Journal of Magnetic Resonance 149, 160–187 (2001)
doi:10.1006/jmre.2000.2239, available online at http://www.idealibrary.com on
ADVANCES IN MAGNETIC RESONANCE
Formal Theory of Spin–Lattice Relaxation
Maurice Goldman
CEA Saclay, DSM/DRECAM/Service de Physique de l’Etat Condensé, F-91191 Gif sur Yvette cedex, France
Received October 25, 2000
I. INTRODUCTION
Spin–lattice relaxation is the irreversible evolution of a spin
system toward thermal equilibrium with the orbital degrees of
freedom of the medium in which the spins are embedded, called
the lattice. This encompasses all spin variables amenable to ob-
servation: longitudinal or transverse magnetizations, spin–spin
energy or multiple-quantum coherences. It is only when the
evolution is due to static spin–spin interactions that we will
speak of spin–spin relaxation. Although it attracted attention
even before the existence of resonance, for electronic spin sys-
tems (1) it is mostly with the development of resonance meth-
ods that spin–lattice relaxation proved to be a central concept,
both as a fundamental topic in thermodynamics and as a tool
of prime importance for investigating the dynamical and struc-
tural properties of condensed matter. This is particularly so
in NMR. The basic understanding and the elaboration of the
main formalisms of nuclear spin–lattice relaxation were devel-
oped from 1948 to about 1960, in parallel with those of elec-
tronic spins. The basic references are Bloembergen, Purcell,
and Pound (BPP) (2), Wangsness and Bloch (3), and Bloch
(4), Solomon (5), Abragam (6), Redfield (7), Bloch (8), and
Tomita (9).
These works were described in detail by Abragam (6) and
Slichter (10). Many developments have and continue to take
place on various extensions and specific applications of spin–
lattice relaxation. A good introduction, with references, can be
found in the “Relaxation” articles of the Encyclopedia of Nuclear
Magnetic Resonance (11).
The purpose of this article is somewhat different. It is to
present ab initio a compact, clear, and complete description
of a general formalism of spin–lattice relaxation theory, with
two objectives: provide the reader with a well-defined proce-
dure for performing relaxation calculations of practical use-
fulness, together with a full and convincing justification of its
derivation. As a corollary, its purpose is not to give credit nor
to analyze in detail the various relaxation mechanisms, except
when needed for pedagogy, nor to provide an extensive bibli-
ography. Although of general applicability, the formalism will
1090-7807/01 $35.00
Copyright ° C 2001 by Academic Press
All rights of reproduction in any form reserved.
concern, by priority, systems where it is a good approximation
to treat the lattice classically and the spin–lattice coupling as
a random perturbation of the spin system: molecules in solu-
tion or insulating solids where the nuclei are relaxed by para-
magnetic impurities. Nothing specific will be said on systems
where a quantum description of the lattice is mandatory: metals,
semiconductors, superconductors, quantum solids, quadrupole
relaxation, etc., for which satisfactory full treatments can be
found in the literature. Nor will we give any qualitative de-
scription of relaxation, which is assumed to be known by the
reader. Such an introduction can be found, e.g., in Ref. (12),
Ch. 9.
The article is organized as follows. In Section II, we use a
classical description of the lattice, and we derive the master
equation for the density matrix and for the expectation value of
an observable, for a spin system subjected to a random pertur-
bation. The treatment is made in turn for a static Hamiltonian
and a time-dependent one, limited to off-resonance RF irradia-
tion. An account for the finite temperature of the lattice is made
phenomenologically. This part follows closely the procedure of
Ref. (13). It departs in several respects from earlier treatments;
in particular it does not use the adiabatic approximation, which
makes it possible to treat on the same footing heteronuclear and
nearly identical homonuclear spins. In Section III, the lattice is
treated quantum-mechanically, so as to take rigorously into ac-
count its finite temperature. For a static spin Hamiltonian, the
treatment is inspired from the description by Abragam (6) of
the Wangsness-Bloch approach (3, 4, 8). The case when the
spin Hamiltonian is time-dependent is treated in several steps:
RF irradiation with an effective field in the rotating frame suc-
cessively much smaller than and comparable with the static field,
and the general case of a time-dependent spin Hamiltonian.
This part does not quite follow the approaches of Bloch (8) or
Redfield (7).
Although an effort has been made toward clarity and sim-
plicity, some calculations are per force somewhat lengthy and
complicated. The last two sections contain such calculations,
and their study is not mandatory in a first reading. They are
indicated by an asterisk.
160
 ADVANCES IN MAGNETIC RESONANCE 161

II. CLASSICAL DESCRIPTION OF THE LATTICE Equation [5] then yields

A. Static Spin Hamiltonian d

σ̃ = −i[H̃1 (t), σ̃ ]. [8]
1. Derivation of the Master Equation dt
We consider a nuclear spin system whose Hamiltonian con-
The Hamiltonian H̃1 (t) has a double time dependence: that
sists of a main, time-independent Hamiltonian H0 , plus a spin–
due to its random character and that due to the passage to the
lattice coupling term H1 (t) of vanishing average value:
interaction representation.
By formal integration of Eq. [8], we obtain
H = H0 + H1 (t). [1]
Z t

We assume in a first step that H0 has only discrete levels, i.e., it is σ̃ (t) = σ̃ (0) − i [H̃1 (t 0 ), σ̃ (t 0 )] dt 0 , [9]
0
a Zeeman interaction in a liquid, so that the dipolar interactions
average out. a form which is inserted into the right-hand side of Eq. [8]. We
We call σ the spin density matrix. Its evolution is given by obtain
the Liouville–von Neumann equation:
Z
d σ̃ t
dσ = −i[H̃1 (t), σ̃ (0)] − [H̃1 (t), [H̃1 (t 0 ), σ̃ (t 0 )]] dt 0 . [10]
= −i[H, σ ]. [2] dt 0
dt
This expression is rigorous. For physical reasons, we subject
We begin by removing the static Hamiltonian H0 , so as to sin- it to two modifications.
gle out the effect of the perturbation H1 (t). This is done by the
use of an interaction representation which, for a purely Zeeman 1. We take an ensemble average of all terms. The reason
interaction, corresponds to the well-known passage to the rotat- is that remote parts of a large system relax independently of
ing frame. We replace all operators Q in the laboratory frame each other. Each of them experiences a random coupling with
by the operators a different evolution history, leading to a different local density
matrix σ̃ . These parts are simulated by different members of a
Gibbs ensemble. All have identical initial σ̃ (0) and, since H1 (t)
Q → Q̃(t) = exp(iH0 t)Q exp(−iH0 t). [3]
has a vanishing average, the first term on the right-hand side of
Eq. [10] vanishes.
The evolution of the density matrix in this representation is 2. In the remaining term on the right-hand side of Eq. [10],
given by we replace σ̃ (t) by

d d
σ̃ = {exp(iH0 t)σ exp(−iH0 t)} σ̃ (t) → σ̃ (t) − σ̃eq ,
dt dt
dσ where σ̃eq is the thermal equilibrium form of the density ma-
= iH0 σ̃ + exp(iH0 t) exp(−iH0 t) − i σ̃ H0 , [4]
dt trix for the Hamiltonian H0 . This is the phenomenological ac-
count for the finite lattice temperature, which will be justified in
that is, according to Eqs. [1] and [2], Section III.
We obtain in place of Eq. [10]
d
σ̃ = i[H0 , σ̃ ] − i exp(iH0 t)[(H0 + H1 ), σ ] exp(−iH0 t). [5] Z t
dt d
σ̃¯ (t) = − [H̃1 (t), [H̃1 (t 0 ), (σ̃ (t 0 ) − σ̃eq )]] dt 0 , [11]
dt 0
When U and U † are hermitian conjugate unitary operators (i.e.,
UU † = 1), we have in full generality where the ensemble average is noted by an overbar.
In the next step, we expand the spin–lattice Hamiltonian in
U [A, B]U † = [U AU † , U BU † ], [6] the form
X X
so that the last term on the right-hand side of Eq. [5] is equal to H1 (t) = Vα Fα (t) = Vα† Fα∗ (t), [12]
α α

exp(iH0 t)[(H0 + H1 ), σ ] exp(−iH0 t) = [(H0 + H̃1 (t)), σ̃ ].

where the Vα are spin operators and the Fα (t) are random func-
[7] tions of time. The equality of the two forms is a consequence
162 MAURICE GOLDMAN

of the fact that H1 , being a Hamiltonian, is hermitian: of the correlation time τc will be given later). We assume that
the quantity hQi we are interested in has a slow evolution on
†
H1 (t) = H1 (t). the time scale τc , and we choose t À τc in Eq. [17]. The ex-
pectation value hQi depends on given matrix elements of σ̃ (t),
The decomposition [12] is made in such a way that and although other physical variables may have a much faster
evolution than hQi, we state for the moment that the evolution of
[H0 , Vα ] = ωα Vα , [13]
σ̃ (t) is slow on the time scale τc . This will soon be justified. We
whence first derive the evolution equation for hQi(t) under the assump-
tion that it is slow, compared with τc−1 , and then we determine a
Ṽ α (t) = exp(iH0 t)Vα exp(−iH0 t) = exp(iωα t)Vα . [14] posteriori which are the conditions for the result to be consistent
with this assumption.
Such a decomposition is always possible. As an example, let The shortness of τc compared to the time scale of evolution
us choose of σ̃ and the choice of t À τc have three main consequences, as
we show next.
Vα = |iihi|H1 | jih j|, [15] The first consequence is that we may replace σ̃ (t 0 ) by σ̃ (t) on
the right-hand side of Eq. [17], since only values of t 0 differing
where |ii and | ji are eigenkets of H0 . Then from t by only a few times τc contribute to the integral. The
second consequence is that, since each member of the Gibbs en-
ωα = hi|H0 |ii − h j|H0 | ji. [16]
semble has experienced its random perturbation for many times
The operator Vα defined by Eq. [15] has only one nonvanishing τc , the effect of the difference in their random evolutions aver-
matrix element. In usual problems, the operators Vα have several ages out, so that the various σ̃ (t) are equal, and we may replace
nonvanishing matrix elements, and there are several possible σ̃¯ (t) by σ̃ (t). As a consequence, there is a decoupling between
choices for the Vα . the average over the spin part and that over the lattice part. Equa-
We can write Eq. [11] under the form tion [17] is then replaced by
X Z t
d d †
σ̃¯ (t) σ̃ (t) = − [Vα (t), [Vβ (t 0 ), (σ̃ (t)
dt dt α,β 0
XZ t †
=− [V˜ α (t), [Ṽβ (t 0 ), (σ̃ (t 0 ) − σ̃eq )]]Fα (t)Fβ∗ (t 0 ) dt 0 . − σ̃eq )]]Fα (t)Fβ∗ (t 0 ) dt 0 . [18]
α,β 0

[17] We limit ourselves to stationary random functions, a realistic

assumption in most practical cases, that is such that
Because of the different density matrices σ̃ (t 0 ) in different
members of the Gibbs ensemble experiencing different time evo- Fα (t)Fβ∗ (t 0 ) = G αβ (|t − t 0 |). [19]
lutions of the random functions Fα (t), Fβ∗ (t 0 ), it is essential to
use a joint average over the spin part (i.e., σ̃ (t 0 )), and the lattice Equation [18] is the master equation for the evolution of the
part (i.e., (Fα (t)Fβ∗ (t 0 ))) under the integral of Eq. [17]. This last density matrix σ̃ (t) in the interaction representation. We use it to
equation is nearly as rigorous as Eq. [10], but in full generality write down the master equation for σ (t) in the initial Schrödinger
it is well nigh insoluble. There is an exception, the case when representation. By inverting Eq. [3], we have
the evolution of H1 (t) corresponds merely to a variation of res-
onance frequency of the spins: modulation of the chemical shift
σ (t) = exp(−iH0 t)σ̃ (t) exp(iH0 t), [20]
or of the indirect interaction by chemical exchange or molecular
reorientations. In this case, the perturbations H1 (t) at different
times commute with each other as well as with the Hamiltonian whence, by differentiating both sides,
H0 . The theory appropriate to this case is specific and it will not
d d σ̃
be treated here (see, e.g., Refs. (13–15)). σ (t) = −i[H0 , σ (t)] + exp(−iH0 t) exp(iH0 t). [21]
For the general case when the random perturbations at dif- dt dt
ferent times do not commute, a clean theory of relaxation is From the expression [18] for d σ̃ /dt, together with the prop-
possible only in the limit when the fluctuation of the random erty [6] and the definition [19], we obtain
perturbation is fast compared with the evolution through relax-
X Z t
ation of the physical variables under study. More specifically, let d †
τc be the time scale for the fluctuation of the random functions σ (t) = −i[H0 , σ (t)] − [Vα , [Ṽβ (t 0 − t),
dt α,β 0
F(t): it is the time scale t − t 0 = τc over which a typical product
Fα (t)Fβ∗ (t 0 ) decays by a substantial amount (a better definition (σ (t) − σeq )]]G αβ (t − t 0 ) dt 0 . [22]
 ADVANCES IN MAGNETIC RESONANCE 163

The term under the integral depends on t 0 only through The double commutators [[Q, Vα ], Vβ ] will often yield
†

t − t 0 = τ , which extends from 0 to t. The third consequence operators differing from Q. Some of them may not correspond
of having t À τc is that we can extend the integral to infinity. to observable physical variables. However, since the observable
From Eq. [14], we have quantity Q depends on them, it is necessary to calculate their
evolution as well. The process must be repeated until reaching
† † †
Ṽβ (t 0 − t) = exp[−iωβ (t 0 − t)]Vβ = exp(iωβ τ )Vβ , [23] a closed system of operators Q j with coupled relaxation evolu-
tions. Some among them may correspond to independently mea-
and we obtain finally surable quantities. One then speaks of cross-relaxation proper,
although there is no fundamental difference between the two
d X † cases as regards the formal calculations.
σ (t) = −i[H0 , σ (t)] − [Vα , [Ṽβ , (σ (t)
dt α,β
Let us consider such a closed set, chosen in such a way that
we have
− σeq )]]Jαβ (ωβ ), [24]
[Q i , H0 ] = −Äi Q i . [30]
where Jαβ (ω), called a spectral density, is defined by
Equation [29] then yields a system of coupled equations of
Z ∞ the form
Jαβ (ω) = G αβ (τ ) exp(iωτ ) dτ. [25]
0 d X
hQ i i(t) = iÄi hQ i i(t) − λi j {hQ j i(t) − hQ j ieq }, [31]
dt j
Being Fourier transforms over positive time only, the spectral
densities are complex. It can be shown quite generally that their
whose solution is straightforward.
real part corresponds to relaxation proper, whereas their imagi-
A case often encountered is that when the difference in the
nary part produces a shift of the resonance frequencies, known
oscillation frequencies (Äi − Ä j ) of two operators Q i and Q j is
as the dynamical shift. We do not consider these shifts in the
much larger than their cross-relaxation rate λi j . It then results,
present article. A comprehensive analysis can be found in (16)
from the general theory of linear differential equations, that the
and references therein.
effect of cross-relaxation on the evolution of either hQ i i or hQ j i
is negligible: it yields a relative contribution of the order of
2. Evolution of Expectation Values: Cross-Relaxation
We are not so much interested in the evolution of the density λi j /|Äi − Ä j | ¿ 1.
matrix itself as in that of measurable spin variables. Let Q be
the operator corresponding to such a variable. We have For such couples of operators, one may then discard the cross-
relaxation terms from Eqs. [31]. This corresponds to the so-
hQi = Tr{Qσ } [26] called adiabatic approximation.
and
3. Evolution in the Interaction Representation
½ ¾
d dσ
hQi = Tr Q . [27] One is often led to calculate an expectation value in the inter-
dt dt action representation. Since most of the time it corresponds to a
We use the following general property: rotating frame, we note it with the subscript r ,

Tr{A[B, C]} = Tr{[A, B]C}. [28] hQir (t) = Tr{Q σ̃ (t)}, [32]

Applied twice in succession to Eq. [27], with dσ/dt given by and

Eq. [24], this yields ½ ¾
d d
d X † hQir (t) = Tr Q σ̃ . [33]
hQi = h−i[Q, H0 ]i(t) − Jαβ (ωβ ){h[[Q, Vα ], Vβ ](t)i dt dt
dt α,β

† On the other hand, by inserting the condition [13] into Eq. [18],
− h[[Q, Vα ], Vβ ]ieq }. [29] we have
X Z t
This is an equation relating expectation values, and there is d
no need to make assumptions as to the form of σ (t). It is the σ̃ (t) = − exp{iωα t} exp{−iωβ (t − τ )}G αβ (τ ) dτ
dt α,β 0
master equation for expectation values in the Schrödinger rep-
†
resentation, that is in the laboratory frame. × [Vα , [Vβ , (σ̃ (t) − σ̃eq )]]. [34]
164 MAURICE GOLDMAN

We extend the integral over τ to infinity and use the definition the frequencies ω I , ω S , ω I + ω S , and ω I − ω S . We write
[25], whence, from Eq. [33], ½
ω I = ω0 + δ
ω S = ω0 − δ [39]
d X
hQir (t) = − exp[i(ωα − ωβ )t]Jαβ (ωβ )
dt α,β with δ ¿ ω0 .
† †
× {h[[Q,Vα ],Vβ ]ir (t) − h[[Q,Vα ],Vβ ]ir eq }. [35] We assume that the correlation functions G(t) are exponential
with a single correlation time τc :
This derivative is a sum of smoothly varying terms, when
G(t) = G(0) exp(−t/τc ). [40]
ωα = ωβ , and of oscillatory terms. When the oscillatory fre-
quencies ωα − ωβ are large compared with the average decay
rate of hQir (t), their contributions are fast oscillations of small It is only with correlation functions of this form that the
amplitude, and they can be ignored. This was the first formula- correlation time τc has a well-defined meaning. The spectral
tion of the adiabatic approximation. When on the other hand the densities (real part) are Lorentzian:
oscillations are not fast, the present treatment is ill adapted to τc
quantifying their effect. By contrast, this is easily done by cal- J (ω) = G(0) . [41]
1 + ω2 τc2
culating the derivatives of expectation values in the Schrödinger
representation, that is in the laboratory frame, as done in We have then
Section IIA2. This is in fact the main advantage of using the
laboratory frame picture. τc
J (ω I ) = G(0)
The simplest way of establishing the connection between 1+ ω0 τ (1 +
2 2
δ/ω0 )2
both formulations is as follows. Let us consider the same set of τc
operators Qi as in Section IIA2. We have ' G(0) . [42]
1 + ω02 τc2

hQ i ir (t) = Tr{Q i σ̃ (t)} Likewise,

= Tr{Q i exp(iH0 t)σ (t) exp(−iH0 t)} τc

J (ω S ) = G(0)
= Tr{[exp(−iH0 t)Q i exp(iH0 t)]σ (t)}, [36] 1 + ω02 τc2 (1 − δ/ω0 )2
τc
' G(0) , [43]
1 + ω02 τc2
or else, according to Eq. [30],
τc
J (ω I − ω S ) = G(0) . [44]
1 + 4δ 2 τc2
hQ i ir (t) = exp(−iÄi t)hQ i i(t). [37]
We assume that δτc ¿ 1, so that
If we have hQ j ieq = 0, Eq. [31] then yields
J (ω I − ω S ) ' J (0), [45]
d X
hQ i ir (t) = − exp[−i(Äi − Ä j )t]λi j hQ j ir (t). [38] a situation very common for homonuclear spins in liquids.
dt j As a consequence of Eqs. [42]–[45], the relaxation is the same
as if both spins had resonance frequencies equal to ω0 .
We see the origin of the oscillatory terms in Eq. [35]: they cor- Instead of using the full interaction representation, defined by
respond to cross-relaxation between operators having different the operator
resonance frequencies in the steady spin Hamiltonian H0 .
U = exp[i(ω I Iz + ω S Sz )t], [46]
4. Intermediate Representation
we could have used a frame rotating at frequency ω0 for both
There are cases when it is convenient to use a representation spins, defined by the operator
intermediate between the Schrödinger and the interaction repre-
sentations. This will be illustrated for a system of homonuclear U 0 = exp[iω0 (Iz + Sz )t]. [47]
spins I and S in a liquid, whose resonance frequencies ω I and ω S
are close, as a result of different chemical shifts. Let us consider If we write
the contribution to their relaxation from the random modulation
of their dipolar interaction (6). It involves spectral densities at Q̃ 0 = U 0 QU 0† [48]
 ADVANCES IN MAGNETIC RESONANCE 165

in this representation, we have fields, with ωβ = ω0 or 2ω0 . We have then

d 0 1 (1ω)2 τc
σ̃ (t) = −i[δ(Iz − Sz ), σ̃ 0 (t)] − i[H̃10 (t), σ̃ 0 (t)]. [49] ∼ . [55]
dt T1 1 + ω02 τc2
The second term is processed as in Section IIA1, and it yields The requirement becomes in that case
the correct result.
If on the other hand the condition δτc ¿ 1 is not fulfilled, τc (1ωτc )2
one must use the full interaction representation, which leads to ∼ ¿ 1. [56]
T1 1 + ω02 τc2
Eq. [34] for d σ̃ /dt. Then it is permissible to limit oneself to the
intermediate representation, defined by It is fulfilled either by condition [54] or by the usually less
stringent condition,
σint (t) = exp[−iδ(Iz − Sz )t]σ̃ (t) exp[iδ(Iz − Sz )t]. [50]
1ω
The evolution equation in this representation is ¿ 1, [57]
ω0

d in the limit when ω0 τc À 1.

σint (t) = −i[δ(Iz − Sz ), σint (t)] + exp[−iδ(Iz − Sz )t]
dt
d 6. Extension to Solids
× σ̃ (t) exp[iδ(Iz − Sz )t]. [51]
dt In NMR, a solid is characterized by the existence of static
spin–spin interactions, usually much smaller than interactions
If we use the approximations [42] and [43], the only oscillating
giving rise to discrete levels: Zeeman or quadrupolar or a com-
terms are with frequencies ω0 and 2ω0 . Since in all resonance
bination of both. The spin–spin interactions, mostly dipolar,
experiments the frequency ω0 is much larger than all relaxation
limited to their secular part, produce a quasi-continuous broad-
rates, we can discard the oscillating terms as a result of the adi-
ening of the otherwise discrete levels. The decay of the transverse
abatic approximation. Equation [51] describes correctly trans-
magnetization is then governed by the spin–spin interactions and
verse cross-relaxation between spins I and S in the presence of
is nonexponential. Spin–lattice relaxation in that case concerns
a differential precession.
other variables such as the longitudinal magnetization or the
The intermediate representation is also useful in other cases:
quadrupole alignment or the dipolar energy.
indirect interactions or RF irradiation, as seen in Section IIB.
We take as an example a Hamiltonian H0 consisting of
Zeeman and secular dipolar interactions, of the form
5. Conditions of Validity of the Theory
As seen above, the theory assumes that the relaxation times H0 = Z + HD0 . [58]
under consideration are much longer than the correlation time
τc . We consider as an example transverse and longitudinal relax- The random perturbation can still be written under the form
ations in a liquid, due to dipolar couplings (see, e.g., Ref. (6)). [12], but with the Vα chosen so as to have
We call (1ω)2 the average square of the dipolar interactions and
ω0 the Larmor frequency. [Z , Vα ] = ωα Vα [59]
Transverse relaxation can be produced by the longitudinal
components of the dipolar field, for which ωβ = 0. Then we in place of [13].
have as an order of magnitude Since the terms Z and HD0 are commuting, we have from
Eq. [3]
1
∼ (1ω)2 τc . [52]
T2 Ṽα (t) = exp(iH0 t)Vα exp(−iH0 t)
= exp(iHD0 t) exp(i Z t)Vα exp(−i Z t) exp(−iHD0 t)
The condition of validity of the theory is for that case
= exp(iωα t)Vα (t), [60]
τc /T2 ¿ 1, [53]
where Vα (t) is defined by
that is,
Vα (t) = exp(iHD0 t)Vα exp(−iHD0 t). [61]
|1ω|τc ¿ 1. [54]
When calculating the evolution of a variable hQ i i, we have
Longitudinal relaxation involves only transverse dipolar by analogy with Eqs. [25] and [28] to calculate integrals of the
166 MAURICE GOLDMAN

form Under these conditions, the calculation of relaxation in a solid

Z ∞ is exactly the same as in a liquid.
†
h[[Q i , Vα ], Vβ (−τ )]i(t) exp(iωβ τ )G αβ (τ ) dτ.
0
7. Synthesis and Discussion
For an exponential correlation function of the form [40], the The derivation leading to the master equation [29] follows a
integral is of the form well-defined succession of steps, which we first recapitulate.
Z ∞
I= K (τ ; t) exp[(iωβ − 1/τc )τ ] dτ [62] 1. Start with the Liouville–von Neumann equation [2] for the
0 evolution of the density matrix.
The variation of K with τ is due to the dipolar interactions 2. Go over to an interaction representation so as to single out
and is therefore complicated. However, the integral [62] has a the effect of the random spin–lattice coupling.
simple solution if the decay of K is slow on the time scale 3. Perform a formal integration of σ and reinject into the
|iωβ − 1/τc |−1 : it is the same as if the dipolar term HD0 were evolution equation, so as to have a two-time product of random
absent, that is, functions.
4. Take an ensemble average jointly over spin and lattice vari-
τc ables.
I ' K (0; t) . [63]
1 − iωβ τc 5. Replace σ̃ by σ̃ − σ̃eq , in a phenomenological treatment of
the finite lattice temperature.
The reason lies in a general characteristics of functions whose 6. Expand the spin–lattice coupling into a sum of products of
decay is due to dipolar interactions, namely that their Fourier spin operators by random lattice functions.
transforms have fast decaying wings, approximately like a 7. Choose the time t much longer than the correlation time τc
Gaussian function. The justification goes as follows. Let the of the random functions. This has three consequences:
Fourier transforms (over positive time) of K (τ ; t) and G(τ ) be • We can replace σ̃ (t 0 ) by σ̃ (t);
K(ω) and • The average of the products spin × lattice decouples into
1 the products of the averages spin × lattice. The latter are corre-
G(ω) = . [64] lation functions of random lattice functions;
1/τc − iω • We can extend the integration limit to infinity, which
The Fourier transform I of the product K (τ ; t)G(τ ) (Eq. [62]) introduces well-defined spectral densities.
is equal to the convolution of their Fourier transforms: 8. Go back to the Schrödinger representation and write the
evolution equations for the expectation value of variables. They
Z +∞
have the following characteristics:
I(ωβ ) ∝ K(ω)G(ωβ − ω) dω • The derivative of an expectation value is a sum of two
−∞
Z +∞
terms: the contribution of the static spin Hamiltonian and that
= K(ω)/[1/τc − i(ωβ − ω)] dω of spin–lattice relaxation;
−∞ • It depends on expectation values, not on the detailed form
Z +∞ of the density matrix;
1 K(ω) dω
= . [65] • The relaxation part contains no oscillatory terms;
1/τc − iωβ −∞ 1 + iω/[1/τc − iωβ ]
• Cross-relaxation shows up without ambiguity.
It is equivalent to say that K (τ ; t) evolves slowly over the time This procedure avoids, explicitly or implicitly, a number of
scale |1/τc − iωβ |−1 , or to say that the width of K(ω) is small traps encountered when performing relaxation calculations. We
compared with |1/τc −iωβ |. As a consequence, only those values come back briefly on some of them.
of ω for which
(a) Use of an interaction representation that does not remove
|ω| ¿ |1/τc − iωβ | all of H0 . It is a representation defined by

contribute to the integral. For these values, the denominator does Q → Q̃ 0 = exp(iHt)Q exp(−iHt)
not differ significantly from unity and the integral is approxi-
mately equal to
with H 6= H0 .
Z +∞ The evolution of the density matrix in this representation is
K(ω) dω ∝ K (0; t), [66]
−∞
d 0
σ̃ (t) − i[(H̃00 − H + H̃10 (t)), σ̃ 0 (t)], [67]
which justifies statement [63]. dt
 ADVANCES IN MAGNETIC RESONANCE 167

i.e., it involves a nonrandom term (H̃00 − H) in addition to the which, through repeated use of property [28], is equal to
random one. This procedure does not allow a clean derivation
† †
of a master equation where the effect of relaxation is properly Tr{Q[Vα , [Vβ , σ (t)]]} = Tr{[[Q, Vα ], Vβ ]σ (t)}
†
accounted for, because the time dependence of Ṽβ (−τ ) is not
†
correct and leads to spectral densities with the wrong frequency. = h[[Q, Vα ], Vβ ]i(t). [71]
However, we have seen in Section IIA4 that it may nevertheless
yield the correct result when the extra frequency corresponding By using [28] only once, it is found that this trace is also equal
to the residual Hamiltonian is much smaller than the inverse to
correlation time. In case of doubt, it is safer to go first to the full
interaction representation and then go back to an intermediate † †
Tr{Q[Vα , [Vβ , σ (t)]]} = Tr{[Q, Vα ][Vβ , σ (t)]}, [72]
representation if it is more convenient than the laboratory frame.
Among more serious wrongdoings is the fact of treating the
effect of the residual term (H̃00 − H) perturbatively, or even i.e., the trace of a product of single commutators. The latter are
worse, to use a term H which does not commute with H0 , so (slightly) simpler to calculate than double commutators, which
that (H̃00 − H) is time dependent. seems to be an advantage. However, Eq. [72] is not equal to the
(b) Make the premature assumption that the average over spin expectation value of an operator, and in order to calculate it, it
and lattice variables is the product of the independent averages. is necessary to guess what the form of σ (t) is. This is usually
As an example, let us consider the derivative of hQir (t) in done by performing a partial decomposition of σ (t) into a set of
the interaction representation. According to Eq. [17], it involves orthogonal operators, limited to those expected to be relevant.
terms of the form We write then
Z t X
† σ (t) = ξi Q i + P, [73]
exp[i(ωα − ωβ )t] Tr{[[Q, Vα ], Vβ ]σ̃ (t 0 )} i
0

× exp[iωβ (t − t 0 )]G(t − t 0 ) dt 0 . with

The integral is of the form Tr(Q i Q j ) = Tr(Q i P) = 0. [74]
Z t
I(t) = C(t 0 )D(t − t 0 ) dt 0 . [68] Only the operators Q i are supposed to intervene in relaxation
0 and are explicited. The remaining term P is supposed not to play
any role.
It is easily solved by the Laplace transform method. Let I(Z), The coefficients ξi are obtained from
C(Z), and D(Z), the Laplace transforms of I (t), C(t), and D(t),
respectively, i.e., for instance, ¡ ¢
hQ i i = Tr{σ (t)Q i } = ξi Tr Q i2 . [75]
Z ∞
I(Z) = I (t) exp(−Zt) dt. [69] The right-hand side of Eq. [72] then reads
0

†
X †
It is a well-known property of Laplace transforms that to Eq. [68] Tr{[Q, Vα ][Vβ , σ (t)]} = ξi Tr{[Q, Vα ][Vβ , Q i ]}. [76]
there corresponds i

I(Z) = C(Z)D(Z),
which would mean that hQir (t) could be calculated rigorously at
any time t, and even in the case when the relaxation rates are not
small compared with τc−1 . This conclusion is totally erroneous,
as results from the formal treatment detailed above. This provide
a proof ad absurdum of the necessity of using a joint average
over spins and lattice at short times.
(c) Make guesses about the form of the density matrix in the
course of relaxation.
We have seen in Section IIA2 that the derivative of hQi
depends on terms of the form
†
Tr{Q[Vα , [Vβ , σ (t)]]},
[70] This procedure is all right provided that the choice of the Q i
is a good one. Some of its triumphs in the past have been:
• to miss cross-relaxation (5, 17),
• to miss cross-correlation effects between dipolar inter-
actions of adjacent spin pairs (18) or between dipolar and
anisotropic nuclear shielding interactions (19),
• to miss the fact that, for spins larger than 1/2, the relaxation
of linear spin components may depend on higher powers of the
spin operators (13).
The advantage of calculating the double commutators in
Eq. [29] is that the operators on which the variation of hQ i i
depends show up naturally. There is no need to make “intelli-
gent” guesses.
168 MAURICE GOLDMAN

B. Time-Dependent Spin Hamiltonian: Off-Resonance We consider the case when Ä is much larger than the relaxation
RF Irradiation rates, or the spin–spin resonance width in solids.
As regards the spin–lattice coupling, we write it in a form
The case of an arbitrary time-dependent spin Hamiltonian is
adapted to the specific nature of the spin Hamiltonian:
complicated. It will be discussed at the end of Section III. We
X
limit ourselves to the practically only case of interest: that of a H1 (t) = Vm Fm (t), [83]
system in a static magnetic field subjected to an irradiation with m
a much smaller periodic field of frequency ω in the vicinity of
the Larmor frequency ω0 corresponding to the static field. Any with
component of the RF field parallel to the dc field can be ignored,
since its effect is negligible. Irradiation is usually made with a [Iz , Vm ] = mVm , [84]
linearly polarized RF field, which can be decomposed into a sum
of circularly polarized fields rotating in opposite directions, that whence, according to Eq. [78],
is with opposite frequencies +ω and −ω. The only effective one X
H̃1 (t) = exp(imωt)Vm Fm (t). [85]
is that rotating at the frequency close to ω0 , say +ω, and we m
discard the other one. The spin Hamiltonian is then a Zeeman
interaction with a static part plus a much smaller normal part (ii) We go over to a doubly rotating frame by the transforma-
rotating at frequency ω: tion

H0 = ω0 Iz + ω1 (Ix cos ωt + I y sin ωt) Q̃(t) → Q̃˜ (t) = exp(iÄI Z t) Q̃ exp(−iÄI Z t), [86]
= ω0 Iz + ω1 exp(−iωIz t)Ix exp(iωIz t)
in which representation the evolution of the density matrix is
= exp(−iωIz t)(ω0 Iz + ω1 Ix ) exp(iωIz t), [77] given by

to which we add the spin–lattice coupling H1 (t). The first step d ˜ (t), σ̃˜ (t)].
σ̃˜ (t) = −i[H̃ 1 [87]
of the formal treatment, passage to an interaction representation dt
which singles out the effect of the spin–lattice coupling, is per-
formed by two successive unitary transformations, as follows. In order to write H̃˜ (t), we express the operators V in a form
1 m
adapted to the tilted axes pertaining to the doubly rotating frame,
(i) Each operator is replaced by
X
Q → Q̃(t) = exp(iωIz t)Q exp(−iωIz t), [78] Vm = λmp V p0 , [88]
m

which corresponds to the passage to a frame rotating at frequency with

ω around the direction 0z of the static field.
Through the same treatment as from Eq. [4] to Eq. [8], the [I Z , V p0 ] = pV p0 . [89]
evolution of the density matrix in this representation is given by
We then obtain from Eqs. [85] and [86]
d
σ̃ (t) = −i[(H̃0 − ωIz + H̃1 (t)), σ̃ (t)]. [79] X
dt ˜ (t) =
H̃ Fm (t)λmp V p0 exp[i(mω + pÄ)t]. [90]
1
The new spin Hamiltonian is m, p

Heff = H̃0 − ωIz = (ω0 − ω)Iz + ω1 Ix
= 1Iz + ω1 Ix = ÄIz , [80]
where 1 is equal to the resonance offset. The effective
Hamiltonian in this rotating frame is a Zeeman interaction with
a static field oriented along an axis 0Z in the 0x z plane at an
angle 2 from the direction 0z of the static field, with
ω1
tan 2 = . [81]
1 hQir (t) = h−iÄ[Q, I Z ]i(t) −
The corresponding Larmor frequency is
¡ ¢1/2
Ä = 12 + ω12 . [82]
To proceed, we solve Eq. [87] exactly as in Section IIA2, for
the case of a static spin Hamiltonian, and then we go back to the
first rotating frame. This is a convenient intermediate representa-
tion, in which the effective Zeeman interaction is static. Possible
periodic terms as a function of t will have frequencies equal to
multiples of ω, and they may safely be discarded. The evolution
equation for hQir (t) is of a form slightly more complicated than
Eq. [35], but otherwise very similar:
d X ∗
λmp λqm Jm,m (mω + qÄ)
dt m, p,q
× {h[[Q, V p0 ], Vq0† ]ir (t) − h[[Q, V p0 ], Vq0† ]ireq }.
[91]
 ADVANCES IN MAGNETIC RESONANCE 169

The only question is about which σ̃eq has to be used in order that of the longitudinal one, with terms proportional to J (ω0 ),
to account for the finite lattice temperature. This is discussed at J (Ä), and J (0).
length in Section III. The answer is that it is not the same for all The only effective time-dependent dipolar interaction be-
terms in the decomposition of the spin–lattice coupling H1 (t). tween an electronic spin S and a nuclear spin I at fixed positions
In the case under study, where the effective frequency Ä in the is equal to
rotating frame is much smaller than the RF frequency ω, the
answer, intuitively correct, is the following:
H1 (t) = Sz (t){AIz + B I+ + B ∗ I− }. [97]
• For the terms Vm which do not commute with Iz , it is the
same as in the absence of RF irradiation, We assume the lattice temperature high enough for the equilib-
rium value hSz ieq to be negligibly small, with
σeq = 1 − βL ω0 Iz , [92]
( γ I γ S h-
where βL is the inverse lattice temperature. A= r3
(1 − 3 cos2 θ )
[98]
• For all other terms, it is equal to unity and yields no con- -
B= − γ Irγ3S h × 32 cos θ sin θ exp(−iϕ),
tribution. Furthermore, for m 6= 0, we have approximately

Jm,m (mω + pÄ) ' Jm,m (mω), [93] where r is their mutual distance, and θ and ϕ are the polar angles
of r in a frame where 0z is along the static part of the field (see,
a result similar to [42] and [43]. Combining these results, the e.g., Ref. (6)).
relaxation due to the terms of H1 with m 6= 0 is exactly the same In the first reference frame, this coupling is
as in the absence of RF irradiation, and that due to the other ones
is toward zero. H̃1 (t) = Sz (t){AIz + B exp(iωt)I+ + B ∗ exp(−iωt)I− }. [99]
For actual calculations, it is also possible to express the V p0 as
a function of the Vm . We give below two examples of relaxation As stated above, the contribution of the terms in I± is calculated
under RF irradiation. They are intended to illustrate the formal- as in the absence of irradiation, that is according to Eq. [35] and
ism for actual physical cases. The complexity of the formulas [92]. The result is
should not conceal the formal simplicity of the calculations.
d S(S + 1)
1. Local Nuclear Relaxation by a Fixed Paramagnetic Center hIz i B = − · 4B B ∗ J (ω)(hIz i − I0 ), [100]
dt 3
In insulating solids, the nuclear relaxation of spins 1/2 is due
to the random modulation of their dipolar coupling with fixed where I0 is equal to hIz ieq for σeq of the form [92]. Here, the
paramagnetic impurities at low concentration. The modulation normalization of J (ω) corresponds to G(0) = 1 in Eq. [40].
is that of the expectation value of the electronic spin compo-
nents under the effect either of their own spin–lattice relaxation d S(S + 1)
or through flip–flop processes among electronic spins. The cor- hI+ i B = − · 2B B ∗ J (ω)hI+ i, [101]
dt 3
responding correlation time is in general not short compared
with the nuclear Larmor period,
which can be written
ω 0 τc >
∼ 1. [94]
d S(S + 1)
hIx,y i B = − · 2B B ∗ J (ω)hIx,y i. [102]
Since the electronic Larmor frequency ωe is about 3 orders of dt 3
magnitude larger than ω0 , we have
This contribution to relaxation along the axes 0X Y Z of the
tilted frame is obtained from the relations
ωe τc À 1. [95]

Iz = cI Z − s I X

We have then for the spectral densities Ix = s I Z + cI X [103]

I = I
y Y
J (ωe ) ¿ J (ω0 ) <
∼ J (Ä) <
∼ J (0) [96] 
I Z = cIz + s I x

and the contribution to relaxation of the transverse electron spin I X = −s Iz + cI x [104]

I = I ,
components, proportional to J (ωe ), is negligible compared with Y y
170 MAURICE GOLDMAN

with c = cos 2 and s = sin 2. We have d S(S + 1) 2 2

hIY i A = − A [c J (0) + s 2 J (Ä)]hIY i. [113]
d d d dt 3
hIz i B = c hIz i + s hIx i
dt dt dt We add both contributions to relaxation, as well as the
S(S + 1) evolution under the effective spin Hamiltonian [80], and we ob-
=− · 2B B ∗ J (ω){2chIz i − 2cI0 + shI x i}
3 tain finally
S(S + 1)
=− · 2B B ∗ J (ω) d S(S + 1) 2 2
3 hI Z i = − {[s A J (Ä)
dt 3
× {2c2 + s 2 )hI Z i − cshI X i − 2cI0 }, [105]
+ 2B B ∗ (2c2 + s 2 )J (ω)]hI Z i − cs[A2 J (0)
and likewise, + 2B B ∗ J (ω)]hI X i − 4cB B ∗ J (ω)I0 } [114]
d d d d S(S + 1) 2 2
hI X i B = −s hIz i + c hIx i hI X i = −ÄhIY i − {[c A J (0)
dt dt dt dt 3
S(S + 1) + 2B B ∗ (2s 2 + c2 )J (ω)]hI X i
=− · 2B B ∗ J (ω){−2shIz i + 2s I0 + chIx i}
3
+ cs[A2 J (Ä) − 2B B ∗ J (ω)]hI Z i
S(S + 1)
=− · 2B B ∗ J (ω){(2s 2 + c2 )hI X i + 4s B B ∗ J (ω)I0 } [115]
3
− cshI Z i + 2s I0 } [106] d S(S + 1) 2 2
hIY i = ÄhI X i − [c A J (0) + s 2 A2 J (Ä)
d S(S + 1) dt 3
hIY i B = − · 2B B ∗ J (ω)hIY i. [107] + 2B B ∗ J (ω)]hIY i. [116]
dt 3

For calculating the contribution of the term in Iz , we write the Since we assume that the effective frequency Ä is much larger
first Eq. [103] under the form than the relaxation rates, we can neglect the cross-relaxation
s between hI X i and hI Z i.
Iz = cI Z − (I+0 + I−0 ). [108] Let us calculate within this approximation the expectation
2
values of the various spin components in the steady state.
We have in the doubly rotating frame Equation [114] yields
s c
I˜ z = cI Z − [I+0 exp(iÄt) + I−0 exp(−iÄt)]. [109] hI Z iss = I0 , [117]
2 c2 + K s 2
Since Iz = I˜z , the calculation of relaxation in the first rotating with
frame is similar to that in the laboratory frame with a static spin
Hamiltonian (Eq. [29]), but with hQieq = 0. We obtain after a 1 A2 J (Ä)
K = + . [118]
trivial calculation 2 4B B ∗ J (ω)
·
d S(S + 1) 2 2 Equation [116] yields
hI Z i A = − A s J (Ä)hI Z i
dt 3
¸ S(S + 1) 2 2
1 hI X iss = [c A J (0) + s 2 A2 J (Ä)
0 0 3Ä
− cs J (0)(hI+ i + hI− i)
2 + 2B B ∗ J (ω)]hIY iss , [119]
S(S + 1) 2 2
=− A [s J (Ä)hI Z i − cs J (0)hI X i] [110]
3 that is,
·
d 0 S(S + 1) 2 2 1 hI X iss ¿ hIY iss .
hI+ i A = − A c J (0)hI+0 i + s 2 J (Ä)(hI+0 i [120]
dt 3 2
¸ Therefore, we neglect hI X iss in Eq. [115] and we replace hI Z iss
− hI−0 i) + cs J (Ä)hI Z i , [111] by the value [117]. We obtain
½
S(S + 1) c2 s[A2 J (Ä) − 2B B ∗ J (ω)]
or else, by separating the real and imaginary parts, hIY iss = −
3Ä c2 + K s 2
¾
d S(S + 1) 2 2
hI X i A = − A [c J (0)hI X i + cs J (Ä)hI Z i] [112] + 4s B B ∗ J (ω) I0 , [121]
dt 3
 ADVANCES IN MAGNETIC RESONANCE 171

that is, malized functions Fm through

 2-
q
hIY iss ¿ hI Z iss , [122] 
A
 = − γr3h × 2
F
3 0



B γ 2 h-
= × 2√1 6 F0 = − 14 A
except for c ¿ 1, where it can be checked that all components r3
[127]

 γ 2 h-
are vanishingly small. 
 C = × 12 F1∗


r3
The physically evident result is that in a large effective field  2-

the steady-state spin orientation is locked along this effective E = − γr3h × 12 F2∗ .
field.
The Fm , of vanishing average value, obey the following
Remark. In the case when Äτc ¿ 1, that is J (Ä) ' J (0), relations:
the relaxation is the same as in the absence of RF irradiation.
6
The corresponding equations are Fm Fm∗ 0 = δm,m 0 . [128]
5

 d
hI i = − T11 (hIz i − I0 ) They are themselves proportional to second-order spherical
dt z rel
[123] harmonics of the polar angles of rIS (see, e.g., (12), (13), (20)).
 d
hI i = − T12 hIx i, Since we content ourselves with a formal treatment, the
dt x rel
alphabet notation is simpler to use and there is no need to detail
whence, according to Eqs. [103] and [104], its specific form. As above, we assume that

µ 2 ¶µ ¶ J (ω ± Ä) ' J (ω).
d c s2 c
hI Z i = − + hIz i − 2 I0 . [124]
dt T1 T2 c + (T1 /T2 )s 2 Furthermore, we assume δ large enough to allow the separate
observation of the spins I and S, but much smaller than the mean
The result turns out to be exactly the same as Eq. [114], ex- effective frequency Ä in the rotating frame (Eq. [82] as well as
cept for J (0) in place of J (Ä). The indiscriminate replacement the correlations rate τc−1 .
of J (0) by J (Ä) under RF irradiation would, however, yield The effective Hamiltonian in the rotating frame (first interac-
erroneous results, since from Eqs. [115] and [116] the deriva- tion representation) is
tives of hI X i and hIY i depend on both spectral densities. It is
H̃(t) = Ä(IZ + SZ ) + δ(Iz − Sz ) + H̃1 (t), [129]
only through the complete calculation given above that one can
obtain the right answer. where I and S have parallel axes Z , defined by Eq. [81]. Since
δ ¿ Ä, we can project Iz and Sz onto I Z and S Z . The effective
2. Dipolar Relaxation of a Homonuclear Spin Pair spin Hamiltonian in this representation is then
in a Liquid
Heff = (Ä + δc)I Z + (Ä − δc)S Z . [130]
We consider two homonuclear spins I and S of resonance
frequencies given by Eq. [39]: As for the spin–lattice coupling H̃1 , it is
½ H̃1 (t) = A(t)Iz Sz + B(t)(I+ S− + I− S+ ) + C(t) exp(iωt)
ω I = ω0 + δ
[125]
ω S = ω0 − δ. × (I+ Sz + Iz S+ ) + c.c. + E(t) exp(i2ωt)I+ S+ + c.c.
[131]
They are subjected to RF excitation at an offset 1 from ω0 (1 = In complete analogy with the preceding section, the relaxation
ω0 − ω), and their relaxation is due to the modulation of their due to terms C and E of Eq. [126] is the same as in the absence of
dipolar interaction by Brownian rotation at constant distance, irradiation. Through the use of Eq. [35], one finds (see, e.g., (6),
with a correlation time τc . The spin–lattice coupling is of the (10), and (12)),
form (6, 12)
d
H1 (t) = A(t)Iz Sz + B(t)(I+ S− + I− S+ ) + C(t)(I+ Sz + Iz S+ ) hIz iC,E = −CC ∗ J (ω){hIz i − I0 } − 2E E ∗ J (2ω)
dt
+ c.c. + E(t)I+ S+ + c.c. [126] × {hIz i + hSz i − 2I0 } [132]
½ ¾
d 3
hIx,y iC,E = −CC ∗ J (ω) hIx,y i + hSx,y i
The notation for the orbital parts corresponds to the Van Vleck dt 2
notation (see, e.g., Ref. (6), Ch. IV). They are related to nor- − E E ∗ J (2ω)hIx,y i [133]
172 MAURICE GOLDMAN

and similar expressions for the spin S, by interchanging the hQieq = 0,

letters I and S. ½·
We can use Eqs. [103] and [104], and similar ones for the d 1
hI Z i A,B = −A2 (3c2 − 1)2 J (0)
spins S, to express the corresponding relaxation rates along the dt 32
axes 0X Y Z . We write explicitely the evolution along 0Z only. ¸
9 2 2 9 4
Through a straightforward calculation, we obtain the following + c s J (Ä) + s J (2Ä) hI Z i
results 16 32
· ¸ ¾
½µ ¶ 1 9
d 3 − (3c2 − 1)2 J (0) − s 4 J (2Ä) hS Z i ,
hI Z iC,E = − c2 + s 2 CC ∗ J (ω) 32 32
dt 2 [138]
¾
+ (2c2 + s 2 )E E ∗ J (2ω) hI Z i and a similar expression for the derivative of hS Z i A,B . We can add
these contributions to those of C, E (Eq. [134]). The equations
− {s 2 CC ∗ J (ω) + 2c2 E E ∗ J (2ω)}hS Z i of evolution of hI X,Y i and hS X,Y i turn out to be much more
+ {cCC ∗ J (ω) + 4s E E ∗ J (2ω)}I0 . [134] complicated.

We have discarded the cross-relaxation terms between hI Z i and C. Alternative Approaches to Relaxation
hI X i. We cite only two of them, which are particularly known and
The evolution equations for the S spin components are obtain of wide use. They concern essentially the case when the spin
by permutation. Hamiltonian is time-independent.
Let us now consider the contribution from the terms in A and
B in Eq. [126]. We rewrite them as 1. Relaxation Matrix of the Density Matrix
½ This approach consists of writing differential equations for the
H1,AB = H̃1,AB = A(t) (cI Z − s I X )(cS Z − s S X ) various matrix elements of the density matrix as a function of
other matrix elements. In the preceding formalism, it amounts to
¾ choosing for the Qi the projections |βihα|, where |αi and |βi are
1
− [(s I Z + cI X )(s S Z + cS X ) + IY SY ] , [135] basis kets of the spin Hilbert space, most of the time eigenkets
2
of the static spin Hamiltonian H0 . Their expectation values are
or else, by using the operators I±0 and S±0 ,
h|βihα|i = Tr(|βihα|σ ) = hβ|βihα|αihα|σ |βi
½ · ¸ = hα|σ |βi = σαβ . [139]
1 2 1
H̃1,AB (t) = A(t) (3c − 1) I Z S Z − (I+0 S−0 + I−0 S+0 )
2 4 The differential system is of the form
3 ¯
− cs(I Z S+0 + S Z I+0 + I Z S−0 + S Z I−0 ) d ¯ X
4
¾ σαβ ¯¯ = Rαβ,γ δ σγ δ . [140]
3 dt rel αδ
+ s 2 (I+0 S+0 + I−0 S−0 ) . [136]
8
The matrix Rαβ,γ δ is called the relaxation matrix. Each of
We neglect δc in Eq. [130] for going to the doubly rotating frame, its indexes corresponds to the labels of two kets of the Hilbert
and we obtain space. If n is the number of dimensions of the Hilbert space,
the number of elements of σ is n 2 and that of the matrix R is
½ · ¸ n 4 , although many of them may vanish. Equation [140] is gen-
˜ (t) = A(t) 1 (3c2 − 1) I S − 1 (I 0 S 0 + I 0 S 0 ) − 3
H̃ 1
2
Z Z
4 + − − +
4 eral, rigorous, and compact. As such, it is often very convenient
for stating general relaxation properties of the system under
× cs[exp(iÄt)(I Z S+0 + S Z I+0 ) study.
+ exp(−iÄt)(I Z S−0 + S Z I−0 )] By contrast, the expectation values hQ i i used in Section IIA2
¾ correspond to linear combinations of such matrix elements of σ .
3 All calculations involve a number of such combinations much
+ s 2 [exp(2iÄt)I+0 S+0 + exp(−2iÄt)I−0 S−0 ] [137]
8 smaller than n 2 . It has sometimes been argued that the determi-
nation of the time evolution of all matrix elements of σ provides
We proceed as in Section IIB1 and we find after a straight- a much better (in fact complete) description of relaxation. Let
forward calculation of the same structure as Eq. [29] with us consider, as an example, the modest case of five spins 1/2
 ADVANCES IN MAGNETIC RESONANCE
linked by relaxation. The number of states in the Hilbert space is
n = 25 = 32, the number of matrix elements is n 2 = 210 = 1024,
and the number of elements of R is 220 = 1,048,576. A collection
of 1024 evolution curves, furthermore dependent on the initial
conditions, does not represent an information, but the burial of
any information. However, this conclusion must be somewhat
moderated. First, the interdependent elements of σ will usually
be grouped into separate sets of dimension much smaller than n 2 .
Second, in systems with not too many dimensions, it may turn
out to be convenient to solve the system [140] by computer and
then to perform the physically relevant combinations. Third, for
systems with few dimensions, the solution of the system [140]
may be simpler and faster than by any other method.
However, in most cases the use of Eq. [29], where the Qi are
observable physical quantities, saves both time and effort.
2. The Memory Function Approach
An excellent and comprehensive description of this method
can be found in Ref. (21). We give but a simplified hint at its
principle.
Let us write the evolution of a function G(t) under the form
Z t
d
G(t) = − K (t, t 0 )G(t 0 ) dt 0 .
dt 0 obtain a system identical with Eq. [31].
The function K (t, t 0 ) is called the memory function of G(t).
Equation [141] can be used for its definition. However, when
G(t) describes the evolution of some property of a system acted
upon by a random interaction f (t), the memory function can be
physically interpreted in terms of the correlation function asso-
ciated to f (t) f ∗ (t 0 ) (22). It often happens that this correlation
function depends only on (t − t 0 ). Then, this is also the case of
the function K , and Eq. [141] becomes
Z t
d
G(t) = − K (t − t 0 )G(t 0 ) dt 0 .
dt 0 a definite example, it can be shown that when the set of Q j
This last equation can be generalized to the case of different
functions G i (t) whose evolutions are coupled. We then get
X Z t there is a fundamental difference between them. In the passage
d
G i (t) = − K i j (t − t 0 )G j (t 0 ) dt 0 .
dt j 0
As a next step, we consider a system involving a large number of
time-dependent functions. It may be possible, by an appropriate
choice, to select a set p of “functions of interest” and to write, for
this set, an equation of the form [143] plus, on each right-hand
side, an extra function Hi (t) which, through physical arguments,
is small and evolves on a much faster time scale than G i (t). Its
effect is then negligible and it can be discarded.
In relaxation problems, these functions are the expectation
values of physical quantities Qi of interest (or rather the depar-
ture of these expectation values from thermal equilibrium).
173
The system [143] then reads
X Z t
d
hQ i i(t) = − K i j (t − t 0 )hQ j i(t 0 ) dt 0 . [144]
dt j 0
The next physical argument is that the decay of the memory
functions K i j is much faster than that of the expectation val-
ues hQ j i. In that case, we may choose t long enough, replace
hQ j i(t 0 ) by hQ j i(t), and extend the integral to infinity. We obtain
d X
hQ i i = − λi j hQ j i, [145]
dt j
with
Z ∞
λi j = K i j (τ ) dτ. [146]
0
Equation [145] describes the evolution due to the sole relaxation.
[141] If the contribution from the static spin Hamiltonian is added, we
This method is sound, elegant, and efficient. Particularly
seducing is the fact that Eq. [144] is rigorous (provided that
the functions Hi (t) have indeed a negligible effect). One is then
tempted to assume that Eqs. [145] and [146] are also rigorous,
that is that all memory functions do indeed quickly decay to
zero.
This will be true provided that the choice of the variables of
interest Q j is correct. There lies the main danger of this method:
the choice of the Q j is made a priori through physical intuition.
If this intuition fails, that is if the set Q j is incomplete, the results
[142] will be wrong although the system [144] is correct. By taking
is incomplete, not all of the restricted set of memory functions
decay fast enough to allow the passage from Eqs. [144] to [145].
Although this method and that of Section IIA look very similar,
[143] from Eqs. [29] to [31], the various variables hQ j i on which the
decay of hQ i i depends are not guessed: they are deduced from
the calculation of the double commutators on the right-hand side
of Eq. [29]. There is no need to exert one’s physical intuition.
As a last remark, one must use a joint ensemble average over
the two functions under the integral in Eq. [144]. The argument
is the same as that in Section IIA1. It is only at large t that it is
decoupled into the product of the averages of K i j and hQ j i. If
this is neglected, one might use Eq. [144] to describe relaxation
at arbitrarily short times, in a way similar to that criticized in
Section IIA7 (Eqs. [68]–[70]).
To sum up, the memory function method is excellent when
used with care.
174 MAURICE GOLDMAN
III. QUANTUM DESCRIPTION OF THE LATTICE
We consider now the system under study from the purely
quantum mechanical point of view. It depends on both spin vari-
ables and lattice variables. Its Hamiltonian is of the form
H = H I + H I F + F. [147]
Here H I is the spin Hamiltonian, which depends only on spin
variables, F is the lattice Hamiltonian, which depends only on
lattice (orbital) variables, and H I F is the spin–lattice coupling,
which depends on both types of variables and commutes with
neither H I nor F.
The density matrix describing the state of the whole system,
including both spin and lattice variables, is called ρ. Its evolution
equation is of the same form as Eq. [2],
d U † (t) = exp[i(H I + F )t],
ρ = −i[H, ρ]. [148]
dt
It acts on a Hilbert space which is the tensorial product of a spin
and of a lattice Hilbert space. What this means is that we can
choose in it basis kets with a double index: |i, f i referring to
spin variables i and lattice variables f .
In the absence of spin–lattice coupling, the main Hamiltonian,
H0 = H I + F, [149]
would consist of two commuting operators whose expectation
values would remain constant in time. It would then be possible
to write the density matrix ρ in the form of a product,
ρ = σ P, [150]
where σ and P depend only on spin and lattice variables, re-
spectively. The role of the term H I F is to couple H I and F and
to induce a mutual evolution with possible exchange of energy.
This is what spin–lattice relaxation consists of.
The theory is made for systems subjected to the following
conditions.
1. The coupling H I F is small enough that it is still meaningful
to speak of separate spin and lattice energy levels.
2. The density matrix ρ can still be written in the form [150].
3. The lattice part is in a statistical equilibrium state charac-
terized by a temperature, i.e., it is of the form
P = exp(−βL F )/Tr{exp(−βL F )}, [151]
where
βL = h-/kTL [152]
is called the lattice inverse temperature.
4. The lattice is an “infinite thermostat,” in the sense that its
temperature is not altered by energy exchanges with the spin
system: form [151] for P does not vary with time.
The development of the theory, very similar to that of Section
IIA2, is made in succession for different cases.
A. Static Spin Hamiltonian
The Hamiltonian H I does not depend on time.
In the absence of H I F , the unitary operator of evolution of
the system would be
U (t) = exp(−iH0 t) = exp[−i(H I + F )t]. [153]
In order to single out the effect of HI F , we go over to an
interaction representation defined by the operator
[154]
a generalization to spin and lattice of the passage to the rotating
frame, whence we get, in place of Eq. [8],
d
ρ̃ = −i[H̃ I F (t), ρ̃]. [155]
dt
The following steps are:
(1) Through formal integration, we obtain an equation similar
to [10]:
Z t
d
ρ̃ = −i[H̃ I F , ρ̃(0)] − [H̃ I F (t), [H̃ I F (t 0 ), ρ̃(t 0 )]] dt 0 .
dt 0
[156]
(2) We take an ensemble average, whose meaning in the
present case will be given later. This allows us to drop the first
term on the right-hand side.
(3) The system is assumed to have two widely different time
scales: the evolution time for ρ̃(t), or more precisely that for the
evolution of the expectation value of an observable, and a much
shorter correlation time τc , whose meaning will be precised later.
We choose the time t much longer than τc . This has the same
three consequences as in Section IIA1. The first two are that we
have a decoupling between spin and lattice ensemble averages
and that we may replace ρ̃(t 0 ) by ρ̃(t) on the right-hand side of
Eq. [156].
(4) We go back to the Schrödinger representation, which
yields
d
ρ(t) = −i[H0 , ρ(t)]
dt
Z t
− [H I F , [H̃ I F (−τ ), ρ(t)]] dτ. [157]
0
We use now the third consequence of having chosen t À τc ,
namely we extend the integral to infinity.
 ADVANCES IN MAGNETIC RESONANCE 175

With the new notation, This provides an unambiguous definition of the spin density
matrix σ , consistently used in Section II:
Ĥ I F (τ ) = H̃ I F (−τ ) = exp(−iH0 τ )H I F exp(iH0 τ ), [158]
σ = TrF ρ. [168]
we obtain
Z ∞
We use this definition in Eq. [162]. The first term on the right-
d
ρ(t) = −i[H0 , ρ] − [H I F , [Ĥ I F (τ ), ρ(t)]] dτ. [159] hand side yields
dt 0
−iTrF {[H0 , ρ]} = −iTrF [H I + F ), σ P]
(5) We expand ρ(t) according to Eq. [150]. As for H I F , we
write it in the form of an expansion similar to that of Eq. [12], = −iTrF {[H I , σ ]P + σ [F, P]}. [169]
X X
HI F = Vα Fα = Vα† Fα† , [160] We use Eq. [167] together with the property that the trace of a
α α commutator vanishes, and we obtain

where the Vα are spin operators and the Fα are time-independent −iTrF {[H0 , ρ]} = −i[H I , σ ], [170]
lattice operators which take the place of the random functions
used in Section II. The Vα are chosen so as to obey the relations which is identical with the first term on the right-hand side of
Eq. [24].
[H I , Vα ] = ωα Vα . [161] (7) We have to take the trace over lattice variables of the
second term on the right-hand side of Eq. [162]. The double
Equation [159] becomes commutator under the sign sum is over products of two opera-
X Z ∞
tors: a spin one and a lattice one. We recall the general formulas
d † for commutators of products,
ρ(t) = −i[H0 , ρ] − [Vα Fα , [V̂ β (τ )
dt α,β 0

†
[AB, C] = A[B, C] + [A, C]B, [171]
× F̂ β (τ ), σ P]] dτ. [162]
whence
† †
Since Vβ and Fβ depend only on spin and lattice variables,
respectively, and owing to the form [149] of H0 , we have [C, AB] = −[AB, C] = A[C, B] + [C, A]B. [172]
† † †
V̂ β (τ ) = exp(−iH I τ )Vβ exp(iH I τ ) = exp(iωβ τ )Vβ [163] For the commutator of two products, we obtain
† †
F̂ β (τ ) = exp(−iFτ )Fβ exp(iFτ ). [164] [AB, C D] = A[B, C D] + [A, C D]B
= AC[B, D] + A[B, C]D
(6) We will use Eq. [162] to calculate the evolution of the ex-
pectation values of spin variables, i.e., of operators Q depending + C[A, D]B + [A, C]D B. [173]
only on the spins, that is
We consider a typical term under the integral on the right-hand
hQi = Tr{Qρ} = Tr{Qσ P}. [165] side of Eq. [162], for which we use provisionally the following
simplified notation:
If one uses reduced traces, the trace of a product of commuting 
operators (such as Qσ and P) is equal to the product of their 
 Vα → V


traces (see, e.g., Ref. (12), Ch. 4). 
 Fα → F
Equation [165] is then † [174]

 V̂ β (τ ) → V̂



 F̂ † (τ ) → F̂,
hQi = Tr I {Qσ } × TrF (P) = Tr(Qσ ). [166] β

In the first term on the right, the traces are on spin variables and and we write schematically, for the evolution of the spin density
lattice variables, respectively. According to Eq. [151], we have matrix σ ,
Z ∞
d
TrF (P) = 1, [167] σ (t) = −i[H I , σ (t)] − TrF [V F, [V̂ F̂, σ P]] dτ.
dt 0
whence the second term. [175]
176 MAURICE GOLDMAN

According to rule [173], the first commutator is equal to term, that is

[V̂ F̂, σ P] = V̂ σ [ F̂, P] (a) TrF (P F̂ F).

+ V̂ [ F̂, σ ]P (b)
[176] We do this by writing explicitely the trace in a basis of eigenkets
+ σ [V̂ , P] F̂ (c) of the lattice Hamiltonian F. We bypass completely the difficulty
+ [V̂ , σ ]P F̂ (d). of defining a reduced trace in a Hilbert space whose dimensions
form a continuous set, as is the case for the lattice. We use
formally a complete trace over numerable basis kets. This will
The terms (b) and (c) contain the commutator of a spin operator
not influence the final result.
by a lattice operator. These operators commute, since they de-
We write then
pend on different variables, and these terms vanish. We are left
with X
Tr([P, F̂]F) = (h f |P| f ih f | F̂| f 0 ih f 0 |F| f i
ff0
[V̂ F̂, σ P] = V̂ σ [ F̂, P] + [V̂ , σ ]P F̂. [177]
− h f | F̂| f 0 ih f 0 |P| f 0 ih f 0 |F| f i)
Now we write its commutator with VF and take the trace with X
respect to lattice variables. We obtain = (h f |P| f ih f | F̂| f 0 ih f 0 |F| f i)
f, f 0
µ ¶
TrF{[V F, [V̂ F̂, σ P]]} = TrF {[V F, (V̂ σ [ F̂, P] + [V, σ ]P F̂)]} h f 0 |P| f 0 i
× 1− . [180]
= TrF {V V̂ σ [F, [ F̂, P]] (a) h f |P| f i

+ [V, V̂ σ ][ F̂, P]F (b) [178] With the notation

+ V [V̂ , σ ][F, P F̂] (c)
h f |F| f i = ω f
+ [V, [V̂ , σ ]]P F̂ F}. (d) [181]
h f 0 |F| f 0 i = ω f 0
The terms (a) and (c) vanish: they involve the trace over F of a
commutator of lattice operators. Reporting the remaining terms we have, according to Eq. [151],
into Eq. [175] yields
h f 0 |P| f 0 i
Z ∞ = exp{βL (ω f − ω f 0 )}. [182]
d h f |P| f i
σ (t) = −i[H I , σ ] − [V, [V̂ , σ ]]TrF (P F̂ F) dτ
dt 0 (9) The third term on the right-hand side of Eq. [179] involves
Z ∞
an integral over τ . The two terms depending on τ are V̂ and
+ [V, V̂ σ ]TrF ([P, F̂]F) dτ. [179] †
0
F̂. The variation of V̂ (i.e.,V̂ β (τ ) in full notation) is given by
Eq. [163]. Let us consider the matrix element,
Let us compare the second term on the right-hand side with †
the corresponding term in Eq. [22] (with the same simplified h f | F̂| f 0 i = h f | F̂ β (τ )| f 0 i.
notation for a typical term). They look very much alike, except
for two differences: According to Eq. [164] it is equal to
The correlation function of random functions F F̂(τ ) is re- †
h f | F̂| f 0 i = exp{−i(ω f − ω f 0 )τ }h f |Fβ | f 0 i. [183]
placed by the trace TrF [P F̂(τ )F]. The latter must therefore be
considered a quantum correlation function. It is this one whose The integral over τ is then
decay time τc is assumed very short compared to that of spin
Z ∞
quantities of interest. Its variation with temperature originates
from that of P (Eq. [151]). J = exp{i(ωβ − ω f + ω f 0 )τ } dτ. [184]
0
The term in σeq added by hand in the classical lattice model
is absent. If the “ansatz” used in Section II is indeed correct, Its real part, the only one we consider, is proportional to
this term must correspond to the third term on the right-hand
side of Eq. [179]. This is what we must analyze next, in three J ∝ δ(ωβ − ω f + ω f 0 ). [185]
successive steps.
(8) In order that they have comparable lattice parts (correla- As a consequence we may replace ratio [182] by
tion functions), we must arrange the trace over F of the third
term on the right to be of the same form as that of the second exp{βL (ω f − ω f 0 )} → exp{βL ωβ }. [186]
 ADVANCES IN MAGNETIC RESONANCE 177

without altering the value of this third term . That is, we may and we have, according to Eq. [191],
use in place of Eq. [180]
hi|σeq |ii
Tr([P, F̂]F) 1 − exp(βL ωβ ) = 1 − exp[βL (ω j − ωi )] = 1 −
X h j|σeq | ji
→ [1 − exp(βL ωβ )] (h f |P| f ih f | F̂| f 0 ih f 0 |F| f i)
f, f 0
1
= [h j|σeq | ji − hi|σeq |ii] × . [194]
= [1 − exp(βL ωβ )]Tr(P F̂ F). [187] h j|σeq | ji

Let us insert this result into Eq. [179], reverse the connection †
According to this last form, the term in hi|Vβ | ji , in the last term
[174], and go back to Eq. [162]. We obtain on the right-hand side of Eq. [188], can be written
d X †
σ (t) = −i[H I , σ ] − Jαβ (ωβ ){[Vα , [Vβ , α]] † † 1
hi|Vβ | ji × (1 − exp(βL ωβ )) = hi|[Vβ , σeq ]| ji × ,
dt α,β h j|σeq | ji
† [195]
− [Vα , Vβ σ ](1 − exp(βL ωβ ))}. [188]
from which Eq. [188] becomes
We can multiply all terms by a spin operator Q and take the
traces, following property [28], We obtain X
d †
σ (t) = −i[H I , σ ] − Jαβ (ωβ ){[Vα , [Vβ , σ ]]
d X ©­£ † ¤®
dt α,β
hQi(t) = h−i[Q, H I ]i(t) − Jαβ (ωβ ) [Q, Vα ], Vβ (t)
dt † −1
α,β − [Vα , [Vβ , σeq ]σeq σ ]}. [196]
­ †® ª
− [Q, Vα ]Vβ (t)(1 − exp(βL ωβ )) . [189]
This last form is still not very palatable and it would not be
Equation [188] is the master equation for the density matrix and very easy to use for actual calculations.
Eq. [189] is the master equation for expectation value. Both are The situation undergoes a qualitative change when going to
compact and of direct usefulness. However, they are puzzling in the limit of high temperature. It corresponds to the situation
two respects. First, it is not immediately clear toward which where for all frequencies ωβ , one has
value spin-lattice relaxation makes the quantity (Q) evolve.
Second, it seems we have not kept our promise: Equations [188] βL ωβ ¿ 1, [197]
and [189] have little resemblance to Eqs. [24] and [29], obtained
by the phenomenological replacement of σ (t) by σ (t)−σeq . This so that only terms linear in βL need be retained (the effect of
is done next. the average dipolar field (23, 24) is a subject in itself, which is
†
(10) Let us consider a specific matrix element of V̂β (τ ), outside the scope of the present article). At high temperature
the density matrices σeq and σ are very close to the unit oper-
† †
hi|V̂ β (τ )| ji = hi| exp(−iH I τ )Vβ exp(iH I τ )| ji ator, with a departure of the first order in βL . The commutator
†
[Vβ , σeq ] contains such a term, because the part proportional to
†
= exp{−i(ωi − ω j )τ }hi|Vβ | ji, [190] the unit operator yields a vanishing contribution. Therefore, in
−1
order to stick to the linear approximation, we need replace σeq σ
where ωi and ω j are the eigenvalues: by unity on the right-hand side of Eq. [196].
( As another way to obtain the same result, we have at high
hi|H I |ii = ωi temperature
[191]
h j|H I | ji = ω j .
1 − exp βL (ω j − ωi ) ' −βL (ω j − ωi )
By comparison with Eq. [163] we have
= (1 − βL ω j ) − (1 − βL ωi )
ωβ = ω j − ωi , [192] = h j|σeq | ji − hi|σeq |ii, [198]
which can be inserted into the last term on the right-hand side
of Eq. [188]. whence
We introduce the thermal equilibrium spin density matrix with
† † †
respect to the spin Hamiltonian H I , hi|Vβ | jiσ → hi|[Vβ , σeq ]| jiσ ' hi|[Vβ , σeq ]| ji [199]

σeq = exp(−βL H I )/Tr{exp(−βL H I )}, [193] to the first order in βL .

178 MAURICE GOLDMAN

−1
As a consequence of discarding σeq σ in Eq. [196], we obtain with
in place of Eq. [188]
H I F (t) = exp(iFt)H I F exp(−iFt). [205]
d X †
σ (t) = −i[H I , σ (t)] − Jαβ (ωβ )[Vα , [Vβ , (σ (t) − σeq )]],
The beginning of the relaxation evolution is when, starting from
dt α,β
spin and lattice thermal equilibrium, the spin part of the system
[200] is disturbed out of equilibrium by a spin excitation. The follow-
ing evolution under relaxation of the spin observables does not
which is identical to Eq. [24] (with H I for the static spin depend on the exact instant of time when the spin excitation is
Hamiltonian instead of H0 ). performed. That is, Eq. [205] can be replaced by
The net result of the present treatment, borrowed from
Refs. (3, 4), is that the replacement H I F (t; T ) = exp[iF(t + T )]H I F exp[−iF(t + T )]. [206]

σ (t) → σ (t) − σeq
is no longer phenomenological: it is proved.
It should be emphatically stressed that Eq. [200] is valid only
in the high-temperature domain. The equations valid at all tem-
peratures are Eqs. [188] and [189]. At high temperature Eq. [189]
becomes
d X
hQi(t) = h−i[Q, H I ]i(t) −
†
Jαβ (ωβ ){h[[Q, Vα ], Vβ ]i(t)
dt α,β
†
+ βL ωβ Tr([Q, Vα ]Vβ )}.
The trace is equal to
†
Tr(Q[Vα , Vβ ]).
† fact the we truncated it: we did not consider its fast, random fluc-
For given terms Vα , Vβ , their commutator tells immediately
which operators Q evolve toward a nonzero equilibrium value.
Remark I—On ensemble averages. We have stressed the im-
portance of taking an ensemble average for the evolution of σ ,
which has two consequences: the impossibility of studying the
density matrix at times comparable with the correlation time τc
and the simplicity brought about by choosing a time t intermedi-
ate between τc and the characteristic evolution time of physical
observables. An argument was given for this procedure in the
classical treatment of the lattice. What is the corresponding ar-
gument when the lattice is treated quantum-mechanically?
The answer is the following. Since we are ultimately inter-
ested in the evolution of the spin density matrix σ , it is unnec-
essary to start from Eq. [148]. We may first go to an interaction
representation with respect to lattice only, that is perform the
connection
0
Q → Q (t) = exp(ihFt)Q exp(−iFt),
whence
d 0
σ (t) = −iTrF [(H I + H I F (t)), σ 0 (t)]
dt
The relaxation evolution must be independent of T , and we may
use in place of [206] an average over T . This corresponds to
choosing a Gibbs ensemble of systems with all possible values
of T . This ensures that the first term on the right-hand side of
Eq. [156] does indeed vanish.
Remark II. It is a general fact that the evolution of a quantum
system, let it be one of its kets or its density matrix, is described
by a unitary operator. This is definitely not the case for the
relaxation contribution to the evolution of σ (e.g., Eq. [200]).
The reason is that σ is not the density matrix of the whole system:
it is merely its projection on the spin variables. It is natural that
[201] a projection, that is a fraction of the total density matrix, should
not behave in the same way as the full density matrix.
In Section II, we have described the density matrix of the
whole system, subjected to both steady (time independent or
[202] varying in a regular fashion) and random interactions. There, the
nonunitary evolution of the density matrix originated from the
tuations because it did not give rise to observable phenomena.
This truncation is the rule in statistical dynamics and thermo-
dynamics. It is at the origin of the transition from microscopic
reversibility to macroscopic irreversibility.
Remark III—On spectral densities. In the general expres-
sion [189], it often happens that most, if not all, contributions
to the relaxation evolution of hQi(t) originate from terms with
α = β. The corresponding spectral density Jββ (ωβ ) is, accord-
ing to Eqs. [179], [181], [183], [184], and [185], of the form
XZ ∞
Jββ (ωβ ) = Re exp(iωβ τ ) dτ h f |P| f i
0
f, f 0
†
× h f | F̂ β (τ )| f 0 ih f 0 |Fβ | f i
XZ ∞
= Re exp[i(ωβ − ω f + ω f 0 )τ ] dτ
f, f 0 0
[203] †
×h f |P| f ih f |Fβ | f 0 ih f 0 |Fβ | f i
X †
= πRe h f |P| f ih f |Fβ | f 0 ih f 0 |Fβ | f i
f, f 0
[204] × δ(ωβ − ω f + ω f 0 ). [207]
 ADVANCES IN MAGNETIC RESONANCE 179

† The spin–lattice coupling must have matrix elements between

In addition to the terms of the form [Q, Vβ ]V̂ β (τ ). . . just
described, let us consider, in the sum [189], those in [Q, |ai and |bi. It can be written in all generality
†
Vβ ]V̂ β (τ ). . . for which
H I F = I+ F + I− F † . [215]
V̂ β (τ ) = exp(−iωβ τ )Vβ . [208]
Through the use of the general Eq. [189], we have
The corresponding spectral density, named for convenience,
† d
Jββ (−ωβ ), is equal to hIz i = −J (ω){h[[Iz , I+ ], I− ]i(t) − h[Iz , I+ ]I− i(t)
dt
Z
†
∞
† × [1 − exp(βL ω)]} − J † (−ω){h[[Iz , I− ], I+ ]i(t)
Jββ (−ωβ ) = Re exp(−iωβ τ )Tr{P F̂ β (τ )Fβ } dτ. [209]
0 − h[Iz , I− ]I+ i(t)[1 − exp(−βL ω)]}
Since we have = −J (ω){h[[Iz , I+ ], I− ]i(t) − h[Iz , I+ ]I− i(t)
†
h f 0 | F̂ β (τ )| f i = h f | F̂ β (τ )| f 0 i∗ , [210] × [1 − exp(βL ω)] + exp(βL ω)h[[Iz , I− ], I+ ]i(t)
− h[Iz , I− ]I+ i(t)[exp(βL ω) − 1]}, [216]
we obtain in place of Eq. [207]

†
XZ ∞ where we have used Eq. [212]. Both double commutators are
Jββ (−ωβ ) = Re exp[−i(ωβ − ω f + ω f 0 )τ ]dτ equal to
f, f 0 0

† [[Iz , I+ ], I− ] = [[Iz , I− ], I+ ] = 2Iz , [217]

× h f 0 |P| f 0 ih f 0 |Fβ | f ih f |Fβ | f 0 i
X † whereas the other terms are
= π Re h f 0 |P| f 0 ih f |Fβ | f 0 i
f, f 0 
[Iz , I+ ]I− = I+ I− = 1
+ Iz
× h f 0 |Fβ | f iδ(ωβ − ω f + ω f 0 ).
2
[211] [218]
[I , I ]I = −I I = − 1 + I .
z − + − + 2 z

By comparison with Eqs. [182] and [207], we obtain

When reported into Eq. [216], they yield
†
Jββ (−ωβ ) = exp(βL ωβ )Jββ (ωβ ). [212] d
hIz i = −J (ω){2[exp(βL ω) + 1]hIz i + exp(βL ω) − 1}
dt
In the high-temperature limit, these spectral densities multi- ½ µ ¶¾
1 1
ply terms linear in inverse temperature βL (term between curly = −J (ω)2[exp(βL ω) + 1] hIz i + tanh βL ω .
brackets on the right-hand side of Eq. [201]). In that case, one 2 2
may forget the exponential in Eq. [212] and use [219]

† According to Eq. [214], we obtain after a little algebra

Jββ (−ωβ ) ' Jββ (ωβ ), [213]
d d
in accordance with the classical treatment of Section IIA. Pu = − Pd = −2J (ω){exp(βL ω)Pu − Pd }
dt dt
= − 2J † (−ω){Pu − exp(−βL ω)Pd }
1. Illustration: Relaxation Transition between Two Quantum
Levels = − {Wu→d Pu − Wd→u Pd }. [220]
We single out in a system two discrete levels of kets |ai and |bi,
This yields
with an energy separation h- ω, and we analyze the relaxation-
driven transition probabilities between them. µ ¶
Pu Wd→u
The simplest way is to describe these two levels by a fictitious = = exp(−βL ω), [221]
Pd Wu→d
spin I = 1/2. The up and down level populations correspond to eq

the following expectation values:

which is consistent with the general results of statistical me-
( ­1 ® chanics.
Pu = 2
+ Iz It can be checked that for obtaining result [221] it is imperative
­1 ® [214] †
Pd = 2
− Iz . to use expectation values for the term [Q, Vα ]Vβ , in Eq. [189].
180 MAURICE GOLDMAN

However, at high temperature, this expectation value is replaced we obtain

by the trace (Eq. [201]). Let us see what it yields.
With the help of Eqs. [217] and [218], Eq. [201] yields d ˜ , ρ̃˜],
ρ̃˜ = −i[H̃ IF [230]
dt
d
hIz i = −J (ω){4hIz i(t) + βL ω}
dt of the same form as Eq. [155], which we transform to a form
½ ¾ analogous to Eq. [156]:
1
= − 4J (ω) hIz i + βL ω . [222]
4 Z t
d ˜ , ρ̃˜ (0)] − ˜ (t), [H̃
˜ (t 0 ), ρ̃˜ (t 0 )]] dt 0 .
The same result is obtained from Eq. [219] expanded to the ρ̃˜ = −i[H̃ IF [H̃ IF IF
dt 0
first order in βL ω. [231]
B. Time-Dependent Spin Hamiltonian
The next steps are:
1. Off-Resonance Irradiation
We take an ensemble average, for which the first term on the
We analyze the same problem as in Section IIB1-1: the relax- right-hand side vanishes.
ation of a nuclear spin subjected to off-resonance irradiation at a We use form [160] for H I F and form [150] for ρ, with P
frequency ω in the vicinity of its Larmor frequency ω0 . The spin given by Eq. [151].
Hamiltonian H I is the same as in Eq. [77], and the spin–lattice We go back to the Schrödinger representation for the lattice
coupling is of the same form as in Eq. [160]. The development variables and to an intermediate representation (first rotating
is a combination of those of Sections IIB and III. We indicate frame) for the spin variables. The corresponding unitary operator
its main steps very succinctly. is
Following the formalism at the beginning of Section III, we
use an interaction representation where the evolution of the den-
UF (t)U I 2 (t).
sity matrix depends only on the spin–lattice coupling. The cor-
responding unitary operator is
We choose t much longer than the correlation time τc , as a
†
U (t) =
† † †
U I 2 (t)U I 1 (t)UF (t), [223] consequence of which we may replace σ̃ (t 0 ) by σ̃ (t) and we
extend all integrals to infinity.
We obtain an equation whose terms have a form similar to
with
Eq. [179],
†
UF (t) = exp(iFt), [224] Z ∞
d
† σ̃ (t) = −i[H̃eff , σ̃ ] − [Ṽ (t), [V̂ (t, τ ), σ̃ (t)]]
U I 1 (t) = exp(iωIz t), [225] dt 0
†
Z ∞
U I 2 (t) = exp(i ÄI Z t), [226]
× TrF (P F̂(τ )F) dτ + [Ṽ (t), V̂ (t, τ )σ̃ (t)]
0
where the orientation of the axis 0Z and the effective frequency
× TrF ([P, F̂]F) dτ, [232]
Ä are given by Eqs. [81] and [82], respectively. As above, we
assume that Ä ¿ ω. We use the following notations.
For a spin operator, where Heff is given by Eq. [80], and

† † † †
U I 1 (t)Q U I 1 (t) = Q̃(t) [227] V̂ (t, τ ) = U I 2 (τ )U I 1 (t − τ )Vβ U I 1 (t − τ )U I 2 (τ )
†
U I 2 (t) Q̃(t)U I 2 (t) = Q̃˜ (t). [228] † † † †
= U I 2 (τ )U I 1 (τ )U I 1 (t)Vβ U I 1 (t)U I 1 (τ )U I 2 (τ ), [233]

For a lattice operator,

that is, according to Eqs. [225] and [227],
†
UF (t)FUF (t) = F̃(t) = F̃˜ (t) = F̂(−t). [229] †
V̂ (t, τ ) = exp(−i ÄI Z τ ) exp(−iωIz τ ) exp(iωIz t)Vβ
Starting from the Liouville–von Neumann equation, × exp(−iωIz t) exp(iωIz τ ) exp(iÄI Z τ ). [234]

d
ρ = −i[H, ρ], We use the formulations [83], [84], [88], and [89] for H I F , and
dt
 ADVANCES IN MAGNETIC RESONANCE 181

we obtain, as a general expression for Eq. [232], of σeq is then given by Eq. [193]. When the spin Hamiltonian
0†
is time dependent, the various operators V̂ q (t, τ ) will have a
d X
σ̃ (t) = −i[Heff , σ̃ (t)] − λmp λqm
∗
dependence on τ produced by several operators (Eq. [234]), and
dt m, p,q the relative dependence on these will not be the same for differ-
0†
Z ∞
ent operators Vq . In order to obtain an expression resembling
× dτ exp[i(mω + qÄ)τ ] Eq. [200], one must therefore invent a fictitious Hamiltonian
0 H I (m, q), tailored so as to yield
× {[V p0 , [Vq0† , σ̃ (t)]]TrF (P F̂ †m (τ )Fm )
V̂ q0† (τ ) = exp[−iH I (m, q)τ ]Vq0† exp[iH I (m, q)τ ]. [241]
− [V p0 , Vq0† σ̃ (t)]TrF ([P, F̂ †m (τ )]Fm )}. [235]
Following the same treatment as from Eqs. [190] to [199], we
The integral involving the first trace over F, in the second are led to express Eq. [239] under the form
term on the right-hand side, yields the quantum spectral density
Jm,m (mω + qÄ). That of the third term is analyzed exactly as in d X ∗
σ̃ (t) = − i[H̃eff , σ̃ (t)] − λmp λqm Jm,m (mω + qÄ)
Section IIIA. It yields, in complete analogy with Eqs. [187], dt m, p,q
Z ∞ × [V p0 , [Vq0† , (σ̃ (t) − σ0 (m, q))]], [242]
exp[i(mω + qÄ)τ ]TrF ([P, F̂ †m (τ )]Fm ) dτ
0
with
= Jm,m (mω + qÄ)[1 − exp(βL (mω + qÄ))]. [236]
σ0 (m, q) = 1 − βL H I (m, q). [243]
We then obtain an expression similar to Eqs. [188] and [189],
†
X Different terms Vq require a priori different pseudo-
d ∗
Hamiltonians H I (m, q), to which there corresponds different
σ̃ (t) = −i[Heff , σ̃ ] − λmp λqm Jm,m (mω + qÄ)
dt m, p,q “equilibrium” density matrices.
× {[V p0 , [Vq0† , σ̃ (t)]] − [V p0 , Vq0† σ̃ (t)] There is still another point. In the absence of a static spin
Hamiltonian, there is no “natural” choice of spin components
× [1 − exp(βL (mω + qÄ))]} [237] of H I F . Instead of using V p0 , Vq0 , one might as well have ex-
d X ∗
hQir (t) = h−i[Q, Heff ]i(t) − λmp λqm Jm,m (mω + qÄ) pressed them as a function of the initial Vm . Using the latter
dt m, p,q would lead to “equilibrium,” density matrices σ0 (m) different
from the σ0 (m, q). Then, there is no single limit for the evolu-
× {h[[Q, V p0 ], Vq0† ]i(t) − h[Q, V p0 ]Vq0† i(t) tion of σ (or σ̃ ) under the effect of relaxation. This of course
× [1 − exp(βL (mω + qÄ))]}. [238] should make no difference for the steady-state limit of observ-
able physical quantities. Under these conditions, there is little
In the high-temperature approximation, they reduce to interest in sticking to form [242]: Equations [237]–[240] are
just as good and simpler. The various points raised above are
d X ∗ illustrated in Section IIIB3.*.
σ̃ (t) = −i[Heff , σ̃ ] − λmp λqm Jm,m (mω + qÄ)
dt m, p,q

× {[V p0 , [Vq0† , σ̃ (t)]] + [V p0 , Vq0† ]βL (mω + qÄ)}
X ∗
hQir (t) = h−i[Q, H̃eff ]i(t) − λmp λqm Jm,m (mω
m, p,q
× {h[[Q, V p0 ], Vq0† ]i(t)
+ Tr(Q[V p0 , Vq0† ])βL (mω + qÄ)}.
These equations are well defined and solve entirely the prob-
lem. However, one might try to express Eq. [239] under a form
similar to Eq. [200]. There is a fundamental difference between
the present case and that of a static spin Hamiltonian analyzed
†
in Section IIB. There, the time evolution of all operators V̂ β (τ )
is due to the static spin Hamiltonian H I (Eq. [163]) and the form
2. Nuclear Relaxation by a Paramagnetic Center:
Simplified Treatment
[239]
We are now in a position to justify the phenomenological treat-
+ qÄ) ment given in Section IIB1, and to its approximations. The con-
ditions correspond to high temperature and effective frequency
Ä much smaller than the Larmor frequency ω0 . We limit our-
[240] selves to a short qualitative discussion.
Let us first consider the terms in B, B ∗ of Eq. [97]. Through
the unitary transformation [234], they yield a dependence on τ
of the form
exp[i(mω + qÄ)τ ],
with m = ±1 and q = 0, ±1.
182 MAURICE GOLDMAN

For the relaxation part of σ̃ (t) (Eq. [242]), they yield spectral still be written under the form [97], but with
densities,
Sz (t) = exp(iFt)Sz exp(−iFt).
J (mω + qÄ) ' J (mω), [244]
It would have been possible to make a different separation: use
as a result of the smallness of Ä, as noted in Section IIB1-1. as H I the Zeeman interactions of the spins I and S plus the I − S
These terms also give rise to “equilibrium” density matrices, dipolar coupling, as F the orbital interactions, and as H I F the
coupling between the electronic spin and the orbital degrees of
σ0 (m, q) = 1 − βL (mω + qÄ)I Z ' 1 − βL (mω)I Z , [245] freedom. For actual relaxation calculations at low temperature,
the latter choice might prove more convenient. Of course, both
for the same reason. choices should yield the same result for the relaxation of the
Since we may neglect Ä in both the J and the σ0 , the result spin I . This comparison will not be analyzed here.
will be the same as if the system had been subjected to a static The calculation will be made successively with spin opera-
spin Hamiltonian: tors in H I F adapted to the axes O X Y Z of the tilted frame and
adapted to the axes O x yz of the first rotating frame.
ωI Z ' ω0 I Z . [246] Axes O X Y Z . We start from the spin–lattice coupling
H̃ I F (t) in the first rotating frame (Eq. [99]). With the help
of Eq. [103] we write the spin operators in terms of I Z and
In that case, there is no point in using operators Iz , I±0 for the
I±0 = I X ± i IY . We obtain
terms in B, B ∗ . Their effect can be calculated with I± , as in
Section IIB1. ½ · ¸
s
As for the terms in AI±0 , they give rise to terms in exp(±i Äτ ), H̃ I F (t) = Sz (t) A cI Z − (I+0 + I−0 ) + B exp(iωt)
as inferred from Eq. [109]. The corresponding matrices σ0 will 2
· ¸
be of the form 1+c 0 1−c 0
× s IZ + I+ − I− + B ∗ exp(−iωt)
2 2
σ0 = 1 ± βL ÄI Z , [247] · ¸¾
1+c 0 1−c 0
× s IZ + I − I . [248]
and they correspond to steady-state limits negligibly small com- 2 − 2 +
pared with those arising from the σ0 (m, q) of Eq. [245].
These brief comments are sufficient to justify the approxima- From Eq. [234], we obtain for Ĥ I F (t, τ )
tions made in Section IIB1. The same approximations apply to
any practical situation involving off-resonance irradiation. Ĥ I F (t, τ )
½ · ¸
s 0 s 0
3. ∗ Nuclear Relaxation by a Paramagnetic Center: = Sz (t) A cI Z − exp(−iÄτ )I+ − exp(iÄτ )I−
2 2
Complete Discussion ·
1+c
We consider the same system as above, we still keep the high- + B exp(iωt) exp(−iωτ )s I Z +
2
temperature approximation, but we lift the preceeding constraint ¸
on the relative magnitudes of Ä and ω0 . We consider explicitely 1−c
× exp(−i(ω + Ä)τ )I+0 − exp(−i(ω − Ä)τ )I−0
the case of a rotating field. With a linearly polarized field which 2
is not small compared to the static field part, one cannot discard ·
1+c
one of its rotating components, and the problem becomes seri- + B ∗ exp(−iωt) exp(iωτ )s I Z + exp(i(ω + Ä)τ )I−0
ously complicated. Although of limited usefulness in existing 2
¸¾
experimental situations, the present discussion is included with 1−c 0
the purpose of clarifying through a specific example the main − exp(i(ω − Ä)τ )I+ . [249]
2
points raised in Section IIIB1.
Let us come back to the spin–lattice coupling [97]. In We can use these expressions in Eq. [239], where we discard
Section IIB1, the electronic spin component Sz (t) was consid- terms oscillating at mω. As a consequence, terms V p0 from
ered a random variable of vanishing average value. In this sec- Eq. [248] are coupled with terms Vq0 from Eq. [249] so as to
tion, we must use it as a static operator, whose evolution is yield products A2 or B B ∗ . The result is then of the form
produced by the lattice. In the present approach, the “lattice”
F consists of the sum of the electronic Zeeman interaction, the d
orbital interactions (for instance the phonons), and the coupling σ̃ (t) = −iÄ[I Z , σ̃ (t)] − A2 λ − B B ∗ µ. [250]
dt
of the electronic spin with the latter. Thanks to its orbital part,
its spectrum is quasi-continuous. The spin–lattice coupling can We do not detail the corresponding lengthy calculation. We look
 ADVANCES IN MAGNETIC RESONANCE 183

instead at the characteristics of the form [242]. We note that we Its contribution to d σ̃ /dt is then of the form
can write Eq. [249] under the form
s
½ · ¸ C1 ∝ − J (ω){[B ∗ . . . , [B(I Z + i IY ), (σ̃ (t) − βL ωI X )]]
s 2
Ĥ I F (t, τ ) = Sz (t) A exp(−iÄI Z τ ) cI Z − (I+0 + I−0 )
2 + [B . . . , [B ∗ (I Z − i IY ), (σ̃ (t) + βL ωI X )]]},
× exp(iÄI Z τ ) + [B exp(iω(t − τ )) [255]
1+c
+ B ∗ exp(−iω(t − τ ))]s I Z + B exp(iωt) whereas the part exp(+iωτ )I Z can be written
2
0
× exp(−i(ω + Ä)I Z τ )I+ exp(i(ω + Ä)I Z τ ) + h.c.
1
1−c exp(iωτ )I Z = {exp(−iωτ I X )(I Z + i IY ) exp(iωτ I X )
0 2
− B exp(iωt) exp(i(ω − Ä)I Z τ )I−
2 + exp(iωτ I X )(I Z − i IY ) exp(−iωτ I X )}. [256]
¾
× exp(−i(ω − Ä)I Z τ ) + h.c. . [251]
Its contribution to d σ̃ /dt is of the form

By comparison with Eqs. [241]–[243], these terms yield con- s

C2 ∝ − J (ω){[B . . . , [B ∗ (I Z + i IY ), (σ̃ (t) + βL ωI X )]]
tributions to d σ̃ /dt of the following forms: 2
the part in I±0 of first term on the right-hand side of Eq. [251]: + [B . . . , [B ∗ (I Z − i IY ), (σ̃ (t) − βL ωI X )]]}. [257]

One might think at first sight that the terms in σ0 would cancel
−A2 J (Ä)[. . . , [. . . , (σ̃ (t) − σ0 (Ä))]],
out in the sum C1 + C2 . This is not so because the first terms in
the double commutator originate from H̃ I F (t), and those pro-
The third term and its Hermitian conjugate: portional to B are not the same as those proportional to B ∗ .
µ ¶2 One may also remark that, say, exp(−iωτ )I Z could be written
∗ 1+c at variance with Eq. [254], as follows,
−B B J (ω + Ä)[. . . , [. . . , (σ̃ (t) − σ0 (ω + Ä))]],
2
exp(−iωτ )I Z
The fifth term and its Hermitian conjugate:
1
= {exp(−iωτ IY )(I Z + i I X ) exp(iωτ IY )
µ ¶2 2
1−c
−B B ∗ J (ω − Ä)[. . . , [. . . , (σ̃ (t) − σ0 (−ω + Ä))]], + exp(iωτ IY )(I Z − i I X ) exp(−iωτ IY )}, [258]
2
and a similar expression for exp(iωτ )I Z . All these forms are
with the notation
correct and they must yield the same result for the evolution of a
physical quantity hQi. It turns out to be much more convenient
σ0 (ω) = 1 − βL ωI Z , [252] to use the formal Eq. [240]. If in the trace on its right-hand side
0†
the operator Vq is I Z , the only operators Q and V p0 for which
which illustrates the fact that different terms of the spin–lattice the trace does not vanish are I+0 and I−0 , in any order. Let us
coupling lead to different limits σ0 . choose Q = I+0 and V p0 = I−0 . The contribution of these terms to
The second term needs special attention. The part the relaxation limit of hI+0 i is proportional to
exp(−iωτ )I Z can be written
½µ ¶
∗ 1+c
exp(−iωτ )I Z = exp(−iωτ ) × 1
2
[(I Z + i IY ) + (I Z − i IY )] −s J (ω)B B Tr([I+0 , I−0 ]I Z ) × −βL ω
2
[253] µ ¶¾
1−c s
− × βL ω = + J (ω)B B ∗ × βL ω. [259]
2 2
or else

1 We consider now explicitely the evolution equation for hI Z i.

exp(−iωτ )I Z = {exp(iωτ I X )(I Z + i IY ) exp(−iωτ I X ) Its calculation is trivial and we will give only its result. We
2
introduce the (arbitrary) expectation value,
+ exp(−iωτ I X )(I Z − i IY ) exp(iωτ I X )}.
¡ ¢
[254] I0 = Tr[I Z σ0 (ω0 )] = −βL ω0 Tr I Z2 , [260]
184 MAURICE GOLDMAN

and we obtain, ignoring cross-relaxation, s

− (1 + c) exp(i(ω + Ä)τ )
2
½ µ ¶ ¸
d 1 Ä s
hI Z i(t) = − S(S + 1) A2 s 2 J (Ä) hI Z i(t) − I0 + (1 − c) exp(i(ω − Ä)τ ) Iz
dt 3 ω0 2
µ ¶ · 2 µ ¶
∗ ω+Ä s 1+c 2
+ B B (1 + c )J (ω + Ä) hI Z i(t) −
2
I0 + exp(iωτ ) +
ω0 2 2
µ ¶¾
ω−Ä µ ¶ ¸
∗
+ B B (1 − c )J (ω − Ä) hI Z i(t) +
2
I0 . 1−c 2
ω0 × exp(i(ω + Ä)τ ) + exp(i(ω − Ä)τ ) I−
2
[261] · 2
s 1−c 2
+ exp(iωτ ) − exp(i(ω + Ä)τ )
In the case treated above, when Ä ¿ ω0 , that is also ω ' ω0 , 2 4
¸ ¸¾
we obtain a result identical to Eq. [114], where the term in hI X i 1 − c2
is ignored. − exp(i(ω − Ä)τ ) I+ . [263]
4
Axes Oxyz. We keep for H̃I F (t) the form [99]. For H I F (t; τ )
we start from Eq. [249] and express the operators I Z , I±0 in terms We have written in extenso this formidable expression (for the
of Iz , I± . We have, from Eq. [104], simplest possible system), but we use it only for a qualitative
discussion.

 I Z = cIz + 2s (I+ + I− ) By simple inspection of the functions of τ , it is immediately

 evident that if we use for the derivative of σ̃ a form similar to
I+0 = −s Iz + 1 +2 c I+ − 1−c
I− [262] Eq. [242], the terms σ0 corresponding to Eq. [243] will not all


2
 0
I− = −s Iz − 1 −2 c I+ + 1+c
2
I− . be equal and furthermore they will be very different from those
arising in the preceding section.
We insert these expressions into Eq. [249] and we obtain There are several lessons to be learned from this analysis, as to
the contrast between a time-independent and a time-dependent
spin Hamiltonian. As regards the latter,
H̃ I F (t, τ )
½ · · (1) There are no privileged axes adapted to the development
cs s
= Sz (t) × A (c + s cos Äτ )Iz +
2 2
− (1 + c) of the spin–lattice coupling. The best choice is that which makes
2 4
¸ the calculation easiest. In the general case analyzed in this sec-
s tion, there is a definite preferences for the axes 0X Y Z . In the
× exp(−iÄτ ) + (1 − c) exp(iÄτ ) I+
4 limit when the approximations made in Section IIIB2 are valid,
· it even turned out that the most convenient was to use different
cs s
+ + (1 − c) exp(−iÄτ ) axes for treating the terms in A and those in B, B ∗ .
2 4 (2) The density matrix σ̃ (t) does not evolve toward a sin-
¸ ¸
s gle limit: there are different σ0 for the different spin terms
− (1 + c) exp(iÄτ ) I− + B exp(iωt) of the spin–lattice coupling, which depend furthermore on the
4
·· choice of its expansion. This deprives that kind of presenta-
s tion of the relaxation equations from much of its “enlightening”
× sc exp(−iωτ ) − (1 + c) exp(−i(ω + Ä)τ )
2 interest.
¸ · 2 (3) The pedestrian use of the direct expression [237] (or its
s s
+ (1 − c) exp(−i(ω − Ä)τ ) Iz + exp(−iωτ ) equivalent Eq. [238]) is both devoid of ambiguity, in general
2 2
much simpler than Eq. [242], and furthermore valid at all tem-
µ ¶ µ ¶
1+c 2 1−c 2 peratures, whereas Eq. [242] is valid only at high temperature.
+ exp(−i(ω + Ä)τ ) + These conclusion will show up even more clearly in the next
2 2
¸ · 2 section.
s 1 − c2
× exp(−i(ω − Ä)τ ) I+ + exp(−iωτ ) −
2 4 4. ∗ The General Problem of a Time-Dependent Spin
¸ Hamiltonian
1 − c2
× exp(−i(ω + Ä)τ ) − exp(−i(ω − Ä)τ ) I− This section is included for completeness. It refers to a physi-
4
·· cal situation which does not seem relevant to experimental inves-
+ B ∗ exp(−iωt) sc exp(iωτ ) tigation at present. However, it might be used for problems not
involving spins, but whose formulation could be made so as to
 ADVANCES IN MAGNETIC RESONANCE 185

formally mimic a spin problem. Furthermore, its very generality IIIB1, and IIIB2. We obtain the by now familiar expression
provides an overview of the problem of relaxation.
We consider a spin system whose spin Hamiltonian undergoes d
ρ(t) = −i[H0 , ρ(t)]
an arbitrary time variation. The canonical example is that of a dt
Zeeman interaction with a field varying both in orientation and Z ∞
in magnitude. All we request is that its spectrum at all times − [H I F , [Ĥ I F (t; τ ), ρ(t)]] dτ, [268]
0
consists of discrete and energy-split levels. In the case when
part of the spectrum is quasi-continuous, as in solids with dipolar
where
interactions, the width of this part should be small compared to
the large splittings, so as to allow the kind of treatment used in
Section IIA6. Ĥ I F (t; τ ) = U (t)H̃ I F (t − τ )U † (t)
The initial development repeats that of the beginning of = U (t)U † (t − τ )H I F U (t − τ )U † (t). [269]
Section III. We begin with the Liouville–von Neumann equa-
tion [148], where the Hamiltonian H0 (Eq. [149]) is the sum of For a general time-dependent Hamiltonian H I , the expression
a fixed lattice part and a time-dependent spin part H I (t). The of U and U † is not so simple as in the preceding sections. We
interaction representation in which the evolution of the density have, however, in full generality,
matrix ρ̃ depends only on the spin–lattice coupling is defined by
the replacement of all operators Q by
U (t) = U (t; t − τ )U (t − τ ), [270]

Q → Q̃(t) = U † (t)QU (t), [264] where U (t; t − τ ) is still defined by Eq. [265] with the initial
condition
with
U (t − τ ; t − τ ) = 1, [271]
( d
U (t) = −i(H I (t) + F )U (t)
dt
[265] that is
d
dt
U † (t) †
= iU (t)(H I (t) + F). Ã Z !
t
U (t; t − τ ) = T exp −i H0 (t 0 ) dt 0 , [272]
We have indeed t−τ

µ ¶ µ ¶ where T is the Dyson chronological operator.

d d † † d
ρ̃(t) = U (t) ρ(t)U (t) + U (t) ρ(t) U (t) Another general property is that
dt dt dt
µ ¶ U (t − τ )U † (t − τ ) = 1, [273]
d
+ U † (t)ρ(t) U (t)
dt
whence, according to Eq. [269],
= iU † (H I + F )ρU − iU † [(H I + F + H I F ), ρ]U
Ĥ I F (t; τ ) = U (t; t − τ )H I F U † (t; t − τ ). [274]
− iU † ρ(H I + F )U
= −iU † [(H I + F + H I F − H I − F ), ρ]U Next, we expand H I F under the same form as [160]. How-
ever, the various spin operators Vα can no longer have a definite
= −i[H̃ I F (t), ρ̃(t)]. [266] relationship with the time-dependent spin Hamiltonian H I (t).
They have to be chosen with reference to the physical quantities
Through formal integration and ensemble average, we obtain one wants to observe.
an equation similar to [156], without the first term on the right- Since H0 = H I + F is a sum of two commuting operators,
hand side: Eq. [272] can be written
Z t U (t; t − τ ) = U I (t; t − τ )UF (τ ), [275]
d
ρ̃(t) = − [H̃ I F (t), [H̃ I F (t − τ ), ρ̃(t − τ )]] dτ. [267]
dt 0
with
à Z !
We choose the time t much longer than the lattice correlation t
0 0
time, which makes it possible to replace ρ̃(t − τ ) by ρ̃(t); we go U I (t; t − τ ) = T exp −i H I (t ) dt [276]
t−τ
back to the Schödinger representations and extend the integral
to infinity, in accordance with the procedure of IIA1 and IIIA, UF (τ ) = exp(−iFτ ), [277]
186 MAURICE GOLDMAN

and Eq. [274] becomes It is also possible, through a simple algebraic manipulation,
X to write
Ĥ I F (t; τ ) = V̂ α (t; τ ) F̂ α (τ )
d
α hQi(t)
X †
dt
= [U I (t; t − τ )Vα U I (t; t − τ )] XZ 1 δ+∞
α = h−i[Q, H I (t)]i(t) − λ (t; ω)Jαβ (ω)
† α,β,δ
2 β
−∞
× [UF (τ )Fα UF (τ )]. [278]
½ µ ¶
­£ † ¤® βL ω
× (exp(βL ω) + 1) dω [Q, Vα ], Vδ (t) + tanh
We use Eqs. [150] and [168] and we obtain for the spin density 2
matrix an expression of the same form as Eq. [179], ¾
­© † ª®
× [Q, Vα ], Vδ (t) (t) , [284]
d
σ (t) = −i[H I (t), σ (t)]
dt
Z ∞X where the symbol {. . . , . . .} has the usual meaning of an anti-
©£ £ † ¤¤
− Vα , V̂ β (t; τ ), σ (t) commutator:
0 α,β
¡ ¢ £ ¤ {A, B} = AB + B A. [285]
† †
× TrF P F̂ β (τ )Fα − Vα , V̂ β (t; τ )σ (t)
¡£ ¤ ¢ ª In the high-temperature limit, it is possible, through the same
†
× TrF P, F̂ β (τ ) Fα dτ . [279] procedure as for Eqs. [241]–[243], to replace in Eq. [282] σ (t)
by σ (t) − σ0 , but there would be a continuous distribution of
Since in the present case the evolution of V̂ β (t; τ ) is
† matrices σ0 , and this would be even less informative than in the
more complicated than Eq. [163], we cannot directly jump to case of a rotating field.
Eqs. [188] and [189]. We do the following. There is one exception, when the spin Hamiltonian H I (t)
We consider a complete set of spin operators Vα , all of which varies very little on the time scale τc . Equation [278] would then
may not be present in H I F . We may then write read
X
†
X † Ĥ I F (t; τ ) ' [exp(−iH I (t)τ )Vα exp(iH I (t)τ )]
V̂ β (t; τ ) = f βδ (t; τ )Vδ , [280] α
δ
× [exp(−iFτ )Fα exp(iFτ )], [286]
†
which is no more than an expansion of over the completeV̂ β
set. and Eq. [279] could be written under the form
Next, we Fourier-analyze the functions f with respect to τ : Z ∞
d
Z σ (t) ' − i[H I (t), σ (t)] − TrF [H I F ,
+∞ dt 0
f βδ (t; τ ) = λδβ (t; ω) exp(iωτ ) dω. [281]
−∞ [Ĥ I F (t; τ ), (σ (t) − σeq (t))]]dτ, [287]

In complete analogy with the passage from Eq. [179] to where

Eq. [188] we obtain
σeq (t) = 1 − βL H I (t). [288]
d X Z +∞
σ (t) = −i[H I (t), σ (t)] − λδβ (t; ω)Jαβ (ω) dω This corresponds to a relaxation of the density matrix toward
dt α,β,δ −∞
its instantaneous equilibrium value. It happens only in a very
† †
× {[Vα , [Vδ , σ (t)]] − [Vα , Vδ σ (t)](1 − exp(βL ω))}. special case.
In the general case, Eq. [283] or [284] provides a complete
[282] solution of the relaxation problem, valid whatever the varia-
tion of the spin Hamiltonian and whatever the lattice temper-
The expectation value of an operator Q evolves according to ature. For a given spin Hamiltonian H I (t), they can be solved
by computer provided that a model is available for the spectral
d X Z +∞ densities.
hQi(t) = h−i[Q, H I (t)]i(t) − λδβ (t; ω)Jαβ (ω) dω
dt α,β,δ −∞
REFERENCES
†
× {h[[Q, Vα ], Vδ ]i(t)
1. H. G. B. Casimir, “Magnetism at Very Low Temperatures,” Ch. VII,
†
− h[Q, Vα ]Vδ i(t)(1 − exp(βL ω))}. [283] Cambridge Univ. Press, Cambridge, 1940, and references therein.
 ADVANCES IN MAGNETIC RESONANCE
2. N. Bloembergen, E. M. Purcell, and R. V. Pound, Phys. Rev. 73, 679 (1948).
3. R. K. Wangsness and F. Bloch, Phys. Rev. 89, 278 (1953).
4. F. Bloch, Phys. Rev. 102, 104 (1956).
5. I. Solomon, Phys. Rev. 99, 559 (1955).
6. A. Abragam, “The Principles of Nuclear Magnetism,” Ch. VIII, Clarendon
Press, Oxford, 1961.
7. A. G. Redfield, in “Advances in Magnetic Resonance” (J. S. Waugh, Ed.),
Vol. 1, p. 1, Academic Press, New York, 1965.
8. F. Bloch, Phys. Rev. 105, 1206 (1957).
9. K. Tomita, Progr. Theor. Phys. 19, 541 (1958).
10. C. P. Slichter, “Principles of Magnetic Resonance,” 3rd ed., Springer-Verlag,
Berlin, 1990.
11. “Encyclopedia of Nuclear Magnetic Resonance” (D. M. Grant and
R. K. Harris, Eds.), Vol. 6, p. 3989–4107, Wiley, Chichester, England, 1996.
12. M. Goldman, “Quantum Description of High-Resolution NMR in
Liquids,” Oxford Univ. Press, Oxford, 1988.
13. M. Goldman, in “Nuclear Magnetic Double Resonance” (B. Maraviglia,
Ed.), p. 1, North-Holland, Amsterdam, 1993.
187
14. S. Zamir, R. Poupko, Z. Luz, and S. Alexander, J. Chem. Phys. 94, 5939
(1991), and references therein.
15. M. Goldman, Mol. Phys. 86, 301 (1995).
16. L. G. Werbelow, Ref. [11], Vol. 3, p. 1776.
17. I. Solomon and N. Bloembergen, J. Chem. Phys. 25, 261 (1956).
18. C. Dalvit and G. Bodenhausen, in “Advances in Magnetic Resonance”
(W. S. Warren, Ed.), Vol. 14, p. 1, Academic Press, New York,
1990.
19. M. Goldman, J. Magn. Reson. 60, 437 (1984).
20. A. Abragam and M. Goldman, “Nuclear Magnetism: Order and
Disorder,” Ch. 1, Oxford Univ. Press, Oxford, 1982.
21. D. Kivelson and K. Ogan, in “Advances in Magnetic Resonance”
(J. S. Waugh, Ed.), Vol. 7, p. 72, Academic Press, New York, 1974.
22. H. Mori, Prog. Theor. Phys. Kyoto 33, 423 (1965).
23. S. Lee, W. Richter, S. Vathyam, and W. S. Warren, J. Chem. Phys. 105, 874
(1996), and references therein.
24. A. Vlassenbroek, J. Jeener, and P. Broekaert, J. Magn. Reson. A 118, 234
(1996), and references therein.