NtTCLEAR MAGNETIC RESONANCE ABSORPTION 679

ground state of Rb~, the value for the average such an extent that with our present lack of
energy might be used to distinguish between knowledge about the level scheme of the trans-
various assumptions regarding the angular dis- formation, such comparisons would be incon-
tribution of the electron and the neutrino. clusive.
Should, however, P-decay to some intermediate Our thanks are due to Professor Niels Bohr
state, followed by y-radiation, occur, a cor- for his kind interest in our work, to Mr. N. O.
rection must be applied to the measured value of Lassen and Mr. K. O. Nielsen for their help in
the average energy to obtain the value ror- the work with the cyclotron, to Mr. B, S. Madsen
responding to a simple P-spectrum. This cor- for the construction of the counters and to Mr.
rection may change the average recoil energy to J. Lindhard for some theoretical discussions.

PH YSI CAL REVIEW VOLUME 73, NUMBER 7 AP R IL 1, 1948

Relaxation Effects in Nuclear Magnetic Resonance Absorytion*

N. BLoEMBERGEN, * E. M. PURcELL, AND R. V. PoUND, ***
I yman Laboratory of Physics, Harvard University, Cambridge, Massachusetts
(Received December 29, 1947)

The exchange of energy between a system of nuclear spins The effect of paramagnetic ions in solution upon the proton
immersed in a strong magnetic 6eld, and the heat reservoir relaxation time and line width has been investigated. The
consisting of the other degrees of freedom {the "lattice" ) relaxation time and line width in ice have been measured
of the substance containing the magnetic nuclei, serves to at various temperatures.
bring the spin system into equibbrium at a 6nite tempera- The results can be explained by a theory which takes
ture. In this condition the system can absorb energy from into account the effect of the thermal motion of the
an applied radiofrequency 6eld. With the absorption of magnetic nuclei upon the spin-spin interaction. The local
energy, however, the spin temperature tends to rise and magnetic field produced at one nucleus by neighboring
the rate of absorption to decrease. Through this "satura- magnetic nuclei, or even by electronic magnetic moments
tion" effect, and in some cases by a more direct method, of paramagnetic ions, is spread out into a spectrum ex-
the spin-lettre relaxation time Ti can be measured. The tending to frequencies of the order of 1 jr„where v, is a
interaction among the magnetic nuclei, with which a correlation time associated with the local Brownian motion
characteristic time Tg' is associated, contributes to the and closely related to the characteristic time which occurs
width of the absorption line. Both interactions have been in Debye's theory of polar liquids. If the nuclear Larmor
studied in a variety of substances, but with the emphasis frequency cu is much less than 1/v„ the perturbations
on liquids containing hydrogen. caused by the local field nearly average out, T& is inversely
Magnetic resonance absorption is observed by means of proportional to v. , and the width of the resonance line, in
a radiofrequenqy bridge; the magnetic field at the sample frequency, is about 1/T&. A similar situation is found in
is modulated at a low frequency. A detailed analysis of hydrogen gas where r, is the time between collisions. In
the method by which T~ is derived from saturation very viscous liquids and in some solids where cur, &1, a
experiments is given. Relaxation times observed range quite different behavior is predicted, and observed. Values
from 10 4 to 10' seconds. In liquids T& ordinarily decreases of r, for ice, inferred from nuclear relaxation measure-
with increasing viscosity, in some cases reaching a mini- ments, correlate well with dielectric dispersion data.
mum value after which it increases with further increase Formulas useful in estimating the detectability of
in viscosity. The line width meanwhile increases mono- magnetic resonance absorption in various cases are derived
tonically from an extremely small value toward a value in the appendix.
determined by the spin-spin interaction in the rigid lattice.

I. INTRODUCTION circuit to a system of nuclear spins immersed in

'N nuclear magnetic resonance absorption, a magnetic held, IID, as a result of transitions
~
energy is transferred from a radiofrequency among the energy levels of the spin system. For
*A brief account of this work has appeared in Nature each of X non-interacting spins, characterized
160, 4/5 (1947). in the usual way by Ik and p, , the maximum
¹¹ Present address: Kamerlingh Onnes Laboratory, Uni- z components of angular momentum and mag-
versity of Leiden.
**¹Society of Fellows. netic moment, respectively, there are 2I+ 1
 BLOV. MBERGEN, l u RCI-. LL, ANr~ l OUNr~

levels spaced in energy by pHD/I. The state of direct method, in the case of the wire, would be
this (non-interacting system) can be described the measurement of the temperature of the
by the magnetic quantum numbers mj m; wire, as a function of time, after switching ofF
~
I
m~, where — «& nz; «& I. Upon the application the heating current. The exact nuclear absorption
of a suitable oscillating magnetic field, transitions analog of this experiment has also been carried
corresponding to stimulated emission (hm, = + 1) out, and aR'ords a striking manifestation of the
and to absorption (hm, = —1) will occur. If relaxation process.
transitions of the latter sort are to preponderate, It may be said at once that nuclear spin-lattice
so that there is a net absorption of energy from relaxation times appear to range between 10 '
the radiation field, it is essential that there be and 10' seconds. Perhaps it seems remarkable at
initially a surplus of spins in the lower states. first that times so long should be associated with
This condition mill be attained eventually if any atomic process. On the contrary, such times
there is some way by which the spin system can were, from the theoretical point of view, unex-
interact with its surroundings and come to pectedly short, for the processes which had been
thermal equilibrium at a finite temperature. At examined theoretically' prior to the first experi-
equilibrium the population of the 2I+1 levels ments in this field yielded much longer relaxation
will be governed by the Boltzmann factor times. In the first experimental detection of
exp(pHom;/IkT), and the requisite surplus in nuclear resonance absorption in a solid, ' it was
lower states will have been established. shown that the relaxation time for protons in
The exposure of the system to radiation, with paraffin, at- room temperature and 7000 gauss,
consequent absorption of energy, tends to upset was certainly less than one minute. Subsequent
the equilibrium state previously attained, by work has shown that T~ for this substance, under
equalizing the population of the various levels. those conditions, is in fact very much shorter.
The new equilibrium state in the presence of the Additional information on nuclear relaxation
radiofrequency field represents a balance between times has come from the work of Bloch and
the processes of absorption of energy by the collaborators, ' and of Rollin4 5 who was the first
spins, from the radiation field, and the transfer to investigate the phenomenon at very low
of energy to the heat reservoir comprising all temperatures. The work here reported began as
other internal degrees of freedom of the substance a more or less cursory survey of typical sub-
containing the nuclei in question. The latter stances and grew, as the theoretical interpreta-
process involves what we shall call for short the tion of the results progressed, into a systematic
spzn-4ttice interaction, and is described by a study of certain types of structures which forms
characteristic time, T~, the spin-lattice relaxation the basis for a comprehensive theory of nuclear
time. It is this time also which measures how magnetic relaxation applicable to liquids, some
long one must wait, after the application of the gases, and certain types of solids.
constant field Ho, for the establishment of It is obvious that the model of non-interacting
thermal equilibrium. spins described above is incomplete, for there
It should be evident that the above-mentioned will necessarily be interaction among the spins
competition between resonance absorption and due to the associated nuclear magnetic moments,
spin-lattice interaction provides a way of meas- not to mention other possible sources of magnetic
uring T&. In a precisely similar way one might interaction such as electronic states of non-
study the heat transfer between a wire and a vanishing magnetic moment. We shall continue
surrounding bath by examining the temperature
of the mire as a function of the power dissipated 'W. Heitler and E. Teller, Proc. Roy. Soc. 155A, 629
in the wire because of a current Rowing through {19363.
~ E. M. Purcell, H. C. Torrey, and R. V. Pound,
Phys.
it. In our case the resonance absorption itself Rev. 69, 37 (1946).
3 F. Bloch, W. W. Hansen, and M. Packard, Phys. Rev.
supplies a convenient thermometer, for the
VO, 474 (1946).
intensity of the absorption, depending as it does 4 B. V. Rollin, Nature 15S, 669
(1946); B. V. Rollin and
upon the surplus of spins in lower states, re8ects J. ~Hatton, Nature 159, 201 (1947).
B. V. Rollin, J. Hatton, A. H. Cooke, and R. J. Benzie,
the "temperature" of the spin system. An equally Nature 1M, 457 (1947).
 NUCLEAR MAGNETI C RESONANCE ABSORPTION

temporarily to ignore interactions of the latter The spin-spin interaction is of course incapable
sort, as well as possible electric quadrupole eSects of bringing about heat transfer between the spin
— it will turn out that we are justi6ed in doing system as a whole and the thermal reservoir, or
"
so for a suf6ciently important class of substances "lattice. However, the spin-spin and spin-
— but we must certainly take into account the lattice interactions are not unrelated, for each
nuclear dipole-dipole interaction. involves the existence at a nucleus of perturbing
The interaction energy of neighboring dipoles fields. Therefore our attention has been directed
separated by the distance r is p'/r' in order of to both eHfects. Broadly speaking, and antici-
magnitude, and if Hay))y'/r'we may expect, since pating results described further on, the perturb-
more remote neighbors are relatively ine8'ective, ing fields responsible for spin-lattice relaxation
that the result of the interactions will be a originate in the thermal motion of neighboring
broadening of each of the 2I+1 levels mentioned magnetic dipoles, whereas the spin-spin inter-
earlier. From a point of view which is somewhat action, although it may be modified by such
naive but useful within limits, we may say that motion, is not dependent upon it for its existence,
the total magnetic field in the s direction at the and is in fact strongest when such motion is
ith nucleus is Ho plus a "local" held Hi. of
order of magnitude
. absent.
The picture which has been sketched in these
introductory remarks will be recognized as an
approximation, obtained by starting with iso-
and that the resonance condition for a transition lated non-interacting spins, then introducing the
Dm; = I is kv = (Ho+H~„)p/I. One would thus
+ interaction as a perturbation, but clinging still
be led to expect the width of the absorption line, to a description in terms of individual m s. That
expressed in gauss, to be of the order p/r'. the picture proves as useful as it does can be
j
I aking another point of view, a spin causes at
its neighbor, i, an oscillating field at the reso-
attributed to the fact that the spin-spin inter-
action is weak compared to Hop and of relatively
nance frequency, because of its own precession short-range character.
about Ho. This fieM is capable of inducing Finally, we review brie0y the phenomeno-
transitions in which i and j
exchange energy.
The lifetime against transitions is about hr'/p, ',
logical theory of magnetic resonance absorption,
before describing the experimental method. The
leading again to a line width, expressed in gauss, phenomenon lends itself to a variety of equiva-
of p/r' Bloch.' introduced the characteristic time lent interpretations. One can begin with static
T, =br'/p' to describe the spin-spin interaction, nuclear paramagnetism and proceed to para-
thus recalling the second process just mentioned. magnetic dispersion or one can follow Bloch's
We were accustomed to use the local 6eld as a analysis, contained in his paper on nuclear
measure of the interaction, which suggests the induction, ' of the dynamics of a system of spins
6rst process. Of course, both effects are present in an oscillating field, which includes the absorp-
ordinarily and are included in any complete tion experiments as a special case. We are
theory of the spin-spin interaction. $ In any case, interested in absorption, rather than dispersion
the examination of the line width is seen to be or induction, in the presence of weak oscillating
important in the experimental investigation of fields, the transitions induced by which can be
this interaction. One of the earliest results of regarded as non-adiabatic. We therefore prefer
resonance absorption experiments was the then to describe the experiment in optical terms.
surprising one that the line width of the proton Consider a substance containing, per cm', No
resonance in liquids is, in general, much less than nuclei of spin I
and magnetic moment p, , placed
p/r in magnitude, a result which receives quanti- in a strong uniform magnetic field Ho along the
tative explanation in the work here reported. z axis, and subjected to a weak oscillating field
F. Bloch, Phys. Rev. VO, 460 (1946}. II, =2H~e '"'' H =0. The probability of a single
f That the two eSects are not one and the same can be transition in which one of the m s changes to
seen by considering a lattice of non-identical nuclei, for
which the second effect is absent but the "dispersion of
local z-6elds" remains. ~ C. J. Gorter and L. J. F. Broer, Physica 9, 591 (1942).
 HLOEM BERGEN, PURCELL, AND POUND

m can be found with the aid of the standard The maximum energy density in the oscillating
formula for magnetic dipole transitions: field is(2Hi)'/Ss. The apparent "Q" of the
Wm;-m =(Ss'/3h') [(m;(M[m, ') )'p. . (1) sample, at frequency v, is therefore given by

M is the magnetic moment operator. Ordinarily (1/Q) = P / vHP = ymhmNOI(I+ 1) vg (v) /6k T. (7)
p„represents the energy density, fre-
in unit
quency range, in the isotropic unpolarized radia- An equivalent statement is that z", the imagi-
tion 6eld. We have to do here with radiation of nary part of the magnetic susceptibility of the
a single frequency from levels of a finite width, substance, is
which we describe by the observed shape of the
1 s Nay'h'I(I+1)
absorption line, g(v). The shape function g(v) is x" = vg(v) (8)
to be normalized so that 4sQ 2 3kT

g(v)dv=1. where h=h/2s. The term in brackets will be

recognized as the static susceptibility go of the
Our radiation field is also unusual in that it spin system. It may be noted in passing that
consists simply of an oscillating magnetic 6eld Eq. (8) is compatible with, and could have been
of one polarization. Only one of the circularly obtained from, the Kramers formula connecting
polarized components (of amplitude Hi) into the real and imaginary parts of the susceptibility,
which this held can be decomposed is e8'ective. which for this special case is
'I he equivalent isotropic unpolarized radiation
2 p" x"(v)dv
density is easily arrived at, and we have for p„ Xo
X~0
p„=3Himg(v)/4s. (2)
Introducing the nuclear gyromagnetic ratio y, If our earlier estimate of the line width is
which will be used frequently hereafter, correct, the maximum value of g(v) should be of
y =2sp/Ih, (3) the order of ms. gnitude hr'/p', 'if so, the maxi-
mum value of x" to be expected, according to
and the matrix elements required in (1), we find
Eq. (8), is of the order of hv/k T, since No = 1/r'
W „ i —(s/3)p'(I+m) (I —m+1) p„ A more accurate estimate based on the theory
= y'HPg(v) (I+m) (I—m+1).
-',
(4) to be given later leads to the formula:
Equation (4) gives the probability for a transi-
tion m~m —1, involving the absorption from
x",„=0.2(hv/kT) LI(I+1)j&.
the radiation field of the energy hv=hyHO/2s. The numerical factor in (10) depends somewhat
If the spin system is initially in equilibrium at on the geometrical arrangement of the nuclei.
the temperature T, the population of each level For v = 10 Mc/sec. , and room temperature,
m exceeds that of the next higher level, m — I, by hv/kT = 1.64 X 10~. If the line is much narrower
than the above estimat- -as we shall find it to
S —S &0 hv
—
be in many cases the maximum value of x" is
2I+1 kT correspondingly larger. The experimental de-
'I'he approximation
(5) is an extremely good one, tection of the absorption thus presents the
for, in the cases with which we shall deal, problem of detecting a small change in the
hv/kT=10 e. The net rate at which energy is susceptibility of the core of a radiofrequency
absorbed is now coil. The absorption is, of course, accompanied
by dispersion, the maximum value of x' being
of the same order of magnitude as the maximum
value of y". The dispersive properties of the
medium, moreover, are those of a substance
y'Hi2Np(hv)'I(I+ 1)g(v)
(6) displaying the Faraday efFect, except that the
6kT efFect is greatly enhanced by resonance. Ad-
 NUCr EAR MAGNETIC RFSONANCE ABSOR. PTrON

vantage is taken of this in the nuclear induction terms of a quantity T2* defined by
method, but not in our experiments.
The formula (8) applies regardless of the origin
2p*= pg(~) . (14)
of the line broadening„even if the line width and The maximum steady-state susceptibility in the
shape are controlled, as is often the case, by presence of the r-f field is thus reduced, relative
accidental inhomogeneities in the supposedly to its normal value, by the "saturation factor"
uniform Hp, prosidsd that the original distribu- [1+y'HiPTiTp"] '. The possibility that Tp~
tion of population among the levels remains itself depends on Hj is not excluded.
substantially unaltered. In other words (8) is The quantity Tp* defined by (14) is a measure
correct for vanishing H~. As II~ is increased, the of the inverse line width. It is strictly connected
thermal contact between spin system and lattice only with the area and peak value of a line-shape
eventually proves unable to cope with the energy curve. In the case of the damped oscillator or
absorbed by the spin system, the spin tempera- single-tuned-circuit curve, 1/prTp" is the full
ture rises, and the relative absorption, measured width between the half-maximum points, on a
by y", diminishes. It is the onset of this satura- frequency scale. We use the notation 12* when
tion effect which has been used to measure the admitting all sources of line broadening, includ-
spin-lattice relaxation time in most of our work. ing inhomogeneity in the magnetic field. The
The interpretation of the data is complicated by symbol T2 will be used to specify the true line
the necessity of distinguishing various cases width which is observable only if the field is
according to the factors which control the line sufFiciently homogeneous.
width, and a detailed discussion of this problem
II. APPARATUS AND EXPERIMENTAL
will be postponed. However, in order to introduce PROCEDURE
at an early stage our working definition of the
spin-lattice relaxation time, we shall now investi- The substance to be investigated was located
within a small coil, in the field of an electro-
gate the occurrence of saturation in a simple
case. magnet. Resonance absorption in the sample
Assume that I=2, so that we have to deal
caused a change in the balance of a radio-
with two levels. Let n denote the surplus popu- frequency bridge containing the coil. In much
lation of the lower level: n=X+q —
of the work a frequency in the neighborhood of
X q, and let
to thermal 30 Mc/sec. was used. This is a convenient
no be the value of n corresponding
region in which to study the H' and F" reso-
equilibrium at the lattice temperature. We
nances which occur at 30 Mc/sec. in fields of
assume that, in the absence of the radiofrequency
7050 and 7487 gauss, respectively. Some experi-
field, the tendency of the spin system to come to
ments were made at 14.5 Mc/sec. and at 4.5
thermal equilibrium with its surroundings is
described by an equation of the form Mc/sec. Considerations of sensitivity alone favor
the highest frequency compatible with available
dn/dt = (1/Ti) (np —n) . (11) magnetic field intensities and the gyromagnetic
The assumption will be justified in Section VII. ratio of the nucleus in question. (See Appendix. )
The characteristic time Ti in Eq. (11) is called A typical r-f coil consists of 12 turns of No. 18
the spin-lattice relaxation time. (Wherever copper wire wound in a helix of inside diameter
possible our definitions and nomenclature follow 7 mm, and 15 mm long, The coil is sometimes
those of Bloch. )' supported on a thin-walled polystyrene form,
The presence of the radiation field requires the but this is not always desirable for, in addition
addition to (11) of another term: to reducing the space available for the sample,
—2nW— i the coil form, if it contains hydrogen, gives rise
dn/dt=(1/Ti)(np n) i. (12)
to an absorption at the proton resonance. $$ The
A steady state is reached when dn/dt=0, or,
using Eq. (4), when
ff An excellent insulating material of negligible hydrogen
content is Teflon, the Dupont fluorine-substituted polymer.
—[1+$y'HPTig(v)] '. %e have not been able to detect a proton resonance in
n/np (13) Teflon. Its mechanical properties, however, are not ideal
for a coil form. Glass would be acceptable, as would
We shall express the maximum value of g(v) in porcelain.
 BLOEMBFRGEN, PURCFLL, AND POUND
coil is surrounded by a cylindrical shield of 17 doubtless arose from magnetic inhomogeneities
mm inside diameter, one end of the coil being in the cold-rolled steel pole caps. More careful
connected to the shield. A hole in the end of the attention to the pole cap material, in future
shield can permits insertion of the sample which, work, is to be recommended. The best position
if a liquid or powder, is contained in a thin-walled in the field was found by exploring with a liquid
glass tube. The effective volume of the sample sample. As will be seen later, the line width in
within the coil is approximately 0.5 cm. It water, for example, is essentially determined by
mould be advantageous to work with larger the field inhomogeneity, and therefore affords a
samples; in the present case the limitation is sensitive test for homogeneity. The location of
imposed by the midth of the magnet gap. For the flattest spot in the field varied somewhat
experiments requiring lowering of the tempera- with field intensity, which is evidence for ferro-
ture, copper tubing was wound around and magnetic inhomogeneity of the poles. The mag-
soldered to the shield can, and acetone cooled by net current was supplied by storage batteries and
dry ice was allowed to flow through the cooling regulated by suitable rheostats, the main rheo-
tube at a controllable rate. This rather clumsy stat consisting of some 15 feet of heavy Advance
arrangement was also necessitated by lack of strip, with a sliding contact. It will be readily
space in the magnet gap. appreciated that the examination of resonance
The magnet is a Socihth Genevoise water- lines a few tenths of a gauss in width in fields of
cooled electromagnet provided with pole pieces several thousand gauss requires an unusually
4 inches in diameter at the gap. The gap width stable source of current. The current could be
was adjustable and was ordinarily between monitored by means of a shunt and a type K
and 1 inch. The pole faces were Bat except for potentiometer. The power required for magnet
a raised rim 0.015 inch high and 0.31 inch wide excitation at 7000 gauss with a 0.75-inch gap
at the periphery of each, which was designed was about 220 watts.
according to the prescription of Rose' to com- A sinusoidal modulation of small amplitude,
pensate the inherent radial inhomogeneity of at 30 c.p. s. , was applied to the magnetic field by
the simple gap. The residual inhomogeneities means of auxiliary coils around the poles, or, in
were apparently of non-geometrical origin, and some cases, by means of a small pair of coils

4' I I
RAOIO I
805-C I I
50 i ~- iI I-
GENERATOR ' I
II
L J
II II
I I II
II II
COMMUN!"
I I PRE- CAT ION S
I I
Ol' J AMPLIFIER RECEIVER
II
I I
I
I' I i
I I I I
N I'Q S II I I
I I I
il ll II
I I 11
II

I
I I I I

II II I

I I I I II CATHOOE I
;'I
I I I I
mm Il RAY I

I
L--~: ii
I I
I
ii&- —--". --J II I
I
I
I
8 i CII Cg I I

so C I
Mi
—.OSCILLOGRAPH
n NARROW
I
I
I

I
30~.
I 3 I I AMPLIFIER
I I

I I BALANCED
I 30qps.
30qp, s. - -- POWER - -- PHASE
I
MIXER
GENERATOR AbilPLIFIER SHIFTER

FIG 1. Block diagram of circuit. The sample and the magnet gap are not drawn in correct proportion.
' M. E. Rose, Phys. Rev. 53, 715 (2938}.
 NUCLEAR MAGNETI C RESONANCE ABSORPTION 685

fitting closely against the shieM surrounding the

r-f coil. The 30-c.p. s. signal was derived from a
small permanent-magnet alternator driven by a
synchronous motor, and a 6L6 served as power
amplifier to drive the modulating coils. The
amplitude of modulation at the sample could be
made as large as 15 gauss peak-to-peak, but was
usually very much smaller.
A block diagram of the radiofrequency circuit
is shown in Fig. 1. The source of the r-f signal
was a General Radio 805-C signal generator.
The output of this instrument covers a con-
venient range in power, the maximum power
available being adequate to saturate most sam-
0 0
ples. The circuit elements are connected together (c) (b)
by standard u.h. f. coaxial cable (50 ohms). Each
of the two branches leading from a tee-junction FIG. 2. Diagram illustrating the dependence of the shape
of the resonance curve upon the nature of the residual
near the signal generator contains a resonant unbalance of the bridge. The diagram is not to scale;
circuit, one of which consists of the coil sur- actuallv, PP'«QP«OP.
rounding the sample and a fixed ceramic con-
denser shunted by a variable condenser of about The latter adjustment alone affects the phase
10 mrnf. The dummy circuit in the other branch balance as well as the amplitude balance, for
is similar in all respects but is not located in the the coupling condenser essentially appears in
field. %hen liquids of high dielectric constant shunt with the tuned circuit. The adjustable
are being investigated it is sometimes worth coupling condenser C4 therefore is ganged with
while to slip a dummy sample into the dummy a trimming tuning condenser C~ in such a way
coil, thus balancing, at least approximately, the as to leave the total tuning capacity unaltered
considerable change in stray capacitance result- during the change in coupling. In this way
ing from the insertion of the sample in the other substantially orthogonal phase and amplitude
coil. adjustments are provided, and the balancing of
An extra half-wave-leng'th of line AB in one the bridge is a simple operation.
branch makes the point D, where the branches In searching for, or examining, a nuclear
join to the line to the amplifier, a voltage node resonance the bridge is eever completely bal-
in the perfectly balanced condition. This is, of anced, but is intentionally unbalanced to such a
course, only one of many ways in which the degree that the voltage appearing at point D is
equivalent of a balanced-pair to unbalanced line large compared to the change in this voltage
transition, essential to any bridge, might be caused by nuclear absorption or dispersion. The
e8'ected, and simplicity is the only virtue claimed only benefits derived from the use of the bridge
for it. The Q of a coil such as that described is are (a) the reduction of the r-f level at the input
about 150, at 30 Mc/sec. , and the shunt imped- to the amplifier to a value low enough to permit
ance of the I.C circuits at resonance is of the considerable r-f amplification before detection,
order of 5000 ohms. The small coupling con- and (b) the reduction in the relative magnitude
densers C2 are chosen to make the impedance of output Quctuations arising from amplitude
seen looking back from the amplifier approximate fluctuations in the signal supplied by the signal
the value for which the amplifier noise figure is generator. With a rather moderate degree of
best. The main tuning capacity in each circuit balance these advantages are already fully ex-
is provided by a fixed ceramic condenser C~. ploited, and further balance does not improve
The bridge is balanced in phase and amplitude matters. We have ordinarily operated with a
by the adjustment of one of the trimming tuning balance of 40 db to 60 db, by which is meant
condensers C3 and the coupling condenser C4. that the voltage appearing at D is between 10—'
 8LOEMBERGEN, PURCELL, AN B POUN I3

Fxe. 3. Proton resonance (ab-

sorption curve} in ferric nitrate
solution.

and 10-' times the value it would have if one 6eld„ in the case of a proton resonance. Pure
branch were removed. phase unbalance, on the other hand, is repre-
It is important to be able to control the sented by the diagram of Fig. 2b, and the
character of the residual unbalance, for upon observed output variation is shown in Fig. 4.
this depends the nature of the signal observed (It must be remembered that actually the re-
as the magnetic held is varied through the region sultant QI" is very much smaller than OP, and
of resonance. The nuclear absorption has the the small circle is in turn very much smaller
eAect of altering the shunt resistance of the than the unbalance voltage QP. ) It is, in fact,
tuned circuit, while the accompanying nuclear the nuclear absorption, y", which is observed in
dispersion changes its resonant frequency slightly. Fig. 3 and the nuclear dispersion, y, which is
The two eA'ects can be described together by observed in Fig. 4. Either can be obtained as
considering, in Fig. 2, the vector OP' as repre- desired by 6rst balancing the bridge to a high
senting the signal arriving at the point D from degree and then setting in an intentional unbal-
I
the branch containing the C circuit with the ance by means of the phase knob or the ampli-
sample in it. As resonance is traversed the end —
tude knob hence the importance of independent
of the vector, P', describes a small circle. hatt' phase and amplitude controls. A "mixed" unbal-
Had the residual unbalance o8 resonance, QI', ance brings about an unwelcome mixture of the
referred to the point D, been in amplitude only, two e6'ects recognizable by the unsymmetrical
so that the signal from the other branch is shape of the output curve. We usually preferred
represented by the vector OP in Fig. 2, the to work with the absorption curve (amplitude
resultant voltage applied to the amplifier, QI", unbalance) because the apparatus is in this
would vary with magnetic field in the manner condition less susceptible to the frequency modu-
illustrated in Fig. 3 which is a trace of the lation present in the signal generator output.
receiver output after detection, vs. magnetic Indeed, absence of frequency modulation effects
in the output is a sensitive criterion for pure
ttf The locus of P' is strictly circular only if the shape amplitude unbalance. The trouble caused b&"
of the absorption curve is that appropriate to a damped
oscillator. frequency modulation is, of course, mainly due
 NUCLEAR MAGNE'I IC RESONANCE ABSORPTION

to the frequency sensitivity of the half-wave following the r-f amplifier and detector, a narrow
line, and could be mitigated by the use of a band amplifier and balanced mixer, a combina-
suitably designed transformer to achieve the tion variously known as a "lock-in" amplifier or
required f 80' phase shift. a phase-sensitive detector. The magnetic held
The r-f amplifier is subject to no special modulation is reduced to a fraction of the line
requirements other than the desirability of a wid th, and the resulting 30-c.p. s. modulation
good noise figure. We have used commercial of the detected receiver output is ampli6ed and
receivers, both Hallicrafters S-36 and National converted by mixing it with a 30-c.p. s. reference
HRO, preceded, when the signal frequency is signal derived from the small alternator. There
near 30 Mc/sec. , by a broad band preamplifier is thus obtained a d. c. output proportional to
of Radiation Laboratory design' which has a the slope of the nuclear absorption curve, and
much smaller noise figure than either of the this is indicated by a meter M3. The relation
commercial receivers. A temperature-limited between meter current and dx"/dH is strictly
diode in a circuit for noise comparison (s"own in linear for sufficiently small amplitudes of mag-
Fig. 1) has been included as a permanent fixture, netic field modulation, because of the relatively
and calibration by this means showed the noise large unbalance voltage in the bridge and the
figure to be 1.5 db when the preampli6er was nature of the lock-in amplifier. The effective
in use. band width of the system is determined by the
Relatively strong signals, such as that dis- time constant of the final d. c. circuit containing
played in Fig. 3, can be observed on an oscillo- the meter, which was usually about 1 second but
scope by merely connecting the detector output could be increased to 10 seconds if desired. The
to the vertical deQection ampli6er, and synchro- amplifier was copied after one designed by
nizing the sweep with the 30-c.p. s. modulation R. H. Dicke. 'o
of the magnetic 6eld, the amplitude of which is The superiority of the narrow-band-amplifier-
adjusted to several times the line width. For plus-meter over the oscilloscope in the matter of
most of our work, however, we have used, detecting weak signals is not as marked as

I IG, 4. Proton resonance (dis-

persion curve) in ferric nitrate
solution

G. E. Valley and H. %'allman, Vacuum Tube Amplifiers, Radiation Iaboratory Series {McGraw-Hill Book Company,
Inc. , New York, in press), Vol. 18.
~o R. H. Dicke, Rev. Sci. Inst. 17, 268 (1946).
 BLOEM BERGEN, P URCELL, AND POUN D

comparison of the band widths alone would gave consistent results. The relation between
suggest, because of the new well recognized H and the voltage applied to the modulating
ability of the observer to integrate noise. How- coils was so nearly linear that a single calibration
ever, for the quantitative examination of weak usually su%ced for a given setting of the gap
lines the narrow-band device is nearly indis-width.
pensable. The band width of the system must, of course,
It will be noted that the output of the lock-inbe adequate to permit a faithful reproduction of
amplifier, in the case of amplitude unbalance, the line shape at the modulation frequency used.
reproduces the derivative of the absorption curve The r-f field applied to the sample must be weak
(which resembles qualitatively a dispersion enough to avoid saturation. If the line is very
curve), while with phase unbalance one obtains narrow a complication can arise from a transient
the derivative of the dispersion curv- ---a large eR'ect to be described brieRy later. The vestige
peak Ranked symmetrically by two dips. of such an efFect is responsible for the asymmetry
at the base of the absorption line in Fig. 3.
III. MEASUREMENT OF LINE VfIDTH
For the study of broad, and hence weak,
When the resonance absorption is intense absorption lines the narrow-band amplifier is
enough to be displayed on the oscilloscope screen used. The modulation amplitude H„ is reduced
the line width can be measured directly on the to a fairly small fraction of the line width, and
trace. A sinusoidal horizontal sweep synchronous the line is explored by slowly changing the
with, and properly phased with respect to, the magnetic field Ho. It usually suffices to note the
modulation of the magnetic held, provides a interval in Hp between points giving extreme
horizontal scale linear in gauss. We denote by deflection of the output meter M3, which are
H the modulation amplitude, and by + the assumed to be the points of maximum slope of
angular frequency of modulation, in these experi- the absorption curve. A plot of the output meter
ments 2s. X30sec.-'. That is, H. =HO+H„sin ur„t. readings in one case is shown in Fig. S. In this
To establish the scale against which the line example, the interval between extreme deRec-
width is measured, H must be known. This tions is AH=0. 5 gauss (curve a). The relation
calibration can be efFected in several ways. The between ddS so determined and T2 defined by
signal generator frequency can be altered by a Eq. (14) would be (1/T2) =(VS/2)yhH, for a
small, known amount, and the resu1ting shift of damped oscillator absorption curve, or (1/T2)
the resonance within the modulation interval = (2s) &ydLH, for a Gaussian curve. The deriva-
noted, or the alternating voltage induced in a tive curve (a) in Fig. 5 indicates an absorption
sma11 coil of known turns&area, at the location line shape of intermediate character. The most
of the sample, can be measured with a vacuum- conspicuous evidence for this is the ratio of the
tube voltmeter. Both methods were used and maximum negative to the maximum positive

(0) UNSATURATED
(b) SATURATED

Frc. 5. Plot of the reading of

the output meter 3III, for a O. SN
solution of ferric nitrate. Curve
(b) was taken at a high power
level and shows the effect of
saturation. The value of the
magnetic field at the line center
was not measured, but merely
computed from the frequency
and the accepted value of the
proton moment.

68I9 6820
Ho GAU$$
 NUCLFAR MAGNFTI C RF SONANCE ABSORPTION 689

slope, which should be 4:1 for the derivative of

a damped oscillator curve, 2.2:1 for the deriva-
tive of a Gaussian, and is in fact about 3:1.
The effect of saturation, evident in the curve (0) fd)

(b) in Fig. 5, is to flatten and spread apart the

extrema of the derivative curve. This comes
about because saturation is more nearly complete
at the center of the line.
A
(b)
IV. MEASUREMENT OF RELAXATION TIME
Two methods by which the spin-lattice relaxa-
tion time T&, defined by Eq. (11), can be meas-
ured have been mentioned in the introduction.
(c)
The direct observation of recovery from satura-
tion was convenient only when Ti was of the Fro. 6. The eHect of local saturation upon a line whose
width is caused by magnetic field inhomogeneity.
order of magnitude of one second; doubtless a
wider range could be covered with suitable
circuit modifications and recording instruments. inhomogeneity of the field Hp. The absorption
The direct method has the advantage that the observed at A was caused by nuclei in one small
interpretation of the results is simple and entirely 8
part of the sample, that at by those in another
unambiguous. The saturation-curve method is part. %'ith the removal of the modulation the
applicable over a much wider range, but the former group was subjected to radiation at its
analysis of the experiment is rather involved resonance frequency all the time, and was more
and absolute values of Tj derived in this way or less thoroughly saturated. The subsequent
are probably less reliable. The two methods recovery from saturation occurred when the
complement one another, as we shall see. modulation was restored. The gash in the curve
Before analyzing special cases, we shall de- betrays, quite literally, a hot spot (in respect to
scribe an instructive experiment which vividly nuclear spin temperature) somewhere in the
demonstrates the saturation phenomenon and sample.
also exposes the source of some of the complica- If the modulation amplitude H is reduced to
tions which are encountered in applying the a small but finite value, rather than zero, the
saturation-curve method to narrow resonance r-f amplitude H~ being meanwhile increased
lines. somewhat, subsequent observation will disclose
The proton resonance in water was observed a hole like that in Fig. 6(f), which then heals at
on the oscilloscope with a weak (non-saturating) the same rate as before. The double dip results,
r-f field. The line appeared as in Fig. 6(a). The of course, from the sinusoidal modulation a —
modulation amplitude H was then reduced to longer time was spent at the turning points C
zero for a few seconds, while the field Hp was and D than at the middle of the hole. f
maintained at a value corresponding to the If the attempt is made to produce the effect
point A; the line was, of course, not visible on shown in Fig. 6 in a substance for which the line
the screen during this interval. Upon turning width is determined by a strong spin-spin inter-
the modulation amplitude back to its original action, rather than field inhomogeneity, it is
value, and restoring the sweep, the line reap- found that the whole curve saturates; it is not
peared bearing a deep, narrow "gash" at the possible to "eat a hole" in the curve. The energy
position A, as in Fig. 6(b). This hole rapidly fille
f. It was the observation of the e8ects just described
up and in a few seconds was gone, as shown in which first made us aware that the true line width in water
the sketches (c) to (e). The explanation of the was much less than the apparent line width. By exploring
the magnet gap, a spot was found where the field was con-
effect is this: the true width of the water reso- siderably more homogeneous and the line correspondingly
nance line was much less than the width of the narrower. The narrowest line obtainable in our magnet at
/000 gauss was about 0.1 gauss wide; at 1100 gauss a
original curve, which was caused entirely by width of 0.015 gauss was achieved in one case.
 BLOK M BERGEN, P URCELL, AN D POUND

absorbed by the spins can no longer be localized, think of IIO as 6xed at the central value Ho~,
but goes to raise the temperature of the spin with ou as the variable. The modulation H
system as a whole. In other words, the applica- sinu t is equivalent to a frequency modulation
tion of an r-f 6eld at a single frequency, with Ho of amplitude yH radians/sec. If the ampli-
fixed, immediately (or at least in a time shorter tude of the modulation is small enough, yII
than Ti) afkcts a region about 2/yTp wide on a (1/Tp, the deflection of the output meter Mp
magnetic field scale, or about 1/prTp wide on a will be nearly proportional to Bx"/BHp, or, in
frequency scale. It is important to keep this terms of the equivalent frequency modulation,
point ln mind. to Bx"/Ba&. The question is, what is to be re-
In any case, it is possible to saturate all parts garded as constant in interpreting Bx"/Bpp in
of the sample by modulating through the whole the presence of saturation.
line, using the highest available r-f 6eld intensity. KVe 6rst consider cases in which the inhomo-
The recovery, observed with a much weaker r-f geneity of the magnetic held, denoted roughly
Field in the sample, appears to follow the simple by bII„has a negligible influence on the line
law (1 — s ""), and we identify ip with Ti, the width. That is, yhH«1/Tp. The apparent sus-
spin-lattice relaxation time. To make a quanti- ceptibility of the sample in the steady state
tative measurement, the oscilloscope trace was (no modulation) at the frequency cubi, relative
photographed with a movie camera operating at to its unsaturated value at coo, the line center, is
a known frame speed, and the 61m analyzed to
determine the time constant of recovery. This is x"(~i, Hi)/x" (~p, 0)
the direct method referred to above. It was =(1+x')-'LI+s/(1+x')]-'=f(x)$(x), (16)
applied to a distilled water sample, to petroleum evaluated at x = xi, were x = (co pip) Tp a— nd
ether, and to a 0.002K solution of CuSO», s = y'HiP Ti Tp. The factor (1+s/(1+ x') $ ' = S(x)
yielding relaxation times of 2.3~0.5 sec. , 3.0 is the saturation factor given by Eq. (13). We
+0.5 sec. , and 0.75&0.2 sec. , respectively. are interested in Bx"/Bx, and, in particular, we
The saturation curve method is based upon ask for the extreme values which the derivative
Eq. (13) in the Introduction. If it were possible attains as x~ is varied. These will be assumed
to measure, in the steady state, the value of
proportional to the extreme meter deflections,
which is reduced from its normal unsatu-
D, , indicated on Fig. 5. There are two cases to
rated value by the factor (1+yPHiPTiTp)-', a
be distinguished according to the relative magni-
knowledge of Hi and T2 would then su%ce to
tude of pp and 1/Ti. If co Ti«1, the saturation
determine Ti. In the modulation method, how-
factor $(x) will vary during a modulation cycle.
ever„we measure By"/BHp, which .is equivalent
If eu„Ti»1, 5 cannot change during the modu-
to measuring Bx "/Bip. This introduces some
lation cycle and may be assigned a value appro-
complications which can be treated brie8y only
priate to the center of the modulation swing.
by con6ning the discussion to typical limiting
The meter deflection D is therefore proportional,
cases.
in the two cases, to difkrent functions of x~,
We assume for analytical convenience that
the shape of the unsaturated absorption curve is namely:
that appropriate for a damped oscillator, Case 1

g(~) =
2T2
(15) (~ Ti&&1), D(xi) ~ —
8
Lf(x)$(x) j
1+»w (p —pp) Tp 1+((al ipp) Tp 8x S=g]
Case Z
In Eq. (15), happ, the center angular frequency, is
"$(»)—
WHO . Although it is actually H, which is varied
8
in the course of the experiment, both by applying
(~ T»&1) D(») f(x)
l9x
the modulation II since t, and by slowly changing
the magnet current in order to move through D. , the extreme excursion of D, is found by an
the resonance curve, it is convenient here to elementary computation to depend in the follow-
 NUCLEAR MAGNETI C RESONANCE ABSORPTION 691

ing way upon the saturation parameter s: amplitude, H, is set at a suitable value and
remains the same throughout a run. The output
Cuse l voltage of the signal generator, to which H~ is
D ~ (1+s)-i; (17)
proportional, is varied in steps by means of the
Case Z
attenuator in the instrument. At each step the
D, 0: 8[2(16+16s+s')&—2s —4]& receiver gain is adjusted to bring the rectified
X [8+8s —s'+ (2+s) (16+16s+s') &)—'. (18) r-f output of the receiver to a fiducial level as
indicated by the voltmeter 352, the bridge bal-
The functions (17) and (18) are plotted in Fig. 7.
ance remaining unaltered. This makes the con-
The saturation effect makes itself evident more
stant of proportionality connecting the deflection
rapidly in Case l, as was to be expected. A
of the final output meter with By"/BHO the same
different line shape would, of course, lead to
for each step in II~. At each signai generator
somewhat diR'erent results.
output setting, after the receiver gain has been
Suppose now that the field is not homogeneous,
readjusted, the resonance curve is slowly tra-
and, in particular, that H &1/yT2 &BH, which
versed and the maximum and minimum readings
we shall label Case 3. The result is then simply
of the output meter noted. The difference be-
a sNPerPosifiow, governed by BH, of curves of the
tween the readings is then plotted against the
sort analyzed in either Case I or Case Z (de-
logarithm of the signal generator voltage.
pending on the magnitude of ur Ti). If the line
To interpret the saturation curves we assume
shape caused by the field inhomogeneity were
that a given degree of saturation is associated
known, the appropriately weighted superposition
with the same value of y'H~'T~T2, or, if the
of curves D(xi) could be constructed. It is clear,
conditions correspond to Case 4, of yHinTi/H .
however, that the decrease in D with s will
This is, of course, not strictly correct if we are
follow much the same course; D, will be sub-
comparing two saturation curves which corre-
stantially reduced when s is of the order of unity.
spond to two diA'erent cases, such as Case I and
Case 4, specified by 1/yT~ &H' &BH, and
Case Z, and allowance can be made if the
~ T»1, is significantly different. It is now the
precision of the data warrants.
modulation amplitude H rather than 1/yT~,
A series of saturation curves for Fe(NOS)3
which determines the width of the region mo- solutions of various concentrations is shown in
mentarily saturated. The situation is, in fact, Fig. 8. This is the roughest set of curves of any
that illustrated by Fig. 6(f). An analysis of this used in deriving the results reported in this
case, which is only feasible if certain simplifying
assumptions are made, shows that the quantity 10
yHi'Ti/H , rather than y'H'i'TiT~, plays the
role of the saturation parameter, a rapid decrease
in D, occurring when (yHi'Ti/H~) = 1. The
indicated dependence of H~ upon H, for the
same degree of saturation, has been confirmed
experimentally. Case 4 is encountered in liquids Dex

with relatively long relaxation times. The meas- .5

urements of D, are particularly diScult here

for the line must be traversed at an extremely
slow rate in order to permit a quasi-stationary
state to develop at e~ery value of x~.
An even more complicated case is defined by
I/yT~&H &BH with ro Ti&1. The discussion
of this case will be omitted as it was not an
important one experimentally.
An experimental saturation curve is obtained FIG. 7. Variation of the extreme output meter deQection
with the saturation parameter s, or y'HQT1T~, for a per-
by the following procedure. The modulation fectly homogeneous Reld.
 8LOEMBERGEN, Pt,'RCF. L1. , 3't||, 8 l O tt NI~

paper, and is exhibited to indicate the experi- in Ti, by large factors, which are most significant
mental limitations. At the same time, it i 11us- in the present investigation.
trates the enormous range which can be covered Discussion of the experimental results on
by the saturation-curve method. The horizontal relaxation time and line width is deferred to
shift of the curves is a measure of the relative Sections XI to XV which follow a theoretical
change in T~2 ~. Of course, T~ must be deter- examination of spin-spin and spin-lattice inter-
mined in each case by the methods already action.
described in order to deduce the relative change
in T~. Figure 9 shows a set of saturation curves V. A TRANSIENT EFFECT: THE "%mGLES"
for ice at various temperatures, together with Rather early in the course of this work a
the function (18) The agreement is good enough;
~
transient e8'ect was noticed which merits a brief
as a matter of fact, the difference between (17) description if only because it can complicate the
and (18), on a logarithmic plot, is quite small, interpretation of nuclear absorption experiments
apart from a lateral shift. All saturation curves in some cases. The efkct is observed when a
which we have observed bend downward with large amplitude of modulation is used to display
about the same slope. the whole absorption curve on the oscilloscope,
To determine the absolute magnitude of T~ and is prominent only when Ti, T~, and 1/ybH
from a saturation curve, 12 having been meas- are all greater than the interval spent at reso-
ured, the relation between H~ and the signal nance during the modulation cycle which is
generator voltage must be known. The relation roughly (II„ar GATI) '. An example is shown in
can be computed from the dimensions of the coil Fig. 10, a photograph of the oscilloscope trace
and the constants of the circuit, with which one in the case of a w ater sample in a very homo-
must include the stray capacitances. We have geneous magnetic field. The prominent "wiggles"
preferred to rely on the direct measurement of occur always after the magnetic field has passed
T~ for this calibration. If higher precision were through resonance, and therefore appear on
required, it would be necessary to make in each opposite sides of the two main peaks in Fig. 10,
case a detailed analysis like that given above for one of which corresponds to increasing, the
the two idealized cases 1 and Z. The accuracy other to decreasing, 6eld.
of the experimental data hardly justifies such a A brief explanation of the effect is most readily
re6nement. Fortunately, it is the relative changes given in terms of the classical model of precessing

I00

Fso. 8. Saturation curves for

ferric nitrate solutions rangining
in concentration from O.ON to
0.6¹ The ordinate is the ex-
treme output meter deflection,
in arbitrary units.

IO-3 IO 2 IO '
GENF&ATOR OUTPUT IN VOI„TS
 NUCLEAR MAGN ETI C RESONAN CE ABSORPTION

o -35'c
x-l7 C
a -5'c
lOO
*~
Fr@. 9. Saturation curves for
ice at -3$'C, —1F'C, and
-5'C. The broken curve is a
plot of Eq. (i8), and may be
shifted horizontally to 6t any of
the experimental curves.

lO-~ lP I

QFNFRATOR OUTPUT IN VOLTS

magnetic gyroscopes. Nuclear spins which have small that the effect was not important. The
been tipped over by the application of the r-f direct measurements of T~ were made in a 6eld
6eld at their precession frequency continue to suSciently inhomogeneous to suppress the
precess in a coherent way after the removal of wiggles, which would not, in any case, have
the r-f 6eld, or after the radiofrequency or interfered with the interpretation of the results.
magnetic 6eld has changed to a non-resonant
value. In our case the magnetic 6eld is changing, VI. SPIN-SPIN INTERACTION
and the nuclear precession rate, following it Of the processes which could conceivably bring
adiabatically, comes alternately in and out of about an exchange of energy between the spins
phase with the applied signal, whose frequency and their surroundings, we can at once discard
has remained at what was formerly the reso- the interaction with the thermal radiation 6eld.
nance value. The effectiveness of this process depends upon
The duration of the "phase memory" is limited the probability of spontaneous emission, A,
by T2. However, if 2/pbbs is smaller than T2, which is of the order of 10 ~ sec. ' for a nuclear
interference between the effects of nuclei in magnetic dipole transition at the frequencies
different parts of the sample smears out the with which we are concerned. The corresponding
wiggles. By slightly detuning the receiver the relaxation time is T~ —(kv/kT) (1/2A) = 10" sec-
wiggles on the forward trace can be enhanced onds. Although this figure must be reduced by a
and those on the back trace suppressed, or vice large factor when the nuclear system is coupleQ
versa; one set of wiggles is produced by spins to a resonant circuit of small dimensions, the "
precessing at a rate higher than vo and the other process is still inadequate to account for the
by precession at a lower frequency. The spacing relaxation effects observed and need not be
between the wiggles also varies in the manner considered further.
suggested by the above explanation. A more
re6ned treatment of the problem, which is most ~ E. M. Purcell, Phys. Rev. 69, 681 (1946). In a sense,
conveniently based on the equations of Bloch, the relaxation of nuclear spins in a metal, which was con-
sidered by Heitler and Teller, reference 1, is a limiting
has been made by H. C. Torrey" case of the coupling of the spins to an electrical circuit,
In our measurements of r~ by the saturation and as such requires an exception to the above dismissal
of radiation damping effects. Although no experimental
curve method, the modulation amplitude was so evidence on nuclear relaxation in a metal has appeared,
it is to be expected that the mechanism treated by Heitler
"H. t . Torrey (private communication). and Teller wi11 play an important role.
 BLQEMBERGEN, PURCELL, AND POUND

FIG. 10. An example of the

transient effect described in
Section IV, observed in water
at 29 Mc/sec. The 6eld modu-
lation and the oscilloscope
sweep are both sinusoidal, but
there is a small phase dif-
ference between them. The
wiggles occur after the mag-
netic 6eld has passed through
resonance.

Electric forces which act during atomic colli- lined in the introduction in which the system of
sions directly perturb the nuclear spins only by spins is characterized by individual spin quantum
virtue of an accompanying variation in the numbers, it is instructive to examine the problem
gradient of the electric 6eld at the nucleus and from a somewhat more general point of view
the existence of a nuclear electric quadrupole by regarding the entire lattice of X spins as a
moment. Direct electrical interactions are thus single quantum-mechanical system. The total
reduced to a rather minor role, quite in contrast s component of angular momentum is speci6ed
to the situation in atomic and molecular spectro- by mh. The Hamiltonian of the system is
scopy, and are entirely absent if the nucleus has
no quadrupole moment. We shall examine one
or two cases in which the nuclear quadrupole-
inhomogeneous electric 6eld interaction is the
important one, but we have been concerned
chieRy with nuclei of spin ~ for which the The second term in (19) is the dipole-dipole
quadrupole moment vanishes. j
interaction, and the sum over -in this term is
Magnetic interactions will occur between the what we have called the local 6eld, Hi„.
In the
nuclear magnetic moments and magnetic mo- absence of the dipole-dipole interaction the sys-
ments associated with electronic states, and tem displays 2NI+1 discrete levels, m= — NI,
among the nuclear magnetic moments them- NI+1, +NI, — separated in energy by hyHO
selves. We shall consider the nuclear dipole- The levels are highly degenerate; for I=-,', for
dipole interaction first, and in greatest detail, example, the degeneracy is 2~N!/(N — 2m) !
not only because it proves to be the dominant X(N+2m)!. The interaction term splits each of
efI'ect in most diamagnetic substances, so far as these levels into a group whose width in engery
spin-lattice relaxation is concerned, but because is of the order, ¹hyHi. , Now ¹yHi. &&WHO
in a practical case, so that the original level
.
it is entirely responsible for the spin-spin relaxa-
tion phenomena. In fact, we shall begin by structure dissolves into what is essentially a
continuum showing none of the previously
considering a spatially rigid lattice of dipoles-
a model necessarily devoid of spin-lattice efkcts. "
prominent periodicity. Broer, who treated in
Before elaborating upon the description out- " L. J. F. Broer, Physica 10, 80j. (1943).
 NUCLEAR MAGNETIC RESONANCE ABSORPTION

this way the general problem of paramagnetic and jth spin; e&, 0.2, and es are the direction
absorption and dispersion, showed that the cosines of the radius vector r in coordinates with
calculation of the absorption for such a system, the z axis parallel to Hp. We now rearrange
assumed to have been brought somehow into Eq. (20) so as to distinguish terms which leave
thermal equilibrium at a temperature T&&hv/k, the total z component, m, unchanged, terms for
could be reduced to the task of 6nding the which 6m=~1, and terms for which des=+2
density function of magnetic moment, f(v), or — 2. This distinction is important because
which he dehned by: the terms involve a time factor e'~~&~o'. The
terms with 6m=0 are time independent and
v+ikv
leave the energy unchanged. We introduce the
polar and azimuthal angles 8;; and q;; instead of
the direction cosines, after which (20) can be
where IrI is the magnetic moment matrix, p and
written in the form:
g refer to two levels separated in energy by hv,
and p„, is the average occupation of these levels. V,;=y'h'r;; '(A+8+C+D+Z+F), (21)
The only rigorous procedure available at this
point is the method of diagonal sums which where
Broer shows is capable of yielding a limited A =Iz;I*,(1 —3 cos'8;;), (Am =0),
amount of information about the function f(v),
8 = —,' [(Iz, iI—v;)(Iz;—
+iIv;)
namely,
+ (I*;+iIv;)(Iz& iIv;) 5—
0
f(v)dv,
Jp
v'f(v)dv, etc. , X (1 —3 cos'8;;), (Am = 0),
C= —', [(Iz, +iIv;-) Iz+ (Iz, +iIv;) Iz~]
the successively higher moments being increas-
ingly di%cult to calculate. An example of the Xsin8;; cos8;;e '&'z'e''r~z', (Am = 1),
application of the diagonal sum method to (22)
D = —', tL(Iz, iIv;)I-z, +(I—
z; iIv;)Iz, 5 sin—
8;;
n. uclear resonance absorption in crystals will be
given in a forthcoming paper by J. H. Van Xcos8;;e+*v'z'e 'z~zz, (Am = —1),
Vleck. '4 One of the results there obtained will be
invoked to complete the following discussion in
,'(Iz, +iIv~)(Iz,+iI—v,) sin'8;;
which we revert to a less rigorous perturbation X e—2'&'ze+mz&«' (gm = 2),
method.
The unperturbed state of the system is speci-
F= ,'(Iz; iIv;) (Izr——iIv;)—
sin'8;;—
fied by the quantum numbers mz, , or for brevity, X e+"v' e "&z"" (hm = —2).
nz;, of the individual spins, the total z component
of spin being given by m=pm;. We assume Terms A and 8 give rise to secular perturba-
that these magnetic quantum numbers still tions; terms C to Ii represent periodic perturba-
characterize a state after the introduction of the tions of small amplitude. That these last four
perturbation V= P;P, ~, V;, , where terms cannot cause appreciable changes is a
consequence of the conservation of energy. For
V;, = y'k'r;; 'LIz;Iz, (1 —
—
3aP) our rigid-lattice model, then, only terms A and
+I„;Iv;(1—3a2') +Iz, Iz, (1 —3 )az
8 need be considered. Classically A corresponds
to the change in the z component of the mag-
3(Iz;Iv, +Iz, Iv;) aga2— netic 6eld by the z components of the neighbor-
3(Iz;Iz, + Iz, Iz;) a&az— ing magnetic moments. This local 6eld will vary
from nucleus to nucleus, slightly changing the
3(Iz;Iz, +Iv;Iz;)a2az5. —(20). Larmor frequency of each. Suppose that the ith
nucleus is surrounded by a number of neighbors,
V;; is the magnetic interaction between the ith
which for simplicity we shall assume to have
"J.H. Van Vleck, Phys. Rev. (in press). spin I=2. Each of these can be parallel or
 BLOEMBERGEN, PURCELL, AND POUND
antiparallel to Ho, and every arrangement leads of a spin in a given state and therefore broadens
to some value of the static part of the 1ocal field. the spectral line. The efII'ect is taken into account
From the nearest neighbors alone, we will usually by applying the proper numerical factor to (24).
obtain several possible values; each of these will The factor has been calculated by Van Vleck by
be further split when the next nearest neighbors the diagonal sum method. " In the case of a
are taken into account, and so on. The result lattice of identical spins it is 9/4, the r. m. s.
will be a continuum of possible values of the (angular) frequency deviation including both
local static field, centered on Ho, whose shape effects then being,
would have to be calculated for the particular
spatial configuration of the lattice involved, and
L(Aa&') hi=—(1/Ti ) = 'Y'hLI(I+1)/3]i
the given direction of Ho. The calculation is X Lpga;(1 —3 cos'8")'r "-']&. (25)
simplified, in all cases of interest, by the near
equality of probability for either spin orientation, If the nuclei are not identical in gyromagnetic
which also insures the symmetry of the distribu- ratio, neighbors which are dissimilar to the
tion about Ho. We have carried out the calcula- nucleus i contribute only through the term A,
tion for a special case of a cubic lattice of spins, and the sum over j in (25) has to be modified
obtaining a roughly |"
aussian shape for the in an obvious way.
distribution. In many cases an approximately We have here introduced the quantity T2",
Gaussian shape is to be expected. An important defined above, as a specification of the line width
exception is found, however, when the nuclei lie in this limiting case of a rigid lattice. We shall
near together in pairs; several such cases have later use T~' to describe the eft'ect upon the line
been studied experimentally by G. E. Pake and width of spin-spin interactions as modified by
are treated in a forthcoming paper. " lattice motion, reserving the unprimed T~ for
the specification of the line width in the general
We can obtain an estimate of the line width
by calculating the mean square contribution of case, to which the spin-lattice interaction also
each neighbor separately and taking the square contributes. If the line shape is Gaussian, the
root of the sum of these contributions, thereby quantity 2/T~" is the width between points of
assuming that the orientation of each spin is maximum slope, on an co scale. In calculating
independent of that of another. The contribution the maximum absorption for such a line shape,
of term A to the mean square local 6eld is given in the rigid lattice, the T2 defined by Eq. (14)
by is to be identified with (s./2) &Ti".
An accurate experimental test of (25) requires,
. —isy2Ii'I(I+ 1)
(II). ') A„— of course, a determination of the line shape. A
)& g;(1 —3 cos'8„)'r,, ' (23) substance which the rigid lattice model might
be expected to represent fairly well is fiuorite,
;in' the mean square deviation CaF~. The measurements of the F'9 resonance
in frequency by
line in CaF~, previously reported, "
and subse-
quent unpublished work on the same substance
'1'o this we must. add the contribution of term 8, by G. E. Pake, are in good agreement with
Eq. (25) (see reference 14).
which corresponds to the simultaneous Hopping
of two antiparallel spins (hm;=+1 and hm; VII. SPIN-LATTICE INTERACTION
= —1, or vice versa). Such a process is energeti-
cally possible and is caused by the precession of The rigid lattice model does not exhibit ther-
the spins around Ho with the Larmor frequency; mal relaxation of the spin system as a whole.
the precession of one spin produces a rotating We shall now show that if the position coordi-
field at the resonance frequency at the location nates, r;;, tII;;, and q;; are made to vary with
of another, resulting in reciprocal transitions. time, transitions are induced in which m changes,
We may say that this process limits the lifetime the spin system and the "lattice" thereby ex-
'i1E. M. Purcell, N. Bloembergen, and R. V. Pound,
' C, F;. Pa, ke, J. Chem. Phys. (in press). Phys. Rev, 'N, 988 (1946).
 NUCLEAR MAGNETIC RF SONANCE 3 BSORPTIAN

changing energy. Not only does this provide 2Ãvo = +Ho are diferent from zero, C and D
thermal contact between the spins and an ade- become secular perturbations because the time
quate heat reservoir, it also modifies, in some factor cancels out. Similarly, E and F become
cases drastically, the spin-spin interaction itself. secular perturbations if J:( —2vo) and Jo(+2vo)
Wa11er'7 in 1932 treated the interaction of do not vanish. Transitions are thus made possible
spins with lattice vibrations; his theory was in which m changes by +1 or ~2, the thermal
applied by Heitler and Teller' to estimate the motion providing, or absorbing, the requisite
eAectiveness of such an interaction in bringing energy.
about thermal equilibrium of a nuclear spin We can form a simple picture of the processes
system. We want to include more general types involved by recalling that the local field at one
of motion, in particular, what we may loosely nucleus i due to a neighbor j, in the rigid-lattice
call Brownian motion in a liquid or non-crystal- model, consists of two parts, a static part
line solid. It will turn out that such internal depending on I~; and a rotating part depending
~Iegrees nf freedom, when active, are far more
eR'ective as a source of spin-lattice interaction
on (I~; — iI»). The motion of the neighbor j
now causes the field arising from its z corn-
than are lattice vibrations, and play an important ponent of magnetic moment to fluctuate; if
role in substances which by many standards the motion contains frequencies synchronous
would be regarded as crystalline. with the precession of the neighbor i, the nucleus
As far as the Hamiltonian for the nuclear spin i will find itself exposed to a radiofrequency field
system is concerned, we regard the motion of capable of inducing a transition. This is the
the molecu}es as produced by external forces, effect described by terms C and D. The double-
the atomic interactions being mainly of an frequency effect comes about as follows: the
electrical nature. The atom or molecule is simply field at i which arises from the precessing compo-
a vehicle by which the nucleus is conveyed from nents of magnetic moment of j consists in
point to point. We thus neglect the reaction of general of both right and left circularly polarized
the magnetic moments of the nuclei upon the components. In the rigid lattice one of these has
motion. The factors in (22) which contain the a negligible effect while the one which rotates in
position coordinates are now to be expanded in the same sense as the precession of i causes the
Fourier integrals, We want to distinguish in our perturbation expressed by term B. If, however,
complex notation between positive and negative
frequencies, and we therefore define the intensi-
we impart to the nucleus j a suitable motion, at
the frequency 2vo we can reverse the sense, at
ties J(v) of the Fourier spectra of the three the nucleus i, of the originally ineffective circular
position functions involved by component, thus permitting it to interact with i.
(P;~ (1 —3 cos'8;;(t))r;, o(t) ')A, An exactly similar situation will necessarily be
created at the nucleus j, with regard to the field
~

arising from the precessing components of mo-

Jo(v)d v,
ment of i, hence a double transition occurs. This
— also shows why still higher multiples of vo do
(P, ~
sin8;;(&!) cos8;;(t)e'&'~&'&r, , o(t) ~')o, not enter the problem.
We can now compute the probability for a
J, (v)d v, (26) change 6m= +1 in the magnetic quantum num-
ber of the ith spin brought about by the fields
associated with the "local field spectrum. " By a
(P;~sino8;;eo*v"&~&r;;-o(/) ~o), „=t J,(v)dv. perturbation calculation, using terms C and E,
and taking an average over all values m; of the
Reconsidering now the terms C, D, 8, and F neighbors, we find for the transition probability,
in the expression for the perturbation (22), we
= ooy4&&oo(I —m;) (I+m;+1)I(I+1)
if Ji( —
Wm, m;+&
observe that vo) and Ji(+ vo) where
'~ I. @hallet, Zeits. f. Physik 79, 370 (1932). X [A( —vo)+-'Jo( —2 vo) j. (27)
 BLOK M BERGEN, PU RCELL, AND POUND
I
If the values of and y are not the same for lattice is then very much greater than that of
all nuclei,Eq. (27) can. be generalized as follows: the spins.
—m~) (I;+m;+1) Consider first the case I=~. Let the total
, +i = poyPko(I; number of spins be N, with N+ and X denoting
X Z~ vs~(I~+1) the population of the lower and upper state,
respectively. For W+ and 8" + we take lV as
X [Ji;(—vo;) +-', Jp, ( —voy —vo;)], (28) given by Eq. (27) modified by the appropriate
where Holtzmann factors:
vpI —
ypFI—
p/2 pr
W+ = We ~sos"' = W[1 —(yipHo/2kT)];
The result can be extended to include the
e8ect of perturbing fields arising from other W ~ = W[1+ (yl'pIIo/2k T) ]
dipole sources such as electronic spins. In such (30)
with
cases the quantization of these spins may be
time dependent under the in8uence of external W= (9/16)pPIpo[J&(vp)+ p Jp(2vp)]
forces (e.g. , paramagnetic relaxation phenomena
Writing e for N+ —N, the surplus in the lower
involving electronic spins) and the function
state, we have
[(I;—m;(t))(I;+m;(t)+1)]& must be included
with the position functions when the Fourier de/dt=2N W + 2N+W+—
expansion is performed, the factor I;(I;+1) in —n) W+
= (N+n) W p (N — . (31)
(28) then being dropped.
The expressions for transitions with hm;= —j. If we assume that ~n~ &&N, as it will be in all
are, of course, the same except that the intensities cases of present interest, and denote by no
have to be taken at positive frequencies in the =NyhHp/2kT the equilibrium value of pp, we
Fourier spectrum. Since J(v) will in all cases be have
an even function of v, it would appear that the dn/dl = 2 W(no —pp) (32)
motion should produce as many transitions up showing that equilibrium is approached accord-
as down. However, the transition probabilities ing to a simple exponential law with a character-
must be weighted by the Boltzmann factors of istic time
the 6nal states, as required by the fact that the T, =1/2W. (33)
system left to itself will come to thermal equi-
librium and the principle of detailed balancing. Ke thus make connection with, and justify,
That is, if N~ and N~ are the equilibrium popu- the manner in which the spin-lattice relaxation
lations of levels p and g which differ in energy time Ti was introduced in Eq. (11).
by EI, — E„we must have, in equilibrium, For I=-' it is always possible to de6ne a spin
,
temperature T, through e~"~«P ~ =N /N+, and
the approach to equilibrium can be described as
W~p/Wp „=N,/N„=e&sv so'"r. (29) a cooling, or warming, of the spin system as a
whole. The corresponding differential equation,
But the probability of a single transition from in which r~ is the lattice temperature, is
p to g, in a system such as we are considering, d T,/dt = (1/T, ) (T,/Tl) (Ti, T,). —
cannot depend on the population of g; it is
therefore necessary to suppose that the lV's are Clearly the transient behavior of the spin system
related by the Boltzmann factor of Eq. (29) is more simply described by the reciprocal
even when the spins are not in equilibrium with temperature, or by e, than by the temperature
the lattice, T being the lattice temperature. We directly.
are thus enabled to trace the approach to The situation is more complicated when I& ~.
equilibrium of the spin system coupled to a A spin temperature cannot be defined unless the
reservoir which remains throughout the process populations of successive levels are related by
at the temperature T. This last assumption is the same factor. Corresponding to this additional
justified if kT&&WHOA for the specific heat of the arbitrariness in the initial state of the system,
 NUCLEAR MAGNETI C RESONANCE ABSORPTION

Eq. (32) has to be replaced by a number of type occur is proportional to N ~. N ~, while the
simultaneous differential equations. However, a rate at which the second type occur is propor-
detailed examination of this question which can tional to No', but otherwise identical. A sta-
be found elsewhere, "
leads to a simple result tionary state is therefore reached only when
when the initial state of the spin system happens No' —N i. N+~, which is precisely the description
to be describable by a temperature T„and when of a Boltzmann distribution over three equi-
kT &&pkIIO&(kTJ. . The approach to equilibrium distant levels.
then resembles that of the I=~ system, the With the above reservations concerning the
surplus number in each level with respect to the case I&-'„wenow write the general formula for
next above increasing or diminishing with the the spin-lattice relaxation due to dipole-dipole
same characteristic time. Any intermediate state interaction in the case of identical nuclei. f $
of the system can then be assigned a temperature,
and (32) applies. It is found, moreover, that (1/Ti) = —,'y'l'i'I(I+1) LJi(vo)+-', Jm(2vo) j. (34)
value of W which must then be used in (33) is VIII. EFFECT OF THE MOTION UPON THE
LINE WIDTH

That is to say, the spin-lattice relaxation time The terms A and 8 of Eq. (22) must now be
for nuclei of spin I
exposed to a given local field
reconsidered to see what effect the Inotion of the
nuclei has upon the line width. These terms will
is the same as the relaxation time for nuclei of
still represent secular perturbations if we take
the same gyromagnetic ratio y, but with I=-,',
exposed to the same field.
the components near zero frequency in Ja(v),
The above argument would apply, for ex- the intensity of the Fourier spectrum of
ample, to a spin system with I&~ which is P;(1 —3 cos'8;;(t))r;; '(t). The question is, which "
allowed to come to equilibrium in a field Ho, frequencies are to be considered as "near zero.
after which Ho is changed adiabatically to a new We may say that the perturbation is secular up
value. A unique T~ can be defined which measures to the frequency v for which hv is of the same
the rate of approach of the populations, N ~, order as the actual splitting of the energy levels
~ -.N~, to their new equilibrium distribution. by the perturbation. The resulting line width
will be specified by 5~=27rhv=1/T, ', and with
Similarly, if the application of an intense radio-
frequency field of frequency vo has completely the application of the above criterion, T2 is to
equalized the level populations, the recovery be found from the following modification of
from such saturation (infinite spin temperature) Eq. (34):
follows the same law. 1/Tg' = -'y'hLI(I+1)/3]i
In general, however, the application of a
+1jxrg~
radiofrequency field alters the distribution to
one which is not precisely a Boltzmann distribu-
Jo(v)d v . (35)
~ —1/~Tg~
—
tion the spin system is itself no longer in
equilibrium. The most powerful agency tending Again an exception must be noted in the case of
to restore internal equilibrium among the spins is non-identical nuclei. If y;&y;, the secular per-
in many cases the spin-spin interaction. By turbation from term 8depends on the intensity
means of the process associated with the term 8 of the Jo spectrum in the neighborhood of
in Eq. (22), in which hm;=+1, hm;= — 1, or v= (y, —y;)IIO/2s. rather than v=0.
vice versa, , the level populations can be read- The observed line width is not determined by
justed. If I= j. , for example, transitions of the T~' alone because the spin-lattice relaxation
type —1-+0 (for m;) accompanied by +1— +0 eAected by the terms C to Ii limits the lifetime
(for m;) augment the population of the central of the nucleus in a given state. For I=-'„
the
level, while the reverse double transition depletes mean lifetime in each of the two levels is 1/W,
it. If N ~, No, N+~ are the populations of the or 2Tt. The resulting lice miChh„ for two levels
levels, the rate at which transitions of the first
f)The case of non-identical nuclei calls for a rather
' L. J. F. Broer, Thesis {Amsterdam 1945), p. 56 ff- obvious generalization which we omit.
 BLOK M BERGEN, P URCELL, AN D POUN D

broadened in this way, is, by a well-known from which to ebtain the functions J~;(v), Ji;(v),
argument'9 twice the width of one level, the JI;(i) governing the intensity of the local field
line shape being given to a high degree of spectrum. By postulating thermal vibrations in
approximation, (Ti»1/vo) by an otherwise rigid lattice, for example, one is led
directly to %aller's result for spin-lattice relaxa-
4Tg
= tion in a crystal. Our interest here is in liquids,
gi(v) (36)
1+16m' T, '(v —va) ' for which the functions Ii will vary in a random
fashion with time, as the particles containing the
a curve whose width between half-value points magnetic nuclei take part in the Brownian
is 1/2s Ti on a frequency scale. With this must motion. The motion will be isotropic on the
be combined the distribution resulting from the average; hence, as is at once evident from (39),
perturbations A and 8, which, if it had a the time average of each of the functions Ii is
Gaussian shape, would be zero.
gp(v) = Tn'(2s) 1 exp[ — —vo)'TPj.
— The statistical character of the motion justi6es
2w'(v (37)
an assumption customary in the theory of
Roughly speaking, we can say that the combina- fluctuation phenomena, that
tion of two bell-shaped curves will yield another
bell-shaped curve the width of which is about (F(~) F*(~+r))" =E(r) (40)
the sum of the widths of the composing curves.
The right side of (40) is called the correlation
Thus,
function of F(t). It is assumed to be an even
(I/2 2) = (2/~)'(I/T2')+(I/2Ti) (38) function of 7 and independent of t. It follows
immediately that E(r) is real. We may assume
Vfe cannot attach a precise meaning to I2 in
also that E(r) approaches zero for sufficiently
general, but the numerical factors in (38) have
large values of z. The relation" between the
been chosen[ ff with two hmiting cases in mind:
Fourier spectrum of a function F and the corre-
(a) Tq'«Ti~i, and (b) T2'=Ti. In the former case
lation function of Ji enables us to write for the
the factor (2/s)& is consistent with Eq. (14) for
spectral intensity,
the Gaussian curve to which, in this case, we
may expect the line to bear some resemblance.
The last term in (38), which is important in case J(v) = E(r)e' '"'dr
(b), is correctly written for the damped oscillator
line shape. However, in this case, an additional Since E(r) is real and even, J(v) is real and even.
uncertainty arises from the arbitrariness of the In order to simplify the following discussion
choice of limits in Eq. (35). we make an assumption about the form of E(r),
namely, that
IX. THE SPECTRUM OF THE SPACE
COORDINATES
E(r) =(F(t)F*(t))~„e~'~"' (42)
In order to apply this general theory of
relaxation time and line width to a particular
The time r, is a characteristic of the random
motion and the function whose correlation E
substance„ it is necessary to develop the Fourier
measures, and is called the correlo, tion time.
spectra of the functions of position coordinates,
Temporarily we shall assume that v, is the same
Fog= (1 —3c ops;, ) r;; ', for each of the functions Iio, F» and I"g, returning
— to this question later. From (41) and (42) we
F~; = sIn8;, cos8;,c'«~r;, ', (39) 6nd, for each of the three J's,
I'"~; = sin28;;e '&'~r;; —',
J(v) =(F(t) F*(t))~,2r, (1+4w'v'r ')-' (43)
"V. %'eisskopf and E. VA'gner, Zeits. f. Physik E8, 54
(~93O); Os, ~8 (I93O). The intensity in the spectrum is nearly constant
ff) It seems ~orth while to preserve internal con-
sistency among the de6nitions and theoretical formulas
even at the risk of making the formulas appear less ap- "See, for example, Ming Chen %'ang and G. E. Uhlen-
proximate than they really are. beck, Rev. Mod. Phys. 17, 323 (1945).
 NUCLEAR MAGNETIC RESONANCE ABSORP TION

at very low frequencies but falls off rapidly a sphere of radius a imbedded in a viscous liquid,
when 2s vs&&1. The subscript
above, but it will be understood
j has been omitted
that, in general,
and thereby obtained the relation r =4svta'/kT,
in which g is the viscosity of the liquid. We
not all neighbors will have the same correlation adopt the same procedure here, noting one
time. di8'erence between the two problems —
a rather
unimportant difference, to be sure, in view of
X. THE THEORY APPLIED TO WATER the much over-simplified model, but a character-
The application of the preceding theory will istic one. 'I he function whose correlation time is
now be illustrated by a calculation of the relaxa- required in the Debye theory is cos8, the sym-
tion time for protons in water. A given proton metry character of which is essentially difkrent
has as its nearest neighbor the other proton in from that of the functions Fo, Fj, and F2 with
the H20 molecule, and we consider first the which we are concerned. Because of this the
effect of this exceptional neighbor. Vibration of correlation time for a given model/ will not be
the molecule can be neglected, because of its the same in our case. In fact, for the sphere in a
high frequency and small amplitude. We regard viscous liquid it is, for each/ ff of the functions F,
the molecule as rigid, and assume that the
(47)
orientation of the vector connecting the two
protons varies randomly, no direction being Adopting Eq. (47) and assuming that a=1.5
preferred. Only the angle coordinates are now X10 ' cm, we find for water at 20'C (it=0.01
variable in time, the functions F being "
poise): ~, =0.35&(10 second. For comparison
we note that the measurements by Saxton~ of
Fa —(1 —3 cos'8)/b',
the dielectric constant of water in the microwave
Fi cos8 sin8e'v/b-—
', (44) region lead to a value 0.81 &&10—"seconds for the
— Debye characteristic time, at 20'C. Although
e"'/b',
F2 sin'8—
this is somewhat less than 3 times our estimated
where the interproton distance. Ke can
b is v„we must remember that with further refine-
substitute a statistical average over the spatial ment of the molecular model a diR'erence between
coordinates for the time average specified in (43). 7, and 1/3m, i,v, would arise from the fact that
The averages are readily found: the H— H line and the polar axis of the molecule
lie in diferent directions.
4/5b';
(Fo(t) Fo*(t))A„— We note at once that 2~vv, &&i, so that the
(Fi(t) Fi*(t))A, = 2/15b', (45) term in bra, ckets in Eq. (46) becomes simply 37..
In other words, we find ourselves at the low
—
(Fm(t) Fm*(t))g, 8/15b'. frequency end of the local field spectrum, where
the spectral intensity is independent of v and
Returning to the general formula (34) with (43)
dhrectly proportional to ~,. For the interproton
and (45) we obtain for the relaxation time (in-
volving the nearest neighbor only)
$ Another model, in some ways a better representation
of the molecule and its environment, is one in which only
(1/Ti) = (3/10) (y'l't'/b') Lr, /(1+4s'v 'r ') a limited number of orientations are permitted, the transi-
+2r, /(1+16s'vo'~. ') j. (46) tions from one to another taking place in a "jump. "
—
Imagine, for example, that the molecular axis in our case
The determination of the correlation time 7, the H-H lin- ~n point toward any of the eight corners
of a cube centered at the molecule, transitions taking place
is closely related to the problem encountered in with the probability co between any such direction and
the Debye theory of dielectric dispersion in polar one of the three adjacent directions only. For this model
the correlation functions of Fo, F1, and Fg are all of the
liquids; there also it is necessary to estimate the form {42) with the correlation time 3/4', whereas the
time, w, during which molecular orientation correlation time of cos8 is 3/2'.
persists. " Debye assumed that in the first
$'Il Note that Fo, F~, and F~ belong to the same spherical
harmonic, Fg{8, rp). It can be shown that the correlation
approximation the molecule could be treated as function of F~(8, q) for the sphere in a viscous liquid, is
exp L-l{l+1)kTw/8~1 j.
~ J. A. Saxton, Radio Research Board Report C115
"P. Debye, I"olar Moleclkes (Dover Publications, New (declassified), D.S.I.R., March 20, 1945. (This is a British
York, 1945), Chapter V. report. )
 8LOEM BERGEN, P URCELL, AND POUN 0

distance we take 1.5 )& j.0 8 cm. Upon substituting Even when an experimental value of D is avail-
numbers into (46), we obtain for (1/Ti)„ the able, the use of the Stokes relation represents no
relaxation rate due to the proton partner only, real sacrifice in accuracy, for the distance of
0.19 sec. '. closest approach, a, can at best be estimated
The effect of the neighboring molecules must only roughly.
now be estimated. The fluctuation in the local Substituting numbers into (50) we obtain for
field arising from the neighbors is mainly caused water at 20'C, (1/Ti) „=0. 10 sec. '. The relaxa-
by their translational motion, and only to a tion time T» is then found by combining both
minor extent by their rotation. %'e regard the effects: (1/Ti) = (1/Ti)„+(1/Ti) „=0. 19 sec
molecules as independent, attributing to each +0.10 sec. ', or Ti —3.4 sec.
either a magnetic moment of 2p„and spin = j., I The efI'ect of the partner thus appears to be
of' else zero spin and moment. Three-fourths of somewhat more important than that of the
the neighboring molecules are of the former neighboring molecules. We have probably over-
(ortho) type. We ask for (F(t)F (t))A, and r, for estimated the e8ect of the neighbors by neglect-
the molecules in a spherical shell between r and ing the rotation of the nearest molecules as well
r+dr The. center of the shell is located at, and as the e8ect upon the local field produced by
moves with, the molecule containing the proton them of the rotation of the molecule containing
i. A reasonable choice for v, is the time it takes the proton i. That is, the appropriate v, for the
the molecule j to move a distance r in any nearest neighbors should be somewhat smaller
than Eq. (48) would predict. Such refinements
direction from its original position on the shell,
for in that time the field at the proton i, due to j, are hardly warranted in view of the obvious
will have changed considerably. The relative shortcomings of the model.
j
motion of i and is simply diffusion, and can be
described most directly by means of the diR'usion
The experimentally determined value of T» for
distilled water at 20'C is 2.3&0.5 seconds,
coefficient D of the liquid. Applying the above obtained by the direct method described in
criterion for v„and remembering that both Section IV. The theoretical estimate of T» is
molecules difII'use, we find, satisfactorily close to the experimental value.
the possibility that the experi-
ro —
Unfortunately,
r'/12D.
mental value was influenced somewhat by dis-
To find (F(t)F~(t))A, the angular functions are solved oxygen cannot be excluded. From subse-
averaged as before and then, treating the mole- quent results with paramagnetic solutions, it is
cules as independent, we sum the efkcts of all estimated that a saturation concentration of 02
neighbors by integrating over the volume from could have contributed at most 0, 3 sec. ' to
the radius of closest approacy, r = 2a, to infinity. (1/Ti).
Here also, 2mvv, is much smaller than one for
all molecules close enough to have an appreciable XI. EXPERIMENTAL RESULTS FOR VARIOUS
LIQUIDS; COMPARISON V@TH THEORY
efkct, so that we have for the relaxation rate
due to the neighbors (1/Ti)„, A more significant test of the theory is pro-
vided by measurements of' the proton relaxation
I" —.
3 r'
&T,l „5
ply =-7'i''¹
3
.4~r'dr
~

&i, r 12D
3+ y4h'Xo
time in a series of similarly constituted liquids
of widely diA'ering viscosity. The conspicuous
role of the viscosity in our description of the
(49) relaxation process will already have been noticed.

¹
cm'.
in
10 aD

Eq. (49) is the number of molecules per

can make use of the well-known relation
VVe
developed by Stokes, D=kT/6m', to write Eq.
Despite the crudeness of a description of the
details of the molecular motion in terms of the
macroscopic viscosity, it can hardly be doubted
that there should be a close connection between
q and r,. Table I lists the values of T» obtained
(49) as for a series of hydrocarbons. These values, and
(1/T, ) „=9w'y'h's¹/Sk T. (50) all relaxation times quoted hereafter, were deter-
 NUCLEAR MAGNETI C RESONANCE ABSORPTION 703

mined by running a saturation curve on the sample could be held constant within 0.5' while
sample, the direct measurements of T~ for the data for a saturation curve were taken. The
water sample, and for the 0.002K CuSO» solution viscosity values were taken from Landolt-
(see Section IV) serving as a calibration of the Bornstein, Physikclish-Chemische Tab@lien.
circuit. The viscosity of each sample was meas- The results obtained with ethyl alcohol at two
ured by conventional methods and is listed for frequencies are shown in Fig. 12, in the form of
comparison. The wide range over which a rough a logarithmic plot of the relaxation time 1»
proportionality exists between s and (1/T~) is against the ratio of the viscosity to the absolute
striking. A more heterogeneous list of substances
is given in Table II. A probable explanation of
temperature, rl/T. According to our theory
and hence (1/T, ), so long as 2m vs, «1, should be
r„
the exceptional behavior of sulfuric acid is the proportional to q/T, as indicated by the line
low density of protons in the material, or, more drawn on the graph with slope —1. Moreover,
precisely, a relatively large value of the shortest the relaxation time should be independent of v,
proton-proton distance (the other constituents the radiofrequency. Both predictions are con-
being non-magnetic). firmed by the experimental results.
A wide range in viscosity is covered by water- A much wider range in q/T is accessible with
glycerin solutions. Figure 11 shows the variation a pure glycerine sample. Glycerine, as is well
of T& with viscosity for the water-glycerine known, ordinarily becomes super-cooled down
system at room temperature. The viscosity to temperatures far below its freezing point of
values were obtained from tables in this case. 18'C, and at the lowest temperatures prevailing
Exact proportionality between (1/T~) and q is in the experiment the liquid was nearly glass-hke.
hardly to be expected for the progressive change The glycerine results are plotted in Fig. 13 which
in the character of the substance must bring includes also the results of line-width measure-
with it a change in the environment of each ments in the low temperature region. The char-
proton and in the details of the local molecular acteristic time T2 associated with the line broad-
motion. Actually a 1000-fold change in g is ening is obtained by setting yhH=2/T~, where
accompanied by a change of only 100-fold in T&. AII is the observed interval in gauss between
A better experiment is one in which the same points of maximum slope on the absorption
substance is used throughout, its viscosity being curve. This procedure, as explained earlier,
varied over the widest possible range by changing would be correct for a Gaussian line shape.
the temperature. Ethyl alcohol and glycerine In the region of low viscosity the Tj values for
are convenient for this purpose; each was studied glycerine display the behavior encountered in
over the temperature range between 60'C and alcohol, namely, inverse proportionality between
—35'C. The temperature of the substance was T~ and q/T, and independence of T~ and ~. The
measured with a copper-constantan thermo- new feature is the minimum in T&, followed by
couple brought into the sample a short distance increasing T~ as g continues to increase. The
away from the r-f coil. The temperature of the inverse line-width T2, on the other hand, de-
whole enclosure containing the coil and sample creases monotonically in the region where it could
was established by circulating a suitable ffuid at be measured, with perhaps a slight tendency to
a controllable rate. The temperature of the flatten oB at the highest rl/T values. The curves

TABLE I. Relaxation time T1, for protons in various TABLE II. Relaxation time T&, for protons in various polar
hydrocarbons at 20'C and 29 Mc/sec. liquids at 20'C and 29 Mc/sec.

Viscosity Tx Viscosity Tl
(centipoises) (seconds) (centipoises) (seconds)

Petroleum ether 0.48 3.5 Diethyl ether 0.25 3.8

Llgl 01Il 0.79 1.7 Water 1.02 2.3
Kerosene 1.55 0.7 Ethyl alcohol 1.2 2.2
Light machine oil 42 0.075 Acetic acid 1.2 2.4
Heavy machine oil 260 0.013 Sulfuric acid 25 0.7
Mineral oil 240 0.007 Glycerine 1000 0.023
 WATER-GI. YGERIN

l GOHOL

Mc/sec
Mc/sec.

e IQ

Ct

IQ
0.5
IQ IQ I lo' 5l0 ~ I04
VISCOSITY IN PQISFS POISES/DEGREE

FIG. 11. Relaxation time for protons in water-glycerin FIG. 12. Relaxation time for protons in ethyl alcohol,
soiutious, measured at 29 Mc/sec. measured at 29 Mc/sec. and at 4.8 Mc/sec.

drawn on the graph are based on the theory, would have the eHect of flattening the minimum
which must now be applied to this more general in the T~ curve.
case, the restriction 2m vr, &(1 being lifted. To discuss the line-width parameter, T2, we
The discussion will be simpli6ed by the go back to Eq. (35) which determines Ts' and
assumption that v„at
a given temperature, is insert for Js(v), (4/5bs)2r, /(1+cs'r, s). We thus
the same for all sources of the local field. We obtain, making use of the constant Cc of Eq. (51),
write Eq. (46) in the form:
(1/Ts')' = (3/s) Cc tan '(2r, /Ts'). (52)
—=C 1
+ 7 (51) For sufficiently large values of v „T~'
attains the
Tl I+co r, 1+4cs r value (2/3Cc), which is identical with Ts", the
where Cj includes the factors which are inde- time characterizing the spin-spin coupling in the
pendent of the temperature and frequency. (The rigid lattice. %hen 7-, is small, we have, from
variation of b and e with temperature is a (52), (1/Ts') = (6/a) Ccr, The da.shed line in
negligible effect )The .asymptotic behavior of Tq Fig. 13 has been drawn, according to this relation
is simple: for rur, «1, (1/Tc) =3C~r, ; for cdrg)1, in a region in which the contribution of (1/T~)
(1/Ti) =3C~/(2cs'r. ) In the i.ntermediate region to the line width is negligible. The agreement
a minimum in T~ occurs for car, =i/V2, with with experiment is satisfactory.
Tc c c —3s&/(2IC~). If we may assume that r, is
&
To the left of the minimum in Tj we should
proportional to ti/T, the shape of the curve have (1/Tc) =3Ccr, so that in this region,
logTc tcs. log(s/T) should be the same for all according to Eq. (38),
substances to which the theory applies; changes (1/Ts) = (2/s) &(1/Ts')+(1/2Tc)
in Cc or in the factor connecting r, and rt/T = 3Ccr, [(2/s) i+-', ]= 1.01/Tc. (53)
merely shift the curve. Two such curves have
been drawn on the graph in Fig. 13. This result signifies only that Ti and T2 are
The salient features of the experimental results approximately equal when 2xc r. «1.
are well reproduced, although there are quanti- The general features of the simplified theory
tative discrepancies. The horizontal shift of the are presented in Fig, 14. The T2 curve is located
minimum is somewhat less than the ratio of the by the value of T2", which depends on the
frequencies, and the experimental minimum is spin-spin interaction in the "frozen" substance,
flatter than predicted. This may be blamed in according to Eq. (25). The break between the
part upon experimental inaccuracy, but it must sloping and the horizontal part of the T2 curve
be remembered that Eq. (51) neglects the varia- occurs near r, = Ts"/V2, according to (52). The'
tion in v, among the neighbors. As we saw minimum in the T~ curve is located by csr, = 1/v2,
earlier, a proper treatment of the neighbors intro- and thus both curves are 6xed. That the transi-
duces a distribution of relaxation times, which tion regions occupy, in most cases, rather small
 NUCLEAR NAG NETI C RESONANCE ABSORPTION

ranges on the logarithmic plot is a consequence relation has again been used to express cD in
of the weakness of the spin-spin interaction terms of s/k T; any difference between the
compared to yHok.
We can now understand why the true line
width in many liquids is unobservably small.
mobility of an ion and that of a water molecule
is thereby neglected.
per cm'.
¹,
is the number of ions

The rapid Auctuations in the perturbing fields According to Eq. (54) the relaxation time
very nearly average out; the intensity in the should be inversely proportional to the ionic
local 6eld spectrum is no greater at v =0 than it concentration and to the square of the magnetic
is at v= vo, resulting in the near equality of Tj moment of the paramagnetic ion. The experi-
and T2 to the left of the Ti minimum. But if T2 mental results presented in Fig. 15 con6rm the
is of the order of one second, as Ti is in many first prediction, except for a marked and unex-
cases, the corresponding line width is of the plained deviation at very Iow concentrations.
'
order of IO gauss. To observe this width, the The predicted proportionality between (1/T~)
applied 6eld Ho would have to be homogeneous and p,,ff' is tested by the data in Table III. The
over the sample to less than 1 part in t0'. ,
values of p, f~ in the middle column were derived
As ~, increases the line width increases, from measurements of Tj in paramagnetic solu-
eventually becoming observable and 6nally at- tions of known concentration by assuming arbi-
taining its asymptotic value, A striking qualita- trarily the value 2.0 Bohr magnetons for Cu++.
tive demonstration of this effect is obtained by These numbers are to be compared with the last
allowing molten paraffin to solidify in the sample column which contains the values of p, ff derived
tube while observing the proton resonance on from susceptibility measurements. It is not sur-
the oscilloscope. If the held is reasonably homo- prising that in the cases of Ni~, Co++, and
geneous, a pronounced broadening of the line is Fe(CN)8 the p, fg obtained from the nuclear
observed when the paraffin freezes. relaxation measurements is too small, for there

XII. RELAXATION TIME AND LINE WIDTH IN

are important contributions from non-diagonal
elements to the magnetic moment of these ions. "
SOLUTIONS CONTAINING PARA- Indeed, Co++ and Fe(CN) ~ show strong
MAGNETIC IONS Curie's ele-
deviations from law. Non-diagonal
The addition of paramagnetic ions to water, ments would give rise to high frequency pertur-
as was 6rst shown by Bloch, Hansen, and bations which would be ineffective in producing
Packard, ' can infIuence markedly the proton nuclear relaxation.
relaxation time. The magnetic moment of such
an ion is of the order of one Bohr magneton, and IQ

the interaction energy p;„p~/r' is thus some 10'

TI
times larger than that of the interactions previ- AND

ously considered. Nevertheless, the discussion of IQ

29 Mc/sec
the last section can be adapted to this case ~
TI Ot
TI at 48 Mc/sec
without much modi6cation. The problem is ~ Tz at 29 Mc/sec.
IQ-3
essentially that treated in estimating the effect
of the neighboring molecules. The appropriate gb

value of 7-, is determined in the same way, with

one possible exception to be mentioned later, -- C/T
and will not be much different in magnitude, so IQ
3 !0 2 IQ I IQ2
that we again have 2svr, ((1. Equation (50), PQISES/DEGREE

adapted to this case, then reads: FIG. 13. The relaxation time T~ and the line-width
.
(1/2 i) = 2~v'~fPW. /5~&
= 12s'y'sÃ~„p,n /5kT.
. (54)
parameter Tg, plotted against the ratio of viscosity to
absolute temperature, for glycerin.
& G. J.
Gorter, Paramageetk Rekuatioe (Elsevier Pub-
ln general, the value of p, tP can be determined lishinN,Company, Inc. , Near York, 1947), p. 4. Also, H. J.
Van Vleck, Electr@ aM Magnetic Smceptibihtiar (Oxford
from data on paramagnetic susceptibility. Stokes' University Press, New York, 1932), Chapter XI.
706 BLOEM BERGEN, PURCELL, AND POUN D

particular interest, for the extremely short spin-

lattice relaxation time observed in concentrated
solutions permits a test of the relation between
T~ and T~. As we have seen, the theory predicts
that T~ and T2 should be about equal when
2xvv, (&1. That this is indeed the case can be
seen by comparing Fig. 15 with Fig. 16, which
gives the measured line width, expressed in
terms of T2, for highly concentrated Fe+++
solutions.
)O-s

XIII. NUCLEAR RELAXATION EXPERIMENTS

T2
+ INVOLVING F" Li', AND D'

Several additional experiments on liquids will

be reported briefly. The resonances of both H'
I (
Z~{s)o= I/s2 pZg~T2$2
and F" were studied in the liquid CHFC1~, a
iO lO-u
SECONDS
iO-' lO s
"
"Freon, which was condensed in a glass tube
FIG. j, 4. Dependence of Ti and lg upon r&, according to and sealed off. Both lines were narrow and the
the simplified theory in which all interactions are assumed
to have the same correlation time. intensities were nearly equal, as would be ex-
pected. (The spin ofF" is , and its gyromagnetic
—',

The magnitude of T~ as predicted by Eq. (54) ratio only a few percent less than that of the
agrees well with experiment. In the case of Fe' ', ' proton. ) The measured relaxation times were
for example, Eq. (54) with g=0.01, y;, =5.9 3.0 sec. and 2.6 sec. for H' and F", respectively.
Bohr magnetons, yields T~=0.08 second, for a According to the theory these times should be
concentration of 10"/cm'. The value observed equal, for the domina, nt effect should be the
at this concentration is 0.06 second. Such close H' — interaction within the molecule, and
F"
agreement must be regarded as accidental. the product pa&'yF»' occurs in the generalized
An effect which has not been taken into form of Eq. (34) in either case. Within the
account in the above discussion is the relaxation experimental error this is confirmed.
of the electron spins (i.e. , paramagnetic relaxa- The compound BeF2 combines with water in
tion) which may in some cases limit the value of any proportion. The behavior of both the proton
7, Paramagnetic relaxation times as short as and the fluorine resonances in BeF~ xH20 was
10 9 second have been observed in crystals at found to resemble that of the proton resonance
room temperature. "
Some measurements of in glycerine. With increasing viscosity (decreas-
paramagnetic dispersion in solutions, ' reported ing x) T~ decreases to a minimum value of about
by Zavoisky, "
indicate relaxation times of the 10 ' second, thereafter increasing to a final value
order of 10 ' to 10 ' second. In applying Eq. of 0.2 second in pure BeF2, an amorphous glassy
(54) we have tacitly assumed that the para- solid. The line width meanwhile increases from
magnetic relaxation time, under the conditions an unobservably small value to approximately
of our experiment was not much shorter than 10 gauss in the anhydrous solid.
"
10 second. It wiII be interesting to examine The resonance absorption of Li~ was examined
nuclear relaxation induced by paramagnetic ions at 14.5 Mc/sec. in solutions of Licl and LiNOs,
under conditions for which the characteristic with the results given in Table IV. The addition
time 7, due to thermal motion is quite long. of paramagnetic ions affects the proton relaxation
The actual v, may then be determined by the time more drastically than it does that of Li~,
paramagnetic relaxation effect. which is probably to be explained by the mutual
The Eire mkQh in paramagnetic solutions is of repulsion of the positive ions, or by the existence
of a "shield" of hydration around the Li ion.
~R. L. Cummerow, D. Halliday, and G. E. Moore, In the absence of paramagnetic ions the lithium
Phys. Rev. 'I2, 1233 (1947}.
"E. Zavoisky, J. Phys. USSR 9, 211 {1945). relaxation time exceeds that of the protons, but
 NUCLEAR MAGNETI C RESONANCE ABSORPTION 707

not by as large a factor as the simplified theory TABLE III. Comparison of magnetic moments of ions
inferred from nuclear relaxation measurements and from
would predict. This may be attributable to the static susceptibility measurements.
eRect of nuclear quadrupole interaction, a phe-
nomenon which is much more conspicuous in petf in Bohr in Bohr
Ijseff
magnetona magnetons
the following example. (from nuclear (from
relaxation susceptibility
The deuteron and proton resonances were Ion experiments) measurements)
examined, at the same frequency, in a sample of Er+++ 8.5 9.4
50 percent heavy water. The relaxation times Fe+++ 5.5 5.9
Cr+++ 4. 1 3.8
observed for D' and H' were 0.5 and 3.0 sec. , Cu+++ 2.0 1.9
respectively. Contrary to what would be expected Ni~ 1.8 3.2
from the ratio of the magnetic moments, the
Co++ 1.1 4.5-5.3
Fe(CN) 0.1 2.4
deuteron relaxation time is the shorter. However,
the interaction of the electric quadrupole mo- s' The numbers in this column are those listed by Gorter, reference
23, pp. 14 and 15, with the exception of the value for Fe(CN)a"
ment of the deuteron with a fluctuating inhomo- which is based on the measurements of L. C. Jackson, Proc. Phys.
Soc. 50, 707 (19'38).
geneous electric field can bring about thermal
relaxation. We omit the analysis of this process,
rate much higher than vp. There are two distinct
which parallels closely the treatment of dipole-
local magnetic fields at the position of one proton
dipole interaction, and point out only that the
in the H2 molecule. One arises from the rotation
interaction energy, in order of magnitude, is
of the molecule as a whole; the other originates,
eQ(BS,/Bs), where Q is the nuclear quadrupole
as in the water molecule, in the magnetic mo-
moment in cm', this is to be compared with
ment of the other proton in the molecule. The
pIIi„.If the fluctuation spectrum of Bb,/Bs is corresponding interaction energies are fairly
assumed similar to that of Hi„, the observed
large and have been accurately measured in
value of 11 for the deuteron in our experiment
molecular beam experiments" where the inter-
can be accounted for by assuming a reasonable
action causes a splitting of the proton resonance
value of Bh,/Bs in the Dqo or HDO molecule. tt ff tt
into six components, in the lowest rotational state
The quadrupole interaction is probably less
of ortho-hydrogen. The interactions involve the
effective in the case of the lithium ion because
the immediate surroundings of the ion are
rotational quantum number and its spatial J
quantization m&. In a gas at ordinary pressures,
electrically more symmetrical.
m~, and, to a lesser extent, J, may be expected
XIV. NUCLEAR RELAXATION IN HYDROGEN GAS to change frequently as a result of molecular
collisions. If m~ were to change with each colli-
We have reported earlier" the observation of
nuclear resonance absorption in hydrogen gas at
room temperature, at pressures between 10 and
30 atmospheres. The low density of nuclei SATURATION OF PARAMJKSETC SOLUTION
++

severely limited the accuracy of measurement, Fr+i

ET 0+0

but it was possible to ascertain that the relaxa-

tion time at 10 atmospheres was approximately
0.015 second. 1& appeared to increase with
increasing pressure. The width of the line was
less than 0.15 gauss.
The interpretation of these results is closely lO

related to our theory of relaxation in liquids.

In each case the key to the problem is the 10
IOl7 IOis ORO IOs I
observation that the local 6eld Quctuates at a
(Oats
NUMSER OF ONS1eC

ttgp A treatment of relaxation by quadrupole coupling Fro. 15. Relaxation time 1), for solutions of paramagnetic
mill be found in the Thesis
's E. M. Purcell, R. V. Pound, and
Phys. Rev. Vo, 986 (1946).
¹
to be presented by one of the
authors (N. B.) at the University of Leiden.
. Bloembergen, "J.
ions, measured at 29 Mc/sec.

M. B. Kellogg, I. I. Rabi, N. F. Ramsey, and

Zacharias, Phys. Rev. 5f, 6 If (1940).
J. R.
 BLOK M BERGEN, P U RCELL, AND POUN D

Tsar. E IV. first obtained by J. S. Schwinger, and can be

reached by a process similar to that followed in
Relaxation time
gram atom H i in sec. Section VII.
Suhetance (gram atom Li J for Li~ for Hi

LiCl+Hgo
LiNog+Hgo 14
$.8 1.I5
2.7
0.4
1.1
—j.
= y'
'1 2r,
H"J(J+1)
—
LiNOIl+Hgo+Fe(N03} 8 0.11 0.0023 3
LiC1+HgG+CrC13
g
6.5 0.24 0.0095
+3II"2
J(J+1)
LiC1+H&+CuS04 6.6 0.18 0.013 (55)
(2 J—1) (2J'+3)
The field at one proton due to the molecular
sion the correlation time for the local 6elds
rotation is speci6ed by means of the constant H'
would be the mean time between collisions, which
in hydrogen at 10 atmospheres is approximately
in Eq. (55), which has the value 27 gauss. "
" H" is a measure of the 6eld at one proton due
10 second. The situation is thus similar to
to the other and has the value 34 gauss. At
that in a liquid of low viscosity, in that 2xw, &&j, .
room temperature we have to do mainly with
We should therefore be prepared to find (a) that
the state J=1; molecules with J=O, 2, etc. ,
Ti~1/r, ; (b) that Ti and Tg are about equal,
(para-hydrogen) are, of course, inert in the
and very much larger than the reciprocal of the
experiment. Identifying v, with the mean time
frequency splitting observed in the molecular
between collisions calculated from gas kinetic
beam experiment. Since we have identi6ed v.
data we find, using Eq. (55), Ti=0.03 seconds,
with the time between collisions, Tl should
at 10 atmospheres, in satisfactory agreement
increase with increasing pressure, as observed.
with experiment. The theoretical value will be
The implication of (b) is that the line should be
reduced if m J does not change in every collision,
very narrow, as observed, and that the line
as assumed.
width should vary inversely with the pressure.
This last efkct would have been masked by XV. NUCLEAR RELAXATION IN ICE
inhomogenelties ln the magnetic 6eld, but the
experimental results on closely analogous cases The process which we have invoked to explain
in liquids leave little room for doubt that nuclear relaxation in liquids has no counterpart
pressure narrowing can occul . in an ideal crystal. Nevertheless, many solids
The effect of the magnetic fields associated show evidence, in the form of specific heat
with neighboring molecules is negligible except anomalies or dielectric dispersion, of internal
at extremely high pressures, and in this respect degrees of freedom other than vibration. We
the theoretical analysis is less uncertain than it shall discuss only one example, which we have
was in the ease of the liquids. %e omit the studied experimentally in some detail, that of ice.
derivation and present only the result, which was Ice shows a marked dielectric dispersion at
rather low frequencies. The residual entropy of
IO ice also provides evidence for internal freedom
Tg
which has been attributed" to the two available
positions for the proton in the —H — bond.
0 0
Without inquiring into the connection between
these two phenomena, or into the details of the
LYt IO
Vi
internal motion, we might assume that our
parameter 7, is proportional to the Debye char-
acteristic time v~, deduced from dielectric dis-
persion data. The nuclear relaxation data for
IO protons in ice have been plotted against v~ in
IOM IO
NUM8KR OF IOUS/cc
Figs. 17 and i8, the value of v~ for each temper-
FIG. 16. The line-width parameter T~, determined from
measurements of Hne width in solutions containing a high
" Linus Pauling, The Nature of the Chemicad Bond
(Cornel1 University Press, Ithaca, New York, 1940), p.
concentration of ferric ions. 302 ff.
 NUCLEAR MAGNETIC RESONANCE ABSORPTION
IO IO

ICE ICE
at 29 Mc/s at 29 Mc/sec

-5
u
&A
IO

I
IO
IO IO' IO IO IO
DEBYE TIME IN SECONDS DEBYE TIME IN SECONOS

Fro. 17. Proton relaxation time in ice, plotted against FIG. 18. T& for ice, plotted against the Debye
the Debye relaxation time derived froxn dielectric dis- relaxation time.
persion data.

The quantitative agreement is as good as we

ature having been taken from the measurements
"
of Wintsch. Figure 17 shows T» for ice, in the
could have expected in view of the simplifying
assumptions which were necessary in applying
temperature range — 2'C to —40'C. It is clear
the theory to a particular substance, and the
that we have to do with the ascending branch considerable uncertainty in the, absolute magni-
of the typical T» curve, and that the assumption tude of the experimental relaxation times. Not
was not a bad one. The line width all of the features of the theory are thereby
increases as the temperature is lowered, as shown
tested; for example, we cannot infer from the
by the results plotted in Fig. 18. The curve in experimental data on water the relative magni-
Fig. 18 was calculated by means Of Eq. (52). tude of the inter- and intramolecular eR'ects
It appears that T2 is approaching, at the lowest which were treated separately in Section X.
temperatures, its asymptotic value. The limit Possibly with improvements in the accuracy of
approached by the line width is indicated as measurement of relaxation times such points
about 16 gauss, which is in good agreement with could be tested. On the other hand, further
the va1ue calculated from the crystal structure re6nement of the theory, in the direction of a
of ice, assuming the nuclei at rest. more faithful representation of the molecule and
It is interesting to note that although the its immediate environment, would appear to be
relaxation time T», for water at 20 C, is about neither easy nor, at present, very pro6table,
equal to that of ice at — 5'C, the two cases except perhaps in the case of a structure like
represent opposite extremes of behavior. In water that of ice which is mell determined.
2xv7, &&j., and the line is narrow; in ice 2xvr, &&1, Since rotational and translational degrees of
and the line is broad. freedom play such an important role in deter-
XVL CONCLUDING REM)QkKS mining line widths and relaxation times, the
study of nuclear resonances provides another
Ke have been able to account for the line means for investigating those phenomena in
widths and spin-lattice relaxation times observed solids which have been ascribed to the onset, or
in a number of hquids, in paramagnetic solutions, cessation, of a particular type of internal motion.
in hydrogen gas, and in ice. In particular, the Experiments of this sort have already been
extremely small width of the resonance line in reported by Bitter and collaborators. ~
ordinary liquids, and the accompanying inverse An understanding of the factors controlling
dependence of T» upon the viscosity of the T» and Tm should be helpful in planning experi-
liquid, appear as consequences of a situation in ments to detect nuclear resonances and to meas-
which all perturbing 6elds Ructuate in a random ure nuclear gyromagnetic ratios. Both T» and
manner at a rate much higher than the nuclear
Larmor frequency —
a situation without an obvi-
ous parallel in atomic or molecular spectroscopy.
¹
"F. Bitter, N. L. Alpert, H. L. Poss, C. G. Lehr, and
S. T. Lin, Phys. Rev. 71, 738 (1947); L. Alpert, Phys.
Rev. 72, 637 (1947).
 aLOEMBERGEN, PURCELL, AND POUND
T~ must be taken into account in estimating by Bloch, ' the main diR'erence between the two
the sensitivity attainable. e8'ects is that the signal caused by the dispersion
Finally, we want to emphasize that our e6ect increases asymptotically to a maximum
understanding of the main features of the spin- as the r-f field H~ is increased, whereas for the
lattice relaxation process does not extend to signal caused by the absorption there is an
tightly bound crystalline structures in which, optimum value for Hi. As a result, one of the
presumably, only vibrational degrees of freedom eA'ects may be the more useful for some purposes
are active. Ke have observed in a fluorite crystal and the other for others. For conducting a search
at room temperature a relaxation time of about for unknown lines, an apparatus using the dis-
9 seconds. This alone is surprising, for the %'aller persion effect might be best since the search
theory applied to this case predicts a relaxation need not be made to cover a range of values of
time of the order of one hour. A more remarkable r-f field strength and modulation amplitude. For
result is that reported by Rollin and Hatton, ' investigating line shape and relaxation phe-
who find that the relaxation time in lithium nomena the absorption effect gives results that
fluoride remains of the order of seconds down are probably easier to interpret.
to 2'K. In all experimental arrangements so far em-
ployed the sample is contained in a resonant
ACKN0%MBGMENT
circuit, most often in the coil of a shunt resonant
The authors have benefited from numerous coil and condenser. In Section I it was shown
illuminating and stimulating discussions with that a quantity, called there the Q of the ab-
Professor H. C. Torrey of Rutgers University, sorption, was given by
Professor J. H. Van Vleck, Professor J. S.
Schwinger, and with Professor C. J. Gorter of the
1/Q=—8 =y'k'NpI(I+1)vo T2*/&AT. (56)
University of Leiden, Visiting Lecturer at This quantity is the ratio of the energy lost per
Harvard in the summer of 1947. radian from the r-f field to the system of nuclear
The work described in this paper was sup- spins to the energy stored in the oscillating r-f
ported by a Frederick Gardner Cottrell Special field. Thus it is the same kind of a quantity as
Grant-in-Aid from the Research Corporation. the ordinary Q used as a figure of merit for
simple coils and condensers. The effect of the
nuclear resonance on the circuit is, then, to
The intensity of the radiofrequency signal reduce the Q of the circuit from Q0 to a value Q„
resulting from nuclear paramagnetic resonance given by
absorption varies over a very wide range de-
pending on the various parameters descriptive
1/Q. = (1/Qo)+f'&. (57)
of the sample. For a large signal such as that The quantity f is a filhng factor for the circuit
produced by the protons in water, little attention determined by the fraction of the total r-f
need be given to the detection technique. Other magnetic energy that is actually stored in the
samples often require extreme care lest a very space occupied by the sample.
small signal be lost in noise. To see what experi- To obtain the largest signal for detection with
mental conditions are imposed by the nature of a circuit of a given Q, it is now apparent that the
the sample it is useful to develop a formula that addition of the loss attributable to 8 should
expresses the signal-to-noise ratio as a function produce the largest possible change in the voltage
of the parameters descriptive of the apparatus across the tuned circuit. This is achieved when
and of the sample. the generator driving the circuit is of the constant
It might be well to point out that, though current type, or, in other words, if the generator
here we work entirely with the nuclear absorp- has a very high internal impedance compared to
tion, the detectability of the nuclear dispersion, the impedance of the tuned circuit at resonance.
observed with an apparatus of the bridge type To achieve this the very small coupling con-
when the residual signal is due to a phase densers are used on the input sides of the circuits
unbalance, is almost exactly the same. As shown described in Section II.
 NUCLEAR MAGNETI C RESONANCE ABSORPTION 7ii
Consider the use of a modulated magnetic field. bering that H~ is one-half the amplitude of the
At times when the field is such that the nuclear oscillating field, Eq. (59) may be used to compute
resonance does not occur the voltage across the the optimum value for Po, giving
resonant circuit is proportional to Qp. At the = pop V,/2wy'TgTpogp,
Pp (60)
peak of the resonance, the voltage is proportional
to Q„which, since 8 is very small compared to where U, is an efFective volume for the part of
1/Qo, may be written the circuit in which magnetic energy is stored.
For a simple solenoid this is very nearly the
Q. = Qo(1 i &—
go)
volume inside the coil, since the field outside is
The resulting modulated voltage may be ex- weak, and the energy is proportional to the
panded in Fourier series if the shape of the square of the field. This equation may be put
absorption line is known, the signal indicating into many other forms. One useful form is
the presence of nuclear absorption being con-
tained in the sidebands. For observation of the
P = HpHp V./2QpTg,
p

effect on an oscillograph all of these should be where H2 is the full width in gauss of the ab-
taken into account but for a system using the sorption line between points of half the peak
"lock-in" amplifier, sensitive only in the imme- absorption, assuming the damped-oscillator line
diate neighborhood of a single frequency, only a shape. This shows that for lines having widths
single pair of sidebands difFering in frequency determined by inhomogeneity in the magnetic
from the driving frequency by just the modula- field, the input power required to produce satu-
tion frequency need be considered. The signal- ration depends on none of the properties of the
to-noise ratio computed from these also gives an sample except T~. Putting Eq. (60) into Eq. (58)
estimate of the usefulness of an oscillograph as for I'0 and using one-half the value of 8 given by
an indicator, if the noise bandwidth of the Eq. (56), one gets, for the total signal power
system is suitably chosen. available,
The voltage across the tuned circuit may be
considered to be given by
P, = U, Qp(A l')'k4Npoy'vp'Too
X P(I+ 1)]'/1152 Tg(k T)'.
8 = (2RP p) &[coaoot+4~i f'bgo cos(roo+&o ) E
If the detection apparatus is described by an
+ ,'Afbgp -cos(&ap op~)t], —
effective noise band width 2I3 (I3 being the
where R is the shunt resistance of the resonant band width at the indicating instrument) and
circuit, I'0 is the power dissipated in the resonant "
an efFective over-all noise figure, F, the resulting
circuit, coo is 2x times the radiofrequency, co is signal-to-noise power ratio is
2m times the modulation frequency, and A is a
constant, of the order of unity, that depends on
P,/P = V.go(A'f)'k'No'y'vopT p*

the exact shape of the line, the amount of the

X [I(I+1)]'/2304(kT)'(k TBF) Tg. (61)
modulation swing and the shape of the modu- The two factors containing k T are kept separate
lating wave. Since this voltage is for the unloaded to emphasize that the noise figure of the detector
circuit, the power available as a signal in the is not independent of the temperature. If Ii is
sidebands is large, most of the noise originates in the appa-
ratus and not in the circuit connected to its
P, = P p(A /ago)'/32. (58)
input terminals.
It follows from the analysis, in Section IV, of In most cases the output indicator mill respond
the saturation e6'ect that there is an optimum linearly to the signal and noise amplitudes and,
strength of the driving signal field II, , given by thus, the ratio of the response to the absorption
line to the r. m. s. noise fluctuation is
(pe)'TgTp* = 1. (59)
For this magnitude of the driving signal the
A./A = VePgoPAik'¹yvoPTp*~I
value of b is reduced to one-half its value in
X (I+1)/48kT(kTBF)&Tg&. (62)
Eq. (56) because of partial saturation. Remem- '0 H. T. Friis, Proc. I.R.E. M, 419 (1944).
To demonstrate completely the dependence on as the concentration is reduced, interactions
the frequency and on the dimensions of the other than the spin-spin interaction among the
circuit an empirical relation showing the de- nuclei under investigation remain negligible,
pendence of Qp on these quantities may be The signals produced by the nuclei in liquids,
inserted. Thus if th'e relaxation time T~ can be made so short
Qp = 0.037 vp& V.', (63) as to be an important contributant to the line
width, are particularly strong. This condition
where ~0 is to be measured in c.p. s, and V. in
cm' may be used. "1'he numerical constant is can be obtained by the use of solutions con-
taining paramagnetic ions to shorten TI sufh-
based on experimental values of Q for coils of
ciently to make Tp*/T, =1. For this case, Kq.
about 1-cm' volume in the region of 30 Mc/sec.
Often the frequency of operation is determined (64) may be written as,
by the magnetic field available. In such a case, A, /A„=1.2 X10 "ItIHo'I'p'M'
using Eq. (63) for Qp, Eq. (62) becomes X (I+1)/ V. PBPP'4, (65)
A, /A =1.62X10~VJiAIp'Roy""Ho'" where p, is the magnetic moment in nuclear
XTp'&I(I+1)/kT(jpTBF)&Tg&. (64) magnetons. This gives a signal-to-noise ratio of
Observe that the signal-to-noise amplitude ratio 1.4X10' for the protons in a 1-cm' water sample
is proportional to y"", for a given field, concen- with a magnetic 6eld of 15,000 gauss and a band
tration, volume of the circuit, filling factor for width of 1 c.p. s. Thus even with an oscillograph
the circuit, and ratio of T~/Tp*. The quantity and a high modulation frequency, requiring a
V.iXp is the total number of nuclei present. noise band width of perhaps 5 kc/sec. , a signal
Thus Eq. (64) shows that the signal-to-noise of the order of 2)&10' times the r.m. s. amplitude
ratio is inversely proportional to V,& for a given of noise may be expected. On the other hand,
number of nuclei and for all the other quantities a 1-molar solution of nuclei having magnetic
constant. This favors as high a concentration as moments of 0.1 nuclear magnetons produces a
possible. Note, however, that if the sample is a signal-to-noise ratio of only about 1.3 when a
crystal with a rigid lattice so that T& is deter- 1-cm' sample in a 15,000-gauss field is used with
rnined by the static dipole-dipole coupling, it is an apparatus having a 1-c.p. s. band width. For

|
preferable to have the nuclei spread out. In this this magnetic field, band width, and volume of
event, the signal-to-noise ratio is proportional circuit, about 10" nuclei having spins of 2 and
to V,"', and the total number of nuclei in the magnetic moments of one nuclear magneton
sample being held constant. This is true only if, would be required to give a signal equal to noise.