0, we
assume this to be true. Then

R(I) aa'f3f3' = 0,
R(2) .... '/3fJ' = 1(2) + II(2) +IIIC2)
= 2: {Aq ...{JAr.{J, ..{t'(q, rj (3'a') +t'(r, q; a(3) J-a.. ,/3' 2:Aq''''YAr.'Y{Jt'(q, r; 1'(3) -a..{J2:A q.{J''YAr.'Y",t'(r, q; (3'1') },
q.r l' l'

(3.Sb)
R(3) " .. '1313' = 1(3) + II(3) + IIIC3) + IV(S)

=i 2: {2:Aq.a'YAr.'Y{JA •.{J'''{t'( q, r, s; 1'(3, (3'a') +t'(s, q, r; aI', 1'(3) +t'( q, s, r; (3'a', 1'(3) J
q,r,8 'Y

- 2:Aq.a{JAr./3''YAs.'Y''{t'(s, q, rj a(3, (3''Y)+t'(q, s, rj "la', (3''Y)+t'(s, r, qj (3'1', a(3)J

l'

+a..{l2:Aq.f3"YArm,A •.'Y''''t'(s, r, q; "II"; (3'1') -a"'{J,2:Aq''''YAro'Y'Y,A •. 'Y,~(q, r, s; 1'''1', "I'(3)}, (3.Sc)

1'1" 1'1"

R(4) .... '{J{l' = IC4)+ II(4) + III(4) + IV(4) + V(4)+ VI(4)

= 2: {2:A q''''Y A r.'Y{JA •.{J''Y,A u 'Y, ..{t'(q, r, u, s; "1(3, 'Y'a', (3'''1') +t'(q, u, r, s; 'Y'a', "1(3, (3'''1')
q,r,8,U 'Y'Y'

+t'(q, u, s, r; "I'a', (3'''1', 'Y(3)+t'(u, q, r, s; a"l, "1(3, (3'''I')+t'(u, q, s, r; a"l, (3'''1', 'Y(3)+t'(u, s, q, r; (3'''1', a"l, "I(3)J
- 2:Aq''''YAr.'Y'Y.As.'Y'{JAu.{J, .. {t'(q, r, s, u; "1"1', "1'(3, (3'a')+t'(q, r, u, s;'Y'Y'; (3'a', 'Y'(3)+t'(q, u, r, Sj (3'a', "1"1', "1'(3)
1'1'

+t'(u, q, r, S; a"l, 1'''1', "I'(3)J- ~Aq,"IlAr./3''YA',n,A".'Y, ..{t'(q, u, s, r; "I'a',"11", (3'''I)+t'(u, q, s, r; a{3, "1"1', (3'''1)
'Y'Y'

+t'(u, s, q, rj "1"1', a(3, (3'''I)+t'(u, s, r, qj "1"1', (3'1', a(3)J+a"'{J,2:Aq''''YArm,A •. 'Y'aAu.6~(q, r, s, Uj "1"1', "1'15, 15(3)
1'1"6
382 JACK H. FREED

The spectral densities t(w) in Eqs. (3.S) are given as one-sided Fourier transforms of correlation functions geT)

t 2(q, f; a(3) = fOO g2(q, f; T) exp( -iw",{3T)dT,

o (3.6a)

ta(q, f, s; Wa , Wb) = fcc dTlfcc dT2 exp[ -i(Wa+Wb)TI] exp( -iwbT2)ga(q, r, S; TI, T2), (3.6b)
o 0

wherel5
g2(q, r; T) = (Fq(t)Fr(t-T», (3.7a)
gs(q, r, S; Tl, T2) = (Fq(t) Fr(t-TI) F.(t-TI-T2) ) (3.7b)
g4( q, r, s, U; TI, T2, Ta) = (F q(t) Fr(t-TI) F.(t-Tl-n) F u(t-TI-T2-Ta) )
- (Fq(t)Fr(t-TI) >(F.(t-TI-T2)F u(t-TI-T2-TS) >. (3.7c)
The R(2)"'''''n' which appears in Eq. (3.Sd) is obtained from R(2)"'''''rr' by replacing the t(q, r; Wa ), etc., by
- fCC (1-exP[i(w{J{3'-W'Y'Y')T])
t2(q, f; wa) = g2(q, r; T) .( ) exp( -iWaT)dT. (3.6d)
o 2 w{J{J'-Wn '

The equations (3.7) define multiple-time correlation
functions, which depend only upon time differences due
to the assumption of stationarity. The spectral densities
of Eq. (3.6) are complex. Thus, for real values of the
products of matrix elements in Eqs. (3.S), it is seen that
the real and imaginary parts of both the second- and
fourth-order spectral densities lead to the linewidth and
frequency-shift contributions, respectively,16 while the
reverse is true for the third-order spectral densities. The
expressions in Eqs. (3.S) involving Roman numerals
are used to number conveniently the R-matrix terms
corresponding to the different combinations of A q
matrix elements which appear in the same order on the
right. We note that if Kn(t) is desired instead of R n,
the infinite upper limits of the time integrals in the
spectral densities given by (3.6) are replaced by the
appropriate finite values as obtained from Eqs. (2.16)
and (2.19). The resulting spectral densities are still
explicit functions of t.
The terms in Eqs. (3.Sc) and (3.Sd) may be regarded
as higher-order time-dependent perturbations. They
include secular and nonsecular effects (as well as
admixtures of the two). Thus, for example, term IV(4)
with a={3 (and neglecting a secular component) may
be regarded as the linewidth contribution due to a
second-order transition probability from state a to
16 Our notation differs from the usual notation (d. Refs. 1, 6),
where g",,(T) "" (Fq(t) Fr*(t+T) ). Our notation generalizes more
easily to the formation of the higher-order time correlation func-
tions, given by Eqs. (3.7). It follows from the properties of
stationary random functions that g2(q,r;T)=g2(r,q;T)=
g2(q, r; -T) and is not necessarily real.
,. G. K. Fraenkel O. Chern. Phys. 42, 4275 (1965)] discusses
dynamic frequency shift effects in ESR spectra arising form the
second-order spectral-density contribution as given by Eq. (3.6a).
state "{' via the whole set of intermediate states "{ and
o. (Such transition probabilities are, of course, fourth
order in the perturbation.) These intermediate states
may be nonresonant or (near) - resonant, where in the
latter case, any small deviations from resonance are
taken up by the spectral densities generated by the
random process. Terms like II(4) are somewhat unusual
in that they may be regarded as cross terms between
third-order and first-order contributions. They are
normally neglected in second-order time-dependent per-
turbation theory which is usually utilized when first-
order terms are unimportant. Term VI(') resembles
[R(2)]2 and shows that powers of the lower-order
R-matrix terms also appear in modified form in higher
order. It indicates, in part, that if the transition defined
by the pair of states (3{3' is uncoupled by R(2) to aa' in
second order but both are coupled to a common transi-
tion "{"{', then the {3{3' and aa' transitions become
coupled in fourth order.
The significance of all these R-matrix terms will be
seen to depend on the nature of the particular problem
being investigated.
IV. APPLICATION TO A TWO-JUMP PROCESS
The two-jump problem, which has been treated by
second-order R-matrix theory,! serves as a simple ex-
ample of the application of the cumulant method to
Markov processes involving discrete values of the
random variable. Also, it is a case which can be solved
exactly,6.8·17,18 so the results of the cumulant method
17 P. W. Anderson, J. Phys. Soc. Japan 9,316 (1954).
18 H. M. McConnell, J. Chern. Phys. 28, 430 (1958).
 SPIN-RELAXATION THEORY 383

may be compared with the exact solution (d. Appendix mI, respectively). Then Eqs. (3.5) become
B).
Consider two states, A and B, with a priori proba-
bilities: (4.8a)
( 4.1)
and with mean lifetimes TA and TB. We may write the
four relevant conditional probabilities1 in the general (4.8b)
form
R (4)* aa'aa' -(4
- \.£ aa- A a'a' )4{(.
'4 a, *
0, 0, 0 )- Tc [R(2) aa' ,aa' J2
P(i!j,T)

=WL1-exp(-!T!/rJJ+exp(-!T!/Tc )('l;;, (4.2)

The spectral densities are easily calculated from Eqs.
where
(4.6) (even for finite times, d. end of Sec. III). We
i,j=A or B, obtain a solution valid to fourth order for the main line
and
associated with the a-'>o/ transition and third order for
(4.3) the subsidiary line, from
We assume that a radical jumping between these states
experiences a modulation of its isotropic hyperfine inter- K(t)"" [[K2(t')+K3(t')Jdt'+R*(4)t
action between the values aA and aBo Then o

ftJC o=g{3e B oSZ+ft! 'Ye! aSJz-ft'YnBolz, (4.4a) = -mI2WAWB~2Tclt+Tc[exp( -tire) -1JI

and
+imJ3(WA-WB)WAWB~3Tc2
ftJC 1 (t) =ft! 'Ye! [aCt) -aJlzSz=F(t)A, (4.4b)
with X[(t-2T )+(t+2Tc ) exp( -tITe)J
C

( 4.5)
+mI4WAWB(1~5WAWB)~4Te3t. (4.9)
Equations (4.4) include only the secular contribution. 1
If we let! 'Ye ! (aA -aB) =~, then it is easy to show Then exp[K(t) J may be expanded according to the
utilizing Eq. (e2) and Eqs. (3.7) that method illustrated by Eqs. (2.26)~(2.31) to yield

(4.6a) exp[K(t) J""[exp(Z) exp( -Xt) -Z exp( - Yt) J,

g2(Tl) exp( +TI/Tc) (4.10)
where
='Ye2 L:Wi [a;(t) -aJ2
i

= WAWB(2~)2,
g3(Tl, T2) exp[(Tl+T2)/rcJ
= ! 'Ye !3L:Wi[ai(t) -aJ3
i ( 4.11c)
= - WAWB(WA-WB) (2~)3,
g4(n, T2, T3) exp[(TI+T2+T3)/rcJ
='Ye4 L:Wi [ai(t) -aJ4- hl~=Wi[ai(t) -aJ2)2
i i
Equations (3.5) and (3.6) may be simplified in the
present case, because we have
t 4(q, r, S, u; 'Y'Y', 'Y{3, (3'Y) =t4(a; 0, 0, 0),
since there is only one perturbing term in JCI(t) which
is also secular. Also, Raa'{3{3' = R aa'aa'Oa{30a'f3' for the a-'>a'
electron-spin transition, (where a and a' represent the
spin states characterized by m.=!, mI and m.= -!,
(4.6b)
-mI4WAWB(1~5WAWB)~4Tc3, (4.11a)
Y"-'Te-l+i(WA-WB)~mI, (4.11b)
(4.6c) These results are in agreement (within the order of this
calculation) with those obtained directly from the
modified Bloch equations17 ,18 as applied to this problem.
(See Appendix B.) The interpretation of this result is
similar to that of the simple example of Sec. II, where
now the stochastic process has been defined properly.
( 4.6d) The first term in Eq. (4.10) gives rise to the main line
for the particular value of mI. Its width contributions
are given by the real part of X in Eq. (4.11a) and the
corrected frequency is given by [Waa' - ImX]. The
(4.7) associated subsidiary line arises from the second term
in Eq. (4.11) so that the width is Tc- 1 and the resonant
frequency is
waa,-i(WA-WB)~mI.
We note that according to Eq. (4.8c), a line-broad-
384 JACK H.
ening mechanism involving modulation of the isotropic
hyperfine interaction introduces higher-order correc-
tions which depend upon powers of mI. Thus, for free
radicals with many equivalent nuclear spins [e.g.,
duroquinone and related radicals19 & with twelve equiva-
lent protons giving a total mI- (max) =6J there may
well be a nonnegligible fourth-order correction to the
widths of the outer lines when the correction is negligible
for the central lines [e.g., for ATc=0.2 and W A = W B =!,
we have
or a 1% correction for mI = 1 but a 16% correction for
mI=4]. Since the relative intensity of the subsidiary
line associated with each of the main lines also depends
on mI2, their effect becomes more pronounced for large
values of mI. (Considerations such as this, could have
important consequences for the relative line shapes of
different hyperfine lines. 19b ) These comments are ex-
pected to apply, in a modified form, to other dynamical
processes which can modulate the isotropic hyperfine
interaction1.2O as well as other relaxation processes
involving nuclear-spin quantum numbers.I.2,19
V. APPLICATION TO ROTATIONAL MODULATION
OF THE SPIN-ORBIT OPERATOR
It is well known in ESR that the g tensor arises from
a second-order time-independent-perturbation cross-
term between the spin-orbit interaction 1i-1AL· Sand
the orbit-field interaction 1i-1{1.L· Bo. Thus
gij= g.Oi,j- 2A1:[L i ,o,nLj ,n,o/ (En - Eo) J, (5.1)
nr'O
where g. = 2.00232 is the free-electron value, gij is the
ijth Cartesian component of the g tensor, and the sum
over n is over all excited states with energy En. This
result is appropriate for free radicals in solids, but it
has also been implicitly carried over into the theory of
spin relaxation in liquids.1.2 That this may be done is
not immediately obvious, because both perturbation
terms must be expected to be random functions of time
in the liquid state. That is, L is strongly coupled to the
molecular frame, while S for large values of Bo is coupled
to the laboratory frame defined by the direction of B o•
The rotational tumbling randomly changes the relative
orientations of the two coordinate systems. We may
then ask what are the limits of validity in using Eq.
(5.1) when "motional narrowing" of the g tensor is
observed.
While this question may be dealt with by utilizing
the formalism discussed below, we note that there are
a number of experimental and theoretical studies which
19 (a) G. K. Fraenkel, J. Phys. Chern. 71, 139 (1967). (b) J.
Gendell, J. H. Freed, and G. K. Fraenkel, J. Chern. Phys. 41,
949 (1964).
20 J. H. Freed and G. K. Fraenkel, J. Chern. Phys. 41, 3623
(1964) .
FREED
bear upon it. Thus, a variety of ESR linewidth studies
summarized by Fraenkel19 & tend to support the conven-
tional approach of calculating g tensors, as at least a
good first approximation. The same may be said of
the comparison between Stone's theory of g tensors in
aromatic free radicals21 and the measurements of liquid
g shifts by Segal, Kaplan, and Fraenkel,22 There is still
the question as to whether there are other spin-orbit
relaxation processes23 ,24 which yield experimental line-
width effects in ESR spectra and are distinguishable
from a g-tensor mechanism. More interesting, perhaps,
are the as yet unresolved questions25-2B relating to the
anomalously large widths characteristic of radicals with
degenerate ground vibronic states (in the gas phase)
such as the benzene anion,29 as well as the lack of corre-
lation of the liquid g shifts of such radicals with those of
other hydrocarbon radical anions. 22 With these consider-
ations in mind, we examine the structure of the spin-
orbit formalism in free radicals as applied to liquids
(Our treatment in the present section is simplified by
neglecting the breakdown of the spin-orbit interaction
into a sum of localized atomic terms appropriate for a
delocalized electron,21 but this should not significantly
affect the over-all nature of our results.)
A. Structure of the Theory
A generalized interpretation of the methods of the
earlier sections must be utilized. Thus, for example, we
write L· S in the notation of Eq. (C6)
1i-1AL· S =1i-1A 1: ;D0)_m,m' (n) L'(l,m) S(l,m l
), (5.2)
m,m'
where both Land S are first rank irreducible tensors
and are operators on the combined electronic and spin
degrees of freedom. ;D(I)-m,m,(n) remains the only
random time varying function. The molecular coordi-
nate svstem defining L'(J ,m) is so chosen that gij = giOij
in Eq~ (5.1), and for simplicity it is taken as the
principal axes of the diffusion tensor. Then the equa-
tions in Secs. II and III and Appendix C may be
applied, provided the following approach is adopted.
Instead of the matrix elements (Aq(l,ml»)aa' which
appear in Eqs. (3.5), we must evaluate (Fq(l,m»)anan'X
(A q(l,ml»)aa', where a and a ' are spin states and an
and a ' n are the associated electronic states. Then each
21 A. J. Stone, Mol. Phys. 6, 509 (1963); 7, 311 (1964).
22 B. G. Segal, M. Kaplan, and G. K. Fraenkel, J. Chern. Phys.
43,4191 (1965). .
23 D. Kivelson and G. Collins, Proc. Intern. Conf. MagnetIc
Resonance 1st, Jerusalem, Israel, 1961, p. 496 (1962).
24 (a) P. W. Atkins and D. Kivelson, J. Chern. Phys. 44, 169
(1966); (b) D. Kiveison, ibid. 45,1324 (1966).
26 H. M. McConnell and A. D. McLachlan, J. Chern. Phys. 34,
1 (1961).
26 H. M. McConnell, J. Chern. Phys. 34, 13 (1961).
27 J. H. Freed, J. Chern. Phys. 43, 1427 (1965).
.. D. Kivelson, J. Chern. Phys. 45,751 (1966).
2D M. G. Townsend and S. 1. Weissman, J. Chern. Phys. 32,
309 (1960).
 SPIN-RELAXATION THEORY 385

F'qO,m) in Eqs. (ClO) is replaced by the appropriate
(F ' q(l,m»ana'n' where the F' q matrix element is chosen
to correspond to the associated A q matrix element which
appears in Eqs. (3.5). This may be regarded as a further
generalization of the index q. The sums over {3, {3', "I, "I',
etc. in Eqs. (3.3) and (3.5) are interpreted as combined
sums over {3, {3n, "I, "In, etc., and the frequencies WafJ,
etc., become WafJ+Wan.fJn' We illustrate with
Since the diagonal matrix elements (L'(l·m»nn vanish
for a polyatomic molecule with a nondegenerate ground
state,30 then 1(2) in Eq. (3.5b) is zero. The next two
terms yield

R(2) (a.a.l.(a' .a'.) = -i L:: Aq.a./l,+ml)Ar.1'a(J,-ml)L:: (- )m+mIF'q,a•. 1'.(J,mlF'r,1'.a.(J,-ml [Tl.m- 1+ i (W1'a+W1'.a.)]-1

q,r,'Y,m l m.'Yn~an

-i L:: A q ,a'1'Cl,m'lA r ,1'a,(J,-m'l L:: (- )m+m'F'q,a'.,1'.(J,-mlF'r,1'.,a,.(l,ml[Tl,m-l+i(Wa'1'+Wa'.1'.)]-1, (5.3)

q,T,',,(,m' m,'Ynrfa'n

where q and r sum over the s-o and O-F type perturbations, as given in Table. I
As long as there are no low-lying electronic states coupled by the operator L to the ground state, and if an and
a'n are taken as the ground electronic state n=O for a spin doublet, then
(5,4)
and Eq. (5.3) becomes
R(2)(
Q,
Ol • Ca 'o)~+!.i
• -
"(-)m'(A q,O:'Y (l,m'lA r,,),Q (I,-m'l_A g,a ,(I,m'lA
3 "--' l' T,j'O!
,(1,-m'l)
q,r.'Y

x L:: (- )m(F'q,O,1'.(I·mlF'r,1'n,OO,-ml/w,,(•.O)' (5.4a)

m,1'n;o!O

If we let a correspond to spin-state m. = +! and a' to free-electron value given from Eq. (5.1) by
ms = -!, then it is found that the only contribution to
this expression comes for qrfr, and one obtains utilizing (g.-g.) II = 12X L:: \ Li,o,n \21, (5,6)
Table I 1 h 3h n;o'O;i-x,y,z En-Eo
R(2) +1/2,-1/2 where the irreducible tensor components L'(m) in Eq.
(5.5) have been reexpressed in Cartesian components.
= (2l*-1/2.+1/2 The sign in Eq, (5.5) depends on whether ~1/2,-1/2 or
=ii(X{3.Bo/h2 ) L:: (- )m(L'o,n(m) L'n.o C- ml /Wn,o) , L 1/ 2 .+1/2 is taken, and leads to the observation of a
m,n¢()
resonance determined by g.=i(gx+gy+g.) , the average
(5.5) g value. Thus the g shift appears naturally as a second-
order dynamic frequency shift and is much larger than
which is just the dynamic frequency shift from the any linewidth contribution, provided that the inequality
(5.4) holds, For
TABLE 1. Terms in the perturbation X, (t).
\ Tl,m- I I ;G I Wo, .. I »wa1"
Spin-Orbit (S-O)
then (Wn.o)-I in Eq. (5.5) for the frequency shift, must
A p (I,m') m'=O S. be replaced bywn,o/(Tl,m-2+Wn}) , and the shift depends
m'=±l =F2- I12 S±
Fp'(I,m) m=O h-IXL. upon the rotational correlation time. There would now
m=±l ±2-'12h- ' XL± be an important linewidth contribution as well. It may
Orbit-Field (O-F)
be seen by utilizing Eq. (5.3) that for RC2\1/2._1/2 there
are nonvanishing contributions only from terms where
A p (I,m') m'=O Bo q=r. The 0-F terms could in principle cause broaden-
m'=±l 0
Fp'(I,m) m=O h- I{3.L. ing because they induce transitions to excited electronic
m=±1 ±2-1I'h-I{3,L± states (without a spin flip). But this effect will be
GTensor unimportant if (1) the ESR resonant frequency (in-
cluding g shift) differs only slightly in the excited state
A (I,m')
• m'=O +i'12Bo S. from that in the ground state, or (2) the equilibrium
m'=±l =FIBoS±
m'=±2 0 population of the excited state is small. For the present
F '(I,m) = ( -I) mh- I{3.g(m)
• g(O) = 6-1I'[2g3 - (gl +g,) ]
g(±I) =0
30 This statement may not be entirely true in the case of the
g(±2) =! (gl - g.) benzene anion. McConnell'6 has suggested a mechanism whereby
a nonvanishing val\le c;o\lld eltist for short intervals of time,
386 JACK H. FREED

we neglect the Sz= S(1·0) components of the S-O terms may now be recovered by examining the appropriately
and consider only the effects of S(1·±l) terms which generalized form of the R(4) given by Eqs. (3.5d) and
do induce spin flips. They yield a contribution to (ClOd). Of the six main groups of terms in Eqs. (3.5d),
141/2,-1/2 for the ground state of only 1(4), 1V(4), and V(4) are found to have the necessary
characteristics that lead to terms of the form of the
2A2
- -3h L: (_)m
2 m.nr'O
7

71,m-1+Wo,n2
-1
I,m L'
O,n
(m)L
'
n,o
(-m)
•
(5.7) g tensor squared, [e.g., for JC4J,

L:A q,a,),A r,'YIlA 8 ,1I''Y,Au,'Y'a' = (AqAr)all(A.Au)lI'a'].

This is very similar to a result obtained by Kivelson 'Y'Y'
and Collins,2a which they call a rotational spin-orbit
Furthermore, only the first and last spectral densities
mechanism. For Wo ,n»71,m-1, spectral densities of the in 1(4) preserve the needed coupling of products of the
form used here begin to lose their significance because
F"s. We illustrate with the term of type
short-time, nondiffusive motions become important, as
do Boltzmann factor corrections to the spectral densities L: L:A q,a'YAr,'YIlA',II''Y,Au,r'a,t( q, r, u, S; -y{3, -y'a' , (31-y').
associated with transitions of large energy. Nevertheless, q,T,8,U "1"1'

the result Eq. (5.5) for the frequency shift, which does (5.8)
not contain any explicit dependence on the tumbling
motion, is expected to hold. Utilizing Eqs. (3.7c) and (ClOd), properly general-
The normal line-broadening terms from the g tensor!,2 ized, we have

t(q, mt'; r, ~'; s, ma'; u, mi; -y{3, -y'a' , (31-y') = L:

i¢,Oim,m '
(1
m/
1

1 1
X L:
~mi' and ( -m
electronic states 1

1
2 ) ( -) m-m'--
Fqr,ano'Y.,Il" (2,m)'-"
F.u,1l n,'Y .. a "(2,-m) [7J,m
. -1 .
+t(WIl "
I
a +WIl ""a J J 1,
m/ -m'
(5.9)
where Eqs. (Cll) and (C12) and the symmetry relation

(jl
ml
j2
m2
ja) = (
m3
-1) i1+i2+ia (
-ml
jl

have been employed, and we have introduced

1
F-BU,/l,l'l'
, ,n,a ,(2,-m)
n
= '"
k...,;
51/ 2 ( 1
m3,m, 1113

2) F' 8,ft" ,1'" (I,m::)F'"

~-~~~~--~-.
u.r n,a 11.
(1,m,)
(5.10)
m W'Y'"~'''

The approximate equality of Eq. (5.10) holds when the inequality (5.4) applies. Then, if excited electronic states
-Yn and -y' n are independent of the particular -y and -y' spin states, etc., and an, a ' n, {3n, {3'n are taken as the ground
electron state for which n=O, we have
n
r ../2,+m) =i5 1/ 2
'"
£.....
(1 1 2) F'.,O,n(1,m3)F' u,n,0(1,m4)
.~-~~~~-- (5.11)
n"'0;m3,m4 ~ m4 1n W".o

Then letting sand U equal S-O and O-F terms and utilizing Table I, we have
2F s _o,O_F(2,-m) =i( -1)m{3eh-1g(2,-m)=iF'.(2,-m). (5.12)
 SPIN-RELAXATION THEORY 387

The 3j symbols may be evaluated using standard references. 31 Then

1 2)( 1
tn-J.' m' ma'
1

x (- )m-m'lF'g(2,-m)F'g(2,m) (T2,m-1+ iwll'a,)-I. (5.13)

When this result is substituted into Eq. (5.8), Eq. (5.9) is again used, and the appropriate summations are per-
formed, we have
.! ~ A
5 L..J g,a
jl(2.-m')A o. fl' a ,(2. m') (_) (m-m')F' 0 (2,-m)F' 0 (2,m)(T2 ,m-1+iwfl 'a,)-I. (5.14)
m,m'

Equation (5.14) is seen to correspond to the first Equation (5.16) just requires that g2(q, r; T) be
spectral density term of 1(2) in Eq. (3.5b), with Eq. expressible as
(e8b) for a simple g-tensor line-broadening mechanism.
g2(Q, r; T) =g2(Q, r; 0) exp( -riTe). (5.16a)
The second spectral density in 1(2) for the g tensor
arises from the last spectral density term in 1(4) of For f»T c , we have
Eq. (3.5d), in the present treatment, etc. The other
four spectral density terms in the first term of Eq. Rew'Ya (t~ 00 ) == Re'lli-ya = W'Ya,
(3.5d) do not contribute to linewidth effects, when where W'Ya is the lattice-induced transition probability
Eq. (5.4) applies, since one may show that the second between states a and ,),.6,33 By analogy to the method of
and third spectral density terms in 1(4) approximately Eqs. (2.22)-(2.28) and Eq. (2.1), we obtain subsidiary
cancel as do the fourth and fifth terms. Terms II(4) and lines whose intensities, widths, and frequencies are
III(4) lead to cross terms between third-order and first- determined by the Fourier transform of
order perturbation theory, and may be neglected in the
approximation that first-order effects [e.g., the result -! ReLlw\a[Tc-1- iw'YaJ)
Eq. (5.7)J are unimportant. 32 Xexp [ - ( Tc- 1 ' .
-'tW'Ya-'tWaa' -*a''Y [Tc- 1 -'tWa''Y.J
) tJ+ W . ""l-l
The method outlined above, wherein the effect of
excited electronic states is explicitly included (as X exp[ - (T c-1- iwa''Y - iw aa ,) tJ} .
opposed to assuming the validity of a "solid-state" spin That is, the widths are again given by Tc- 1, but the
Hamiltonian for liquids) could be applied to spin- frequency shifts are W'Y a and Wa''Y' In the present section,
rotational mechanisms24 & as well as various higher-order when !w'Y a and !wa''Y are identified with the two sets of
processes24b including those which involve stochastic terms in Eq. (5.3), then the frequency shifts are of the
processes other than rotational diffusion. order of Wo,n or of electronic transitions, and they are
B. Subsidiary Lines not relevant for ESR.

As will be seen shortly, the subsidiary line associated C. Discussion

with the perturbation result given by Eqs. (5.7) and
(5.5) can be neglected. However, we analyze it as a
simple example of a subsidiary line arising from a non-
secular term. We need only consider K 2 (t), which for
nonsecular-only terms is given (in general notation) by
[K2(t) aa'aa,Jnonsee = -! L[W\a( t) +w*a''Y(t) J,
'Y
where from Eq. (3,Sb)
W'Ya(t) =2LAq,a'YAr.'Yag2(q, r; 0) (Tc-J+iw'Ya)-1
g,T
X {l-exp[ -h-l+iw'Ya)t]l.
It is seen from the above analysis that for free radicals
with nondegenerate ground states, the use of a solid-
like g tensor in the liquid state is justified on theoretical
grounds, and furthermore, there exists no other rota-
tionally-modulated spin-orbit relaxation mechanism of
(5.15) importance (at least to fourth order). The g-tensor
relaxation terms may be regarded as a second-order
time-dependent process (from R(4») , via virtual excited
electronic states. This is then an example where a
"higher-order" process makes a dominant linewidth
contribution, and the generalized condition for motional
(5.16) narrowing Eq. (2.21) must be employed.

31 M. Rotenberg et aI., The 3j and 6j Symbols (Technology Press, Cambridge, Mass., 1959).
32 It is readily shown, using an analysis similar to that above, that when Eq. (5.4) applies, the only contribution for a poly-
atomic molecule from the S-O and O-F terms in third order (Le., from R(3») is a small dynamic frequency shift that is one order
higher in essentially )."/;u,,O,n than the g shift from R(2) and can therefore be neglected. Such dynamic frequency shifts that are of
order ).,,4/h'wo,n3 have been neglected in the above discussion. They arise from terms IV(4), V(4), and VI(4) in Eq. (3.5d). (They are the
principal terms from VI(4).)
33 J. H. Freed, J. Chern. Phys. 43,2312 (1965).
388 JACK H.
On the other hand, there could be relevant implica-
tions for the benzene anion problem. In the liquid state,
it is unreasonable to expect that the benzene anion will
retain its vibronic degeneracy, except perhaps for a
small fraction of the time,26 although the actual mag-
nitude of the (random) -solvent-induced splitting has
only been guessed at. 25 It could be possible, then, that
for at least short intervals in time (which are still
longer than rotational correlation times), the inequality
Eq. (7.4') does not hold, where W o•n is taken as the
splitting of the degenerate states. This suggestion is
still insufficient to yield even a partial breakdown of
the conventional formulation, because, to the extent
that the degenerate benzene orbitals are pure p7r or-
bitals, the operator L will have no appreciable matrix
elements between them. 26 However, McConnell has sug-
gested a mechanism involving distortion of these or-
bitals with a small amount of atomic d character that
leads to orbital angular momentum about the ring. 26
Another, perhaps more obvious, way of obtaining a non-
vanishing (L)o.n would be to include the effects of 0"--'1f"
interaction (which is, in fact, the accepted mechanism
generating the benzene-anion proton splitting34) mixing
some 0"* character into the 7r orbitals. It should be
noted that from Eq. (5.5) and the discussion following,
the g shift could have a component that depends on the
rotational correlation time, hence on temperature. 35
Also, the rotational spin-orbit mechanism of Eq. (5.7)
could become effective. It is clear that further work is
necessary to analyze the importance of such possibilities
as compared to the suggested mechanisms of
McConne1l26 and Kivelson. 28 In any event, all these
spin-orbit mechanisms involve rotational modulation
in an important way, and the formalism of this section
should prove useful in analyzing them in a more rigorous
fashion.
VI. HIGHER-ORDER g-TENSOR EFFECTS
The general applicability and limiting conditions of
the g-tensor formalism in the case of liquids was dis-
cussed in the preceding section. Rather than continue
with the spin-orbit operator to fifth and higher order,
we now use the g-tensor formalism directly in the
R-matrix expansion.
Since the higher order effects will begin to be impor-
tant for slower tumbling rates, we consider the case of
W;»TI. m-l, so that nonsecular effects are negligible com-
pared to secular effects. Thus, the only relevant spin
operator is Agl.O (see Table I). We shall further simplify
the treatment by letting E l •m = TZ-1 independent of m,
(i.e., isotropic rotational diffusion), although this is
not a necessary assumption. These assumptions plus
the inclusion of only g-tensor terms lead to the following
34 H. M. McConnell and D. B. Chestnut, J. Chern. Phys. 28,
107 (1958).
3. In Ref. 22 there is no specific mention of any temperature
variation of the g value for the benzene anion.
FREED
simplified spectral densities [from Eqs. (3.6) and
(ClO) ]:
C2(g, l=2; 0) = !T2[(g(O»)2+2(g(2»)2]Be2h-2 (6.1a)
C3(g; l=2; 0,0) = (2/3sh2g(O{(g(O») 2_6(g(2») 2]B.3h-3
(6.1b)
C4(g, l=2; 0, 0, 0) = (2/35) n2[(g(O»)2+2 (g(2»)2J2,B.4fi- 4
X [(2/7)T2+ (18/3s)T4J (6.1c)
and
( 6.1d)
where the g(m) are given in Table I and the 3j symbols
have been evaluated from standard tables. 3! Note that
Eqs. (Cll) and (C12) have simplified the analysis of
Eq. (6.1c). The first set of Eqs. (4.8) again apply, and
we obtain
R(2)*aa' .aa' = - (2/15) Bo2,Be2h-2T2L.(gi 2- gl), (6.2a)
i
R(3)*aa'.aa' = -i(¥)3/2Bo3Ca(g, l=2; 0), (6.2b)
and
R(4) *aa' .aa' = - (13/35) T2[R(2) *aa' .aa,J2. (6.2c)
Thus, the fourth-order term leads to a small increase in
the calculated width of the main line, while the third-
order term gives its frequency shift.36 The subsidiary
line arising from the secular g-tensor perturbation has a
width of T2- 1 and a resonant frequency of
W aa + (f) . (¥) 1/2,BeBJi,-lg(O)
X I [(g(o»)L6(g(2»)2J/[ (g(O»)2+ 2(g(2») 2J I
and is obtained using Eq. (2.31).
The secular perturbation terms arising from the
intramolecular electron-nuclear dipolar interaction
could be treated in a manner analogous to the g tensor.
However, the pseudosecular contributions from this
interaction would not be negligible compared to the
secular terms, because they involve low-frequency
nuclear-spin transitions. One may note [d. Eq. (s.17)J
that the ESR subsidiary lines from these terms will be
shifted by (=F'YnBo-'Yed), Le., the NMR frequencies
of the radical. However, for large dipolar interaction
terms, the frequency shift from the K 3 (t) term would
have to be included as well.
ACKNOWLEDGMENT
We wish to thank Mr. Daniel S. Leniart for his
help with some of the calculations in this work.
APPENDIX A
We wish to outline a proof showing that the operator
nn in Eq. (2.20) is independent of index n. In Sec. III,
36 If the nonsecular g-tensor terms were retained, then there
would be a width contribution from the real part of R(3)*",,,,',,,,,,',
d. Sec. III.
 SPIN-RELAXATION THEORY 389

the detailed results for n:::; 4 are given, and it is found terms differ only in the ordering of the tj and in the
that in all cases exact groupings of the operators over which the average

[exp( -inl)]aa'w=exp[ -i(Waa'-WIlW)t], (A1)

operation is performed .ccf.
Eq. (2.16c)]. We then
rewrite Eqs. (2.19) for Kn(t-Ht:J) as
where a, a', {3, and (3' represent eigenstates of Xo. We Kn( (0) =inlnAnPnXllCt)xXll(t1)x" ,X1t(tn_l) x, (A3)
wish to show that Eq. (A1) applies for all n. This is
readily done provided the operators in K(t) are given where the operator
matrix representations utilizing as a basis the eigen-
states of Xo. Then we may write
Xlt(tj) Xaa 'IltJ ,
=XICtj)xaa'tJtJ' exp[ -i(waa'-WW)tiJ
dT1['" dT2'"
-'>['"
o
[0
dTn-l. (A3a)
= [Xl (t j ) a~a'P' -Xl (tj) Wa'OatJ] 0 0

X exp[-i(waa,-wtJW) tj] (A2) P n is a permutation operator which generates the

where the definitions (2.Sb) and (2.6a) have been used.
Note that in Eq. (A2) only the XI(t) matrix elements,
and not the exponents, are affected by the stationary-
random process. Note also that from the discussion of
Sec. II and the implied form of Kn(t) given by Eq.
(2.17), Kn(t) consists of a sum of terms each involving
products of exactly n operators Xlt(tj)x. Each of these
required sums of products of Xl t (tJ ) x, while the opera-
tor An performs the needed averaging operations. Now
by letting tj=t-tj in Eq. (A3), the operations In, An,
and P n relate only to the fj and the functions of tj.
The fj decompose into proper sums of the Tj of Eq.
(2.19b) after P n is applied, so that the second form of
Eq. (A3a) becomes appropriate. Thus we may rewrite
Eq. (A3) in terms of matrix elements as

Kn( (0) aa'tJtJ' =inlnAnPn L: Xl (t) xaa''Yl'Yl/exp[ -i(waa' -W'Y11"I) t]X l (t-f1) X'Yl'Y'l'Y2'Y'2
all 'Yi,"'/i/

X exp[ -i(wm' I- Wm'2) (t- t I) ]- . ,Xl (t-f n-l) x'Y._11" .-1!lW exp[ -i(w'Y._l'Y' ._I-WPtJ') (t-f .-1) ]
=exp[ -i(waa'-WP{J')tJinlnAnPn L: X1(t)xaa''Yl'Y'IX1(t-f1)x'Yl'Y'l'Y2'Y'2 exp[i(w'Yl'Y'I- Wm'2)f1]-"
all 1'1.1'1'

XX1 (t-f n-l) x'Yn_l'Y'n_1!lP' exp[i(w'Yn_l'Y'._I-WP{J' ) tn-I]

= [exp( -irU) ]aa'WRCn)*aa'PIl', (A4)
which is the desired result.
APPENDIX B
The well-known solution to the modified Bloch equations for a two-jump systeml7 •18 may be applied directly
to the formulation in Sec. IV to give a line-shape function
( ) Tc- 1+i(WA-WB).1mI+iw
(B1)
I W a: -w2+iw[T.-1+i(WA-WB).1mI]+WAWB.12mI2'

The frequency W=O is taken at the center of the particular hyperfine line of resonant frequency:
fz.w aa, = g.{3J3.+ Ii I 'Y. I amI.
The inverse transform of Eq. (B1), which gives G(t) of Eq. (2.1), may be obtained by straightforward application
of Laplace transforms and yields
G(t) a: ['2 exp( -'It) -'1 exp( -'2t)]/ ('2-'1), (B2)
where -'1 and -'2 are the two values of ![-c±(c2-4d)I/2] while
C=Tc-l+i(WA - W B).1mI
and
d= W AW B(mI.1)2.
For I mIllTc I < 1, then (c2-4d) 1/2 may be expanded in powers of mIllTc. Then the approximate form of Eq. (B2)
has
'1 =mI21l2TcWAWB-imI81l3Tc2(WA - W B) WAWB - WAWBmI41l"rc 3(1-SWAW B) ••• (B3a)
1
'2=T c- +i(WA - WB)mrll-mr21l2TcWAWB···. (B3b)
390 JACK H. FREED

When Eqs. (B3) are substituted in Eq. (B2) and terms of proper order are retained, one obtains Eqs. (4.10)
and (4.11).

APPENDIX C: CORRELATION FUNCTIONS FOR MARKOV PROCESSES

AND ROTATIONAL DIFFUSION

A stationary Markov process yCt) is completely determined by specifying the a priori probability W(Y2)dY2 of
finding Y2 in the range (Y2, Y2+dY2) and the conditional probability P(YI I Y2, T) dY2 that given YI at an initial time,
one finds y in the range (Y2, Y2+dY2) at a time T later. One may then write an nth order joint probability density37:

n-I

=W(Yn) IIP(Yn-i+l I Yn-i, Tn--i) , (Cl)

i=l

where
11> 12' , ,> tn-I> In.
Then, for the nth-order time correlation of the random function Fq(y) :
n-I
(Fqj(/)Fq2(t-Tl)" 'Fq.(t- LTn»
~I

= f dYnW(Yn) Fqn(Yn) if-" f dYn-l" ,dY2dYlgP(Yn-i+l I Yn-" Tn-i) Fqn _1 , "F q2Fq l" (C2)

The integrals in Eq. (C2) may be changed to summations when the variable Yi takes on a discrete set of values.
The conditional probability function for rotational diffusion, p(nn I 0; T), (where 0 0 and 0 represent values
of Euler angles for a tumbling molecular axis with respect to a fixed laboratory axis system) is well known even
for anisotropic diffusion. For simplicity we assume axially symmetric diffusion, where38
P(Oo I 0; T) = L [(2L+l)/81r 2]:DKM(L)*(00):Dn/ L)(0) exp( -EL.KT) (C3)
L,K,M
and
(C4)
The :DK~ are the Wigner rotation matrices and
(CS)

where <Rl and <R2=<R3 the principal values of the rotational diffusion tensor. When <Ra-<Rl =0 the isotropic tumbling
result is obtained.
The perturbation XI(t) may be written as39
Xl(t) = L :D-m,m,(L) (fl)F'q(L,m)Aq(L,m'), (C6)
L,m,m',q

where both the F' q(L,m) and the Aq(L,-m) are irreducible tensor components of rank L and component m. The F'
in Eq. (C6) are expressed in molecule-fixed coordinates, while A is a spin operator quantized in space-fixed axes.
The D-m,m,(L) (0) terms include the transformation from space-fixed to molecule-fixed axes and are the random
functions of time. The needed correlation functions [Eqs. (3.7) and (C2) ] may be obtained by utilizing the rela-
tions40
(C7)

37 M. C. Wang and G. E. Uhlenbeck, Rev. Mod. Phys. 17,323 (1945).

38 J. H. Freed, J. Chern. Phys. 41, 2077 (1964).
3. Equation (C6) follows the notation utilized in Refs. 1 and 38. It differs from that of Edmonds~ and others where
3Ct (t) = ~ :D-m ...'CL)( _1)mF'CL,m)ACL.m')
L,m,m l

by incorporating (_I)m into the definition of F'CL,m).

40 A. R. Edmonds, AnguJar Momentum In Quantum Mechanics (Princeton University Press, Princeton, N.J., 1957).
 SPIN-RELAXATION THEORY 391

and
1a) , (C8)
m'a
where the terms in parenthesis in Eq. (C8) are the 3j symbols. al.4O Also
",I
,jJ m,m
,*= (_)m-m'",1 ,
cJJ -m,m . (C9)
We then obtain
gl(r) =0 (ClOa)
g2(llm'!q, 12m'2r; T) = (2lt+ 1)-IOll,i20m'2,-m'1~) - )m+m1'F'q(ll,m)F'/lt,-m) exp[ - (Eh,m)T] (C10b)
m

1a )
-ma

= . L,. ( ml
11
,
j)( 13 j ) (2j+l) (- )m-m'
1,m,m;1;><O m' m'a -m'

j)( 13
m -ma

X exp[ - (EI,,-m,)r3] exp[ - (E j ,m)T2] exp[ - (Eh,ml)Tl]. (ClOd)

Note that in Eqs. (ClO) , each subscript q has been expanded to include explicitly the particular irreducible tensor
component by the indices I and m. The restriction of the summation in Eq. (ClOd) to j~O comes about because
the second term in Eq. (3.7c) leads to the cancellation of thej=O component from the first term.
In applying these results, it is useful to employ the spherical tensor coupling expression40

(Cll)

It may be shown from this expression and the symmetry relation40

(Cl2)

that if Tall and Tb lt have only real components, or if a=b, then for I-odd Ta,b(l,m) =0.