Oxford University Press, Amen House, London E.C.4 GLASGOW NEW YORE TORONTO MELBOURNE WELLINGTON BOMBAY CALCUTTA MADRAS EARACHT KUALA LUMPUR CAPE TOWN IBADAN NAIROBI ACCRA © Oxford University Press 1961 PRINTED IN GREAT BRITAINPREFACE People have now-a-days got a strange opinion that everything should be taught by lectures. Now, I cannot see that lectures can do so much good as reading the books from which the lectures are taken. Dr. Joungon (1766) Since the first successful detection of nuclear resonance signals late in 1945, nuclear magnetism has developed at a pace which after fifteen years still shows no sign of slackening. Besides its first and obvious application to the measurement of nuclear moments, it has become a major tool in the study of the finer properties of matter in bulk. Struc- ture of molecules, reaction rates and chemical equilibria, chemical bond- ing, crystal structures, internal motions in solids and liquids, electronic densities in metals, alloys, and semiconductors, internal fields in ferro- magnetic and antiferromagnetic substances, density of states in super- conductors, properties of quantum liquids, are some of the topics where nuclear magnetism has so far provided specific and detailed information. Detailed studies of nuclear relaxation processes combined with electron spin resonance have led to original methods of nuclear alignment which may in the near future open up new experimental possibilities to nuclear physicists as well as to the specialists of very low temperatures. On a more fundamental level, systems of nuclear spins provide examples of problems in statistical mechanics and in the theory of irreversible processes that are sufficiently simple to allow a clean and unambiguous solution both theoretically and experimentally, but are still sufficiently complex to be far from trivial. Thus the hitherto purely academic concept of negative temperature has found its full physical significance in nuclear magnetism. Thus also the trend toward thermal equilibrium of a system (the spin system) coupled to a thermal bath (the lattice) can be investigated, theoretically and experimentally, more fully than is usually possible in other fields of physics. One of the more remarkable features of nuclear magnetism and, at least for the author of this book, the most deeply satisfying, is the very close connexion between theory and experiment that leaves little room for a theory that could not be tested by a suitable experiment or for an experiment that does not admit of a well-defined theoretical interpreta- tion; this perhaps to a greater extent than in any other branch of modern physics.vi PREFACE Such at any rate was the view that has presided over the writing of this book. It is an attempt to provide an account of the principles and methods of nuclear magnetism that would be acceptable to the theorist and of use to the experimentalist ; a wish often expressed, seldom ful- filled, that stands a better chance, it is hoped, in a field where theorist and experimentalist as often as not are the same person. On the contents of this book the following comments can be made. The experimental methods are discussed very briefly and in very general terms. The lack of space and the lack of competence of the author have precluded any detailed description of the experimental techniques currently in use. One can only hope that the definite need for a monograph dealing with these problems will be satisfied in the future. There are no tables of important constants such as nuclear moments, quadrupole couplings, chemical shifts, relaxation times, and spin-spin couplings. The amount of labour involved in the compiling of such data, the space requirements, and also the danger of rapid obsolescence have led to the conclusion that their presentation in review articles in a form comparable to that currently used for electron resonance data is much more suitable. For the same reason it did not seem possible nor indeed desirable to include an extensive bibliography of the published work in the field of nuclear magnetism, which at the time of writing contains well over 2,000 titles and is continually increasing. The selection of the references included in this book was made on the following basis. The viewpoint adopted is that in a book, in contrast to an article, no references need be given for credit and that it is the privilege and indeed the role of the author to blend together the contri- butions of the various workers in the field and to modify them as he sees fit, for the benefit of the reader. In fairness to all concerned it should therefore be made perfectly clear that the references given in each case are not necessarily considered by the author as the most original or the most important contributions in the field but rather as convenient sources of information to supplement that given in the book. To give one example among many, the important work of Kubo and Tomita, which has influenced the author very strongly, is not quoted because it was felt that the gist of its contents was dealt with in sufficient detail to make further reference to the original paper less useful for the reader than that to some other, perhaps less important, work. Under such circumstances it was clear that a name index would serve no useful purpose and it has accordingly been omitted.PREFACE vii To conclude, one may ask what category of readers I had in mind in writing this book. This is a difficult question which may be answered in part by saying what kind of book I would have liked to have written. What I had in mind was to attempt for nuclear magnetism what Van Vleck has done for the theory of electric and magnetic suscepti- bilities in a book which thirty years after its publication is still the book on the subject. A demi-succés in this undertaking would be good enough. It is not possible to acknowledge in detail my great indebtedness for illuminating discussions and constructive criticism to so many friends and colleagues and especially to the members of our group at Saclay and to our visitors from abroad. I would like, however, to thank parti- cularly ‘the long legg’d spinner’, my good friend Ionel Solomon who has patiently coached me in the mysteries of the lock-in detector and who has kindly prepared some beautiful photographs with the express purpose of illustrating this book. The planning of the book and the writing of many chapters have benefited from the help and advice of Professor R. V. Pound more than I can say. Although I must take full responsibility for the entire book, those familiar with his work will have no difficulty in discerning his influence. Let him be assured of my deep gratitude. I am very grateful to many colleagues and to the editors of the following journals for permission to reproduce a number of figures from these journals: Acta Metallurgica—Comptes Rendus de V Academie des Sciences—Journal of the Physical Society of Japan—Molecular Physics— Nature—Physica—Physical Review—Physical Review Letters—Proceed- ings of the Physical Society—Proceedings of the Royal Society—Science— The Journal of Chemical Physics—The Physics and Chemistry of Solids. My special thanks go to those authors who kindly took the trouble of sending original photographs. I am greatly indebted to Madame Houzé and Madame Gugenberger for their help with the preparation of the manuscript and to Madame Gugenberger for her competent handling of the proof. _Mr. Timothy Hall has kindly prepared the subject index. A.A. Saclay October 1960CONTENTS I. GENERAL INTRODUCTION A. B. Cc. D. Nuclear paramagnetism Special features of radio-frequency spectroscopy The phenomenon of resonance; ‘resonant’ and ‘non-resonant’ methods (a) Boam measurements of atomic, molecular, and nuclear mag- netic moments (b) Measurement of the magnetic moment of the neutron (c) The fine structure of the hydrogen atom (d) The fine structure of positronium (e) The magnetic moment of the -meson (f) Resonance detected by optical means (g) Perturbed angular correlations Nuclear magnetic resonance (a) The principle (b) Various extensions and generalizations REFERENCES If. MOTION OF FREE SPINS A. B. Cc. Classical treatment Quantum mechanical treatment Quantum mechanical description of a statistical ensemble of free spins, Density matrix D. Relation with the perturbation method E. F. ‘Transient effects (a) Free precession (b) Spin echoes (c) The adiabatic theorem, adiabatic passage The general problem of two levels coupled by an rf. fisld. The fictitious spin } : REFERENCES III. MACROSCOPIC ASPECTS OF NUCLEAR MAGNETISM L A. B. IL. A. B. Cc InrropuctTIon Static susceptibility Resonant absorption of r.f. energy . THE PHENOMENOLOGICAL Equations or BLocH Steady-state solutions—saturation, Steady-state solutions in an inhomogeneous field Modified Bloch equations in low fields 24 27 32 32 33 34 36 38 39 39 39 40 44 45 49 53x CONTENTS Ill. Transrent MetHops 1n Nucrear MacGnetism 57 A, The method of the spin echoes 58 {a} Spin diffusion 59 (8) Coherent and incoherent pulses 62 B, Free precession 63 (a) Free precession in the earth's field 64 C. Adiabatic passage 65 Applications of the fast-passage methods 66 (a) Measurement of the relaxation time T, 66 (0) Measurement of 7 in liquids 67 (c) Other applications of adiabatic fast passage 68 D. The method of transient nutation 68 IV. Detection METHODS 7 A. General 71 B. Detection of steady-state nuclear signals 75 {a) Q-meter detection 75 (b) Bridge and crossed coils methods 76 (c) Marginai oscillator 7 (d) Audio-modulation, narrow band amplification, phase sensi- tive detection, signal-to-noise ratio 78 (1) Lock-in detection and signal-to-noise ratio 79 (2) Signal-to-noise ratio 82 (e) Transient: effects in steady-state detection 85 C, Transient methods of detection 86 (a) Adiabatic fast passage 86 (b) Pulse methods, coherent and incoherent pulses 87 D. Negative absorption—masers 89 AppENDIX. Proof of the Kramers-Krénig relations 93 REFERENCES 96 IV, DIPOLAR LINE WIDTH IN A RIGID LATTICE 97 I. IntTRopvuctTion 97 A. The local field 97 B, General theory of magnetic absorption 98 Il, Broapeyine sy Lixe Spins 108 A, Dipole-dipole interaction. 103 B, Definition of the moments 106 C. Principle of the calculation of momenta 108 D. Caleulation of the second and fourth moments ait E. Relationship between the line shape and the free precession signal 114 F. A comparison between theory and experiment 15 ITI. Dirrotar BRoaDENING BY UNLIKE Srins 122CONTENTS IV. DipotaR BROADENING IN MAGNETICALLY DILUTED SuBpsrances A. The method of moments B. The statistical theory V. MopiricaTions IN THE DIPOLAR BROADENING CAUSED BY THE EXISTENCE OF QUADRUPOLE CouPLINGs REFERENCES V. SPIN TEMPERATURE A. Non-interacting spins B. Interacting spins in high field C. Interacting spins in low fields D. Zeeman system with more than one spin species E, Dynamics of thermal spin-spin processes REFERENCES VI. ELECTRON-NUCLEUS INTERACTIONS I. Exzerrosratic CoupLines A. The Hamiltonian B. Tonic crystals C. Molecular crystals Il. Maenetic INTERACTIONS A, The coupling Hamiltonian B. The effect of electron-nucleus coupling in diamagnetic substances (a) General (b) Calculation of the chemical shift (c) Indirect interaction between nuclear spins in diamagnetic substances (1) The orbital coupling (2) The Heitler-London approximation (3) The method of molecular orbitals C. The effect of electron-nucleus coupling in paramagnetic substances (a) Non-metals (1) Nature of the coupling (2) Observability of nuclear resonance (6) Metals (1) The frequency shift in metals (2) The indirect interactions in metals D. Nuclear resonance in antiferromagnetic and ferromagnetic substances REFERENCES xi 125 125 126 128 132 133 133 136 144 150 154 158 159 159 166 169 170 73 173 175 183 184 186 190 191 191 191 193 199 199 206 210 214xii CONTENTS VII. FINE STRUCTURE OF RESONANCE LINES— QUADRUPOLE EFFECTS 216 I. Fine SrrucruRE CAUSED RY DipoLaR CourLine 216 A. Rigid lattice 216 (a) Two identical spins (two protons) ' and 1? 216 (8) Systems of more than two spins 292 B. Nuclear resonance in solid hydrogen 223 (a) Introduction. System of two interacting protons 223 (b) Solid hydrogen 295 (c) Ortho- and para-hydrogen 226 (a) Crystalline potential 227 (e) Magnetic resonance in a strong field 228 (f) Magnetic resonance in zero field 231 (g) Magnetic resonance in HD and D, 231 Il, Ensroy Levets or Nucuzar Spins 1n THE PRESENCE oF QuADRUPOLE INTERACTIONS 232 A. High magnetic fields 233 (a) Energy levels in single crystals 233 (b) Imperfect eubie crystals 237 (1) Powder pattern 237 (2) First-order broadening in imperfect crystals 237 (3) Transient methods, multiple echoes 241 (4) Second-order quadrupole broadening in imperfect crystals 246 B. Low magnetic fields 249 (a) Zero field spectra 249 (2) Integer spins 250 (2) Half-integer spins 251 (b) Zeeman splittings of quadrupole levels 253 (1) Integer spins 253 (2) Half-integer spins 254 (c) Transient methods 257 (1) Transient magnetization in zero field 257 (2) Transient magnetization in a small magnetic field H, 260 APPENDIX, Sign of the quadrupole coupling 261 REFERENCES 263 VIII, THERMAL RELAXATION IN LIQUIDS AND GASES 264 I. Intropuction 264 A. Coupling of the nuclear spins with the radiation field 264 B. Coupling of the spin system with the lattice 267 II. Retaxation 1n Liguips anp Gases 268 A. General 268 B. Definitions 270CONTENTS C. Motion of a system subject to a perturbation which is a random function of time (a) Transition probability (b) The master equation for populations (c) The master equation for the density matrix (2) The master equation in operator form (e) Macroscopic differential equations (f) Summary of the notation introduced in this section, (g) Justification of the four assumptions leading to the generalized master equation D. Quantum mechanical formulation of the problem E. Relaxation by dipolar coupling (a) Like sping (b) Unlike spins (c) Correlation functions resulting from random molecular rota- tion or translation (1) Rotation (2) Translation. F, Other mechaniams of relaxation in liquids (a) General (6) (1) Scalar spin-spin coupling (2) Scalar relaxation of the first kind (3) Scalar relaxation of the second kind (c) Quadrupole relaxation in liquids through molecular reorienta- tion (d) Relaxation through anisotropic chemical shift combined with molecular reorientation G. Nuclear relaxation in gages (a) The H, molecule—diatomic molecules (b) Relaxation in monatomic gases ILL, Comparison BETWEEN THEORY AND EXPERIMENT A. Dipolar coupling between like spins (a) Short correlation times, relative values of T, (b) Absolute values of 7, (c) Long correlation times B, Coupling between unlike spins (a) Single irradiation methods (8) Double irradiation methods (1) The HF molecule (2) Coupling between a nuclear spin and an electronic spin ©. Electric quadrupole relaxation in liquids D. Nuclear relaxation in gases (a) Nuclear relaxation in hydrogen gas xiii 272 274 276 278 280 281 232 283 289 290 294 297 298 300 305 305 306 308 309 313 315 316 316 322 323 324 324 326 327 328 328 333 333 346 349 349xiv CONTENTS (b) Nuclear relaxation in liquid hydrogen (c) Relaxation in monatomic gases REFERENCES IX. THERMAL RELAXATION AND DYNAMIC POLARIZA.- TION IN SOLIDS I, Conpuction Evecrrons and Spin-Lattice RELAXATION in METALS A. An elementary calculation of the relaxation time B. Nuclear relaxation time and spin temperature C. Dynamic nuclear polarization in metals (the Overhauser effect) (a) Fermi statistics and non-equilibrium electron spin distribution (b) Dynamic polarization (c) Coupled equations for nuclear and electron spin polarization D. Comparison with experiment (@) Measurements of 7; (’) Dynamic polarization experiments (c) Dynamic nuclear polarization in metals at the temperatures of liquid helium II. Nuctear RELAXATION CAUSED BY Fixep PaRramaa- NETIC IMPURITIES A. Theory B. Comparison with experiment III. Magnetic Renaxation anp Dynamic PorarizaTION In SEMICONDUCTORS AND INSULATORS A. Relaxation by conduction electrons in semiconductors B. Dynamic polarization by fixed paramagnetic impurities—solid state effect IV. RELAXATION BY THERMAL VIBRATIONS IN A CRYSTAL- LIne Larrice A. Lattice vibrations and phonons B, Transition probabilities induced by the spin-phonon coupling C. Magnetic and quadrupole relaxation by spin-phonon coupling (a) Magnetic relaxation (b) Quadrupole relaxation D. Ultrasonic experiments (a) Quadrupole transitions (6) Magnetic transitions REFERENCES x. THEORY OF LINE WIDTH IN THE PRESENCE OF MOTION OF THE SPINS I. INTRODUCTION 350 352 353 354 355 356 359 364 364 367 368 370 370 313 375 378 379 386 389 389 392 401 402 404 409 409 411 417 419 421 423 424CONTENTS Il, Tre Aprapatic Case A. General theory B. Exchange nerrowing C, Brownian motion narrowing III, Tue Non-Aprasatic Line WiptH A, Line width and transverse relaxation time B. General case IV. Destruction oF FINE STRUCTURES THROUGH MOTION V. InrLuence or InteRNAaL Morrons IN SoxrrDs ON THE Wipth AND RELAXATION PROPERTIES OF ZEEMAN RESONANCE LINES A. Rotational motions B. Translational diffusion in solids VI. Inrivence or InrErRNAL Motions IN SOLIDS ON THE WiprH AND RELAXATION OF QUADRUPOLE RESONANCE LInEs A. Torsion oscillations (a) The spin Hamiltonian (b) The line width (c) Relaxation time (a) The spectral densities B. Hindered rotations (a) Fast motion (8) Slow motion REFERENCES XI. MULTIPLET STRUCTURE OF RESONANCE LINES IN LIQUIDS I. Evercy LEVELS opseRvED BY ConTINUOUS WavE MetTHops A S lorl- In the frame rotating with the velocity —w the applied field H, and the then static component of the r-f. field can be replaced by a large effective static field (Fig. II, 1) H, = —[(wtoy)?+att = —2, lyl Y making with H, the small angle 6, where tan = ,/(w)+w). In this frame the remaining rotating component +w is the sum of three fields: an oscillating field parallel to H,, of frequency 2w and amplitude H, sin 6, and in the plane perpendicular to H, a field of amplitude 4H,(1+-cos 0) rotating with the frequency 2w, and a very small field of amplitude 4H,(1—cos #) rotating with a frequency —2w. If we neglect the effect22 MOTION OF FREE SPINS Ch. IL of the parallel field, the resonance condition for these two rotating components is Q? = 4w* or -{w—w)(3w-+ 09) = w?. (8) One solution of (8) for Q = +2w is to a first approximation 2 H, 2 = or aa). weet ge oft +(2) | (9) The resonance is shifted by a relative amount (H,/2H,)*. The existence of this shift was established theoretically at an early date (1). The second approximate solution of (8) corresponds to Q=-2%w or we —hwy. (10) Thus there is a second resonance for a frequency approximately one- third of the Larmor frequency of the spin. This resonance corresponds to a much weaker phenomenon than the principal one because the magnitude of the field rotating at the frequency —2w is 1—cos6 @ Af oo, \ 3H,\? h— ~~ tues TaH} * A more elaborate calculation shows that higher order resonances exist for |wo/w| = 2n+1, where » is an integer. The existence of these subsidiary resonances was demonstrated theoretically (2) and experi- mentally (3) in connexion with optical pumping experiments. For the ratios (H,/H)) usually met in nuclear resonance these effects are very small and will be neglected from now on in this discussion. B. Quantum mechanical treatment In the Heisenberg representation the equations of motion of the angular momentum operator AI are = [#,1] =[—yaH.1,0, (11) where 4 = —yii(H.1) is the Hamiltonian describing the coupling of the spin with the applied field. For the z component Bel RCH e]+ Hilly 1) =, Hy—1, He) = AEH i dt (12) This equation has the same form as the classical equation (1). This identity has the following consequences: the expectation value I,> taken over the wave function of a free spin I also obeys the classicalCh. ID MOTION OF FREE SPINS 23 equation and can therefore be calculated classically. This result will first be applied to the problem of a spin J = } in a magnetic field, sum of a dic. field Hy along the z-axis and a field H, rotating in the zy plane. Let p, and p_ be the probabilities of finding the spin in the states I= +} and [= —4} respectively: (L) = 4(p,—p-) = H1—2_). If at time t = 0, p, = 1, p_ = 0, at time ¢, (1,()) will be $ cosa, where « is the angle calculated classically by the formula (7), and hence p- = P(t) = }(1—cos«) = sin*@sin® fat = sine gat. (13) ‘This is the well-known formula of Rabi for a spin } in a rotating field. Since equations (12) are linear, it follows that they are also obeyed by the total moment of a large number of identical, non-interacting spins. Furthermore, when the components of angular momentum, measured in units of 4, are large numbers, it is permissible to treat them as classical quantities. It follows that the classical equation (1) describes the observable behaviour of a sample containing a large number of identical spins provided interactions among them can be neglected. An explicit solution of the Schrédinger equation describing the motion of a spin J in a rotating field will now be derived: i = a = —y{Hy L+H (1,008 wt +1, sin wt) = fool phoy(L, 4 T etetyyy, (14) where I, = I,+iJ,. Transformation to the rotating frame of coordinates is equivalent to the substitution ~ = Uy, = exp(—iwi,t)y,. The Schrédinger equation in the new frame is if, = (0% V-i> 0}, (18) From the commutation relations one obtains UAL, U = eletl: I, e-toth = J, gold], e-twld — giot] ells] g-iwlt = g-twty_ U4 #U = Wey ],+-o;1,) ihe = (oo, toy Lhe (18) This integrates immediately to te = expl—if{(o— 4) Lt LE} (0), . (15')24 MOTION OF FREE SPINS Ch. IL where ¥,(0) = (0), p = exp(—iwl, 1) exp{—ia(n.1)9y(0), (17) where a is given by (5’) and n is a unit vector of components n, = cos 6, n, = sin, n, = 0. From (17) it is easy to compute the probability amplitude 4,,,, and the probability Pym = |Ajvm|* for finding the spin in the state m’ at time ¢, knowing that it was at time t = 0 in the state m: Amnm = (m' W(t) = (m'lexp{—(iwl,t)}exp{—ia(n.I)}|m), (18) Pam = {(m' |exp{—ia(n.1)¢}|m) |?. (18’) (18’) is easily computed in the case of a spin }. In that case I = }o where o,, oy, o, are the Pauli matrices which have the following properties: (no)? = |n? = 1, (n.)??4 = (no), exp{—ia(n.I)t} = exp{—}ia(n.o)t} = cos fat—i(n.o)sin fat, and therefore P_43 = |(—4]cos fat—i sin 4at(cos 0 o,+8in 8 c,,)|$)|? = sin?@ sin? tat, (19) which is the same as (13) already proved by a semi-classical argument. In the general case J > 3, A,,, can be obtained explicitly by using the theory of the irreducible representations of the group of rotations. Rather than reproduce the long derivation here only the formula for Pim i8 given (4)? Pam = (G08 $x)*7(I-+-m)!(I-++-m’)! (I—m)! (I—m')! x or 2mm! x [2 (tan $a)" NOcmpm\(I--moay(iom—aif? 9) where sin? ta = sin*@sin®(4ad). C. Quantum mechanical description of a statistical ensemble of free spins. Density matrix Consider a sample containing a large number N of identical nuclear spins, for simplicity spins 1 = }, which has been placed in a static field H, for a long time. A measurement of the macroscopic nuclear mag- netization will, as is well known, yield 2 value M, = M, = x. Hy, where Xo is the static nuclear susceptibility, in a direction parallel to the field, and values M, = M, = 0 at right angles to the field. One may investi- gate whether this behaviour is compatible with a description of all individual nuclear spins by the same wave function: =a,£+a,, where |a,|?+Ja,/*? =1 (20) 1 »Ch. IT MOTION OF FREE SPINS 25 and é and 7 are eigenstates of the operator i, with eigenvalues +4 and —}. One finds easily 3- 2 = NGi,) = NWligify = Na, -—loa\ TE = Nidlig) = Natay - = = Na, ay The first equation (21) is self-evident and has been used already in the present discussion: |a,|? and |a,|* are the probabilities p, and p_ for finding each spin in the state +} or, as one often says, the relative populations of these states. However, the second equation (21) shows clearly that unless a, or a, = 0 (which is a case of complete polariza- tion for all the spins of the sample), the magnetization of the sample should have a component in the equatorial plane in contradiction with experiments. Thus the hypothesis that the coefficients a, and a, are the same for all the spins of the sample is untenable. Instead one must assume that they are different for different spins and that the average of the product at a, taken over the whole sample oa 1 et asd, = yeu =0. In the same way one must use the averages |a,|? = p, and ja, = p_ for the relative populations of the two levels. If now a rotating field Ai, is introduced suddenly at time t = 0, for a duration ¢, the coefficients @,(t) and a,(t) can be calculated as linear combinations of a,(0) and a,(0) using the probability amplitude (18), and the average a¥(t)a,(f) can be computed. Its value can actually be predicted very simply since we know from the equations (1) the motion of the macroscopic magnetiza- tion of the sample. Thus at resonance, for instance, for w = w, the magnetic moment will have precessed around Oz by an angle w»t and at the same time around the axis Oz of the rotating frame by an angle w,t. The result is M(t) = My cosa, t, M,(t) = Mysin(w, thetot and, comparing this with the equations (21), one obtains AF (E)ag(8) = 4{]a(OV7*—[ag(0) F}sin(a, tem, [ay(#)[?— |ag(2)|? = {|a4(0)?— |a2(0) "Joos w, t. (22)26 MOTION OF FREE SPINS Ch. IT From this very simple example one can conclude that the description of an ensemble of spins i by the relative populations p,, = |d_/* of their energy levels Z,, is in general incomplete and should be supple- mented by the knowledge of cross-products such a8 a,,a%,. This result is well known in quantum statistics where one introduces a statistical operator p represented by a so-called density matrix with matrix elements — Pan’ = (rtlp|m’) = Ody. (23) p contains all the information necessary for the description of a statistical ensemble of identical systems. In particular, the predicted value of any observable Q of the system is given by the equation = trace{pQ}, (24) of which the equations (21) are a special case. The variation of p with time is given by the equation ii dp za, = —[*#,p], 2 ih {#,p] (25) where # is the Hamiltonian of an individual system. This equation is very similar to Heisenberg’s equations of motion (note, however, the change in sign) and p(f) is related to p(0) through p(t) = U(t)p(0)U-(t), (26) where U(t) is a unitary operator. In the case of a time-independent Hamiltonian #%, U can be written U= exp| 11), (26’) % A complication, often encountered in practice, results from the fact that the systems (spins here) of the ensemble sometimes differ statis- tically in their Hamiltonians, as well as in their wave functions (coeffi- cients a,,) at t = 0. For instance, because of the inevitable inhomogeneity of the applied field over the dimensions of the ensemble, the Larmor frequencies of all the spins will be distributed over a corresponding range. In that case one should subdivide the statistical ensemble into sub-ensembles sufficiently large for each of them to contain a large number of individual systems (spins) but at the same time sufficiently small for the variations of the individual Hamiltonians to be negligible inside each of them.Ch. IT MOTION OF FREE SPINS 27 For each of the sub-ensembles one can then define density matrices pi(t) and p,(0) related by pilt) = Up, 0)07 *(2), (27) and then define an average density matrix p for the whole ensemble. It is quite clear that the relation between p(t) and p(0), thus defined, will not in general be unitary. A more detailed discussion and a generalization of the properties of the density matrix will be deferred until later. In the next section the results obtained so far will be compared with those of the conventional theory of transition probabilities which uses the perturbation method. D. Relation with the perturbation method In the perturbation method, the unperturbed system is the spin in the d.c. magnetic field Hj, described by the unperturbed Hamiltonian Hh = — phy I, = heyy (28) and the perturbation is the coupling of this spin with the rotating field, H, = H, cos wt, H, = H,sin wt, described by a perturbation Hamiltonian H, = — yh H, L+H, 1) = —AyhH (Le Leet), 28") The eigenvalues of % are of the form mfw, where the integer or half integer m takes 27+ 1 values from —J to +. Because of the properties of the operators J, 4%, has only matrix elements between states for which m differs by one unit. It is well known in quantum mechanics that a time-dependent perturbation of frequency w can only induce transitions between states separated by an energy interval AH = fiw, if w is in the neighbourhood of w». Physically this result expresses the conservation of energy, the change in energy fiw, of the system being compensated by the energy fiw of the photon which is absorbed or emitted. If H, is a true rotating field the equality w = w, has to be satisfied with respect to the sign as well. For an oscillating field 2H, cos wt, if |w| = |wo|, one of the two rotating components of the oscillating field has the right sense of rotation. In practice, the energy levels E, and E, will have a certain width and the energy difference AZ = E,—F, = hwy will be described by a distribution function p(AZ) around a central value AE° = fiw$, with { p(AB)d(AB) == 1. Instead of p(AZ) one can introduce distribution functions g(v9) = hp(hvy) or flw,) = Fip(iew,) also normalized through J 9(vo)dve = | flwe)dw = 1.28 MOTION OF FREE SPINS Ch. IT The width 5 of the distribution is then of the order of [f(w,)]-! on the w, scale. The transition probability per unit time induced by the rf. field of frequency w hetween two states |) and |b) is given by the well-known formula Way = = \(b\26 1a) | *9(A.2), 29) where we make AZ = fiw, which for the Hamiltonians (28) and (28’) yields Woamaca = 22 PE HB (iL, (m—1) PLE) = pat(T+-myT—m+ Vgbr). (30) In particular, for spin 4, Woy = Wag = W = goign) = dotafle). (30") It is interesting to compare the exact solution (19), (19’) with the approximate solutions (30), (30). For simplicity the comparison will be made only for spin $. According to (30') a spin in the state 4-4 at time t = 0 should, at time #, have a probability P = Wt of being found in the state —4, at least for such small values of ¢ that Wi < 1, since this result has been obtained by a first-order perturbation theory. On the other hand, the exact formula (19) gives, at resonance, for small values of t, the entirely different result, P = jw} #?, It is therefore very important to reconcile these results in order to understand why the perturbation method, widely used in the theory of nuclear resonance, does give results in agreement with experiment. Since formula (30’) was established under the assumption of a distribution f{w,) of the Larmor frequencies, the same distribution should be introduced in the exact treatment by multiplying the expression (19) by the shape func- tion f(w,) and integrating over w». This procedure gives Pe = ot [ Fy) dy (31) and a transition probability per unit time i in fete? = tot [Et flog) diy = ot [ EEE ot ua, “ (31) where u = wy—w. If it is assumed that the width 8 of the distribution is much larger than ,, for values of ¢ such that 8¢ > 1, f(w+) will be practically constant over the range of values of u which contribute significantlyCh. IT MOTION OF FREE SPINS 29 to the integral and can be taken out of the integral. One then gets dP aT 41? f(w) Jg(w t), (82) where J, is a Bessel function. Fur small values of the argument J) ~ 1 and P = trai f(w), as in formula (30’). (32’) One must remember, however, that (32’) is valid only under the assump- tion w,t <1 <8¢t and, since § = 1/f(w) > w, was assumed, one also has 2 We xe Wt Set é1 Although the equivalence of the exact treatment and of the perturba- tion theory for short times such that Wt < 1 has thus been established, the validity of the method of the transition probabilities over long times (Wt > 1) is still not established. Consider a statistical ensemble of spins with relative populations p,(t) and p_(t), yielding Ut) = WWryh{p,Q—p_(0}. If an r-f. field is applied at time zero the customary theory of transition probabilities leads to the following equations: dp, _ eB) dp. _ aE VPs), (33) which integrates to ‘ Pi —P_ = {p+(0)—p_(0)}exp(—2W2), M(t) = M,(0)exp(—2 Wt), (34) aM, dt On the other hand, from (32) one easily obtains aM, a —2WM,(0)Io(w, #)- (35) The two expressions (34) and (35) coincide for small ¢ such that Wt <1 (although not too small, since (35) has been established under the assumption 8¢ > 1) and they both vanish for ¢ very large. However, in the intermediate region the two results behave quite differently and, since (35) is the exact formula, one is inclined to question the validity of equations (33). or = —2WM,(0)exp(—2W1).30 MOTION OF FREE SPINS Ch. IT It is not really surprising that the transition-probability method could lead to an incorrect, or at least an incomplete, description since it deliberately overlooks the coherence of the applied rf. field and deals only with populations, that is, with diagonal elements of the density matrix p ignoring the off-diagonal ones. On the other hand, the comparison made above between the predictions of the formulae (34) and (35) relative to the destruction of the magnetization existing at time t = 0 by an rf. field applied at that time did not correspond to a usual physical situation. It was assumed that at time ¢ = 0 the sample had a magnetization M, = xq A) parallel to the applied field 1, which, as was shown earlier, corresponded to the absence of off-diagonal matrix elements pi, and p,; of p and to the inequality of the diagonal lement: he FT eemens Pu = [a [? =p, and. py = lagi? = p_. This situation is created by a relaxation mechanism which does two things: it tends to destroy the off-diagonal elements, if any, of p and to give to the diagonal elements, or populations, the values correspond- ing to thermal equilibrium. The two processes of which the first is usually called the spin-spin relaxation and the second the spin-lattice relaxation need not, and in fact usually do not, proceed with the same speed, but can be associated with time constants T, and 7, with T, > 7. Under the combined effect of the relaxation and of the rf. field a steady state is reached where the magnetization is different from zero and where the departure M,—M,, from the equilibrium value M), can be measured, In particular, if ther.f. field is sufficiently weak, this departure, proportional to the r.f. transition probability W, gives a direct measure~ ment of the shape function f(w) which describes the distribution of the energy levels of the spin system. It will be shown for a simple model that the transition probability method and the exact formula (19) lead in that case to the same result. Let us assume that each spin undergoes at random intervals with a mean value 7;, collisions of such a nature that there is no correlation between the values of the wave function (i) of the spin before and after the collision (strong collisions). Then, immediately after a collision, aa, = pe = ph = 0, [ay = pur = (Pado» igi? = Paz = (P-)o» where (p,)y and (p_), are the equilibrium populations.Ch. If MOTION OF FREE SPINS 31 These hypotheses, applied to the transition probability method, lead. to equations qj 1 Fe = — We P-)— 770d . (36) t-_ wep.) (bd dt an nia which have a steady-state solution =-_% "= Towa M-M, _ _2WT, Then ‘i, 1427, (37) For WY, <1, a 2—~2WT, (condition of no saturation). Qn the othe hand, to apply the exact formula (19) for the transition probability P(t) = (w?/a)sin*4at one must be careful to choose the starting time as immediately after a collision, since it is only then that the off-diagonal elements p,. vanish and (19) is valid. The equation giving p,(t) can then be written p= ool *\e(OPO+P.(O1—POD+ + ool (SE Oe Pet (ent Pe 1 1 ° (38) The first term is the contribution to p(t) from the spins which under- went no collision between the switching on of the rf. field and the time t, and the integral is the sum of the contributions from the spins which had their Zast collision at time ¢’. A similar relation can be written for p_(t). A steady-state solution can be obtained by making 1 go to infinity. It gives M,= 2 | [ exo ra 2PU)} 2 = M1 ———___*1 ___}. 39) xf annie 6) This may be written MoM, _ __—iT My ei T}+-(w—,)*T7+132 MOTION OF FREE SPINS Ch. IT To be equivalent to (37) and to establish thereby the correctness of the perturbation theory method for this ‘strong-collision’ model, one must ah inl 2W = (uf Gye TE+ 1, This is indeed correct for the model chosen and in fact the well-known Lorentzian shape factor qt, i Se) = * Fe aT characteristic of spectroscopic lines emitted with mean life 7), can be recognized. The model is far too special to be characteristic of the general situation in magnetic resonance spectroscopy. For example, such a situation as a distribution of Larmor frequencies w, in the en- semble must be treated by multiplying both sides of (39) by the shape function f(w,) and integrating over w,, If it is assamed that the width 8 of this function is much larger than both w, and 1/7,, such integration gives (39) 2 ow M, = ul - ay = Mh — py . ON eat TD | er 0/ TD} For strengths of the rf. field so small that w,7, <1, = 2W, This is the same.result as by the perturbation theory for the condition Wt, <1. In addition, under restriction allowing the application of the ‘strong collision’ model, that is, under the assumption that the shape function S(@) is given by (39), it is found that both methods give M-M,_ 27, M, «1+2W7, without restriction on WT7,. E. Transient effects (a) Free precession If the relaxation phenomena are neglected, the solutions of the equations of motion in an applied d.c. field Hy = —wo/y are Ud) = M0), M(t) = M,(O)exp(iwg t). (40) For a sample in thermal equilibrium at time ¢ = 0, M.(0) = xo Ho M,(0) = 0, and no transverse magnetization exists. Such a magnetiza- tion can be created by applying a rotating field H, for a time +. If itsCh. 1 MOTION OF FREE SPINS 33 frequency w is equal to the Larmor frequency wa, at the end of the r.f. pulse, one will have Mz) = My 008.47, U(r) = Mysin wy7 expliagr). (41) After the rf. field is cut off, taking (41) as initial condition for (40), there results a precessing magnetization M,(t) = Mysin wr exp(iwy ¢) which keeps its amplitude permanently. This amplitude is a maximum following what is called a 90° pulse (wr = 47). It vanishes after a 180° pulse (w,7 = 7). In practice there may be a scatter in the Larmor frequencies described by the shape function f(w). At the end of the rf. pulse the magnetization corresponding to each value of w» can be computed using the rotating coordinates device. The result becomes very simple if one assumes that the amplitude w, of the r-f. field is much larger than the width 8 of the shape function and that the duration of the pulse is of the order of 1/,, for then the resonance condition is approximately fulfilled for all the spins of the sample, the effective field H, has the same direction OX perpendicular to the applied field, the angle of precession around H, has the same value (cw, 7), and (41) is still valid. Once the pulse is over the precessing magnetization will be M,() = Mysinwyr f Sleaplexp iat) deny = Mysin(w, r)exp(iws t) f f{(w+-w)exp(iut) du, (42) where w} is the central Larmor frequency. The transverse magnetiza- tion still precesses at the mean Larmor frequency but with a time- dependent amplitude G,(¢) which is the Fourier transform of the shape functign. As ¢ goes to infinity, because of the destructive interference among the contributions of the different parts of the sample to the transverse magnetization, @,(¢) goes to zero, and the observation of its decay gives the same information on the shape function as the observation of the resonance with a vanishingly small r.f. field. For instance, if the shape function is a Lorentz curve f(w8+u) = (b/m)[1/(b?+4*)], the decay shape will be the exponential G(t) = G(0)exp(—bt). It will be shown later how this result, established under very special assumptions, can be generalized to apply to real systems. (b) Spin echoes In spite of the fact that the precessing magnetization has disappeared after a time 6, which, because of the general properties of the Fourier 9592.97 D34 MOTION OF FREE SPINS Ch. IT transform is of the order of 1/5 (8 = width of the shape function), it is possible to restore it to its initial value by application of a second appropriately dimensioned r.f. pulse. Studies based on this technique are known as spin echoes. Consider what happens at time t in the frame S' rotating at the central frequency w} = w of the spectral distribution. With respect to that frame, the spins which have a Larmor frequency Wy = w8-+-u will have precessed by an angle ut. Suppose now that at time t a 180° pulse is applied, corresponding to a rotation of 7 around an effective field of direction X’ in the equatorial plane, of the frame S’ (which may or may not coincide with the direction X of the effective field for the first pulse). At the end of the pulse the spins of frequency w§-+-u will have with respect to X’ the phase —u(é+7) (apart from a constant) and therefore at time 2(t--7) they will all be aligned again. ‘Thus the transverse magnetization of all parts of the sample will add constructively to a value equal to the one immediately following the first pulse. Again a more detailed study of this phenomenon and its possible applications will be deferred until later. (c) The adiabatic theorem, adiabatic passage From the equation of motion (1) we may immediately deduce days aM = = 2M—— = 0. 43) dt (ue) dt ° (48) The magnitude of the magnetization Jf is a constant of the motion, whatever the variation of H with time. It will now be shown that if this variation is sufficiently slow, the angle of the magnetization with the instantaneous direction of the field is also a constant of the motion. The variation with time of the vector H can be described quite generally by the vector equation a =Q\H+0,H, (44) where the vector 2 and the scalar Q, have the dimensions of a fre- quency. Consider a moving frame S’ where the z-axis is continuously aligned along the instantaneous direction of the field H. According to (44) the relative motion of S’ with respect to the laboratory will be a rotation about an instantaneous axis 2. In that frame the magnetization will change in time according to eM Me Ma (a+9). (45)ch. I MOTION OF FREE SPINS 35 By definition, in this frame H, = H, = 0 and a om = M,0,—M,0,. If \Q| < |yA| then, approximately, aM, aM, ta —! = —yHM,. a = vHM, a yl M, and M, are approximately sinusoidal functions with instantaneous frequency w(t) = —yH(t). After a long time #, the change in Uf, will be AM, = M,{t)—M,(0) = i [00,0 — M,e)O,0Y] at. If the variation of Q with time is sufficiently slow, or to be precise, if its Fourier expansion has negligible components at frequencies of the order of |yH(t)|, then, for any ¢, MOQ, laa ~ | H,, which is not incompatible with the previous ones, the duration + = |H,/H,| of the passage through the resonance will be a small fraction of the modulation period. If we start from a value of H, say far above resonance where the effective field is practically parallel to Hy, and go through the resonance to the other side, far below the resonance, the magnetic moment M, initially parailel to Hy, will remain continuously parallel to H, and, thus, end up antiparallel to Hy. At the passage through the resonance there will be a transverse magnetization equal to the initial value M). If there is a distribution of Larmor frequencies, the magnetic moment of the sample can still be reversed by adiabatic passage, since the con- dition (48) is independent of the width 8 of the shape function. How- ever, the maximum value of the transverse magnetization during the passage is reduced in a ratio of the order of w,/5 if 5 > {w,|. The modification caused in the theory of adiabatic passage by the existence of relaxation and the practical applications of this method will be discussed later. F. The general problem of two levels coupled by an r.f. field. The fictitious spin } If we apply to a system a perturbation which is a sinusoidal function of time with a frequency w in the neighbourhood of the frequency @ = (H,—E,)/h, where H, and B, are the energy levels of two states \a) and |b) of the system, it is permissible to a good approximation to disregard the other levels of the system. We then have a system with two degrees of freedom where all the relevant physical observables can. be represented by 2x2 matrices. Each operator Q can then be written as Q = $9.+q.8 ,where q, = tr(Q} inside the manifold of the two levels, and s = }o, where o;, o,, oy are the usual Pauli matrices. Thus the total Hamiltonian of the system including the r.f. perturbation can be written as H = YE, —y'K(H'.s), where the four quantities Ey, y'H%, y'H,, y'H; are easily determined from the matrix elements of of, and H’ and y‘ are a fictitious magnetic field and a fictitious gyromagnetic ratio, Similarly, the density matrix p can be written as p = 34-m.s. The equation of motion (h/i)(dp/dt) = —[H#, p] leads immediately to am oe ymAW, (49)Ch. IT MOTION OF FREE SPINS 37 which enables us to treat m as a fictitious magnetic moment. The expectation value of any operator Q will be = tr(pQ) = H{got(a-m)}. (50) As an example consider the case of a spin J in a magnetic field where the levels J, = m are not equidistant because of a small quadrupole interaction and therefore the various transitions |Am| = 1 can be in- duced separately. For instance, for the transition —4-— }, y=y H= MH, Hp+il’, = (1U+1)+4} em, Hye =[I+4]he, L=s, Ly =[1+4]5¢. +il,> = [+4 \(m_+im,). Since the transverse magnetization Ji, other terms such as dipolar couplings between the spins or : quadrupolar couplings with local gradients, which we shall describe as an extra term, 4, in the Hamiltonian. The assumption of thermal equilibrium between the spin system and the lattice is expressed by a statistical spin operator rool) The macroscopic nuclear magnetic moment of the sample will be40 MACROSCOPIC ASPECTS OF NUCLEAR MAGNETISM Ch. DI M = tr{yhJ,p} or, if we assume that the ratio (%+%)/kT is small, we ttlyhI (Het) 1 Mee) EE ® If #, is such that tr{s4 J,} = 0, which happens to be the case for both spin-spin couplings and quadrupole couplings, it will be seen that Mn PRA tJ} 1 ~ Ny I(I+1)H, = try kT 3kT in accordance with (1) and irrespective of the relative magnitude of % and %4, provided that both are much smaller than k7. The conditions under which a nuclear magnetic resonance is observed are manifold. A first distinction can be drawn between the use of very small r.f. fields which can be assumed to leave the populations of the spin states practically unperturbed, and that of large r-f. fields which decrease appreciably the differences between the populations of the spin states, a phenomenon known as saturation. Another distinction can be made between so-called slow-passage methods where at each stage of the experiment a quasi steady-state is reached, and transient methods where non-equilibrium situations occur. Yet another can be drawn between the behaviour of solid samples and fluid (liquid or gaseous) samples. In the solids there is usually a tight coupling between the nuclear spins which complicates consider- ably the phenomenon of nuclear resonance. The effects of this coupling will be examined at some length in the next two chapters. In fluids this coupling is quenched to a considerable extent by the rapid relative motions of the spins, as will be explained in Chapters VIII and X, and as @ consequence the dynamics of nuclear magnetism are a good deal simpler there. B. Resonant absorption of r.f. energy The amount of r.f. energy absorbed per unit time by a sample con- taining per unit volume W spins I of magnetic moment yhJ is easily computed using the formula (30) of Chapter IT, which gives the transi- tion probability per unit time induced by a rotating r-f. field of fre- quency w = 2av and amplitude H, = —w,/y. If the assumption of negligible saturation can be made, the difference in populations between the states I, = mand I, = m—1 is for each spin: Pyaa Pn = P,loxe(Fa)—1) ~ open relist) 1) = parle = ery er = (IF) eT .Ch. I MACROSCOPIC ASPECTS OF NUCLEAR MAGNETISM 41 The total energy absorbed per unit time will thus be fy 1 P= hwo OTE) 7 iy *\(nIZ,lm—1) Nfl) _ Baswy mad = 2QT+ HEP Paw PH? = ot AU DNflw)2n. (4) tr{L, LIN f(w) ‘The origin of the finite width of spin levels described by the shape function f(w) occurring in (4) need not be specified in those formulae. It may be caused by dipolar interactions between the spins, by the inhomogeneity of the applied field, by fluctuating local magnetic fields such as those that exist in metals and are due to the spins of conduction electrons, etc. It is sufficient to know that a relaxation mechanism of unspecified nature maintains the spins at the temperature of the lattice and thus the populations P,, of the spin levels at their Boltzmann values. It is important to realize, however, that the very fact that r.f. energy is being absorbed by the spin system requires a transverse component of nuclear magnetization, incompatible with a rigorous description of the spin system by the populations of its levels. As explained in Chapter II, such a description would imply the absence of off-diagonal] matrix elements for the statistical operator and thus no transverse magnetization. Let us assume that the rotating field of amplitude H, is actually produced by a linearly polarized field H, = 2H,coswt, the counter- rotating component having a negligible effect as explained previously. The r.f. power absorbed by the spin system is uy ee, = —m —u, ue ira (5) If the r.f. excitation 2H, cosw4 of the spin system is sufficiently small, the response M,(t) of the spin system may be assumed proportional to it and can be written. M(t) = 2H,fx'(w)eos wt +-x"(w)sin wt}, (8) where y'(w) and y"(w), independent of H,, are the so-called real and imaginary parts of the r.f. susceptibility y = y’—iy”, defined by the relations 7, = 2Hyrefet™}, — M, = 2Hhref xe}, (2) where ‘re’ means real part.42 MACROSCOPIC ASPECTS OF NUCLEAR MAGNETISM Ch. IIT The method of calculating x’ and y” from the microscopic structure of the spin system will be explained in Chapter Iv. Carrying (6) over into (5), we find P = 2H} yw which, by comparison with (4) and using the formula (1) for xo, gives x"(w) = drxoeoflo). (8) The fact that, with our sign conventions, w) = —yH), and thus also x’(w) may take negative values, should not worry ‘the reader since tine absorbed power P = 2H?y"w will have the sign of yyw or, since w & wp, the sign of xq 3. It will be noticed that from the relation (8) between x"(w) and the static susceptibility yo, all quantum-mechanical quantities have dis- appeared, This is a consequence of the so-called Kramers-Krinig relations, which exist for linear systems between the real and imaginary parts of their response to a sinusoidal excitation and will be derived at the end of this chapter. These relations are xW)— Xe -19 ft cede’, yw) le f exe da’, (8") In (8’) 2 fags means ti ( f + I} we ake Some caution should be exercised in applying these formulae to nuclear magnetism. According to the definition (6) of x’ and x", x’ is clearly an even and y” an odd function of w. In nuclear magnetism we often calculate the response to rotating rather than oscillating fields and the precessing magnetization thus calculated can be considered as a response to an oscillating field only because the effect of the counter-rotating component is neglected. Let M(t) = 2Hiy{yp(w)00s ot-+-y’p(w)sin ot} be the response to a field rotating at a frequency w: H, = H, cos wt, H, = H,sinat. The response to a linearly polarized field: H, = 2H, coswt, H, = 0, sum of two rotating fields with opposite frequencies, will be if the system is linear: M(t) = 2Hy{y’(w)cos wt+- y"(w)sin wt},Ch. IIT MACROSCOPIC ASPECTS OF NUCLEAR MAGNETISM 43 with x0) = xalo)+xn(—2), x") = xo) —xR(—#)- (8") Since x‘g(w) and xz(w) are very small unless w = wo, which is the resonance frequency, replacing x’ and y” in (8’) by their values (8”), writing w = w)+y, w' = wo+y' and neglecting small terms, we get the K.-K. relations into a form better adapted for later purposes: © , wl Xp(@ot+y') dy! Haleaty) ax oe | Xt, —e Lig f xalooty’)dy’ . mw hp f Xnleoty') dy’ a klorty) ac 2g | Maleate 8”) ~~ TE x%(wwp+y’) is an even function of y’ (symmetrical resonance curve) it is clear from the first equation (8”) that y‘p(w)+y) is an odd function of y, and x‘p(wo) = 0. In order to demonstrate (8), we make w = 0 in the first relation (8’), getting lg f xkle—xe—o'), 0-27 | w ‘dew’ 25 fxpw')de’ 2 f =29 face af J riled. se eo Writing xg(w) = af(w) where a is a constant and f(w) is a shape function normalized to unity, we get @ = 42(w xo) whence (8). The shape function f(w) will in general be a bell-shaped narrow curve with a maximum at the Larmor frequency w, of the spin system and, since f(w) is normalized to unity, it will have a width A such that Af(w) ~ 1. For instance, the Lorentz shape already met with in Chapter II is described by T, 1 $= Te arTy ® ‘The half width at half intensity is equal to A = 1/7, and f(wy) = 7/7. For such a shape, according to (8), x'(o0) = deolaxo = bro'Z- (19) The same relation is valid with a factor of order unity for other line shapes. “It is because resonance lines can be very narrow in nuclear44 MACROSCOPIC ASPECTS OF NUCLEAR MAGNETISM Ch. lI magnetism, and thus w,/A can be a large number, that the resonant r.f. susceptibility is much larger than the static one, and resonant methods can be so much more sensitive than static methods. The simplicity of the results thus obtained for vanishingly small r-f. fields has to be contrasted with the complications that arise as soon as the rf. field is sufficiently large for the saturation to set in. Definite assumptions have to be made then about the internal structure of the spin system, the origin of the line width and the relaxation mechanisms, in order to be able to predict the behaviour of the spin system submitted to large r.f. perturbations. For a very special model (no interactions between the spins, and strong collisions) a calculation of such behaviour was given in Chapter IT. Il. Tue PHENoMENOLOGICAL EQUATIONS OF BLOCH In 1946 Felix Bloch proposed for the description of magnetic pro- perties of ensembles of nuclei in external magnetic fields a set of very simple equations derived from phenomenological arguments, that have proved exceedingly fruitful, and, for liquid samples at least, have provided in most cases a correct quantitative description of the detailed behaviour of the phenomena. The heuristic argument for obtaining these equations is as follows. First, in an arbitrary homogeneous field the equation of motion of the nuclear magnetization for an ensemble of free spins was shown in Chapter II to be dM/dt = yM aH. Second, in a static field H, = Hy, the trend of the magnetization towards its equilibrium value M, = M, = x) Hy can often be described with good accuracy by the equation dM,/dt = —{(M,—M,)/T,}. 7, is calied the longitudinal relaxation time. Third, if by any means such as an rf. pulse, the nuclear magnetization is given a component at right angles to the applied field Hj, the various local fields, owing to the fact that the spins are actually not free but interact with each other and with their surroundings, cause the transverse magnetization to decay at a rate which can often be represented by the equations dM, _ _M, dM, _ _M, dt de” where 7, is called the transverse relaxation time. Fourth, and this is a new assumption rather than a consequence of the three previous points, in the presence of an applied field, the sum of a d.c. field and a much smaller r-f. field, the motion due to relaxationCh. IIT MACROSCOPIC ASPECTS OF NUCLEAR MAGNETISM 45 can be superposed on the motion of the free spins, leading to the equation dM U4, WM, <— = yMAH——2 1 _ 2 oy, (11) a” T i where i’, j’, k’ are the unit vectors of the laboratory frame of reference. Postponing until later the question of the validity of the equations (11), we shall examine some of their consequences. A. Steady-state solutions—saturation Assume that the applied field is the sum of a d.c. field H,= I= — and of an rf. field H, of amplitude 7, = —w,/y rotating at a frequency w in the neighbourhood of w». This field will usually be one of the rotating components of an applied field H, = 2H,coswt, linearly polarized along the OX axis of the laboratory frame, the effect of the counter-rotating component being neglected. In the frame rotating around H, at the frequency w there is an effective static field Her = (+2 ie HA jo Wredk—wi _ dok-wi (9) Y Y where i, j, k=k’ are the unit vectors of the rotating frame and wy = —yH, w, = —yH,. In the rotating frame the equation of mo- tion is + y dM H,i+i,j M,—M, ay = MA Hen) — hk k, where J, and M, are the transverse components of M in that frame. ae (13) can be rewritten as ae at (13) Wei Me Moy, 13°) i te Me aA doko na 2 or a, Te ay oa, 18, Bind —M, OMe = a, Me T (14)46 MACROSCOPIC ASPECTS OF NUCLEAR MAGNETIPM Ch. IIT The general solution of (14), for fixed values of the parameters, is a sum of decreasing exponential terms and of a steady-state solution obtained by setting aM; aH, _ dil, _ 9, dt dt dt If a sufficiently long time has elapsed for the transient exponentials to have decayed, the steady-state solution can be written _ Aw yH, TR Mt. ° T+ (f,d0)}+y HiT T 7H, T, y= Tea ARTI _ 1+(Aw7,)? = TEyen (18) The fact that the three components of M are proportional to M, is not surprising: in the absence of initial polarization, that is, of in- equalities of populations among the magnetic energy levels, the nuclear Tesonance cannot be observed. From the values (15) of the components J, and M,, the transverse components M, and M, in the laboratory frame can be calculated through MAiM, = (iM eet, MM, = H,,cos ot—M, sin wt, M, = MW, sin wt+ M, cos wt. (16) A notation introduced by Felix Bloch and widely used in the litera- ture is Wy x w=, v= —-Y =O 4m, (16’) ly leo] The components of the magnetization in the laboratory frame are functions of time and can induce in a coil a detectable voltage at the frequency w. Experimental devices to be described briefly later in this chapter permit observation of u and v separately or of various combinations of these two quantities. For negligible saturation, that is, for y*H?7,7, <1, v, which for reasons to appear shortly is called the absorption, can be written v= —a|y\H, Mofz,(Au), (17) where fp,(Aw) is the normalized Lorentz shape function (9) with half width at half #tensity 1/T,.Ch. III MACROSCOPIC ASPECTS OF NUCLEAR MAGNETISM 47 For appreciable saturation, v can be written =a yl My = OR PMER THOS “ where the half width 7’ is 1 1 1 7a A DY (18') In presence of saturation the resonance curve representing v still has the Lorentz shape but becomes broader in the ratio # = (14°22). The value of |v/ is maximum at resonance and equal to le lmax = ly Ts . mx THD,” For small values of H,, |Y|max is proportional to H,; then it passes through a maximum reached when y?H?7, 7, = 1 and equal to HMloaxsmax = 48éo,/ (7): (19) and decreases towards zero as H, is increased further. In practice 7, is sometimes shorter than 7; so that the steady state transverse magnetization is smaller than M,. The behaviour of the function w is different from that of the function v. It is an odd rather than even function of Aw, vanishes at resonance, and has a maximum and a minimum on either side of the resonance for (18”) ho = PERE D1 Ts tT; with Unex = Umi = $y |F Ts My. As H, increases |upax! = [Umin| grows steadily to an asymptotic value T, bnaalaeray = Imex = 5°, / (pe) (20) Finally M,, which at resonance and for small values of H, differs from M, only to the second order in H,, decreases steadily towards zero as H, increases, A convenient geometrical representation of the solutions (15) is obtained if, through elimination of Aw and H, from (15), a relation is derived between M,, M,, and M,. It shows with a little algebra that the locus of the vector M as a function of H, and Aw is an ellipsoid with48 MACROSCOPIC ASPECTS OF NUCLEAR MAGNETISM Ch. III axes 2a, = 2a, = Myy(T,/T,), 2a, = My, centred at the pointx = y = 0, z= 4m. All the preceding results for u and » are easily visualized in this manner. The fact that for large H, the dispersion remains finite, whereas the absorption goes down to zero, makes it preferable, while searching for signais in samples with unknown relaxation times, to use strong rf. fields and detection devices to be described later, sensitive to the dis- persion rather than to the absorption. The r.f. power absorbed by the spin system is easily calculated from the formula ae _Mat . The vector dH/dt has in the rotating frame components: R=Z=0, VY=ok, and P= -m@ = —ollt, H, or, by (15), Pe —wyHiT, = Xoo TD _ 1+ (fT, AoP+yHeT, 7,” ° (2, Aw)*+7H7 TF, T, (21) Since v = (p/|wo|)M,, P can be written as P = —|w{/t,v, whence the name of absorption component given to the quantity », For large values of H,, P approaches asymptotically at resonance the maximum value xo Hj/T, = M,H,/T,. This can be interpreted by saying that the maximum energy that can be transferred to the lattice in the time 7, is the energy y Hj, twice the paramagnetic energy 4x H3 stored in the sample at thermal equilibrium. It must be pointed out, however, and will be shown later that the ‘signal’ available for the detection of the absorption is proportional to v rather than to the absorbed power P. From the definition (6) of the r.f. susceptibilities y’ and x” and the equation (16) we obtain 2, . vo)=sR xe -= or, replacing in (15) M, by x» — = X07 a Jo) dw TF XR ~ TH Aw) AAT ‘ ___ beets (22) xh TTC Ra yACh. Il MACROSCOPIC ASPECTS OF NUCLEAR MAGNETISM 49 Xp and yp can be rewritten as xe = — dx doT fr; (Aw)oo, xr = 4ex(1+YART, Tr) hoof; (Aw), (23) where 1, = Ti[1--yH} 7, T)]-* and fp; (Aw) is the normalized Lorentz shape function t% L a 1+(T Aw? For negligible saturation, Xr = —brxowo(Aw Te) fr,(Ae), xR == BX ofr,(Aw). (24) The reader can easily satisfy himself that the susceptibilities yp(w) and x‘n(w) defined by (24) do verify the KK. relations (8”). On the other hand, it is important to notice that the susceptibilities (23) describing the existence of saturation do not verify the K.-K. relations as indeed they should not, since the response of the spin system does not depend linearly on Hj. It is actually easy to verify that in that case relations of the type (8”) exist between yp and yp[1+y*H?7, 7]t as defined by (23). A well-known, but sometimes forgotten, point is that the suscepti- bility of an assembly of spins is a tensor, correlating the magnetization with the applied field by a relation of the form M,= & xu Hy Thus yz and xp given by (22) are the real and imaginary part of the component x,., of the susceptibility tensor. In some schemes of detec- tion, to be described later, we are interested in the component x,, that gives the values of M, when the rf. field is applied along ox. Since (neglecting the effect of the counter-rotating component of the r.f. field) the magnetization precesses with the velocity w around the axis oz, it is clear that M,(t) = M,(t—7/2w), whence Xpx COS wt+ XH, 8in wt = yp, 008(wt—47)+ y7_ Sin(wt—47), Xye = Xo Xp = Kew B. Steady-state solutions in an inhomogeneous field It often happens, especially for liquid samples, that the spread y AH in Larmor frequencies due to the inhomogeneity of the applied field across the sample results for the observed absorption curve in a width much greater than that predicted by the equations (17) and (18), 9805.97 zE50 MACROSCOPIC ASPECTS OF NUCLEAR MAGNETISM (Ch. III namely 1/7, for negligible saturation and 1/7’, if saturation is present. Suppose that h(x) da (normalized to J Ma) dx = 1) represents the rela- tive weight of spins with Larmor frequencies w, between w}+2 and w8+a-+dx (w§ being the value of the Larmor frequency with the maximum weight). For those spins the steady-state solutions (15) are valid with —Aw = w.—w = w§+a2—w and the total response of the sample will be obtained by multiplying these solutions by h(x)dx and integrating over «. Thus for v as given by (18), we obtain _ _—7h\.M, i ok = TF yHiT Tt Srilop—e+2)h(x) dx. (25) If the width is entirely determined by the inhomogeneity of the field for all obtainable values of 1/7; = (1/Q))[1+y*Hi 7, T,}*, A(x) has a much slower dependence on « than f7;(w}—w-+-<), which has a relatively sharp maximum for w3—w-+-x = 0, and v can be written = 7M — a? cerry Oe —)- eo) The dependence of |v.) = 7/y|Z, 14 4(0)[1-+7°H? 7, T,]-* on A, that is, the saturation, is quite different from that predicted previously from equation (18*) since |v,,,,| increases steadily towards a maximum value . ah(0) I?maxlasymp = yt (26') quite different from (19). At the same time, according to (26), the shape of o(c), entirely contained in h(w—w}), does not change with Hy. ‘To pursue the discussion farther, it is helpful for the sake of simplicity to assume that the shape function h(x) of the field inhomogeneity also has the Lorentz form: 1 Th T+ GT a’ where 1/7} = yAH is the half width at half intensity, due to the field inhomogeneity. The integral (25) can then be performed exactly. It is the folding (Faltung) of two Lorentz distributions, well known to result into a Lorentz distribution with a width that is the sum of the widths of the two components. If we define 7;* through Re) = (26") 1 1 1 OT enCh. IIT MACROSCOPIC ASPECTS OF NUCLEAR MAGNETISM 51 aoe 1 1,1 or, for negligible H,, 7 = Zt , (25) can be rewritten —Tlyl\y Me eo ayn)! pirrle— ef), (28) or, for small H, and negligible saturation, v= —tly|, Mo fri(o—o). (28’) 1 qf Aen t (28”) it is possible to give to H, values, large enough to have y*H?7, 7, > 1, but at the same time small enough to have a1 ve 1 =A TT 1 ‘ Te OT; Ty =r The quantity |%pax! then, according to (28), reaches an asymptotic value “un 1, uy ith ouclaome & te = BE (re)? (ae) |1/y7;|, due for instance to a poor magnet, causes a considerable decrease in the steady-state r.f. magnetization.Ch. II MACROSCOPIC ASPECTS OF NUCLEAR MAGNETISM 53 Suppose that by some means the real 7, of the sample is reduced well below the value 1/|yAH|. The broadening due to the field in- homogeneity is then a small perturbation on the real width and the maximum absorption and dispersion signals become }M,,/(7;/7,). If in the process of shortening 7}, 7, is also shortened and becomes com- parable to 7, the transverse magnetization becomes 2» M,, a consider- able improvement. It will be shown in Chapter VIII that dissolving paramagnetic impurities in an otherwise diamagnetic liquid results precisely in short and comparable values of 7, and T,. This procedure known as ‘doping’ is widely used to improve the steady-state signals. CG. Modified Bloch equations in low fields The validity of the Bloch equations (11) and of their steady-state solutions (15) becomes questionable when the d.c. applied field Hy is comparable either to the line width in gauss 1/yT, or to the amplitude Hy of the rf. field. It has been suggested on theoretical grounds that the assumption of a magnetization relaxing towards the d.c. equilibrium value My = xp Ho should be replaced, when an r.f. field H,(t) is present, by that of relaxa- tion towards the instantaneous value y)(H)+H,(t)). The distinction, academic in high fields, becomes important when H, and H, are comparable. The assumption of the relaxation of M towards y(H)+H,(¢)) per- mits us, as will appear presently, to get rid of an unsatisfactory feature of the Bloch equations. The r.f. power absorbed, given by formula (21), goes to zero with Hy = —wo/y, even if the r.f. frequency w has a finite value. This result is difficult to accept on theoretical grounds and is in conflict with the low-frequency relaxation measurements of the Dutch school. The justification from first principles of the validity of modified Bloch equations will be given in Chapter XII and we shall be content for the time being to explore some of their consequences. It will be shown in Chapter XII that if H, is not small compared with Hy, the condi- tions for the validity of the simple assumption of relaxation towards xo(H+H,(t)) are such that they also result in the equality 7, = 7, = T. The modified equations adopted are thus dM _ M—y)H / an yM AH— — It becomes necessary now to discriminate more carefully between the response to an oscillating and to a rotating field, the effect of the (34)54 MACROSCOPIC ASPECTS OF NUCLEAR MAGNETISM Ch. IIT counter-rotating component in the former case being appreciable if H, is comparable with Hy. We first look for steady-state solutions in a rotating field. With the same notations as in (14) except for the simplification 7, = 7), (34) can be written in the rotating frame am, (iy) Hh) a = FF +doll, aM, It, “g= Sell Gio Me, dM, M,—M™, F = ott, - 2. (35) We express the transverse components MH, and J, the steady-state solution of (35), by means of the r.f. susceptibilities yj}4(w) and x34(w). The subscript & stands for rotating and the superscript M is to remind us that (35) are the modified Bloch equations. We find mH, _ le AoTM—[1+(-H, FI Xe 3H, 21 (Pho +O T? * 1 wT Mt _l a= oH 3 TF Thao TP 68) For negligible saturation (ly|H, 7 <1), (36) can be rewritten , , li, . ~ 1, xe = ets pxk xet = xh Gp xe (37) Xe phe T™ Xp ey , where Xe 21+Phoe’ %* > 2 TT PAape (67) are the rotating field susceptibilities (24) for the unmodified Bloch equations (11) (with 7, = Z,). The susceptibilities yf and yf (37) satisfy the K.—K. relations. This follows immediately from the fact that yp and yp (37’) verify these relations. For M, we obtain M=ylH (wiy\yH, TY) = aa ray rT “8 which. is to be contrasted with the value (15) previously obtained which can be rewritten as = (woly)yH, T)* . M= xe| Hot EM ol Tyy (38) (Remember the definition w) = —yH,.)Ch. III MACROSCOPIC ASPECTS OF NUCLEAR MAGNETISM 55 From a comparison between (36) and (22) on the one hand, (38) and (38') on the other, interesting features of the modified Bloch equations appear. The most striking is the fact that for H, = —w/y = 0, (38), in contradistinction to (38'), predicts a non-vanishing magnetization M,, This is plausible physically. In the rotating frame the spins ‘see’ a static field H.» with components H, = —w/y, H, = H,, around which 0 1 2 3 4 $ 6 7 H, (Gauss) —> Fic. III, 1. Observed signal in a d.c. field Hy as a function of r.f. amplitude in solid diphenyl, picril hydrazil. The full curves give the theoretical prediction of eqn. (38), assuming 7, = T, = T = 62x 10-* sec, measured independently. they precess, while relaxing towards M2 = xoH,. There is clearly a lack of symmetry between positive and negative values of M,, resulting in a non-zero steady value of M,. The predictions of formula (38) have been tested experimentally (1). Fig. III, 1 shows AM, = M,— x, H, predicted by (38), as a function of the strength H, of the rotating field, for (2) Hy = —w/y, (b) Hy = 0, together with the experimental points. The agreement with the theory is gratifying and demonstrates clearly that the unmodified Bloch equa- tions, which predict M, = 0 for H, = 0, are inadequate. The other interesting feature of the modified Bloch equations is that the absorbed power P = 2y"H?w does not vanish with Hy since yj" given by (36) is proportional to w rather than to wp. Referring for further details to reference (1), we pass on to the steady- state solution of (34) for a linearly polarized r-f. field.