Spectroscopy
James Keeler
University of Cambridge, Department of Chemistry
c James Keeler, 2002
1 What this course is about
This course is aimed at those who are already familiar with using NMR on a
daytoday basis, but who wish to deepen their understanding of how NMR
experiments work and the theory behind them. It will be assumed that you are
familiar with the concepts of chemical shifts and couplings, and are used to
interpreting proton and
13
C spectra. It will also be assumed that you have at
least come across simple twodimensional spectra such as COSY and HMQC
and perhaps may have used such spectra in the course of your work. Similarly,
some familiarity with the nuclear Overhauser effect (NOE) will be assumed.
That NMR is a useful for chemists will be taken as self evident.
This course will always use the same approach. We will ﬁrst start with
something familiar – such as multiplets we commonly see in proton NMR
spectra – and then go deeper into the explanation behind this, introducing
along the way new ideas and new concepts. In this way the new things that
we are learning are always rooted in the familiar, and we should always be
able to see why we are doing something.
In NMR there is no escape from the plain fact that to understand all but
the simplest experiments we need to use quantum mechanics. Luckily for us,
the quantum mechanics we need for NMR is really rather simple, and if we
are prepared to take it on trust, we will ﬁnd that we can make quantum me
chanical calculations simply by applying a set of rules. Also, the quantum
mechanical tools we will use are quite intuitive and many of the calculations
can be imagined in a very physical way. So, although we will be using quan
tum mechanical ideas, we will not be using any heavyduty theory. It is not
necessary to have studied quantum mechanics at anything more than the most
elementary level.
Inevitably, we will have to use some mathematics in our description of
NMR. However, the level of mathematics we need is quite low and should not
present any problems for a science graduate. Occasionally we will use a few
ideas from calculus, but even then it is not essential to understand this in great
detail.
Course structure
The course is accompanied by a detailed set of handouts, which for conve
nience is divided up into “chapters”. You will notice an inconsistency in the
style of these chapters; this comes about because they have been prepared (or
at least the early versions of them) over a number of years for a variety of
purposes. The notes are sufﬁciently complete that you should not need to take
many extra notes during the lectures.
Each chapter has associated with it some exercises which are intended to
illustrate the course material; unless you do the exercises you will not under
stand the material. In addition, there will be some practical exercises which
1–2 What this course is about
involve mainly data processing on a PC. These exercises will give you a feel
for what you can do with NMR data and how what you see relates to the
theory you have studied. Quite a lot of the exercises will involve processing
experimental data.
Chapter 2 considers how we can understand the form of the NMR spec
trum in terms of the underlying nuclear spin energy levels. Although this
approach is more complex than the familiar “successive splitting” method for
constructing multiplets it does help us understand how to think about multi
plets in terms of “active” and “passive” spins. This approach also makes it
possible to understand the form of multiple quantum spectra, which will be
useful to us later on in the course. The chapter closes with a discussion of
strongly coupled spectra and how they can be analysed.
Chapter 3 introduces the vector model of NMR. This model has its limita
tions, but it is very useful for understanding how pulses excite NMR signals.
We can also use the vector model to understand the basic, but very impor
tant, NMR experiments such as pulseacquire, inversion recovery and most
importantly the spin echo.
Chapter 4 is concerned with data processing. The signal we actually
record in an NMR experiment is a function of time, and we have to convert
this to the usual representation (intensity as a function of frequency) using
Fourier transformation. There are quite a lot of useful manipulations that we
can carry out on the data to enhance the sensitivity or resolution, depending
on what we require. These manipulations are described and their limitations
discussed.
Chapter 5 is concerned with how the spectrometer works. It is not nec
essary to understand this is great detail, but it does help to have some basic
understanding of what is going on when we “shim the magnet” or “tune the
probe”. In this chapter we also introduce some important ideas about how the
NMR signal is turned into a digital form, and the consequences that this has.
Chapter 6 introduces the product operator formalism for analysing NMR
experiments. This approach is quantum mechanical, in contrast to the semi
classical approach taken by the vector model. We will see that the formalism
is well adapted to describing pulsed NMR experiments, and that despite its
quantum mechanical rigour it retains a relatively intuitive approach. Using
product operators we can describe important phenomena such as the evolution
of couplings during spin echoes, coherence transfer and the generation of
multiple quantum coherences.
Chapter 7 puts the tools from Chapter 6 to immediate use in analysing
and understanding twodimensional spectra. Such spectra have proved to be
enormously useful in structure determination, and are responsible for the ex
plosive growth of NMR over the past 20 years or so. We will concentrate on
the most important types of spectra, such as COSY and HMQC, analysing
these in some detail.
Chapter 8 considers the important topic of relaxation in NMR. We start
out by considering the effects of relaxation, concentrating in particular on
the very important nuclear Overhauser effect. We then go on to consider the
sources of relaxation and how it is related to molecular properties.
1–3
Chapter 9 does not form a part of the course, but is an optional advanced
topic. The chapter is concerned with the two methods used in multiple pulse
NMR to select a particular outcome in an NMR experiment: phase cycling
and ﬁeld gradient pulses. An understanding of how these work is helpful in
getting to grips with the details of how experiments are actually run.
Texts
There are innumerable books written about NMR. Many of these avoid any
serious attempt to describe how the experiments work, but rather concentrate
on the interpretation of various kinds of spectra. An excellent example of
this kind of book is J. K. M. Sanders and B. K. Hunter Modern NMR
Spectroscopy (OUP).
There are also a number of texts which take a more theorybased approach,
at a number of different levels. Probably the best of the more elementary
books if P. J. Hore Nuclear Magnetic Resonance (OUP).
For a deeper understanding you can do no better that the recently pub
lished M. H. Levitt Spin Dynamics (Wiley).
Acknowledgements
Chapters 2 to 5 have been prepared especially for this course. Chapters 6, 7
and 8 are modiﬁed from notes prepared for summer schools held in Mishima
and Sapporo (Japan) in 1998 and 1999; thanks are due to Professor F Inagaki
for the opportunity to present this material.
Chapter 9 was originally prepared (in a somewhat different form) for an
EMBO course held in Turin (Italy) in 1995. It has been modiﬁed subsequently
for the courses in Japan mentioned above and for another EMBO course held
in Lucca in 2000. Once again I am grateful to the organizers and sponsors of
these meetings for the opportunity to present this material.
Finally, I wish to express my thanks to Professor AJ Shaka and to the
Department of Chemistry, University of California, Irvine, for the invitation
to give this course. The University of Cambridge is acknowledged for a period
of study leave to enable me to come to UC Irvine.
James Keeler
University of Cambridge, Department of Chemistry
March 2002
James.Keeler@ch.cam.ac.uk
wwwkeeler.ch.cam.ac.uk
2 NMR and energy levels
E
2
E
1
hν = E
2
–E
1
ν
energy
levels
spectrum
Fig. 2.1 A line in the spectrum
is associated with a transition
between two energy levels.
The picture that we use to understand most kinds of spectroscopy is that
molecules have a set of energy levels and that the lines we see in spectra
are due to transitions between these energy levels. Such a transition can be
caused by a photon of light whose frequency, ν, is related to the energy gap,
E, between the two levels according to:
E = hν
where h is a universal constant known as Planck’s constant. For the case
shown in Fig. 2.1, E = E
2
− E
1
.
In NMR spectroscopy we tend not to use this approach of thinking about
energy levels and the transitions between them. Rather, we use different rules
for working out the appearance of multiplets and so on. However, it is use
ful, especially for understanding more complex experiments, to think about
how the familiar NMR spectra we see are related to energy levels. To start
with we will look at the energy levels of just one spin and them move on
quickly to look at two and three coupled spins. In such spin systems, as they
are known, we will see that in principle there are other transitions, called
multiple quantum transitions, which can take place. Such transitions are not
observed in simple NMRspectra, but we can detect themindirectly using two
dimensional experiments; there are, as we shall see, important applications of
such multiple quantum transitions.
Finally, we will look at strongly coupled spectra. These are spectra in
which the simple rules used to construct multiplets no longer apply because
the shift differences between the spins have become small compared to the
couplings. The most familiar effect of strong coupling is the “rooﬁng” or
“tilting” of multiplets. We will see how such spectra can be analysed in some
simple cases.
2.1 Frequency and energy: sorting out the units
NMR spectroscopists tend to use some rather unusual units, and so we need
to know about these and how to convert from one to another if we are not to
get into a muddle.
Chemical shifts
It is found to a very good approximation that the frequencies at which NMR
absorptions (lines) occur scale linearly with the magnetic ﬁeld strength. So,
if the line from TMS comes out on one spectrometer at 400 MHz, doubling
the magnetic ﬁeld will result in it coming out at 800 MHz. If we wanted
to quote the NMR frequency it would be inconvenient to have to specify the
exact magnetic ﬁeld strength as well. In addition, the numbers we would have
Chapter 2 “NMR and energy levels” c James Keeler, 2002
2–2 NMR and energy levels
to quote would not be very memorable. For example, would you like to quote
the shift of the protons in benzene as 400.001234 MHz?
ν
TMS
ν
frequency
δ
TMS
= 0 δ
chemical shift
Fig. 2.2 An NMR spectrum can
be plotted as a function of
frequency, but it is more
convenient to use the chemical
shift scale in which frequencies
are expressed relative to that of
an agreed reference compound,
such as TMS in the case of
proton spectra.
We neatly sidestep both of these problems by quoting the chemical shift
relative to an agreed reference compound. For example, in the case of proton
NMR the reference compound is TMS. If the frequency of the line we are
interested in is ν (in Hz) and the frequency of the line from TMS is ν
TMS
(also in Hz), the chemical shift of the line is computed as:
δ =
ν −ν
TMS
ν
TMS
. (2.1)
As all the frequencies scale with the magnetic ﬁeld, this ratio is independent
of the magnetic ﬁeld strength. Typically, the chemical shift is rather small
so it is common to multiply the value for δ by 10
6
and then quote its value
in parts per million, or ppm. With this deﬁnition the chemical shift of the
reference compound is 0 ppm.
δ
ppm
= 10
6
×
ν −ν
TMS
ν
TMS
. (2.2)
Sometimes we want to convert from shifts in ppm to frequencies. Suppose
that there are two peaks in the spectrum at shifts δ
1
and δ
2
in ppm. What is
the frequency separation between the two peaks? It is easy enough to work
out what it is in ppm, it is just (δ
2
−δ
2
). Writing this difference out in terms
of the deﬁnition of chemical shift given in Eq. 2.2 we have:
(δ
2
−δ
1
) = 10
6
×
ν
2
−ν
TMS
ν
TMS
−10
6
×
ν
1
−ν
TMS
ν
TMS
= 10
6
×
ν
2
−ν
1
ν
TMS
.
Multiplying both sides by ν
TMS
now gives us what we want:
(ν
2
−ν
1
) = 10
−6
×ν
TMS
×(δ
2
−δ
1
).
It is often sufﬁciently accurate to replace ν
TMS
with the spectrometer reference
frequency, about which we will explain later.
If we want to change the ppmscale of a spectruminto a frequency scale we
need to decide where zero is going to be. One choice for the zero frequency
point is the line from the reference compound. However, there are plenty of
other possibilities so it is as well to regard the zero point on the frequency
scale as arbitrary.
Angular frequency
Frequencies are most commonly quoted in Hz, which is the same as “per
second” or s
−1
. Think about a point on the edge of a disc which is rotating
about its centre. If the disc is moving at a constant speed, the point returns
to the same position at regular intervals each time it has competed 360
◦
of
rotation. The time taken for the point to return to its original position is called
the period, τ.
2.2 Nuclear spin and spin states 2–3
t
i
m
e
start
one
period
Fig. 2.3 A point at the edge of a
circle which is moving at a
constant speed returns to its
original position after a time
called the period. During each
period the point moves through
2π radians or 360
◦
.
The frequency, ν, is simply the inverse of the period:
ν =
1
τ
.
For example, if the period is 0.001 s, the frequency is 1/0.001 = 1000 Hz.
There is another way of expressing the frequency, which is in angular
units. Recall that 360
◦
is 2π radians. So, if the point completes a rotation in
τ seconds, we can say that it has rotated though 2π radians in τ seconds. The
angular frequency, ω, is given by
ω =
2π
τ
.
The units of this frequency are “radians per second” or rad s
−1
. ν and ω are
related via
ν =
ω
2π
or ω = 2πν.
We will ﬁnd that angular frequencies are often the most natural units to use in
NMR calculations. Angular frequencies will be denoted by the symbols ω or
whereas frequencies in Hz will be denoted ν.
Energies
A photon of frequency ν has energy E given by
E = hν
where h is Planck’s constant. In SI units the frequency is in Hz and h is in
J s
−1
. If we want to express the frequency in angular units then the relation
ship with the energy is
E = h
ω
2π
=
¯
hω
where
¯
h (pronounced “h bar” or “h cross”) is Planck’s constant divided by
2π.
The point to notice here is that frequency, in either Hz or rad s
−1
, is di
rectly proportional to energy. So, there is really nothing wrong with quoting
energies in frequency units. All we have to remember is that there is a factor
of h or
¯
h needed to convert to Joules if we need to. It turns out to be much
more convenient to work in frequency units throughout, and so this is what we
will do. So, do not be concerned to see an energy expressed in Hz or rad s
−1
.
2.2 Nuclear spin and spin states
NMR spectroscopy arises from the fact that nuclei have a property known as
spin; we will not concern ourselves with where this comes from, but just take
it as a fact. Quantum mechanics tells us that this nuclear spin is characterised
by a nuclear spin quantum number, I . For all the nuclei that we are going
2–4 NMR and energy levels
to be concerned with, I =
1
2
, although other values are possible. A spinhalf
nucleus has an interaction with a magnetic ﬁeld which gives rise to two energy
levels; these are characterised by another quantum number m which quantum
mechanics tells us is restricted to the values −I to I in integer steps. So, in
the case of a spinhalf, there are only two values of m, −
1
2
and +
1
2
. Strictly, α is the low energy state
for nuclei with a positive
gyromagnetic ratio, more of
which below.
By tradition in NMR the energy level (or state, as it is sometimes called)
with m =
1
2
is denoted α and is sometimes described as “spin up”. The state
with m = −
1
2
is denoted β and is sometimes described as “spin down”. For
the nuclei we are interested in, the α state is the one with the lowest energy.
If we have two spins in our molecule, then each spin can be in the α or
β state, and so there are four possibilities: α
1
α
2
, α
1
β
2
,β
1
α
2
and β
1
β
2
. These
four possibilities correspond to four energy levels. Note that we have added a
subscript 1 or 2 to differentiate the two nuclei, although often we will dispense
with these and simply take it that the ﬁrst spin state is for spin 1 and the second
for spin 2. So αβ implies α
1
β
2
etc.
We can continue the same process for three or more spins, and as each spin
is added the number of possible combinations increases. So, for example, for
three spins there are 8 combinations leading to 8 energy levels. It should be
noted here that there is only a onetoone correspondence between these spin
state combinations and the energy levels in the case of weak coupling, which
we will assume from now on. Further details are to be found in section 2.6.
2.3 One spin
There are just two energy levels for a single spin, and these can be labelled
with either the m value of the labels α and β. From quantum mechanics we
can show that the energies of these two levels, E
α
and E
β
, are:
E
α
= +
1
2
ν
0,1
and E
β
= −
1
2
ν
0,1
where ν
0,1
is the Larmor frequency of spin 1 (we will need the 1 later on, but
it is a bit superﬂuous here as we only have one spin). In fact it is easy to see
that the energies just depend on m:
E
m
= mν
0,1
.
You will note here that, as explained in section 2.1 we have written the
energies in frequency units. The Larmor frequency depends on a quantity
known as the gyromagnetic ratio, γ , the chemical shift δ, and the strength of
the applied magnetic ﬁeld, B
0
: This negative Larmor frequency
for nuclei with a positive γ
sometimes seems a bit
unnatural and awkward, but to
be consistent we need to stick
with this convention. We will
see in a later chapter that all
this negative frequency really
means is that the spin
precesses in a particular sense.
ν
0,1
= −
1
2π
γ
1
(1 +δ
1
)B
0
(2.3)
where again we have used the subscript 1 to distinguish the of nucleus. The
magnetic ﬁeld is normally given in units of Tesla (symbol T). The gyromag
netic ratio is characteristic of a particular kind of nucleus, such as proton or
carbon13; its units are normally rad s
−1
T
−1
. In fact, γ can be positive or
negative; for the commonest nuclei (protons and carbon13) it is positive. For
such nuclei we therefore see that the Larmor frequency is negative.
2.4 Two spins 2–5
To take a speciﬁc example, for protons γ = +2.67 × 10
8
rad s
−1
T
−1
,
so in a magnetic ﬁeld of 4.7 T the Larmor frequency of a spin with chemical
shift zero is
ν
0
= −
1
2π
γ (1 +δ)B
0
= −
1
2π
×2.67 ×10
8
×4.7 = −200 ×10
6
Hz.
In other words, the Larmor frequency is −200 MHz.
We can also calculate the Larmor frequency in angular units, ω
0
, in which
case the factor of 1/2π is not needed:
ω
0
= −γ (1 +δ)B
0
which gives a value of 1.255 × 10
9
rad s
−1
.
Spectrum
As you may knowfromother kinds of spectroscopy you have met, only certain
transitions are allowed i.e. only certain ones actually take place. There are
usually rules – called selection rules – about which transitions can take place;
these rules normally relate to the quantum numbers which are characteristic
of each state or energy level.
E
β
E
α
ν
αβ
= E
β
–E
α
m
= +
1
/
2
m
= –
1
/
2
ν
αβ
= ν
0,1
energy
levels
spectrum
Fig. 2.4 The transition between
the two energy levels of a
spinhalf is allowed, and results
in a single line at the Larmor
frequency of the spin.
In the case of NMR, the selection rule refers to the quantum number m:
only transitions in which m changes by one (up or down) are allowed. This is
sometimes expressed as
m = m(initial state) −m(ﬁnal state)
= ±1.
Another way as saying this is that one spin can ﬂip between “up” and “down”
or vice versa.
In the case of a single spinhalf, the change in m between the two states is
(+
1
2
− (−
1
2
)) = 1 so the transition is allowed. We can now simply work out
the frequency of the allowed transition:
ν
αβ
= E
β
− E
α
= −
1
2
ν
0,1
−(+
1
2
ν
0,1
)
= −ν
0,1
.
Note that we have taken the energy of the upper state minus that of the lower
state. In words, therefore, we see one transition at the minus the Larmor
frequency, −ν
0,1
.
You would be forgiven for thinking that this is all an enormous amount of
effort to come up with something very simple! However, the techniques and
ideas developed in this section will enable us to make faster progress with the
case of two and three coupled spins, which we consider next.
2.4 Two spins
2–6 NMR and energy levels
ν
0,2
ν
0,1
frequency
J
12
J
12
Fig. 2.5 Schematic spectrum of
two coupled spins showing two
doublets with equal splittings.
As indicated by the dashed
lines, the separation of the
Larmor frequencies is much
larger than the coupling
between the spins.
We know that the spectrum of two coupled spins consists of two doublets,
each split by the same amount, one centred at the chemical shift of the ﬁrst
spin and one at the shift of the second. The splitting of the doublets is the
scalar coupling, J
12
, quoted in Hz; the subscripts indicate which spins are
involved. We will write the shifts of the two spins as δ
1
and δ
2
, and give the
corresponding Larmor frequencies, ν
0,1
and ν
0,2
as:
ν
0,1
= −
1
2π
γ
1
(1 +δ
1
)B
0
ν
0,2
= −
1
2π
γ
2
(1 +δ
2
)B
0
.
If the two nuclei are of the same type, such a proton, then the two gyromag
netic ratios are equal; such a two spin system would be described as homonu
clear. The opposite case is where the two nuclei are of different types, such
as proton and carbon13; such a spin system is described as heteronuclear.
Energy levels
As was already described in section 2.2, there are four possible combinations
of the spin states of two spins and these combinations correspond to four
energy levels. Their energies are given in the following table:
number spin states energy
1 αα +
1
2
ν
0,1
+
1
2
ν
0,2
+
1
4
J
12
2 αβ +
1
2
ν
0,1
−
1
2
ν
0,2
−
1
4
J
12
3 βα −
1
2
ν
0,1
+
1
2
ν
0,2
−
1
4
J
12
4 ββ −
1
2
ν
0,1
−
1
2
ν
0,2
+
1
4
J
12
The second column gives the spin states of spins 1 and 2, in that order. It is
easy to see that these energies have the general form:
E
m
1
m
2
= m
1
ν
0,1
+m
2
ν
0,2
+m
1
m
2
J
12
where m
1
and m
2
are the m values for spins 1 and 2, respectively.
For a homonuclear system ν
0,1
≈ ν
0,2
; also both Larmor frequencies are
much greater in magnitude than the coupling (the Larmor frequencies are of
the order of hundreds of MHz, while couplings are at most a few tens of Hz).
Therefore, under these circumstances, the energies of the αβ and βα states are
rather similar, but very different from the other two states. For a heteronuclear
system, in which the Larmor frequencies differ signiﬁcantly, the four levels
are all at markedly different energies. These points are illustrated in Fig. 2.6.
Spectrum
The selection rule is the same as before, but this time it applies to the quantum
number M which is found by adding up the m values for each of the spins. In
this case:
M = m
1
+m
2
.
The resulting M values for the four levels are:
2.4 Two spins 2–7
αα
αα
αβ
βα
ββ
ββ
βα
αβ
1
H –
1
H
13
C –
1
H
1
1
2
2
3
3
4
4
Fig. 2.6 Energy levels, drawn approximately to scale, for two spin systems. On the left is shown a
homonuclear system (two protons); on this scale the αβ and βα states have the same energy. On the
right is the case for a carbon13 – proton pair. The Larmor frequency of proton is about four times that
of carbon13, and this is clear reﬂected in the diagram. The αβ and βα states now have substantially
different energies.
number spin states M
1 αα 1
2 αβ 0
3 βα 0
4 ββ −1
The selection rule is that M = ±1, i.e. the value of M can change up or
down by one unit. This means that the allowed transitions are between levels
1 & 2, 3 & 4, 1 & 3 and 2 & 4. The resulting frequencies are easily worked
out; for example, the 1–2 transition: Throughout we will use the
convention that when computing
the transition frequency we will
take the energy of the upper
state minus the energy of the
lower: E = E
upper
− E
lower
.
ν
12
= E
2
− E
1
= +
1
2
ν
0,1
−
1
2
ν
0,2
−
1
4
J
12
−(
1
2
ν
0,1
+
1
2
ν
0,2
+
1
4
J
12
)
= −ν
0,2
−
1
2
J
12
.
The complete set of transitions are:
transition spin states frequency
1 →2 αα →αβ −ν
0,2
−
1
2
J
12
3 →4 βα →ββ −ν
0,2
+
1
2
J
12
1 →3 αα →βα −ν
0,1
−
1
2
J
12
2 →4 αβ →ββ −ν
0,1
+
1
2
J
12
The energy levels and corresponding schematic spectrum are shown in
Fig. 2.7. There is a lot we can say about this spectrum. Firstly, each allowed
transition corresponds to one of the spins ﬂipping from one spin state to the
other, while the spin state of the other spin remains ﬁxed. For example, tran
sition 1–2 involves a spin 2 going from α to β whilst spin 1 remains in the α
state. In this transition we say that spin 2 is active and spin 1 is passive. As
2–8 NMR and energy levels
αα
αβ
ββ
βα
1
2 3
4
−ν
0,1
−ν
0,2
frequency
13
24
12
34
24
13
34
12
spin 1 flips
flips spin 2
α β
α β
Fig. 2.7 On the left, the energy levels of a twospin system; the arrows show the allowed transitions:
solid lines for transitions in which spin 1 ﬂips and dotted for those in which spin 2 ﬂips. On the right, the
corresponding spectrum; it is assumed that the Larmor frequency of spin 2 is greater in magnitude than
that of spin 1 and that the coupling J
12
is positive.
spin 2 ﬂips in this transition, it is not surprising that the transition forms one
part of the doublet for spin 2.
Transition 3–4 is similar to 1–2 except that the passive spin (spin 1) is
in the β state; this transition forms the second line of the doublet for spin 2.
This discussion illustrates a very important point, which is that the lines of a
multiplet can be associated with different spin states of the coupled (passive)
spins. We will use this kind of interpretation very often, especially when
considering twodimensional spectra.
The two transitions in which spin 1 ﬂips are 1–3 and 2–4, and these are
associated with spin 2 being in the α and β spin states, respectively. Which
spin ﬂips and the spins states of the passive spins are shown in Fig. 2.7.
What happens is the coupling is negative? If you work through the table
you will see that there are still four lines at the same frequencies as before. All
that changes is the labels of the lines. So, for example, transition 1–2 is now
the right line of the doublet, rather than the left line. From the point of view of
the spectrum, what swaps over is the spin state of the passive spin associated
with each line of the multiplet. The overall appearance of the spectrum is
therefore independent of the sign of the coupling constant.
Multiple quantum transitions
There are two more transitions in our twospin system which are not allowed
by the usual selection rule. The ﬁrst is between states 1 and 4 (αα →ββ) in
which both spins ﬂip. The M value is 2, so this is called a doublequantum
transition. Using the same terminology, all of the allowed transitions de
scribed above, which have M = 1, are singlequantumtransitions. From the
table of energy levels it is easy to work out that its frequency is (−ν
0,1
−ν
0,2
)
i.e. the sum of the Larmor frequencies. Note that the coupling has no effect
on the frequency of this line.
The second transition is between states 2 and 3 (αβ → βα); again, both
2.5 Three spins 2–9
spins ﬂip. The M value is 0, so this is called a zeroquantum transition,
and its frequency is (−ν
0,1
+ ν
0,2
) i.e. the difference of the Larmor frequen
cies. As with the doublequantum transition, the coupling has no effect on the
frequency of this line.
αα
βα
ββ
αβ
1
2 3
4
14
23
Fig. 2.8 In a twospin system
there is one double quantum
transition (1–4) and one
zeroquantum transition (2–3);
the frequency of neither of
these transitions are affected by
the size of the coupling between
the two spins.
In a two spin system the double and zeroquantum spectra are not espe
cially interesting, but we will see in a threespin system that the situation is
rather different. We will also see later on that in twodimensional spectra we
can exploit to our advantage the special properties of these multiplequantum
spectra.
2.5 Three spins
If we have three spins, each of which is coupled to the other two, then the
spectrum consists of three doublets of doublets, one centred at the shift of
each of the three spins; the spin topology is shown in Fig. 2.9. The appearance
of these multiplets will depend on the relative sizes of the couplings. For
example, if J
12
= J
13
the doublet of doublets from spin 1 will collapse to a
1:2:1 triplet. On the other hand, if J
12
= 0, only doublets will be seen for spin
1 and spin 2, but spin 3 will still show a doublet of doublets.
J
13
J
12
J
23
1
2 3
ν
0,1
J
13
J
12
J
13
Fig. 2.9 The topology – that is
the number of spins and the
couplings between them – for a
threespin system in which each
spin is coupled to both of the
others. The resulting spectrum
consists of three doublets of
doublets, a typical example of
which is shown for spin 1 with
the assumption that J
12
is
greater than J
13
.
Energy levels
Each of the three spins can be in the α or β spin state, so there are a total of
8 possible combinations corresponding to 8 energy levels. The energies are
given by:
E
m
1
m
2
m
3
= m
1
ν
0,1
+m
2
ν
0,2
+m
3
ν
0,3
+m
1
m
2
J
12
+m
1
m
3
J
13
+m
2
m
3
J
23
where m
i
is the value of the quantum number m for the i th spin. The energies
and corresponding M values (= m
1
+m
2
+m
3
) are shown in the table:
number spin states M energy
1 ααα
3
2
+
1
2
ν
0,1
+
1
2
ν
0,2
+
1
2
ν
0,3
+
1
4
J
12
+
1
4
J
13
+
1
4
J
23
2 αβα
1
2
+
1
2
ν
0,1
−
1
2
ν
0,2
+
1
2
ν
0,3
−
1
4
J
12
+
1
4
J
13
−
1
4
J
23
3 βαα
1
2
−
1
2
ν
0,1
+
1
2
ν
0,2
+
1
2
ν
0,3
−
1
4
J
12
−
1
4
J
13
+
1
4
J
23
4 ββα −
1
2
−
1
2
ν
0,1
−
1
2
ν
0,2
+
1
2
ν
0,3
+
1
4
J
12
−
1
4
J
13
−
1
4
J
23
5 ααβ
1
2
+
1
2
ν
0,1
+
1
2
ν
0,2
−
1
2
ν
0,3
+
1
4
J
12
−
1
4
J
13
−
1
4
J
23
6 αββ −
1
2
+
1
2
ν
0,1
−
1
2
ν
0,2
−
1
2
ν
0,3
−
1
4
J
12
−
1
4
J
13
+
1
4
J
23
7 βαβ −
1
2
−
1
2
ν
0,1
+
1
2
ν
0,2
−
1
2
ν
0,3
−
1
4
J
12
+
1
4
J
13
−
1
4
J
23
8 βββ −
3
2
−
1
2
ν
0,1
−
1
2
ν
0,2
−
1
2
ν
0,3
+
1
4
J
12
+
1
4
J
13
+
1
4
J
23
We have grouped the energy levels into two groups of four; the ﬁrst group
all have spin 3 in the α state and the second have spin 3 in the β state. The en
ergy levels (for a homonuclear system) are shown schematically in Fig. 2.10.
Spectrum
The selection rule is as before, that is M can only change by 1. However, in
the case of more than two spins, there is the additional constraint that only
2–10 NMR and energy levels
ααα
αβα
ββα
βαα
1
2 3
4
ααβ
αββ
βββ
βαβ
5
6 7
8
Fig. 2.10 Energy levels for a homonuclear threespin system. The levels can be grouped into two sets of
four: those with spin 3 in the α state (shown on the left with solid lines) and those with spin 3 in the β state,
shown on the right (dashed lines).
one spin can ﬂip. Applying these rules we see that there are four allowed
transitions in which spin 1 slips: 1–3, 2–4, 5–7 and 6–8. The frequencies of
these lines can easily be worked out from the table of energy levels on page 2–
9. The results are shown in the table, along with the spin states of the passive
spins (2 and 3 in this case).
transition state of spin 2 state of spin 3 frequency
1–3 α α −ν
0,1
−
1
2
J
12
−
1
2
J
13
2–4 β α −ν
0,1
+
1
2
J
12
−
1
2
J
13
5–7 α β −ν
0,1
−
1
2
J
12
+
1
2
J
13
6–8 β β −ν
0,1
+
1
2
J
12
+
1
2
J
13
ααα
αβα
ββα
βαα
1
2 3
4
ααβ
αββ
βββ
βαβ
5
6 7
8
−ν
0,1
J
13
J
12
J
12
J
13
96 92 108 104 100
13 57 24 68
spin 2
spin 3
α
α
α β
β α β
β
Fig. 2.11 Energy levels for a threespin system showing by the arrows the four allowed transitions which
result in the doublet of doublets at the shift of spin 1. The schematic multiplet is shown on the right, where
it has been assuming that ν
0,1
= −100 Hz, J
12
= 10 Hz and J
13
= 2 Hz. The multiplet is labelled with the
spin states of the passive spins.
These four transitions form the four lines of the multiplet (a doublet of
doublets) at the shift of spin 1. The schematic spectrum is illustrated in
Fig. 2.11. As in the case of a twospin system, we can label each line of the
2.5 Three spins 2–11
multiplet with the spin states of the passive spins – in the case of the multiplet
from spin 1, this means the spin states of spins 2 and 3. In the same way, we
can identify the four transitions which contribute to the multiplet from spin
2 (1–2, 3–4, 5–6 and 7–8) and the four which contribute to that from spin 3
(1–5, 3–7, 2–6 and 4–8).
Subspectra
ααα
αβα
ββα
βαα
1
2 3
4
ααβ
αββ
βββ
βαβ
5
6 7
8
−ν
0,1
−ν
0,2
J
13
J
12
spin 3
in β state
spin 3
in α state
J
12
J
12
J
12
J
23
−ν
0,1
β
3
−ν
0,2
β
3
−ν
0,1
α
3
−ν
0,2
α
3
Fig. 2.12 Illustration of the division of the two multiplets from spins 1 and 2 into subspectra according to
the spin state of spin 3. The transitions associated with spin 3 in the α state (indicated by the full lines
on the energy level diagram) give rise to a pair of doublets, but with their centres shifted from the Larmor
frequencies by half the coupling to spin 3. The same is true of those transitions associated with spin 3
being in the β state (dashed lines), except that the shift is in the opposite direction.
One was of thinking about the spectrum from the threespin system is
to divide up the lines in the multiplets for spins 1 and 2 into two groups or
subspectra. The ﬁrst group consists of the lines which have spin 3 in the α
state and the second group consists of the lines which have spin 3 in the α
state. This separation is illustrated in Fig. 2.12.
There are four lines which have spin3 in the α state, and as can be seen
from the spectrum these form two doublets with a common separation of J
12
.
However, the two doublets are not centred at −ν
0,1
and −ν
0,2
, but at (−ν
0,1
−
1
2
J
13
) and (−ν
0,2
−
1
2
J
23
). We can deﬁne an effective Larmor frequency for
spin 1 with spin 3 in the α spin state, ν
α
3
0,1
, as
ν
α
3
0,1
= ν
0,1
+
1
2
J
13
and likewise for spin 2:
ν
α
3
0,2
= ν
0,2
+
1
2
J
23
.
The two doublets in the subspectrum corresponding to spin 3 being in the α
state are thus centred at −ν
α
3
0,1
and −ν
α
3
0,2
. Similarly, we can deﬁne effective
Larmor frequencies for spin 3 being in the β state:
ν
β
3
0,1
= ν
0,1
−
1
2
J
13
ν
β
3
0,2
= ν
0,2
−
1
2
J
23
.
2–12 NMR and energy levels
The two doublets in the β subspectrum are centred at −ν
β
3
0,1
and −ν
β
3
0,2
.
We can think of the spectrum of spin 1 and 2 as being composed of two
subspectra, each from a two spin system but in which the Larmor frequencies
are effectively shifted one way of the other by half the coupling to the third
spin. This kind of approach is particularly useful when it comes to dealing
with strongly coupled spin systems, section 2.6
Note that the separation of the spectra according to the spin state of spin
3 is arbitrary. We could just as well separate the two multiplets from spins 1
and 3 according to the spin state of spin 2.
Multiple quantum transitions
There are six transitions in which M changes by 2. Their frequencies are
given in the table.
transition initial state ﬁnal state frequency
1–4 ααα ββα −ν
0,1
−ν
0,2
−
1
2
J
13
−
1
2
J
23
5–8 ααβ βββ −ν
0,1
−ν
0,2
+
1
2
J
13
+
1
2
J
23
1–7 ααα βαβ −ν
0,1
−ν
0,3
−
1
2
J
12
−
1
2
J
23
2–8 αβα βββ −ν
0,1
−ν
0,3
+
1
2
J
12
+
1
2
J
23
1–6 ααα αββ −ν
0,2
−ν
0,3
−
1
2
J
12
−
1
2
J
13
3–8 βαα βββ −ν
0,2
−ν
0,3
+
1
2
J
12
+
1
2
J
13
These transitions come in three pairs. Transitions 1–4 and 5–8 are centred
at the sum of the Larmor frequencies of spins 1 and 2; this is not surprising as
we note that in these transitions it is the spin states of both spins 1 and 2 which
ﬂip. The two transitions are separated by the sum of the couplings to spin 3
(J
13
+ J
23
), but they are unaffected by the coupling J
12
which is between the
two spins which ﬂip.
ααα
αβα
ββα
βαα
1
2 3
4
ααβ
αββ
βββ
βαβ
5
6 7
8
−ν
0,1
−ν
0,2
J
13
+ J
23
spin 3
58 14
β α
frequency
Fig. 2.13 There are two double quantum transitions in which spins 1 and 2 both ﬂip (1–4 and 5–8). The
two resulting lines form a doublet which is centred at the sum of the Larmor frequencies of spins 1 and
2 and which is split by the sum of the couplings to spin 3. As with the singlequantum spectra, we can
associate the two lines of the doublet with different spin states of the third spin. It has been assumed that
both couplings are positive.
We can describe these transitions as a kind of double quantum doublet.
Spins 1 and 2 are both active in these transitions, and spin 3 is passive. Just as
2.6 Strong coupling 2–13
we did before, we can associate one line with spin 3 being in the α state (1–4)
and one with it being in the β state (5–8). A schematic representation of the
spectrum is shown in Fig. 2.13.
There are also six zeroquantum transitions in which M does not change.
Like the double quantum transitions these group in three pairs, but this time
centred around the difference in the Larmor frequencies of two of the spins.
These zeroquantumdoublets are split by the difference of the couplings to the
spin which does not ﬂip in the transitions. There are thus many similarities
between the double and zeroquantum spectra.
In a three spin system there is one triplequantum transition, in which M
changes by 3, between levels 1 (ααα) and 8 (βββ). In this transition all of
the spins ﬂip, and from the table of energies we can easily work out that its
frequency is −ν
0,1
−ν
0,2
−ν
0,3
, i.e. the sum of the Larmor frequencies.
We see that the singlequantumspectrumconsists of three doublets of dou
blets, the doublequantum spectrum of three doublets and the triplequantum
spectrum of a single line. This illustrates the idea that as we move to higher
orders of multiple quantum, the corresponding spectra become simpler. This
feature has been used in the analysis of some complex spin systems.
Combination lines
There are three more transitions which we have not yet described. For these,
M changes by 1 but all three spins ﬂip; they are called combination lines.
Such lines are not seen in normal spectra but, like multiple quantum transi
tions, they can be detected indirectly using twodimensional spectra. We will
also see in section 2.6 that these lines may be observable in strongly coupled
spectra. The table gives the frequencies of these three lines:
transition initial state ﬁnal state frequency
2–7 αβα βαβ −ν
0,1
+ν
0,2
−ν
0,3
3–6 βαα αββ +ν
0,1
−ν
0,2
−ν
0,3
4–5 ββα ααβ +ν
0,1
+ν
0,2
−ν
0,3
Notice that the frequencies of these lines are not affected by any of the
couplings.
2.6 Strong coupling
We need to be careful here as
Larmor frequencies, the
differences between Larmor
frequencies and the values of
couplings can be positive or
negative! In deciding whether or
not a spectrum will be strongly
coupled we need to compare
the magnitude of the difference
in the Larmor frequencies with
the magnitude of the coupling.
So far all we have said about energy levels and spectra applies to what are
called weakly coupled spin systems. These are spin systems in which the
differences between the Larmor frequencies (in Hz) of the spins are much
greater in magnitude than the magnitude of the couplings between the spins.
Under these circumstances the rules from predicting spectra are very sim
ple – they are the ones you are already familiar with which you use for con
structing multiplets. In addition, in the weak coupling limit it is possible to
work out the energies of the levels present simply by making all possible com
binations of spin states, just as we have done above. Finally, in this limit all
of the lines in a multiplet have the same intensity.
2–14 NMR and energy levels
If the separation of the Larmor frequencies is not sufﬁcient to satisfy the
weak coupling criterion, the systemis said to be strongly coupled. In this limit
none of the rules outlined in the previous paragraph apply. This makes pre
dicting or analysing the spectra much more difﬁcult than in the case of weak
coupling, and really the only practical approach is to use computer simulation.
However, it is useful to look at the spectrum from two strongly coupled spins
as the spectrum is simple enough to analyse by hand and will reveal most of
the of the crucial features of strong coupling. You will often ﬁnd that people
talk of two spins being strongly
coupled when what they really
mean is the coupling between
the two spins is large. This is
sloppy usage; we will always
use the term strong coupling in
the sense described in this
section.
It is a relatively simple exercise in quantum mechanics to work out the
energy levels and hence frequencies and intensities of the lines froma strongly
coupled twospin system
1
. These are given in the following table.
transition frequency intensity
1–2
1
2
D −
1
2
−
1
2
J
12
(1 +sin 2θ)
3–4
1
2
D −
1
2
+
1
2
J
12
(1 −sin 2θ)
1–3 −
1
2
D −
1
2
−
1
2
J
12
(1 −sin 2θ)
2–4 −
1
2
D −
1
2
+
1
2
J
12
(1 +sin 2θ)
In this table is the sum of the Larmor frequencies:
= ν
0,1
+ν
0,2
and D is the positive quantity deﬁned as
D
2
= (ν
0,1
−ν
0,2
)
2
+ J
2
12
. (2.4)
The angle θ is called the strong coupling parameter and is deﬁned via
sin 2θ =
J
12
D
.
The ﬁrst thing to do is to verify that these formulae give us the expected
result when we impose the condition that the separation of the Larmor fre
quencies is large compared to the coupling. In this limit it is clear that
D
2
= (ν
0,1
−ν
0,2
)
2
+ J
2
12
≈ (ν
0,1
−ν
0,2
)
2
and so D = (ν
0,1
− ν
0,2
). Putting this value into the table above gives us
exactly the frequencies we had before on page 2–7.
When D is very much larger than J
12
the fraction J
12
/D becomes small,
and so sin 2θ ≈ 0 (sin φ goes to zero as φ goes to zero). Under these cir
cumstances all of the lines have unit intensity. So, the weak coupling limit is
regained.
Figure 2.14 shows a series of spectra computed using the above formulae
in which the Larmor frequency of spin 1 is held constant while the Larmor
frequency of spin 2 is progressively moved towards that of spin 1. This makes
the spectrum more and more strongly coupled. The spectrum at the bottom
is almost weakly coupled; the peaks are just about all the same intensity and
where we expect them to be.
1
See, for example, Chapter 10 of NMR: The Toolkit, by P J Hore, J A Jones and S Wim
peris (Oxford University Press, 2000)
2.6 Strong coupling 2–15
34 24 12 13
12 34
0 20 100 40 60 80
frequency (Hz)
–ν
0,1
ν
0,2
= –90
ν
0,2
= –50
ν
0,2
= –20
ν
0,2
= –10
–ν
0,2
–ν
0,2
–ν
0,2
–ν
0,2
J
12
J
12
~
Fig. 2.14 A series of spectra of a two spin system in which the Larmor frequency of spin 1 is help constant
and that of spin 2 is moved in closer to spin 1. The spectra become more and more strongly coupled
showing a pronounced roof effect until in the limit that the two Larmor frequencies are equal only one line
is observed. Note that as the “outer” lines get weaker the “inner” lines get proportionately stronger. The
parameters used for these spectra were ν
0,1
= −10 Hz and J
12
=5 Hz; the peak in the top most spectrum
has been truncated.
Fig. 2.15 The intensity
distributions in multiplets from
stronglycoupled spectra are
such that the multiplets “tilt”
towards one another; this is
called the “roof” effect.
However, as the Larmor frequencies of the two spins get closer and closer
together we notice two things: (1) the “outer” two lines get weaker and the
“inner” two lines get stronger; (2) the two lines which originally formed the
doublet are no longer symmetrically spaced about the Larmor frequency; in
fact the stronger of the two lines moves progressively closer to the Larmor
frequency. There is one more thing to notice which is not so clear from the
spectra but is clear if one looks at the frequencies in the table. This is that the
two lines that originally formed the spin 1 doublet are always separated by
J
12
; the same is true for the other doublet.
These spectra illustrate the socalled roof effect in which the intensities
of the lines in a strongly coupled multiplet tilt upwards towards the multiplet
from the coupled spin, making a kind of roof; Fig. 2.15 illustrates the idea.
The spectra in Fig 2.14 also illustrate the point that when the two Larmor
frequencies are identical there is only one line seen in the spectrum and this
is at this Larmor frequency. In this limit lines 1–2 and 2–4 both appear at the
Larmor frequency and with intensity 2; lines 1–3 and 3–4 appear elsewhere
but have intensity zero.
The “take home message” is that from such strongly coupled spectra we
can easily measure the coupling, but the Larmor frequencies (the shifts) are
no longer midway between the two lines of the doublet. In fact it is easy
2–16 NMR and energy levels
enough to work out the Larmor frequencies using the following method; the
idea is illustrated in Fig. 2.16.
J
12
J
12
D
13 12 24 34
D
Fig. 2.16 The quantities J
12
and
D are readily measurable from
the spectrum of two strongly
coupled spins.
If we denote the frequency of transition 1–2 as ν
12
and so on, it is clear
from the table that the frequency separation of the lefthand lines of the two
multiplets (3–4 and 2–4) is D
ν
34
−ν
24
= (
1
2
D −
1
2
+
1
2
J
12
) −(−
1
2
D −
1
2
+
1
2
J
12
)
= D.
The separation of the other two lines is also D. Remember we can easily
measure J
12
directly from the splitting, and so once we know D it is easy to
compute (ν
0,1
−ν
0,2
) from its deﬁnition, Eqn. 2.4.
D
2
= (ν
0,1
−ν
0,2
)
2
+ J
2
12
therefore (ν
0,1
−ν
0,2
) =
D
2
− J
2
12
.
Now we notice from the table on 2–14 that the sum of the frequencies of
the two stronger lines (1–2 and 2–4) or the two weaker lines (3–4 and 2–4)
gives us −:
ν
12
+ν
24
= (
1
2
D −
1
2
−
1
2
J
12
) +(−
1
2
D −
1
2
+
1
2
J
12
)
= −.
Now we have a values for = (ν
0,1
+ν
0,2
) and a value for (ν
0,1
−ν
0,2
) we
can ﬁnd ν
0,1
and ν
0,2
separately:
ν
0,1
=
1
2
( +(ν
0,1
−ν
0,2
)) ν
0,2
=
1
2
( −(ν
0,1
−ν
0,2
)).
In this way we can extract the Larmor frequencies of the two spins (the shifts)
and the coupling from the strongly coupled spectrum.
Notation for spin systems
There is a traditional notation for spin systems which it is sometimes useful to
use. Each spin is given a letter (rather than a number), with a different letter
for spins which have different Larmor frequencies (chemical shifts). If the
Larmor frequencies of two spins are widely separated they are given letters
which are widely separated in the alphabet. So, two weakly coupled spins are
usually denoted as A and X; whereas three weakly coupled spins would be
denoted AMX.
If spins are strongly coupled, then the convention is to used letters which
are close in the alphabet. So, a strongly coupled twospin systemwould be de
noted AB and a strongly coupled threespin system ABC. The notation ABX
implies that two of the spins (A and B) are strongly coupled but the Larmor
frequency of the third spin is widely separated from that of A and B.
The ABX spin system
We noted in section 2.5 that we could think about the spectrum of three cou
pled spins in terms of subspectra in which the Larmor frequencies were re
placed by effective Larmor frequencies. This kind of approach is very useful
for understanding the AB part of the ABX spectrum.
2.6 Strong coupling 2–17
40 50 30 20 10 0
–ν
0,A
–
1
/
2
J
AX
–ν
0,B
–
1
/
2
J
BX
–ν
0,A
+
1
/
2
J
AX
–ν
0,B
+
1
/
2
J
BX
–ν
0,B
–ν
0,A
frequency (Hz)
full spectrum
β subspectrum
α subspectrum
Fig. 2.17 AB parts of an ABX spectrum illustrating the decomposition into two subspectra with different
effective Larmor frequencies (indicated by the arrows). The parameters used in the simulation were
ν
0,A
= −20 Hz, ν
0,B
= −30 Hz, J
AB
= 5 Hz, J
AX
= 15 Hz and J
BX
= 3 Hz.
As spin X is weakly coupled to the others we can think of the AB part of
the spectrum as two superimposed AB subspectra; one is associated with the
X spin being in the α state and the other with the spin being in the β state. If
spins A and B have Larmor frequencies ν
0,A
and ν
0,B
, respectively, then one
subspectrum has effective Larmor frequencies ν
0,A
+
1
2
J
AX
and ν
0,B
+
1
2
J
BX
.
The other has effective Larmor frequencies ν
0,A
−
1
2
J
AX
and ν
0,B
−
1
2
J
BX
.
The separation between the two effective Larmor frequencies in the two
subspectra can easily be different, and so the degree of strong coupling (and
hence the intensity patterns) in the two subspectra will be different. All we
can measure is the complete spectrum (the sum of the two subspectra) but
once we know that it is in fact the sum of two ABtype spectra it is usually
possible to disentangle these two contributions. Once we have done this, the
two subspectra can be analysed in exactly the way described above for an
AB system. Figure 2.17 illustrates this decomposition.
The form of the X part of the ABX spectrum cannot be deduced from
this simple analysis. In general it contains 6 lines, rather than the four which
would be expected in the weak coupling limit. The two extra lines are combi
nation lines which become observable when strong coupling is present.
2–18 NMR and energy levels
2.7 Exercises
E 2–1
In a proton spectrum the peak from TMS is found to be at 400.135705 MHz.
What is the shift, in ppm, of a peak which has a frequency of
400.136305 MHz? Recalculate the shift using the spectrometer frequency,
ν
spec
quoted by the manufacturer as 400.13 MHz rather than ν
TMS
in the de
nominator of Eq. 2.2:
δ
ppm
= 10
6
×
ν −ν
TMS
ν
spec
.
Does this make a signiﬁcant difference to the value of the shift?
E 2–2
Two peaks in a proton spectrum are found at 1.54 and 5.34 ppm. The spec
trometer frequency is quoted as 400.13 MHz. What is the separation of these
two lines in Hz and in rad s
−1
?
E 2–3
Calculate the Larmor frequency (in Hz and in rad s
−1
) of a carbon13 res
onance with chemical shift 48 ppm when recorded in a spectrometer with
a magnetic ﬁeld strength of 9.4 T. The gyromagnetic ratio of carbon13 is
+6.7283 ×10
7
rad s
−1
T
−1
.
E 2–4
Consider a system of two weakly coupled spins. Let the Larmor frequency Of course in reality the Larmor
frequencies out to be tens or
hundreds of MHz, not 100 Hz!
However, it makes the numbers
easier to handle if we use these
unrealistic small values; the
principles remain the same,
however.
of the ﬁrst spin be −100 Hz and that of the second spin be −200 Hz, and let
the coupling between the two spins be −5 Hz. Compute the frequencies of
the lines in the normal (single quantum) spectrum.
Make a sketch of the spectrum, roughly to scale, and label each line with
the energy levels involved (i.e. 1–2 etc.). Also indicate for each line which
spin ﬂips and the spin state of the passive spin. Compare your sketch with
Fig. 2.7 and comment on any differences.
E 2–5
For a three spin system, draw up a table similar to that on page 2–10 showing
the frequencies of the four lines of the multiplet from spin 2. Then, taking
ν
0,2
= −200 Hz, J
23
= 4 Hz and the rest of the parameters as in Fig. 2.11,
compute the frequencies of the lines which comprise the spin 2 multiplet.
Make a sketch of the multiplet (roughly to scale) and label the lines in the
same way as is done in Fig. 2.11. How would these labels change if J
23
=
−4 Hz?
On an energy level diagram, indicate the four transitions which comprise
the spin 2 multiplet, and which four comprise the spin 3 multiplet.
E 2–6
For a three spin system, compute the frequencies of the six zeroquantum
2.7 Exercises 2–19
transitions and also mark these on an energy level diagram. Do these six
transitions fall into natural groups? How would you describe the spectrum?
E 2–7
Calculate the line frequencies and intensities of the spectrum for a system of
two spins with the following parameters: ν
0,1
= −10 Hz, ν
0,2
= −20 Hz,
J
12
= 5 Hz. Make a sketch of the spectrum (roughly to scale) indicating
which transition is which and the position of the Larmor frequencies.
E 2–8
The spectrum from a stronglycoupled two spin system showed lines at the Make sure that you have your
calculator set to “radians” when
you compute sin 2θ.
following frequencies, in Hz, (intensities are given in brackets): 32.0 (1.3),
39.0 (0.7), 6.0 (0.7), 13.0 (1.3). Determine the values of the coupling constant
and the two Larmor frequencies. Show that the values you ﬁnd are consistent
with the observed intensities.
frequency (Hz)
15 10 20 25 30 35 40
Fig. 2.18 The AB part of an ABX spectrum
E 2–9
Figure 2.18 shows the AB part of an ABX spectrum. Disentangle the two
subspectra, mark in the rough positions of the effective Larmor frequencies
and hence estimate the size of the AX and BX couplings. Also, give the value
of the AB coupling.
3 The vector model
For most kinds of spectroscopy it is sufﬁcient to think about energy levels
and selection rules; this is not true for NMR. For example, using this energy
level approach we cannot even describe how the most basic pulsed NMR ex
periment works, let alone the large number of subtle twodimensional exper
iments which have been developed. To make any progress in understanding
NMR experiments we need some more tools, and the ﬁrst of these we are
going to explore is the vector model.
This model has been around as long as NMR itself, and not surprisingly
the language and ideas which ﬂow from the model have become the language
of NMR to a large extent. In fact, in the strictest sense, the vector model can
only be applied to a surprisingly small number of situations. However, the
ideas that ﬂow from even this rather restricted area in which the model can
be applied are carried over into more sophisticated treatments. It is therefore
essential to have a good grasp of the vector model and how to apply it.
3.1 Bulk magnetization
magnetization
vector
m
a
g
n
e
t
i
c
f
i
e
l
d
z
x
y
Fig. 3.1 At equilibrium, a
sample has a net magnetization
along the magnetic ﬁeld
direction (the z axis) which can
be represented by a
magnetization vector. The axis
set in this diagram is a
righthanded one, which is what
we will use throughout these
lectures.
We commented before that the nuclear spin has an interaction with an applied
magnetic ﬁeld, and that it is this which gives rise the energy levels and ul
timately an NMR spectrum. In many ways, it is permissible to think of the
nucleus as behaving like a small bar magnet or, to be more precise, a mag
netic moment. We will not go into the details here, but note that the quantum
mechanics tells us that the magnetic moment can be aligned in any direction
1
.
In an NMR experiment, we do not observe just one nucleus but a very
large number of them (say 10
20
), so what we need to be concerned with is the
net effect of all these nuclei, as this is what we will observe.
If the magnetic moments were all to point in random directions, then the
small magnetic ﬁeld that each generates will cancel one another out and there
will be no net effect. However, it turns out that at equilibrium the magnetic
moments are not aligned randomly but in such a way that when their contri
butions are all added up there is a net magnetic ﬁeld along the direction of the
applied ﬁeld (B
0
). This is called the bulk magnetization of the sample.
The magnetization can be represented by a vector – called the magnetiza
tion vector – pointing along the direction of the applied ﬁeld (z), as shown in
Fig. 3.1. From now on we will only be concerned with what happens to this
vector.
This would be a good point to comment on the axis systemwe are going to
use – it is called a righthanded set, and such a set of axes is show in Fig. 3.1.
The name righthanded comes about fromthe fact that if you imagine grasping
1
It is a common misconception to state that the magnetic moment must either be aligned
with or against the magnetic ﬁeld. In fact, quantum mechanics says no such thing (see Levitt
Chapter 9 for a very lucid discussion of this point).
Chapter 3 “The vector model” c James Keeler, 2002
3–2 The vector model
the z axis with your right hand, your ﬁngers curl from the x to the y axes.
3.2 Larmor precession
m
a
g
n
e
t
i
c
f
i
e
l
d
z
x
y
Fig. 3.2 If the magnetization
vector is tilted away from the z
axis it executes a precessional
motion in which the vector
sweeps out a cone of constant
angle to the magnetic ﬁeld
direction. The direction of
precession shown is for a
nucleus with a positive
gyromagnetic ratio and hence a
negative Larmor frequency.
Suppose that we have managed, some how, to tip the magnetization vector
away from the z axis, such that it makes an angle β to that axis. We will see
later on that such a tilt can be brought about by a radiofrequency pulse. Once
tilted away from the z axis we ﬁnd is that the magnetization vector rotates
about the direction of the magnetic ﬁeld sweeping out a cone with a constant
angle; see Fig. 3.2. The vector is said to precesses about the ﬁeld and this
particular motion is called Larmor precession.
If the magnetic ﬁeld strength is B
0
, then the frequency of the Larmor
precession is ω
0
(in rad s
−1
)
ω
0
= −γ B
0
or if we want the frequency in Hz, it is given by
ν
0
= −
1
2π
γ B
0
where γ is the gyromagnetic ratio. These are of course exactly the same
frequencies that we encountered in section 2.3. In words, the frequency at
which the magnetization precesses around the B
0
ﬁeld is exactly the same as
the frequency of the line we see from the spectrum on one spin; this is no
accident.
As was discussed in section 2.3, the Larmor frequency is a signed quantity
and is negative for nuclei with a positive gyromagnetic ratio. This means that
for such spins the precession frequency is negative, which is precisely what is
shown in Fig. 3.2.
z
x
y
Fig. 3.3 The precessing
magnetization will cut a coil
wound round the x axis, thereby
inducing a current in the coil.
This current can be ampliﬁed
and detected; it is this that forms
the free induction signal. For
clarity, the coil has only been
shown on one side of the x axis.
We can sort out positive and negative frequencies in the following way.
Imagine grasping the z axis with your right hand, with the thumb pointing
along the +z direction. The ﬁngers then curl in the sense of a positive preces
sion. Inspection of Fig. 3.2 will show that the magnetization vector is rotating
in the opposite sense to your ﬁngers, and this corresponds to a negative Lar
mor frequency.
3.3 Detection
The precession of the magnetization vector is what we actually detect in an
NMR experiment. All we have to do is to mount a small coil of wire round
the sample, with the axis of the coil aligned in the xyplane; this is illustrated
in Fig. 3.3. As the magnetization vector “cuts” the coil a current is induced
which we can amplify and then record – this is the socalled free induction
signal which is detected in a pulse NMR experiment. The whole process is
analogous to the way in which electric current can be generated by a magnet
rotating inside a coil.
3.4 Pulses 3–3
z
x
M
0
sin β
β
Fig. 3.4 Tilting the
magnetization through an angle
θ gives an xcomponent of size
M
0
sin β.
Essentially, the coil detects the xcomponent of the magnetization. We can
easily work out what this will be. Suppose that the equilibrium magnetization
vector is of size M
0
; if this has been tilted through an angle β towards the x
axis, the xcomponent is M
0
sin β; Fig. 3.4 illustrates the geometry.
Although the magnetization vector precesses on a cone, we can visualize
what happens to the x and ycomponents much more simply by just thinking
about the projection onto the xyplane. This is shown in Fig. 3.5.
At time zero, we will assume that there is only an xcomponent. After a
time τ
1
the vector has rotated through a certain angle, which we will call
1
.
As the vector is rotating at ω
0
radians per second, in time τ
1
the vector has
moved through (ω
0
× τ
1
) radians; so
1
= ω
0
τ
1
. At a later time, say τ
2
, the
vector has had longer to precess and the angle
2
will be (ω
0
τ
2
). In general,
we can see that after time t the angle is = ω
0
t .
τ
1
ε
1 ε
2
ε
τ
2
x
y
Fig. 3.5 Illustration of the precession of the magnetization vector in the xyplane. The angle through which
the vector has precessed is given by ω
0
t. On the righthand diagram we see the geometry for working out
the x and y components of the vector.
time
M
y
M
x
Fig. 3.6 Plots of the x and
ycomponents of the
magnetization predicted using
the approach of Fig. 3.5.
Fourier transformation of these
signals will give rise to the usual
spectrum.
We can now easily work out the x and ycomponents of the magnetiza
tion using simple geometry; this is illustrated in Fig. 3.5. The xcomponent
is proportional to cos and the ycomponent is negative (along −y) and pro
portional to sin . Recalling that the initial size of the vector is M
0
sin β, we
can deduce that the x and ycomponents, M
x
and M
y
respectively, are:
M
x
= M
0
sin β cos(ω
0
t )
M
y
= −M
0
sin β sin(ω
0
t ).
Plots of these signals are shown in Fig. 3.6. We see that they are both
simple oscillations at the Larmor frequency. Fourier transformation of these
signals gives us the familiar spectrum – in this case a single line at ω
0
; the
details of how this works will be covered in a later chapter. We will also
see in a later section that in practice we can easily detect both the x and
ycomponents of the magnetization.
3.4 Pulses
We now turn to the important question as to how we can rotate the magneti
zation away from its equilibrium position along the z axis. Conceptually it is
3–4 The vector model
easy to see what we have to do. All that is required is to (suddenly) replace
the magnetic ﬁeld along the z axis with one in the xyplane (say along the
x axis). The magnetization would then precess about the new magnetic ﬁeld
which would bring the vector down away from the z axis, as illustrated in
Fig. 3.7.
magnetic
field
z
x
y
z
x
y
Fig. 3.7 If the magnetic ﬁeld
along the z axis is replaced
quickly by one along x, the
magnetization will then precess
about the x axis and so move
towards the transverse plane.
Unfortunately it is all but impossible to switch the magnetic ﬁeld suddenly
in this way. Remember that the main magnetic ﬁeld is supplied by a powerful
superconducting magnet, and there is no way that this can be switched off;
we will need to ﬁnd another approach, and it turns out that the key is to use
the idea of resonance.
The idea is to apply a very small magnetic ﬁeld along the x axis but one
which is oscillating at or near to the Larmor frequency – that is resonant with
the Larmor frequency. We will show that this small magnetic ﬁeld is able to
rotate the magnetization away from the z axis, even in the presence of the very
strong applied ﬁeld, B
0
.
Conveniently, we can use the same coil to generate this oscillating mag
netic ﬁeld as the one we used to detect the magnetization (Fig. 3.3). All we
do is feed some radiofrequency (RF) power to the coil and the resulting oscil
lating current creates an oscillating magnetic ﬁeld along the xdirection. The
resulting ﬁeld is called the radiofrequency or RF ﬁeld. To understand how this
weak RF ﬁeld can rotate the magnetization we need to introduce the idea of
the rotating frame.
Rotating frame
When RF power is applied to the coil wound along the x axis the result is a
magnetic ﬁeld which oscillates along the x axis. The magnetic ﬁeld moves
back and forth from +x to −x passing through zero along the way. We will
take the frequency of this oscillation to be ω
RF
(in rad s
−1
) and the size of
the magnetic ﬁeld to be 2B
1
(in T); the reason for the 2 will become apparent
later. This frequency is also called the transmitter frequency for the reason
that a radiofrequency transmitter is used to produce the power.
It turns out to be a lot easier to work out what is going on if we replace,
in our minds, this linearly oscillating ﬁeld with two counterrotating ﬁelds;
Fig. 3.8 illustrates the idea. The two counter rotating ﬁelds have the same
magnitude B
1
. One, denoted B
+
1
, rotates in the positive sense (from x to y)
and the other, denoted B
−
1
, rotates in the negative sense; both are rotating at
the transmitter frequency ω
RF
.
At time zero, they are both aligned along the x axis and so add up to give
a total ﬁeld of 2B
1
along the x axis. As time proceeds, the vectors move away
from x, in opposite directions. As the two vectors have the same magnitude
and are rotating at the same frequency the ycomponents always cancel one
another out. However, the xcomponents shrink towards zero as the angle
through which the vectors have rotated approaches
1
2
π radians or 90
◦
. As the
angle increases beyond this point the xcomponent grows once more, but this
time along the −x axis, reaching a maximum when the angle of rotation is π.
The ﬁelds continue to rotate, causing the xcomponent to drop back to zero
and rise again to a value 2B
1
along the +x axis. Thus we see that the two
3.4 Pulses 3–5
x
x
y
y
2B
1
B
1

B
1
+
time
f
i
e
l
d
a
l
o
n
g
x
Fig. 3.8 Illustration of how two counterrotating ﬁelds (shown in the upper part of the diagram and marked
B
+
1
and B
−
1
) add together to give a ﬁeld which is oscillating along the x axis (shown in the lower part). The
graph at the bottom shows how the ﬁeld along x varies with time.
counterrotating ﬁelds add up to the linearly oscillating one.
Suppose now that we think about a nucleus with a positive gyromagnetic
ratio; recall that this means the Larmor frequency is negative so that the sense
of precession is from x towards −y. This is the same direction as the rotation
of B
−
1
. It turns out that the other ﬁeld, which is rotating in the opposite sense
to the Larmor precession, has no signiﬁcant interaction with the magnetiza
tion and so from now on we will ignore it.
x
x
ω
RF
y
y
B
1

ω
RF
Fig. 3.9 The top row shows a ﬁeld rotating at −ω
RF
when viewed in a ﬁxed axis system. The same ﬁeld
viewed in a set of axes rotating at −ω
RF
appears to be static.
We now employ a mathematical trick which is to move to a coordinate
system which, rather than being static (called the laboratory frame) is rotating
about the z axis in the same direction and at the same rate as B
−
1
(i.e. at
−ω
RF
). In this rotating set of axes, or rotating frame, B
−
1
appears to be static
3–6 The vector model
and directed along the x axis of the rotating frame, as is shown in Fig. 3.9.
This is a very nice result as the time dependence has been removed from the
problem.
Larmor precession in the rotating frame
We need to consider what happens to the Larmor precession of the magne
tization when this is viewed in this rotating frame. In the ﬁxed frame the
precession is at ω
0
, but suppose that we choose the rotating frame to be at the
same frequency. In the rotating frame the magnetization will appear not to
move i.e. the apparent Larmor frequency will be zero! It is clear that moving
to a rotating frame has an effect on the apparent Larmor frequency.
The general case is when the rotating frame is at frequency ω
rot. fram.
; in
such a frame the Larmor precession will appear to be at (ω
0
−ω
rot. fram.
). This
difference frequency is called the offset and is given the symbol :
= ω
0
−ω
rot. fram.
. (3.1)
We have used several times the relationship between the magnetic ﬁeld
and the precession frequency:
ω = −γ B. (3.2)
Fromthis it follows that if the apparent Larmor frequency in the rotating frame
is different from that in ﬁxed frame it must also be the case that the apparent
magnetic ﬁeld in the rotating frame must be different from the actual applied
magnetic ﬁeld. We can use Eq. 3.2 to compute the apparent magnetic ﬁeld,
given the symbol B, from the apparent Larmor frequency, :
= −γ B
hence B = −
γ
.
This apparent magnetic ﬁeld in the rotating frame is usually called the reduced
ﬁeld, B.
If we choose the rotating frame to be at the Larmor frequency, the offset
will be zero and so too will the reduced ﬁeld. This is the key to how the
very weak RF ﬁeld can affect the magnetization in the presence of the much
stronger B
0
ﬁeld. In the rotating frame this ﬁeld along the z axis appears to
shrink, and under the right conditions can become small enough that the RF
ﬁeld is dominant.
The effective ﬁeld
In this discussion we will
assume that the gyromagnetic
ratio is positive so that the
Larmor frequency is negative.
From the discussion so far we can see that when an RF ﬁeld is being applied
there are two magnetic ﬁelds in the rotating frame. First, there is the RF ﬁeld
(or B
1
ﬁeld) of magnitude B
1
; we will make this ﬁeld static by choosing the
rotating frame frequency to be equal to −ω
RF
. Second, there is the reduced
ﬁeld, B, given by (−/γ ). Since = (ω
0
− ω
rot. fram.
) and ω
rot. fram.
=
−ω
RF
it follows that the offset is
= ω
0
−(−ω
RF
)
= ω
0
+ω
RF
.
3.4 Pulses 3–7
This looks rather strange, but recall that ω
0
is negative, so if the transmitter
frequency and the Larmor frequency are comparable the offset will be small.
z
x
∆B
θ
B
eff
B
1
Fig. 3.10 In the rotating frame
the effective ﬁeld B
eff
is the
vector sum of the reduced ﬁeld
B and the B
1
ﬁeld. The tilt
angle, θ, is deﬁned as the angle
between B and B
eff
.
In the rotating frame, the reduced ﬁeld (which is along z) and the RF or
B
1
ﬁeld (which is along x) add vectorially to give an effective ﬁeld, B
eff
as
illustrated in Fig. 3.10. The size of this effective ﬁeld is given by:
B
eff
=
B
2
1
+B
2
. (3.3)
The magnetization precesses about this effective ﬁeld at frequency ω
eff
given
by
ω
eff
= γ B
eff
just in the same way that the Larmor precession frequency is related to B
0
.
By making the offset small, or zero, the effective ﬁeld lies close to the
xyplane, and so the magnetization will be rotated from z down to the plane,
which is exactly what we want to achieve. The trick is that although B
0
is
much larger than B
1
we can affect the magnetization with B
1
by making it
oscillate close to the Larmor frequency. This is the phenomena of resonance.
The angle between B and B
eff
is called the tilt angle and is usually given
the symbol θ. From Fig. 3.10 we can see that:
sin θ =
B
1
B
eff
cos θ =
B
B
eff
tan θ =
B
1
B
.
All three deﬁnitions are equivalent.
The effective ﬁeld in frequency units
For practical purposes the thing that is important is the precession frequency
about the effective ﬁeld, ω
eff
. It is therefore convenient to think about the
construction of the effective ﬁeld not in terms of magnetic ﬁelds but in terms
of the precession frequencies that they cause.
z
x
ω
eff
θ
Ω
ω
1
Fig. 3.11 The effective ﬁeld can
be thought of in terms of
frequencies instead of the ﬁelds
used in Fig 3.10.
For each ﬁeld the precession frequency is proportional to the magnetic
ﬁeld; the constant of proportion is γ , the gyromagnetic ratio. For example, we
have already seen that in the rotating frame the apparent Larmor precession
frequency, , depends on the reduced ﬁeld:
= −γ B.
We deﬁne ω
1
as the precession frequency about the B
1
ﬁeld (the positive sign
is intentional):
ω
1
= γ B
1
and we already have
ω
eff
= γ B
eff
.
Using these deﬁnitions in Eq. 3.3 ω
eff
can be written as
ω
eff
=
ω
2
1
+
2
.
Figure 3.10 can be redrawn in terms of frequencies, as shown in Fig. 3.11.
Similarly, the tilt angle can be expressed in terms of these frequencies:
sin θ =
ω
1
ω
eff
cos θ =
ω
eff
tan θ =
ω
1
.
3–8 The vector model
Onresonance pulses
The simplest case to deal with is where the transmitter frequency is exactly
the same as the Larmor frequency – it is said that the pulse is exactly on
resonance. Under these circumstances the offset, , is zero and so the reduced
ﬁeld, B, is also zero. Referring to Fig. 3.10 we see that the effective ﬁeld is
therefore the same as the B
1
ﬁeld and lies along the x axis. For completeness
we also note that the tilt angle, θ, of the effective ﬁeld is π/2 or 90
◦
.
x
y
z
B
1
x
y
z
B
1
(a) (b)
Fig. 3.12 A “grapefruit” diagram in which the thick line shows the motion of a magnetization vector during
an onresonance pulse. The magnetization is assumed to start out along +z. In (a) the pulse ﬂip angle
is 90
◦
. The effective ﬁeld lies along the x axis and so the magnetization precesses in the yzplane. The
rotation is in the positive sense about x so the magnetization moves toward the −y axis. In (b) the pulse
ﬂip angle is 180
◦
and so the magnetization ends up along −z.
In this situation the motion of the magnetization vector is very simple. Just
as in Fig. 3.7 the magnetization precesses about the ﬁeld, thereby rotating in
the zyplane. As we have seen above the precession frequency is ω
1
. If the RF
ﬁeld is applied for a time t
p
, the angle, β, through which the magnetization
has been rotated will be given by
β = ω
1
t
p
.
β is called the ﬂip angle of the pulse. By altering the time for which the pulse
has been applied we can alter than angle through which the magnetization is
rotated.
In many experiments the commonly used ﬂip angles are π/2 (90
◦
) and π
(180
◦
). The motion of the magnetization vector during onresonance 90
◦
and
180
◦
pulses are shown in Fig. 3.12. The 90
◦
pulse rotates the magnetization
from the equilibrium position to the −y axis; this is because the rotation is
in the positive sense. Imagine grasping the axis about which the rotation is
taking place (the x axis) with your right hand; your ﬁngers then curl in the
sense of a positive rotation.
If the pulse ﬂip angle is set to 180
◦
the magnetization is taken all the way
from +z to −z; this is called an inversion pulse. In general, for a ﬂip angle β
3.5 Detection in the rotating frame 3–9
simple geometry tells us that the z and ycomponents are
M
z
= M
0
cos β M
y
= −M
0
sin β;
this is illustrated in Fig. 3.13.
z
y
M
0
sin β
M
0
c
o
s
β
β
Fig. 3.13 If the pulse ﬂip angle
is β we can use simple
geometry to work out the y and
zcomponents.
Hard pulses
In practical NMR spectroscopy we usually have several resonances in the
spectrum, each of which has a different Larmor frequency; we cannot there
fore be on resonance with all of the lines in the spectrum. However, if we
make the RF ﬁeld strong enough we can effectively achieve this condition.
By “hard” enough we mean that the B
1
ﬁeld has to be large enough that it
is much greater than the size of the reduced ﬁeld, B. If this condition holds,
the effective ﬁeld lies along B
1
and so the situation is identical to the case
of an onresonance pulse. Thinking in terms of frequencies this condition
translates to ω
1
being greater in magnitude than the offset, .
transmitter
frequency
0 5 10
ppm
Fig. 3.14 Illustration of the
range of offsets one might see
in a typical proton spectrum. If
the transmitter frequency is
placed as shown, the maximum
offset of a resonance will be 5
ppm.
It is often relatively easy to achieve this condition. For example, consider
a proton spectrum covering about 10 ppm; if we put the transmitter frequency
at about 5 ppm, the maximum offset is 5 ppm, either positive or negative;
this is illustrated in Fig. 3.14. If the spectrometer frequency is 500 MHz the
maximum offset is 5 ×500 = 2500 Hz. A typical spectrometer might have a
90
◦
pulse lasting 12 µs. From this we can work out the value of ω
1
. We start
from
β = ω
1
t
p
hence ω
1
=
β
t
p
.
We know that for a 90
◦
pulse β = π/2 and the duration, t
p
is 12 × 10
−6
s;
therefore
ω
1
=
π/2
12 ×10
−6
= 1.3 ×10
5
rad s
−1
.
The maximum offset is 2500 Hz, which is 2π × 2500 = 1.6 × 10
4
rad s
−1
.
We see that the RF ﬁeld is about eight times the offset, and so the pulse can
be regarded as strong over the whole width of the spectrum.
3.5 Detection in the rotating frame
To work out what is happening during an RF pulse we need to work in the
rotating frame, and we have seen that to get this simpliﬁcation the frequency
of the rotating frame must match the transmitter frequency, ω
RF
. Larmor
precession can be viewed just as easily in the laboratory and rotating frames;
in the rotating frame the precession is at the offset frequency, .
It turns out that because of the way the spectrometer works the signal that
we detect appears to be that in the rotating frame. So, rather than detecting an
oscillation at the Larmor frequency, we see an oscillation at the offset, .
2
2
Strictly this is only true if we set the receiver reference frequency to be equal to the
transmitter frequency; this is almost always the case. More details will be given in Chapter 5.
3–10 The vector model
It also turns out that we can detect both the x and ycomponents of the
magnetization in the rotating frame. From now on we will work exclusively
in the rotating frame.
3.6 The basic pulse–acquire experiment
RF
acq
90˚
{ { {
1 2 3
Fig. 3.15 Timing diagram or
pulse sequence for the simple
pulse–acquire experiment. The
line marked “RF” shows the
location of the pulses, and the
line marked “acq” shows when
the signal is recorded or
acquired.
At last we are in a position to describe how the simplest NMR experiment
works – the one we use every day to record spectra. The experiment comes in
three periods:
1. The sample is allowed to come to equilibrium.
2. RF power is switched on for long enough to rotate the magnetization
through 90
◦
i.e. a 90
◦
pulse is applied.
3. After the RF power is switched off we start to detect the signal which
arises from the magnetization as it rotates in the transverse plane.
The timing diagram – or pulse sequence as it is usually known – is shown in
Fig. 3.15.
x
y
time
M
x
M
y
Fig. 3.16 Evolution during the acquisition time (period 3) of the pulse–acquire experiment. The magne
tization starts out along −y and evolves at the offset frequency, (here assumed to be positive). The
resulting x and ymagnetizations are shown below.
During period 1 equilibrium magnetization builds up along the z axis. As
was described above, the 90
◦
pulse rotates this magnetization onto the −y
axis; this takes us to the end of period 2. During period 3 the magnetization
precesses in the transverse plane at the offset ; this is illustrated in Fig. 3.16.
Some simple geometry, shown in Fig. 3.17, enables us to deduce how the
x and ymagnetizations vary with time. The offset is so after time t the
vector has precessed through an angle (×t ). The ycomponent is therefore
proportional to cos t and the xcomponent to sin t . In full the signals are:
M
y
= −M
0
cos(t )
M
x
= M
0
sin(t ).
As we commented on before, Fourier transformation of these signals will give
the usual spectrum, with a peak appearing at frequency .
3.7 Pulse calibration 3–11
x
y
Ωt
Fig. 3.17 The magnetization
starts out along the −y axis and
rotates through an angle t
during time t.
Spectrum with several lines
If the spectrum has more than one line, then to a good approximation we can
associate a magnetization vector with each. Usually we wish to observe all
the lines at once, so we choose the B
1
ﬁeld to be strong enough that for the
range of offsets that these lines cover all the associated magnetization vectors
will be rotated onto the −y axis. During the acquisition time each precesses
at its own offset, so the detected signal will be:
M
y
= −M
0,1
cos
1
t − M
0,2
cos
2
t − M
0,3
cos
3
t . . .
where M
0,1
is the equilibrium magnetization of spin 1,
1
is its offset and so
on for the other spins. Fourier transformation of the free induction signal will
produce a spectrum with lines at
1
,
2
etc.
3.7 Pulse calibration
It is crucial that the pulses we use in NMR experiments have the correct ﬂip
angles. For example, to obtain the maximum intensity in the pulse–acquire
experiment we must use a 90
◦
pulse, and if we wish to invert magnetization we
must use a 180
◦
pulse. Pulse calibration is therefore an important preliminary
to any experiment.
flip angle
s
i
g
n
a
l
i
n
t
e
n
s
i
t
y
90º 180º 270º
π 3π/2 π/2
Fig. 3.18 Illustration of how pulse calibration is achieved. The signal intensity varies as (sin β) as shown by
the curve at the bottom of the picture. Along the top are the spectra which would be expected for various
different ﬂip angles (indicated by the dashed lines). The signal is a maximum for a ﬂip angle of 90
◦
and
goes through a null at 180
◦
; after that, the signal goes negative.
If we imagine an onresonance or hard pulse we have already determined
from Fig. 3.13 that the ycomponent of magnetization after a pulse of ﬂip an
gle β is proportional to sin β. If we therefore do a pulseacquire experiment
(section 3.6) and vary the ﬂip angle of the pulse, we should see that the inten
sity of the signal varies as sin β. A typical outcome of such an experiment is
shown in Fig. 3.18.
The normal practice is to increase the ﬂip angle until a null is found; the
ﬂip angle is then 180
◦
. The reason for doing this is that the null is sharper
than the maximum. Once the length of a 180
◦
pulse is found, simply halving
the time gives a 90
◦
pulse.
3–12 The vector model
Suppose that the 180
◦
pulse was found to be of duration t
180
. Since the
ﬂip angle is given by β = ω
1
t
p
we can see that for a 180
◦
pulse in which the
ﬂip angle is π
π = ω
1
t
180
hence ω
1
=
π
t
180
.
In this way we can determine ω
1
, usually called the RF ﬁeld strength or the
B
1
ﬁeld strength.
It is usual to quote the ﬁeld strength not in rad s
−1
but in Hz, in which
case we need to divide by 2π:
(ω
1
/2π) =
1
2t
180
Hz.
For example, let us suppose that we found the null condition at 15.5 µs; thus
ω
1
=
π
t
180
=
π
15.5 ×10
−6
= 2.03 ×10
5
rad s
−1
.
In frequency units the calculation is
(ω
1
/2π) =
1
2t
180
=
1
2 ×15.5 ×10
−6
= 32.3 kHz.
In normal NMR parlance we would say “the B
1
ﬁeld is 32.3 kHz”. This is
mixing the units up rather strangely, but the meaning is clear once you know
what is going on!
3.8 The spin echo
x
y
time
p
h
a
s
e
,
φ
0
0
τ/2 τ τ
τ
3τ/2 2τ
2τ
1
8
0
p
u
l
s
e
π
π/2
π−Ωτ
Ωτ Ωτ
Fig. 3.19 Vector diagrams showing how a spin echo refocuses the evolution of the offset; see text for
details. Also shown is a phase evolution diagram for two different offsets (the solid and the dashed line).
3.9 Pulses of different phases 3–13
RF
acq
90˚ 180˚
τ τ
2τ
Fig. 3.20 Pulse sequence for
the spin echo experiment. The
180
◦
pulse (indicated by an
open rectangle as opposed to
the closed one for a 90
◦
pulse)
is in the centre of a delay of
duration 2τ, thus separating the
sequence into two equal
periods, τ. The signal is
acquired after the second delay
τ, or put another way, when the
time from the beginning of the
sequence is 2τ. The durations
of the pulses are in practice very
much shorter than the delays τ
but for clarity the length of the
pulses has been exaggerated.
We are now able to analyse the most famous pulsed NMR experiment, the
spin echo, which is a component of a very large number of more complex ex
periments. The pulse sequence is quite simple, and is shown in Fig. 3.20. The
special thing about the spin echo sequence is that at the end of the second τ
delay the magnetization ends up along the same axis, regardless of the values
of τ and the offset, .
We describe this outcome by saying that “the offset has been refocused”,
meaning that at the end of the sequence it is just as if the offset had been
zero and hence there had been no evolution of the magnetization. Figure 3.19
illustrates how the sequence works after the initial 90
◦
pulse has placed the
magnetization along the −y axis.
During the ﬁrst delay τ the vector precesses from −y towards the x axis.
The angle through which the vector rotates is simply (t ), which we can
describe as a phase, φ. The effect of the 180
◦
pulse is to move the vector to a
mirror image position, with the mirror in question being in the xzplane. So,
the vector is now at an angle (τ) to the y axis rather than being at (τ) to
the −y axis.
During the second delay τ the vector continues to evolve; during this time
it will rotate through a further angle of (τ) and therefore at the end of the
second delay the vector will be aligned along the y axis. A few moments
thought will reveal that as the angle through which the vector rotates dur
ing the ﬁrst τ delay must be equal to that through which it rotates during
the second τ delay; the vector will therefore always end up along the y axis
regardless of the offset, .
The 180
◦
pulse is called a refocusing pulse because of the property that
the evolution due to the offset during the ﬁrst delay τ is refocused during the
second delay. It is interesting to note that the spin echo sequence gives exactly
the same result as the sequence 90
◦
– 180
◦
with the delays omitted.
Another way of thinking about the spin echo is to plot a phase evolution
diagram; this is done at the bottom of Fig. 3.19. Here we plot the phase, φ,
as a function of time. During the ﬁrst τ delay the phase increases linearly
with time. The effect of the 180
◦
pulse is to change the phase from (τ) to
(π − τ); this is the jump on the diagram at time τ. Further evolution for
time τ causes the phase to increase by (τ) leading to a ﬁnal phase at the end
of the second τ delay of π. This conclusion is independent of the value of the
offset ; the diagram illustrates this by the dashed line which represents the
evolution of vector with a smaller offset.
As has already been mentioned, the effect of the 180
◦
pulse is to reﬂect
the vectors in the xzplane. The way this works is illustrated in Fig. 3.21.
The arc through which the vectors are moved is different for each, but all the
vectors end up in mirror image positions.
3.9 Pulses of different phases
So far we have assume that the B
1
ﬁeld is applied along the x axis; this does
not have to be so, and we can just as easily apply it along the y axis, for
example. A pulse about y is said to be “phase shifted by 90
◦
” (we take an
3–14 The vector model
y
y
z
x
Fig. 3.21 Illustration of the effect of a 180
◦
pulse on three vectors which start out at different angles from
the −y axis (coloured in black, grey and light grey). All three are rotated by 180
◦
about the x axis on the
trajectories indicated by the thick lines which dip into the southern hemisphere. As a result, the vectors
end up in mirror image positions with respect to the xzplane.
x
y
y
z
B
1
x
z
B
1
(a) (b)
Fig. 3.22 Grapefruit plots showing the effect on equilibrium magnetization of (a) a 90
◦
pulse about the y
axis and (b) a 90
◦
pulse about the −x axis. Note the position of the B
1
ﬁeld in each case.
xpulse to have a phase shift of zero); likewise a pulse about −x would be
said to be phase shifted by 180
◦
. On modern spectrometers it is possible to
produce pulses with arbitrary phase shifts.
If we apply a 90
◦
pulse about the y axis to equilibrium magnetization we
ﬁnd that the vector rotates in the yzplane such that the magnetization ends
up along x; this is illustrated in Fig. 3.22. As before, we can determine the
effect of such a pulse by thinking of it as causing a positive rotation about the
y axis. A 90
◦
pulse about −x causes the magnetization to appear along y, as
is also shown in Fig. 3.22.
We have seen that a 180
◦
pulse about the x axis causes the vectors to
move to mirror image positions with respect to the xzplane. In a similar way,
3.10 Relaxation 3–15
a 180
◦
pulse about the y axis causes the vectors to be reﬂected in the yzplane.
3.10 Relaxation
RF
acq
90º 180º
τ
Fig. 3.23 The pulse sequence
for the inversion recovery
experiment used to measure
longitudinal relaxation.
We will have a lot more to say about relaxation later on, but at this point we
will just note that the magnetization has a tendency to return to its equilibrium
position (and size) – a process known as relaxation. Recall that the equilib
rium situation has magnetization of size M
0
along z and no transverse (x or
y) magnetization.
So, if we have created some transverse magnetization (for example by ap
plying a 90
◦
pulse) over time relaxation will cause this magnetization to decay
away to zero. The free induction signal, which results from the magnetization
precessing in the xy plane will therefore decay away in amplitude. This loss
of x and ymagnetization is called transverse relaxation.
Once perturbed, the zmagnetization will try to return to its equilibrium
position, and this process is called longitudinal relaxation. We can measure
the rate of this process using the inversion recovery experiment whose pulse
sequence is shown in Fig. 3.23. The 180
◦
pulse rotates the equilibrium mag
netization to the −z axis. Suppose that the delay τ is very short so that at the
end of this delay the magnetization has not changed. Now the 90
◦
pulse will
rotate the magnetization onto the +y axis; note that this is in contrast to the
case where the magnetization starts out along +z and it is rotated onto −y.
If this gives a positive line in the spectrum, then having the magnetization
along +y will give a negative line. So, what we see for short values of τ is a
negative line.
increasing τ
Fig. 3.24 Visualization of the outcome of an inversion recovery experiment. The size and sign of the z
magnetization is reﬂected in the spectra (shown underneath). By analysing the peak heights as a function
of the delay τ it is possible to ﬁnd the rate of recovery of the zmagnetization.
As τ gets longer more relaxation takes place and the magnetization shrinks
towards zero; this result is a negative line in the spectrum, but one whose size
is decreasing. Eventually the magnetization goes through zero and then starts
to increase along +z – this gives a positive line in the spectrum. Thus, by
recording spectra with different values of the delay τ we can map out the
recovery of the zmagnetization from the intensity of the observed lines. The
3–16 The vector model
whole process is visualized in Fig. 3.24.
3.11 Offresonance effects and soft pulses
y
z
x
b
c
d
D
C
B
A
a
Fig. 3.25 Grapefruit diagram showing the path followed during a pulse for various different resonance
offsets. Path a is for the onresonance case; the effective ﬁeld lies along x and is indicated by the dashed
line A. Path b is for the case where the offset is half the RF ﬁeld strength; the effective ﬁeld is marked B.
Paths c and d are for offsets equal to and 1.5 times the RF ﬁeld strength, respectively. The effective ﬁeld
directions are labelled C and D.
So far we have only dealt with the case where the pulse is either on res
onance or where the RF ﬁeld strength is large compared to the offset (a hard
pulse) which is in effect the same situation. We now turn to the case where
the offset is comparable to the RF ﬁeld strength. The consequences of this are
sometimes a problem to us, but they can also be turned to our advantage for
selective excitation.
As the offset becomes comparable to the RF ﬁeld, the effective ﬁeld be
gins to move up from the x axis towards the z axis. As a consequence, rather
than the magnetization moving in the yz plane from z to −y, the magne
tization starts to follow a more complex curved path; this is illustrated in
Fig. 3.25. The further off resonance we go, the further the vector ends up
from the xy plane. Also, there is a signiﬁcant component of magnetization
generated along the x direction, something which does not occur in the on
resonance case.
We can see more clearly what is going on if we plot the x and y magne
tization as a function of the offset; these are shown in Fig. 3.26. In (a) we
see the ymagnetization and, as expected for a 90
◦
pulse, on resonance the
equilibrium magnetization ends up entirely along −y. However, as the offset
increases the amount of ymagnetization generally decreases but imposed on
this overall decrease there is an oscillation; at some offsets the magnetiza
tion is zero and at others it is positive. The plot of the xmagnetization, (b),
shows a similar story with the magnetization generally falling off as the offset
increases, but again with a strong oscillation.
Plot (c) is of the magnitude of the magnetization, which is given by
M
abs
=
M
2
x
+ M
2
y
.
3.11 Offresonance effects and soft pulses 3–17
(a) (b)
(c)
20 10
10 20
1
0.5
0.5
1
20 10 0 10 20
0.5
1
20 10 10 20
1
0.5
0.5
1
M
x
M
y
M
abs
Ω / ω
1
Ω / ω
1
Ω / ω
1
Fig. 3.26 Plots of the magnetization produced by a pulse as a function of the offset. The pulse length has
been adjusted so that on resonance the ﬂip angle is 90
◦
. The horizontal axes of the plots is the offset
expressed as a ratio of the RF ﬁeld strength, ω
1
; the equilibrium magnetization has been assumed to be
of size 1.
This gives the total transverse magnetization in any direction; it is, of course,
always positive. We see from this plot the characteristic nulls and subsidiary
maxima as the offset increases.
What plot (c) tells us is that although a pulse can excite magnetization
over a wide range of offsets, the region over which it does so efﬁciently is
really rather small. If we want at least 90% of the full intensity the offset
must be less than about 1.6 times the RF ﬁeld strength.
Excitation of a range of shifts
There are some immediate practical consequences of this observation. Sup
pose that we are trying to record the full range of carbon13 shifts (200 ppm)
on an spectrometer whose magnetic ﬁeld gives a proton Larmor frequency
of 800 MHz and hence a carbon13 Larmor frequency of 200 MHz. If we
place the transmitter frequency at 100 ppm, the maximum offset that a peak
can have is 100 ppm which, at this Larmor frequency, translates to 20 kHz.
According to our criterion above, if we accept a reduction to 90% of the full
intensity at the edges of the spectrum we would need an RF ﬁeld strength of
20/1.6 ≈ 12.5 kHz. This would correspond to a 90
◦
pulse width of 20 µs. If
the manufacturer failed to provide sufﬁcient power to produce this pulse width
we can see that the excitation of the spectrum will fall below our (arbitrary)
90% mark.
We will see in a later section that the presence of a mixture of x and y
magnetization leads to phase errors in the spectrum which can be difﬁcult to
correct.
Selective excitation
Sometimes we want to excite just a portion of the spectrum, for example just
the lines of a single multiplet. We can achieve this by putting the transmitter
in the centre of the region we wish to excite and then reducing the RF ﬁeld
3–18 The vector model
transmitter
offset
strong pulse
selective pulse
excitation
Fig. 3.27 Visualization of the use of selective excitation to excite just one line in the spectrum. At the top
is shown the spectrum that would be excited using a hard pulse. If the transmitter is placed on resonance
with one line and the strength of the RF ﬁeld reduced then the pattern of excitation we expect is as shown
in the middle. As a result, the peaks at nonzero offsets are attenuated and the spectrum which is excited
will be as shown at the bottom.
strength until the degree of excitation of the rest of the spectrum is essentially
negligible. Of course, reducing the RF ﬁeld strength lengthens the duration
of a 90
◦
pulse. The whole process is visualized in Fig. 3.27.
Such pulses which are designed to affect only part of the spectrum are
called selective pulses or soft pulses (as opposed to nonselective or hard
pulses). The value to which we need to reduce the RF ﬁeld depends on the
separation of the peak we want to excite from those we do not want to excite.
The closer in the unwanted peaks are the weaker the RF ﬁeld must become
and hence the longer the 90
◦
pulse. In the end a balance has to be made be
tween making the pulse too long (and hence losing signal due to relaxation)
and allowing a small amount of excitation of the unwanted signals.
Figure 3.27 does not portray one problem with this approach, which is
that for peaks away from the transmitter a mixture of x and ymagnetization
is generated (as shown in Fig. 3.26). This is described as a phase error, more
of which in a later section. The second problem that the ﬁgure does show is
that the excitation only falls off rather slowly and “bounces” through a series
of maximum and nulls; these are sometimes called “wiggles”. We might
be lucky and have an unwanted peak fall on a null, or unlucky and have an
unwanted peak fall on a maximum.
Much effort has been put into getting round both of these problems. The
key feature of all of the approaches is to “shape” the envelope of the RF
pulses i.e. not just switch it on and off abruptly, but with a smooth variation.
Such pulses are called shaped pulses. The simplest of these are basically bell
shaped (like a gaussian function, for example). These suppress the “wiggles”
at large offsets and give just a smooth decay; they do not, however, improve
the phase properties. To attack this part of the problem requires an altogether
more sophisticated approach.
3.11 Offresonance effects and soft pulses 3–19
Selective inversion
Sometimes we want to invert the magnetization associated with just one res
onance while leaving all the others in the spectrum unaffected; such a pulse
would be called a selective inversion pulse. Just as for selective excitation,
all we need to do is to place the transmitter on the line we wish to invert
and reduce the RF ﬁeld until the other resonances in the spectrum are not af
fected signiﬁcantly. Of course we need to adjust the pulse duration so that the
onresonance ﬂip angle is 180
◦
.
M
z
M
z
Ω / ω
1
Ω / ω
1
20 10 10
5 5
20
0.5 0.5
0.5 0.5
1 1
1 1
(a) (b)
Fig. 3.28 Plots of the zmagnetization produced by a pulse as a function of the offset; the ﬂip angle
onresonance has been set to 180
◦
. Plot (b) covers a narrower range of offsets than plot (a). These plots
should be compared with those in Fig. 3.26.
Figure 3.28 shows the zmagnetization generated as a function of offset
for a pulse. We see that the range over which there is signiﬁcant inversion
is rather small, and that the oscillations are smaller in amplitude than for the
excitation pulse.
This observation has two consequences: one “good” and one “bad”. The
good consequence is that a selective 180
◦
pulse is, for a given ﬁeld strength,
more selective than a corresponding 90
◦
pulse. In particular, the weaker
“bouncing” sidelobes are a useful feature. Do not forget, though, that the
180
◦
pulse is longer, so some of the improvement may be illusory!
The bad consequence is that when it comes to hard pulses the range of
offsets over which there is anything like complete inversion is much more
limited than the range of offsets over which there is signiﬁcant excitation.
This can be seen clearly by comparing Fig. 3.28 with Figure 3.26. Thus, 180
◦
pulses are often the source of problems in spectra with large offset ranges.
3–20 The vector model
3.12 Exercises
E 3–1
A spectrometer operates with a Larmor frequency of 600 MHz for protons.
For a particular set up the RF ﬁeld strength, ω
1
/(2π) has been determined to
be 25 kHz. Suppose that the transmitter is placed at 5 ppm; what is the offset
(in Hz) of a peak at 10 ppm? Compute the tilt angle, θ, of a spin with this
offset.
For the normal range of proton shifts (0 – 10 ppm), is this 25 kHz ﬁeld
strong enough to give what could be classed as hard pulses?
E 3–2
In an experiment to determine the pulse length an operator observed a positive
signal for pulse widths of 5 and 10 µs; as the pulse was lengthened further the
intensity decreased going through a null at 20.5 µs and then turning negative.
Explain what is happening in this experiment and use the data to determine
the RF ﬁeld strength in Hz and the length of a 90
◦
pulse.
A further null in the signal was seen at 41.0 µs; to what do you attribute
this?
E 3–3
Use vector diagrams to describe what happens during a spin echo sequence in
which the 180
◦
pulse is applied about the y axis. Also, draw a phase evolution
diagram appropriate for this pulse sequence.
In what way is the outcome different from the case where the refocusing
pulse is applied about the x axis?
What would the effect of applying the refocusing pulse about the −x axis
be?
E 3–4
The gyromagnetic ratio of phosphorus31 is 1.08 × 10
8
rad s
−1
T
−1
. This
nucleus shows a wide range of shifts, covering some 700 ppm.
Estimate the minimum 90
◦
pulse width you would need to excite peaks
in this complete range to within 90% of the their theoretical maximum for a
spectrometer with a B
0
ﬁeld strength of 9.4 T.
E 3–5
A spectrometer operates at a Larmor frequency of 400 MHz for protons and
hence 100 MHz for carbon13. Suppose that a 90
◦
pulse of length 10 µs is
applied to the protons. Does this have a signiﬁcant effect of the carbon13
nuclei? Explain your answer carefully.
E 3–6
Referring to the plots of Fig. 3.26 we see that there are some offsets at which
the transverse magnetization goes to zero. Recall that the magnetization is ro
tating about the effective ﬁeld, ω
eff
; it follows that these nulls in the excitation
3.12 Exercises 3–21
come about when the magnetization executes complete 360
◦
rotations about
the effective ﬁeld. In such a rotation the magnetization is returned to the z
axis. Make a sketch of a “grapefruit” showing this.
The effective ﬁeld is given by
ω
eff
=
ω
2
1
+
2
.
Suppose that we express the offset as a multiple κ of the RF ﬁeld strength:
= κω
1
.
Show that with this values of the effective ﬁeld is given by:
ω
eff
= ω
1
1 +κ
2
.
(The reason for doing this is to reduce the number of variables.)
Let us assume that onresonance the pulse ﬂip angle is π/2, so the duration
of the pulse, τ
p
, is give from
ω
1
τ
p
= π/2 thus τ
p
=
π
2ω
1
.
The angle of rotation about the effective ﬁeld for a pulse of duration τ
p
is
(ω
eff
τ
p
). Show that for the effective ﬁeld given above this angle, β
eff
is given
by
β
eff
=
π
2
1 +κ
2
.
The null in the excitation will occur when β
eff
is 2π i.e. a complete rota
tion. Show that this occurs when κ =
√
15 i.e. when (/ω
1
) =
√
15. Does
this agree with Fig. 3.26?
Predict other values of κ at which there will be nulls in the excitation.
E 3–7
When calibrating a pulse by looking for the null produced by a 180
◦
rotation,
why is it important to choose a line which is close to the transmitter frequency
(i.e. one with a small offset)?
E 3–8
Use vector diagrams to predict the outcome of the sequence:
90
◦
−delay τ −90
◦
applied to equilibrium magnetization; both pulses are about the x axis. In
your answer, explain how the x, y and z magnetizations depend on the delay
τ and the offset .
E 3–9
Consider the spin echo sequence to which a 90
◦
pulse has been added at the
end:
90
◦
(x) −delay τ −180
◦
(x) −delay τ −90
◦
(φ).
3–22 The vector model
The axis about which the pulse is applied is given in brackets after the ﬂip
angle. Explain in what way the outcome is different depending on whether
the phase φ of the pulse is chosen to be x, y, −x or −y.
E 3–10
The socalled “1–1” sequence is:
90
◦
(x) −delay τ −90
◦
(−x)
For a peak which is on resonance the sequence does not excite any observable
magnetization. However, for a peak with an offset such that τ = π/2 the
sequence results in all of the equilibrium magnetization appearing along the
x axis. Further, if the delay is such that τ = π no transverse magnetization
is excited.
Explain these observations and make a sketch graph of the amount of
transverse magnetization generated as a function of the offset for a ﬁxed delay
τ.
The sequence has been used for suppressing strong solvent signals which
might otherwise overwhelmthe spectrum. The solvent is placed on resonance,
and so is not excited; τ is chosen so that the peaks of interest are excited. How
does one go about choosing the value for τ?
E 3–11
The socalled “1–1” sequence is:
90
◦
(x) −delay τ −90
◦
(y).
Describe the excitation that this sequence produces as a function of offset.
How it could be used for observing spectra in the presence of strong solvent
signals?
4 Fourier transformation and data
processing
In the previous chapter we have seen how the precessing magnetization can
be detected to give a signal which oscillates at the Larmor frequency – the
free induction signal. We also commented that this signal will eventually
decay away due to the action of relaxation; the signal is therefore often called
the free induction decay or FID. The question is how do we turn this signal,
which depends on time, into the a spectrum, in which the horizontal axis is
frequency.
time
frequency
Fourier
transformation
Fig. 4.1 Fourier transformation
is the mathematical process
which takes us from a function
of time (the time domain) – such
as a FID – to a function of
frequency – the spectrum.
This conversion is made using a mathematical process known as Fourier
transformation. This process takes the time domain function (the FID) and
converts it into a frequency domain function (the spectrum); this is shown in
Fig. 4.1. In this chapter we will start out by exploring some features of the
spectrum, such as phase and lineshapes, which are closely associated with
the Fourier transform and then go on to explore some useful manipulations of
NMR data such as sensitivity and resolution enhancement.
4.1 The FID
In section 3.6 we saw that the x and y components of the free induction sig
nal could be computed by thinking about the evolution of the magnetization
during the acquisition time. In that discussion we assumed that the magneti
zation started out along the −y axis as this is where it would be rotated to by
a 90
◦
pulse. For the purposes of this chapter we are going to assume that the
magnetization starts out along x; we will see later that this choice of starting
position is essentially arbitrary.
x
y
time
M
y
M
x
Fig. 4.2 Evolution of the magnetization over time; the offset is assumed to be positive and the magnetization
starts out along the x axis.
Chapter 4 “Fourier transformation and data processing” c James Keeler, 2002
4–2 Fourier transformation and data processing
S
x
S
y
S
0
Ωt
Fig. 4.3 The x and y
components of the signal can
be thought of as arising from
the rotation of a vector S
0
at
frequency .
If the magnetization does indeed start along x then Fig. 3.16 needs to be
redrawn, as is shown in Fig. 4.2. From this we can easily see that the x and y
components of the magnetization are:
M
x
= M
0
cos t
M
y
= M
0
sin t.
The signal that we detect is proportional to these magnetizations. The con
stant of proportion depends on all sorts of instrumental factors which need not
concern us here; we will simply write the detected x and y signals, S
x
(t ) and
S
y
(t ) as
S
x
(t ) = S
0
cos t and S
y
(t ) = S
0
sin t
where S
0
gives is the overall size of the signal and we have reminded ourselves
that the signal is a function of time by writing it as S
x
(t ) etc.
It is convenient to think of this signal as arising from a vector of length S
0
rotating at frequency ; the x and y components of the vector give S
x
and S
y
,
as is illustrated in Fig. 4.3.
time
S
x
(
t
)
o
r
r
e
a
l
p
a
r
t
S
y
(
t
)
o
r
i
m
a
g
.
p
a
r
t
Fig. 4.4 Illustration of a typical
FID, showing the real and
imaginary parts of the signal;
both decay over time.
As a consequence of the way the Fourier transform works, it is also con
venient to regard S
x
(t ) and S
y
(t ) as the real and imaginary parts of a complex
signal S(t ):
S(t ) = S
x
(t ) +i S
y
(t )
= S
0
cos t +i S
0
sin t
= S
0
exp(i t ).
We need not concern ourselves too much with the mathematical details here,
but just note that the timedomain signal is complex, with the real and imag
inary parts corresponding to the x and y components of the signal.
We mentioned at the start of this section that the transverse magnetization
decays over time, and this is most simply represented by an exponential decay
with a time constant T
2
. The signal then becomes
S(t ) = S
0
exp(i t ) exp
−t
T
2
. (4.1)
A typical example is illustrated in Fig. 4.4. Another way of writing this is to
deﬁne a (ﬁrst order) rate constant R
2
= 1/T
2
and so S(t ) becomes
S(t ) = S
0
exp(i t ) exp(−R
2
t ). (4.2)
The shorter the time T
2
(or the larger the rate constant R
2
) the more rapidly
the signal decays.
4.2 Fourier transformation
Fourier transformation of a signal such as that given in Eq. 4.1 gives the fre
quency domain signal which we know as the spectrum. Like the time domain
signal the frequency domain signal has a real and an imaginary part. The real
4.2 Fourier transformation 4–3
part of the spectrum shows what we call an absorption mode line, in fact in
the case of the exponentially decaying signal of Eq. 4.1 the line has a shape
known as a Lorentzian, or to be precise the absorption mode Lorentzian. The
imaginary part of the spectrum gives a lineshape known as the dispersion
mode Lorentzian. Both lineshapes are illustrated in Fig. 4.5.
time
frequency
Fig. 4.6 Illustration of the fact that the more rapidly the FIDdecays the broader the line in the corresponding
spectrum. A series of FIDs are shown at the top of the ﬁgure and below are the corresponding spectra,
all plotted on the same vertical scale. The integral of the peaks remains constant, so as they get broader
the peak height decreases.
frequency
Ω
absorption
dispersion
Fig. 4.5 Illustration of the
absorption and dispersion mode
Lorentzian lineshapes.
Whereas the absorption
lineshape is always positive, the
dispersion lineshape has
positive and negative parts; it
also extends further.
This absorption lineshape has a width at half of its maximum height of
1/(πT
2
) Hz or (R/π) Hz. This means that the faster the decay of the FID
the broader the line becomes. However, the area under the line – that is the
integral – remains constant so as it gets broader so the peak height reduces;
these points are illustrated in Fig. 4.6.
If the size of the time domain signal increases, for example by increasing
S
0
the height of the peak increases in direct proportion. These observations
lead to the very important consequence that by integrating the lines in the
spectrum we can determine the relative number of protons (typically) which
contribute to each.
The dispersion line shape is not one that we would choose to use. Not
only is it broader than the absorption mode, but it also has positive and nega
tive parts. In a complex spectrum these might cancel one another out, leading
to a great deal of confusion. If you are familiar with ESR spectra you might
recognize the dispersion mode lineshape as looking like the derivative line
shape which is traditionally used to plot ESR spectra. Although these two
lineshapes do look roughly the same, they are not in fact related to one an
other.
Positive and negative frequencies
As we discussed in section 3.5, the evolution we observe is at frequency
i.e. the apparent Larmor frequency in the rotating frame. This offset can be
positive or negative and, as we will see later, it turns out to be possible to
determine the sign of the frequency. So, in our spectrum we have positive and
negative frequencies, and it is usual to plot these with zero in the middle.
4–4 Fourier transformation and data processing
Several lines
What happens if we have more than one line in the spectrum? In this case,
as we saw in section 3.5, the FID will be the sum of contributions from each
line. For example, if there are three lines S(t ) will be:
S(t ) =S
0,1
exp(i
1
t ) exp
−t
T
(1)
2
+ S
0,2
exp(i
2
t ) exp
−t
T
(2)
2
+ S
0,3
exp(i
3
t ) exp
−t
T
(3)
2
.
where we have allowed each line to have a separate intensity, S
0,i
, frequency,
i
, and relaxation time constant, T
(i )
2
.
The Fourier transform is a linear process which means that if the time
domain is a sum of functions the frequency domain will be a sum of Fourier
transforms of those functions. So, as Fourier transformation of each of the
terms in S(t ) gives a line of appropriate width and frequency, the Fourier
transformation of S(t ) will be the sum of these lines – which is the complete
spectrum, just as we require it.
4.3 Phase
So far we have assumed that at time zero (i.e. at the start of the FID) S
x
(t ) is
a maximum and S
y
(t ) is zero. However, in general this need not be the case
– it might just as well be the other way round or anywhere in between. We
describe this general situation be saying that the signal is phase shifted or that
it has a phase error. The situation is portrayed in Fig. 4.7.
In Fig. 4.7 (a) we see the situation we had before, with the signal starting
out along x and precessing towards y. The real part of the FID (corresponding
to S
x
) is a damped cosine wave and the imaginary part (corresponding to S
y
)
is a damped sine wave. Fourier transformation gives a spectrum in which the
real part contains the absorption mode lineshape and the imaginary part the
dispersion mode.
In (b) we see the effect of a phase shift, φ, of 45
◦
. S
y
now starts out at a
ﬁnite value, rather than at zero. As a result neither the real nor the imaginary
part of the spectrum has the absorption mode lineshape; both are a mixture of
absorption and dispersion.
In (c) the phase shift is 90
◦
. Now it is S
y
which takes the form of a
damped cosine wave, whereas S
x
is a sine wave. The Fourier transform gives
a spectrum in which the absorption mode signal now appears in the imaginary
part. Finally in (d) the phase shift is 180
◦
and this gives a negative absorption
mode signal in the real part of the spectrum.
What we see is that in general the appearance of the spectrum depends on
the position of the signal at time zero, that is on the phase of the signal at time
zero. Mathematically, inclusion of this phase shift means that the (complex)
signal becomes:
S(t ) = S
0
exp(i φ) exp(i t ) exp
−t
T
2
. (4.3)
4.3 Phase 4–5
S
x
S
y
S
x
S
y
S
x
S
y
S
x
S
y
S
x
S
y
S
x
S
y
φ
S
x
S
y
φ
S
x
S
y
φ
real imag imag
imag imag
real
real real
(a) (b)
(d) (c)
Fig. 4.7 Illustration of the effect of a phase shift of the time domain signal on the spectrum. In (a) the
signal starts out along x and so the spectrum is the absorption mode in the real part and the dispersion
mode in the imaginary part. In (b) there is a phase shift, φ, of 45
◦
; the real and imaginary parts of the
spectrum are now mixtures of absorption and dispersion. In (c) the phase shift is 90
◦
; now the absorption
mode appears in the imaginary part of the spectrum. Finally in (d) the phase shift is 180
◦
giving a negative
absorption line in the real part of the spectrum. The vector diagrams illustrate the position of the signal at
time zero.
Phase correction
It turns out that for instrumental reasons the axis along which the signal ap
pears cannot be predicted, so in any practical situation there is an unknown
phase shift. In general, this leads to a situation in which the real part of the
spectrum (which is normally the part we display) does not show a pure ab
sorption lineshape. This is undesirable as for the best resolution we require
an absorption mode lineshape.
Luckily, restoring the spectrum to the absorption mode is easy. Suppose
with take the FID, represented by Eq. 4.3, and multiply it by exp(i φ
corr
):
exp(i φ
corr
)S(t ) = exp(i φ
corr
) ×
S
0
exp(i φ) exp(i t ) exp
−t
T
2
.
This is easy to do as by now the FID is stored in computer memory, so the
4–6 Fourier transformation and data processing
multiplication is just a mathematical operation on some numbers. Exponen
tials have the property that exp(A) exp(B) = exp(A + B) so we can rewrite
the time domain signal as
exp(i φ
corr
)S(t ) = exp(i (φ
corr
+φ))
S
0
exp(i t ) exp
−t
T
2
.
Now suppose that we set φ
corr
= −φ; as exp(0) = 1 the time domain signal
becomes:
exp(i φ
corr
)S(t ) = S
0
exp(i t ) exp
−t
T
2
.
The signal now has no phase shift and so will give us a spectrum in which
the real part shows the absorption mode lineshape – which is exactly what we
want. All we need to do is ﬁnd the correct φ
corr
.
It turns out that the phase correction can just as easily be applied to the
spectrum as it can to the FID. So, if the spectrum is represented by S(ω) (a
function of frequency, ω) the phase correction is applied by computing
exp(i φ
corr
)S(ω).
Such a correction is called a frequency independent or zero order phase cor
rection as it is the same for all peaks in the spectrum, regardless of their offset. Attempts have been made over
the years to automate this
phasing process; on well
resolved spectra the results are
usually good, but these
automatic algorithms tend to
have more trouble with
poorlyresolved spectra. In any
case, what constitutes a
correctly phased spectrum is
rather subjective.
In practice what happens is that we Fourier transform the FID and dis
play the real part of the spectrum. We then adjust the phase correction (i.e.
the value of φ
corr
) until the spectrum appears to be in the absorption mode –
usually this adjustment is made by turning a knob or by a “click and drag”
operation with the mouse. The whole process is called phasing the spectrum
and is something we have to do each time we record a spectrum.
In addition to the phase shifts introduced by the spectrometer we can of
course deliberately introduce a shift of phase by, for example, altering the
phase of a pulse. In a sense it does not matter what the phase of the signal
is – we can always obtain an absorption spectrum by phase correcting the
spectrum later on.
Frequency dependent phase errors
We saw in section 3.11 that if the offset becomes comparable with the RF
ﬁeld strength a 90
◦
pulse about x results in the generation of magnetization
along both the x and y axes. This is in contrast to the case of a hard pulse,
where the magnetization appears only along −y. We can now describe this
mixture of x and y magnetization as resulting in a phase shift or phase error
of the spectrum.
Figure 3.25 illustrates very clearly how the x component increases as the
offset increases, resulting in a phase error which also increases with offset.
Therefore lines at different offsets in the spectrum will have different phase
errors, the error increasing as the offset increases. This is illustrated schemat
ically in the upper spectrum shown in Fig. 4.8.
If there were only one line in the spectrum it would be possible to ensure
that the line appeared in the absorption mode simply by adjusting the phase
4.3 Phase 4–7
p
h
a
s
e
frequency
0
frequency
0
(a)
(b)
(c)
Fig. 4.8 Illustration of the appearance of a frequency dependent phase error in the spectrum. In (a) the
line which is on resonance (at zero frequency) is in pure absorption, but as the offset increases the phase
error increases. Such an frequency dependent phase error would result from the use of a pulse whose
RF ﬁeld strength was not much larger than the range of offsets. The spectrum can be returned to the
absorption mode, (c), by applying a phase correction which varies with the offset in a linear manner, as
shown in (b). Of course, to obtain a correctly phased spectrum we have to choose the correct slope of the
graph of phase against offset.
in the way described above. However, if there is more than one line present
in the spectrum the phase correction for each will be different, and so it will
be impossible to phase all of the lines at once.
Luckily, it is often the case that the phase correction needed is directly
proportional to the offset – called a linear or ﬁrst order phase correction. Such
a variation in phase with offset is shown in Fig. 4.8 (b). All we have to do is
to vary the rate of change of phase with frequency (the slope of the line) until
the spectrum appears to be phased; as with the zeroorder phase correction the
computer software usually makes it easy for us to do this by turning a knob
or pushing the mouse. In practice, to phase the spectrum correctly usually
requires some iteration of the zero and ﬁrstorder phase corrections.
The usual convention is to express the frequency dependent phase correc
tion as the value that the phase takes at the extreme edges of the spectrum.
So, for example, such a correction by 100
◦
means that the phase correction is
zero in the middle (at zero offset) and rises linearly to +100
◦
at on edge and
falls linearly to −100
◦
at the opposite edge.
For a pulse the phase error due to these offresonance effects for a peak
with offset is of the order of (t
p
), where t
p
is the length of the pulse. For
a carbon13 spectrum recorded at a Larmor frequency of 125 MHz, the max
imum offset is about 100 ppm which translates to 12500 Hz. Let us suppose
that the 90
◦
pulse width is 15 µs, then the phase error is
2π ×12500 ×15 ×10
−6
≈ 1.2 radians
which is about 68
◦
; note that in the calculation we had to convert the offset
from Hz to rad s
−1
by multiplying by 2π. So, we expect the frequency de
pendent phase error to vary from zero in the middle of the spectrum (where
the offset is zero) to 68
◦
at the edges; this is a signiﬁcant effect.
For reasons which we cannot go into here it turns out that the linear phase
4–8 Fourier transformation and data processing
correction is sometimes only a ﬁrst approximation to the actual correction
needed. Provided that the lines in the spectrum are sharp a linear correction
works very well, but for broad lines it is not so good. Attempting to use a
ﬁrstorder phase correction on such spectra often results in distortions of the
baseline.
4.4 Sensitivity enhancement
Inevitably when we record a FID we also record noise at the same time. Some
of the noise is contributed by the ampliﬁers and other electronics in the spec
trometer, but the major contributor is the thermal noise from the coil used
to detect the signal. Reducing the noise contributed by these two sources
is largely a technical matter which will not concern use here. NMR is not
a sensitive technique, so we need to take any steps we can to improve the
signaltonoise ratio in the spectrum. We will see that there are some manip
ulations we can perform on the FID which will give us some improvement in
the signaltonoise ratio (SNR).
By its very nature, the FID decays over time but in contrast the noise just
goes on and on. Therefore, if we carry on recording data for long after the FID
has decayed we will just measure noise and no signal. The resulting spectrum
will therefore have a poor signaltonoise ratio.
(a) (b) (c)
Fig. 4.9 Illustration of the effect of the time spent acquiring the FID on the signaltonoise ratio (SNR) in the
spectrum. In (a) the FID has decayed to next to nothing within the ﬁrst quarter of the time, but the noise
carries on unabated for the whole time. Shown in(b) is the effect of halving the time spent acquiring the
data; the SNR improves signiﬁcantly. In (c) we see that taking the ﬁrst quarter of the data gives a further
improvement in the SNR.
This point is illustrated in Fig. 4.9 where we see that by recording the
FID for long after it has decayed all we end up doing is recording more noise
and no signal. Just shortening the time spent recording the signal (called
the acquisition time) will improve the SNR since more or less all the signal
is contained in the early part of the FID. Of course, we must not shorten the
acquisition time too much or we will start to miss the FID, which would result
in a reduction in SNR.
Sensitivity enhancement
Looking at the FID we can see that at the start the signal is strongest. As
time progresses, the signal decays and so gets weaker but the noise remains at
the same level. The idea arises, therefore, that the early parts of the FID are
4.4 Sensitivity enhancement 4–9
“more important” as it is here where the signal is the strongest.
This effect can be exploited by deliberately multiplying the FID by a func
tion which starts at 1 and then steadily tails away to zero. The idea is that this
function will cut off the later parts of the FID where the signal is weakest, but
leave the early parts unaffected.
weighting function weighting function original FID
weighted FID weighted FID
(f) (g) (h)
(i) (j) (k)
(b) (a)
(d) (e)
(c)
Fig. 4.10 Illustration of how multiplying a FID by a decaying function (a weighting function) can improve
the SNR. The original FID is shown in (a) and the corresponding spectrum is (f). Multiplying the FID by a
weighting function (b) gives (c); Fourier transformation of (c) gives the spectrum(g). Note the improvement
in SNR of (g) compared to (f). Multiplying (a) by the more rapidly decaying weighting function (d) gives
(e); the corresponding spectrum is (h). The improvement in SNR is less marked. Spectra (f) – (h) are all
plotted on the same vertical scale so that the decrease in peak height can be seen. The same spectra
are plotted in (i) – (k) but this time normalized so that the peak height is the same; this shows most clearly
the improvement in the SNR.
A typical choice for this function – called a weighting function – is an
exponential:
W(t ) = exp(−R
LB
t ) (4.4)
where R
LB
is a rate constant which we are free to choose. Figure 4.10 illus
trates the effect on the SNR of different choices of this decay constant.
Spectrum (g) shows a large improvement in the SNR when compared to
spectrum (f) simply because the long tail of noise is suppressed. Using a more
rapidly decaying weighting function, (d), gives a further small improvement
in the SNR (h). It should not be forgotten that the weighting function also
acts on the signal, causing it to decay more quickly. As was explained in
section 4.2 a more rapidly decaying signal leads to a broader line. So, the use
4–10 Fourier transformation and data processing
of the weighting function will not only attenuate the noise it will also broaden
the lines; this is clear from Fig. 4.10.
Broadening the lines also reduces the peak height (remember that the in
tegral remains constant) – something that is again evident from Fig. 4.10.
Clearly, this reduction in peak height will reduce the SNR. Hence there is a
tradeoff we have to make: the more rapidly decaying the weighting function
the more the noise in the tail of the FID is attenuated thus reducing the noise
level in the spectrum. However, at the same time a more rapidly decaying
function will cause greater line broadening, and this will reduce the SNR.
It turns out that there is an optimum weighting function, called the matched
ﬁlter.
Matched ﬁlter
Suppose that the FID can be represented by the exponentially decaying func
tion introduced in Eq. 4.2:
S(t ) = S
0
exp(i t ) exp(−R
2
t ).
The line in the corresponding spectrum is of width (R
2
/π) Hz. Let us apply a
weighting function of the kind described by Eq. 4.4 i.e. a decaying exponen
tial:
W(t ) × S(t ) = exp(−R
LB
t )
S
0
exp(i t ) exp(−R
2
t )
.
The decay due to the weighting function on its own would give a linewidth
of (R
LB
/π) Hz. Combining the two terms describing the exponential decays
gives
W(t ) × S(t ) = S
0
exp(i t ) exp (−(R
LB
+ R
2
)t ) .
From this we see that the weighted FID will give a linewidth of
(R
LB
+ R
2
)/π.
In words, the linewidth in the spectrum is the sum of the linewidths in the
original spectrumand the additional linebroadening imposed by the weighting
function.
It is usual to specify the weighting function in terms of the extra line
broadening it will cause. So, a “linebroadening of 1 Hz” is a function which
will increase the linewidth in the spectrum by 1 Hz. For example if 5 Hz of
linebroadening is required then (R
LB
/π) = 5 giving R
LB
= 15.7 s
−1
.
It can be shown that the best SNR is obtained by applying a weight
ing function which matches the linewidth in the original spectrum – such a
weighting function is called a matched ﬁlter. So, for example, if the linewidth
is 2 Hz in the original spectrum, applying an additional line broadening of 2
Hz will give the optimum SNR.
We can see easily from this argument that if there is a range of linewidths
in the spectrum we cannot ﬁnd a value of the linebroadening which is the
optimum for all the peaks. Also, the extra line broadening caused by the
matched ﬁlter may not be acceptable on the grounds of the decrease in reso
lution it causes. Under these circumstances we may choose to use sufﬁcient
line broadening to cut off the excess noise in the tail of the FID, but still less
than the matched ﬁlter.
4.5 Resolution enhancement 4–11
4.5 Resolution enhancement
We noted in section 4.2 that the more rapidly the time domain signal decays
the broader the lines become. A weighting function designed to improve the
SNR inevitably leads to a broadening of the lines as such a function hastens
the decay of the signal. In this section we will consider the opposite case,
where the weighting function is designed to narrow the lines in the spectrum
and so increase the resolution.
The basic idea is simple. All we need to do is to multiply the FID by a
weighting function which increases with time, for example a rising exponen
tial:
W(t ) = exp(+R
RE
t ) R
RE
> 0.
This function starts at 1 when t = 0 and then rises indeﬁnitely.
×
× ×
=
=
(a)
(b) (e)
(f)
(g)
(h)
(i)
(j)
(a)
(c) (d)
Fig. 4.11 Illustration of the use of weighting functions to enhance the resolution in the spectrum. Note that
the scales of the plots have been altered to make the relevant features clear. See text for details.
The problem with multiplying the FID with such a function is that the
noise in the tail of the FID is ampliﬁed, thus making the SNR in the spectrum
very poor indeed. To get round this it is, after applying the positive expo
nential we multiply by a second decaying function to “clip” the noise at the
4–12 Fourier transformation and data processing
tail of the FID. Usually, the second function is chosen to be one that decays
relatively slowly over most of the FID and then quite rapidly at the end – a
common choice is the Gaussian function:
W(t ) = exp(−αt
2
),
where α is a parameter which sets the decay rate. The larger α, the faster the
decay rate.
Lorentzian
Gaussian
Fig. 4.12 Comparison of the
Lorentzian and Gaussian
lineshapes; the two peaks have
been adjusted so that their peak
heights and widths at half height
are equal. The Gaussian is a
more “compact” lineshape.
The whole process is illustrated in Fig. 4.11. The original FID, (a), con
tains a signiﬁcant amount of noise and has been recorded well beyond the
point where the signal decays into the noise. Fourier transformation of (a)
gives the spectrum (b). If (a) is multiplied by the rising exponential function
plotted in (c), the result is the FID (d); note how the decay of the signal has
been slowed, but the noise in the tail of the FID has been greatly magniﬁed.
Fourier transformation of (d) gives the spectrum (e); the resolution has clearly
been improved, but at the expense of a large reduction in the SNR.
Referring now to the bottom part of Fig. 4.11 we can see the effect of
introducing a Gaussian weighting function as well. The original FID (a) is
multiplied by the rising exponential (f) and the decaying Gaussian (g); this
gives the timedomain signal (h). Note that once again the signal decay has
been slowed, but the noise in the tail of the FID is not as large as it is in (d).
Fourier transformation of (h) gives the spectrum (i); the resolution has clearly
been improved when compared to (b), but without too great a loss of SNR.
Finally, plot (j) shows the product of the two weighting functions (f) and
(g). We can see clearly from this plot how the two functions combine together
to ﬁrst increase the timedomain function and then to attenuate it at longer
times. Careful choice of the parameters R
RE
and α are needed to obtain the
optimum result. Usually, a process of trial and error is adopted.
If we can set R
RE
to cancel exactly the original decay of the FID then the
result of this process is to generate a timedomain function which only has
a Gaussian decay. The resulting peak in the spectrum will have a Gaussian
lineshape, which is often considered to be superior to the Lorentzian as it
is narrower at the base; the two lineshapes are compared in Fig. 4.12. This
transformation to a Gaussian lineshape is often called the LorentztoGauss
transformation.
Parameters for the LorentztoGauss transformation
The combined weighting function for this transformation is
W(t ) = exp(R
RE
t ) exp(−αt
2
).
The usual approach is to specify R
RE
in terms of the linewidth that it would
create on its own if it were used to specify a decaying exponential. Recall that
in such a situation the linewidth, L, is given by R
RE
/π. Thus R
RE
= πL.
The weighting function can therefore be rewritten
W(t ) = exp(−πLt ) exp(−αt
2
).
For compatibility with the linebroadening role of a decaying exponential it
is usual to deﬁne the exponential weighting function as exp(−πLt ) so that
4.6 Other weighting functions 4–13
a positive value for L leads to line broadening and a negative value leads to
resolution enhancement.
From Fig. 4.11 (j) we can see that the effect of the rising exponential and
the Gaussian is to give an overall weighting function which has a maximum
in it. It is usual to deﬁne the Gaussian parameter α from the position of this
maximum. A little mathematics shows that this maximum occurs at time t
max
given by:
t
max
= −
Lπ
2α
(recall that L is negative) so that
α = −
Lπ
2t
max
.
We simply select a value of t
max
and use this to deﬁne the value of α. On some
spectrometers, t
max
is expressed as a fraction, f , of the acquisition time, t
acq
:
t
max
= f t
acq
. In this case
α = −
Lπ
2 f t
acq
.
4.6 Other weighting functions
Many other weighting functions have been used for sensitivity enhancement
and resolution enhancement. Perhaps the most popular are the sine bell are
variants on it, which are illustrated in Fig. 4.13.
t
acq
t
acq
/2 0
0 π/4 π/8 π/2 phase
Fig. 4.13 The top row shows sine bell and the bottom row shows sine bell squared weighting functions for
different choices of the phase parameter; see text for details.
The basic sine bell is just the ﬁrst part of a sin θ for θ = 0 to θ = π;
this is illustrated in the top lefthand plot of Fig. 4.13. In this form the func
tion will give resolution enhancement rather like the combination of a rising
exponential and a Gaussian function (compare Fig. 4.11 (j)). The weighting
function is chosen so that the sine bell ﬁts exactly across the acquisition time;
mathematically the required function is:
W(t ) = sin
πt
t
acq
.
4–14 Fourier transformation and data processing
The sine bell can be modiﬁed by shifting it the left, as is shown in
Fig. 4.13. The further the shift to the left the smaller the resolution enhance
ment effect will be, and in the limit that the shift is by π/2 or 90
◦
the function
is simply a decaying one and so will broaden the lines. The shift is usually
expressed in terms of a phase φ (in radians); the resulting weighting function
is:
W(t ) = sin
(π −φ)t
t
acq
+φ
.
Note that this deﬁnition of the function ensures that it goes to zero at t
acq
.
The shape of all of these weighting functions are altered subtly by squar
ing them to give the sine bell squared functions; these are also shown in
Fig. 4.13. The weighting function is then
W(t ) = sin
2
(π −φ)t
t
acq
+φ
.
Much of the popularity of these functions probably rests of the fact that
there is only one parameter to adjust, rather than two in the case of the
LorentztoGauss transformation.
4.7 Zero ﬁlling
Before being processed the FID must be converted into a digital form so that
it can be stored in computer memory. We will have more to say about this
in Chapter 5 but for now we will just note that in this process the signal is
sampled at regular intervals. The FID is therefore represented by a series
of data points. When the FID is Fourier transformed the spectrum is also
represented by a series of data points. So, although we plot the spectrum as a
smooth line, it is in fact a series of closely spaced points.
t
acq
t
acq
t
acq
(a) (b) (c)
Fig. 4.14 Illustration of the results of zero ﬁlling. The FIDs along the top row have been supplemented
with increasing numbers of zeroes and so contain more and more data points. Fourier transformation
preserves the number of data points so the line in the spectrum is represented by more points as zeroes
are added to the end of the FID. Note that the FID remains the same for all three cases; no extra data has
been acquired.
This is illustrated in Fig. 4.14 (a) which shows the FID and the correspond
ing spectrum; rather than joining up the points which make up the spectrum
we have just plotted the points. We can see that there are only a few data
points which deﬁne the line in the spectrum.
4.8 Truncation 4–15
If we take the original FID and add an equal number of zeroes to it, the
corresponding spectrum has double the number of points and so the line is
represented by more data points. This is illustrated in Fig. 4.14 (b). Adding a
set of zeroes equal to the number of data points is called “one zero ﬁlling”.
We can carry on with this zero ﬁlling process. For example, having added
one set of zeroes, we can add another to double the total number of data
points (“two zero ﬁllings”). This results in an even larger number of data
points deﬁning the line, as is shown in Fig. 4.14 (c).
Zero ﬁlling costs nothing in the sense that no extra data is required; it
is just a manipulation in the computer. Of course, it does not improve the
resolution as the measured signal remains the same, but the lines will be better
deﬁned in the spectrum. This is desirable, at least for aesthetic reasons if
nothing else!
It turns out that the Fourier transform algorithm used by computer pro
grams is most suited to a number of data points which is a power of 2. So,
for example, 2
14
= 16384 is a suitable number of data points to transform,
but 15000 is not. In practice, therefore, it is usual to zero ﬁll the time domain
data so that the total number of points is a power of 2; it is always an option,
of course, to zero ﬁll beyond this point.
4.8 Truncation
In conventional NMR it is virtually always possible to record the FID until it
has decayed almost to zero (or into the noise). However, in multidimensional
NMR this may not be the case, simply because of the restrictions on the
amount of data which can be recorded, particularly in the “indirect dimen
sion” (see Chapter X for further details). If we stop recording the signal be
fore it has fully decayed the FID is said to be “truncated”; this is illustrated in
Fig. 4.15.
Fig. 4.15 Illustration of how truncation leads to artefacts (called sinc wiggles) in the spectrum. The FID
on the left has been recorded for sufﬁcient time that it has decayed almost the zero; the corresponding
spectrum shows the expected lineshape. However, if data recording is stopped before the signal has fully
decayed the corresponding spectra show oscillations around the base of the peak.
4–16 Fourier transformation and data processing
As is shown clearly in the ﬁgure, a truncated FID leads to oscillations
around the base of the peak; these are usually called sinc wiggles or truncation
artefacts – the name arises as the peak shape is related to a sinc function. The
more severe the truncation, the larger the sinc wiggles. It is easy to show that
the separation of successive maxima in these wiggles is 1/t
acq
Hz.
Clearly these oscillations are undesirable as they may obscure nearby
weaker peaks. Assuming that it is not an option to increase the acquisition
time, the way forward is to apply a decaying weighting function to the FID so
as to force the signal to go to zero at the end. Unfortunately, this will have the
side effects of broadening the lines and reducing the SNR.
Highly truncated time domain signals are a feature of three and higher
dimensional NMR experiments. Much effort has therefore been put into ﬁnd
ing alternatives to the Fourier transform which will generate spectra without
these truncation artefacts. The popular methods are maximum entropy, linear
prediction and FDM. Each has its merits and drawbacks; they all need to be
applied with great care.
4.9 Exercises 4–17
4.9 Exercises
E 4–1
In a spectrum with just one line, the dispersion mode lineshape might be ac
ceptable – in fact we can think of reasons why it might even be desirable
(what might these be?). However, in a spectrum with many lines the disper
sion mode lineshape is very undesirable – why?
E 4–2
Suppose that we record a spectrum with the simple pulseacquire sequence
using a 90
◦
pulse applied along the x axis. The resulting FID is Fourier trans
formed and the spectrum is phased to give an absorption mode lineshape.
We then change the phase of the pulse from x to y, acquire an FID in the
same way and phase the spectrum using the same phase correction as above.
What lineshape would you expect to see in the spectrum; give the reasons for
your answer.
How would the spectrum be affected by: (a) applying the pulse about −x;
(b) changing the pulse ﬂip angle to 270
◦
about x?
E 4–3
The gyromagnetic ratio of phosphorus31 is 1.08 × 10
8
rad s
−1
T
−1
. This
nucleus shows a wide range of shifts, covering some 700 ppm.
Suppose that the transmitter is placed in the middle of the shift range and
that a 90
◦
pulse of width 20 µs is used to excite the spectrum. Estimate the
size of the phase correction which will be needed at the edges of the spectrum.
(Assume that the spectrometer has a B
0
ﬁeld strength of 9.4 T).
E 4–4
Why is it undesirable to continue to acquire the FID after the signal has de
cayed away?
How can weighting functions be used to improve the SNR of a spectrum?
In your answer described how the parameters of a suitable weighting function
can be chosen to optimize the SNR. Are there any disadvantages to the use of
such weighting functions?
E 4–5
Describe how weighting functions can be used to improve the resolution in
a spectrum. What sets the limit on the improvement that can be obtained in
practice? Is zero ﬁlling likely to improve the situation?
E 4–6
Explain why use of a sine bell weighting function shifted by 45
◦
may enhance
the resolution but use of a sine bell shifted by 90
◦
does not.
E 4–7
In a proton NMR spectrum the peak from TMS was found to show “wig
4–18 Fourier transformation and data processing
gles” characteristic of truncation of the FID. However, the other peaks in the
spectrum showed no such artefacts. Explain.
How can truncation artefacts be suppressed? Mention any difﬁculties with
your solution to the problem.
5 How the spectrometer works
NMR spectrometers have now become very complex instruments capable of
performing an almost limitless number of sophisticated experiments. How
ever, the really important parts of the spectrometer are not that complex to
understand in outline, and it is certainly helpful when using the spectrometer
to have some understanding of how it works.
Broken down to its simplest form, the spectrometer consists of the follow
ing components:
• An intense, homogeneous and stable magnetic ﬁeld
• A “probe” which enables the coils used to excite and detect the signal
to be placed close to the sample
• A highpower RF transmitter capable of delivering short pulses
• A sensitive receiver to amplify the NMR signals
• A digitizer to convert the NMR signals into a form which can be stored
in computer memory
• A “pulse programmer” to produce precisely times pulses and delays
• A computer to control everything and to process the data
We will consider each of these in turn.
5.1 The magnet
Modern NMR spectrometers use persistent superconducting magnets to gen
erate the B
0
ﬁeld. Basically such a magnet consists of a coil of wire through
which a current passes, thereby generating a magnetic ﬁeld. The wire is of a
special construction such that at low temperatures (less than 6 K, typically)
the resistance goes to zero – that is the wire is superconducting. Thus, once
the current is set running in the coil it will persist for ever, thereby generating a
magnetic ﬁeld without the need for further electrical power. Superconducting
magnets tend to be very stable and so are very useful for NMR.
To maintain the wire in its superconducting state the coil is immersed in
a bath of liquid helium. Surrounding this is usually a “heat shield” kept at
77 K by contact with a bath of liquid nitrogen; this reduces the amount of
(expensive) liquid helium which boils off due to heat ﬂowing in from the
surroundings. The whole assembly is constructed in a vacuum ﬂask so as
to further reduce the heat ﬂow. The cost of maintaining the magnetic ﬁeld
is basically the cost of liquid helium (rather expensive) and liquid nitrogen
(cheap).
Chapter 5 “How the spectrometer works” c James Keeler, 2002
5–2 How the spectrometer works
Of course, we do not want the sample to be at liquid helium temperatures,
so a room temperature region – accessible to the outside world – has to be
engineered as part of the design of the magnet. Usually this room temperature
zone takes the form of a vertical tube passing through the magnet (called the
bore tube of the magnet); the magnetic ﬁeld is in the direction of this tube.
Shims
The lines in NMR spectra are very narrow – linewidths of 1 Hz or less are
not uncommon – so the magnetic ﬁeld has to be very homogeneous, meaning
that it must not vary very much over space. The reason for this is easily
demonstrated by an example.
Consider a proton spectrum recorded at 500 MHz, which corresponds to
a magnetic ﬁeld of 11.75 T. Recall that the Larmor frequency is given by
ν
0
= −
1
2π
γ B
0
(5.1)
where γ is the gyromagnetic ratio (2.67 × 10
8
rad s
−1
T
−1
for protons). We
need to limit the variation in the magnetic ﬁeld across the sample so that the
corresponding variation in the Larmor frequency is much less that the width
of the line, say by a factor of 10.
With this condition, the maximumacceptable change in Larmor frequency
is 0.1 Hz and so using Eq. 5.1 we can compute the change in the magnetic ﬁeld
as 2π0.1/γ = 2.4 × 10
−9
T. Expressed as a fraction of the main magnetic
ﬁeld this is about 2 × 10
−10
. We can see that we need to have an extremely
homogeneous magnetic ﬁeld for work at this resolution.
On its own, no superconducting magnet can produce such a homogeneous
ﬁeld. What we have to do is to surround the sample with a set of shim coils,
each of which produces a tiny magnetic ﬁeld with a particular spatial proﬁle.
The current through each of these coils is adjusted until the magnetic ﬁeld has
the required homogeneity, something we can easily assess by recording the
spectrum of a sample which has a sharp line. Essentially how this works is
that the magnetic ﬁelds produced by the shims are cancelling out the small
residual inhomogeneities in the main magnetic ﬁeld.
Modern spectrometers might have up to 40 different shim coils, so ad
justing them is a very complex task. However, once set on installation it is
usually only necessary on a day to day basis to alter a few of the shims which
generate the simplest ﬁeld proﬁles.
The shims are labelled according to the ﬁeld proﬁles they generate. So, for
example, there are usually shims labelled x, y and z, which generate magnetic
ﬁelds varying the corresponding directions. The shim z
2
generates a ﬁeld
that varies quadratically along the z direction, which is the direction of B
0
.
There are more shims whose labels you will recognize as corresponding to the
names of the hydrogen atomic orbitals. This is no coincidence; the magnetic
ﬁeld proﬁles that the shims coils create are in fact the spherical harmonic
functions, which are the angular parts of the atomic orbitals.
5.2 The probe 5–3
5.2 The probe
The probe is a cylindrical metal tube which is inserted into the bore of the
magnet. The small coil used to both excite and detect the NMR signal (see
sections 3.2 and 3.4) is held in the top of this assembly in such a way that
the sample can come down from the top of the magnet and drop into the coil.
Various other pieces of electronics are contained in the probe, along with some
arrangements for heating or cooling the sample.
to
tx/rx
tune
match
Fig. 5.1 Schematic of the key
parts of the probe. The coil is
shown on the left (with the
sample tube in grey); a tuned
circuit is formed by a capacitor
(marked “tune”). The power
transfer to the transmitter and
receiver (tx. and rx.) is
optimized by adjusting the
capacitor marked “match”. Note
that the coil geometry as shown
is not suitable for a
superconducting magnet in
which the main ﬁeld is parallel
to the sample axis.
The key part of the probe is the small coil used to excite and detect the
magnetization. To optimize the sensitivity this coil needs to be as close as
possible to the sample, but of course the coil needs to be made in such a way
that the sample tube can drop down from the top of the magnet into the coil.
Extraordinary effort has been put into the optimization of the design of this
coil.
The coil forms part of a tuned circuit consisting of the coil and a capacitor.
The inductance of the coil and the capacitance of the capacitor are set such
that the tuned circuit they form is resonant at the Larmor frequency. That the
coil forms part of a tuned circuit is very important as it greatly increases the
detectable current in the coil.
Spectroscopists talk about “tuning the probe” which means adjusting the
capacitor until the tuned circuit is resonant at the Larmor frequency. Usually
we also need to “match the probe” which involves further adjustments de
signed to maximize the power transfer between the probe and the transmitter
and receiver; Fig. 5.1 shows a typical arrangement. The two adjustments tend
to interact rather, so tuning the probe can be a tricky business. To aid us, the
instrument manufacturers provide various indicators and displays so that the
tuning and matching can be optimized. We expect the tuning of the probe to
be particularly sensitive to changing solvent or to changing the concentration
of ions in the solvent.
5.3 The transmitter
The radiofrequency transmitter is the part of the spectrometer which generates
the pulses. We start with an RF source which produces a stable frequency
which can be set precisely. The reason why we need to be able to set the
frequency is that we might want to move the transmitter to different parts of
the spectrum, for example if we are doing experiments involving selective
excitation (section 3.11).
Usually a frequency synthesizer is used as the RF source. Such a device
has all the desirable properties outlined above and is also readily controlled
by a computer interface. It is also relatively easy to phase shift the output
from such a synthesizer, which is something we will need to do in order to
create phase shifted pulses.
As we only need the RF to be applied for a short time, the output of the
synthesizer has to be “gated” so as to create a pulse of RF energy. Such a gate
will be under computer control so that the length and timing of the pulse can
be controlled.
5–4 How the spectrometer works
to probe
synthesizer
gate
amplifier
attenuator
computer
control
Fig. 5.2 Typical arrangement of
the RF transmitter. The
synthesizer which is the source
of the RF, the attenuator and the
gate used to create the pulses
all under computer control.
The RF source is usually at a low level (a few mW) and so needs to be
boosted considerably before it will provide a useful B
1
ﬁeld when applied to
the probe; the complete arrangement is illustrated in Fig. 5.2. RF ampliﬁers
are readily available which will boost this small signal to a power of 100 W
or more. Clearly, the more power that is applied to the probe the more intense
the B
1
ﬁeld will become and so the shorter the 90
◦
pulse length. However,
there is a limit to the amount of power which can be applied because of the
high voltages which are generated in the probe.
When the RF power is applied to the tuned circuit of which the coil is part,
high voltages are generated across the tuning capacitor. Eventually, the volt
age will reach a point where it is sufﬁcient to ionize the air, thus generating
a discharge or arc (like a lightening bolt). Not only does this probe arcing
have the potential to destroy the coil and capacitor, but it also results in un
predictable and erratic B
1
ﬁelds. Usually the manufacturer states the power
level which is “safe” for a particular probe.
Power levels and “dB”
The spectrometer usually provides us with a way of altering the RF power
level and hence the strength of the B
1
ﬁeld. This is useful as we may wish to
set the B
1
ﬁeld strength to a particular level, for example for selective excita
tion.
The usual way of achieving this control is add an attenuator between the
RF source (the synthesizer) and the ampliﬁer. As its name implies, the atten
uator reduces the signal as it passes through.
The attenuation is normal expressed is decibels (abbreviated dB and pro
nounced “deebee”). If the input power is P
in
and the output power is P
out
the
attenuation in dB is
10 ×log
10
P
out
P
in
;
note that the logarithm is to the base 10, not the natural logarithm. The factor
of 10 is the “deci” part in the dB.
For example, if the output power is half the input power, i.e. P
in
= 2×P
out
,
the power ratio in dB is
10 ×log
10
P
out
P
in
= 10 ×log
10
1
2
= −3.0
So, halving the power corresponds to a change of −3.0 dB, the minus indi
cating that there is a power reduction i.e. an attenuation. An attenuator which
achieves this effect would be called “a 3 dB attenuator”.
Likewise, a power reduction by a factor of 4 corresponds to −6.0 dB. In
fact, because of the logarithmic relationship we can see that each 3 dB of
attenuation will halve the power. So, a 12 dB attenuator will reduce the power
by a factor of 16.
It turns out that the B
1
ﬁeld strength is proportional to the square root of
the power applied. The reason for this is that it is the current in the coil which
is responsible for the B
1
ﬁeld and, as we know from elementary electrical the
ory, power is equal to (resistance × current
2
). So, the current is proportional
to the square root of the power.
5.3 The transmitter 5–5
Therefore, in order to double the B
1
ﬁeld we need to double the current
and this means multiplying the power by a factor of four. Recall from the
above that a change in the power by a factor of 4 corresponds to 6 dB. Thus,
6 dB of attenuation causes the B
1
ﬁeld to fall to half its original value; 12 dB
would cause it to fall to one quarter of its initial value, and so on.
Usually the attenuator is under computer control and its value can be set
in dB. This is very helpful to us as we can determine the attenuation needed
for different B
1
ﬁeld strengths. Suppose that with a certain setting of the
attenuator we have determined the B
1
ﬁeld strength to be ω
init
1
/2π, where we
have expressed the ﬁeld strength in frequency units as described in section 3.4.
However, in our experiment we want the ﬁeld strength to be ω
new
1
/2π. The
ratio of the powers needed to achieve these two ﬁeld strengths is equal to the
square of the ratio of the ﬁeld strengths:
power ratio =
ω
new
1
/2π
ω
init
1
/2π
2
.
Expressed in dB this is
power ratio in dB = 10 log
10
ω
new
1
/2π
ω
init
1
/2π
2
= 20 log
10
ω
new
1
/2π
ω
init
1
/2π
.
This expression can be used to ﬁnd the correct setting for the attenuator.
As the duration of a pulse of a given ﬂip angle is inversely proportional to
ω
1
/2π the relationship can be expressed in terms of the initial and new pulse
widths, τ
init
and τ
new
:
power ratio in dB = 20 log
10
τ
init
τ
new
.
For example, suppose we have calibrated the pulse width for a 90
◦
pulse to
be 15 µs but what we actually wanted was a 90
◦
pulse of 25 µs. The required
attenuation would be:
power ratio in dB = 20 log
10
15
25
= −4.4dB.
We would therefore need to increase the attenuator setting by 4.4 dB.
Phase shifted pulses
In section 3.9 we saw that pulses could be applied along different axes in the
rotating frame, for example x and y. At ﬁrst sight you might think that to
achieve this we would need a coil oriented along the y axis to apply a pulse
along y. However, this is not the case; all we need to do is to shift the phase
of the RF by 90
◦
.
5–6 How the spectrometer works
x
x
y
y
φ
(b)
(a)
S
x
S
x
Fig. 5.3 Illustration of how a 90
◦
phase shift takes us from cosine
to sine modulation. In (a) the
vector starts out along x; as it
rotates in a positive sense the x
component, S
x
, varies as a
cosine wave, as is shown in the
graph on the right. In (b) the
vector starts on −y; the x
component now takes the form
of a sine wave, as is shown in
the graph. Situation (b) is
described as a phase shift, φ, of
−90
◦
compared to (a).
In section 3.4 we described the situation where RF power at frequency
ω
RF
is applied to the coil. In this situation the B
1
ﬁeld along the x axis can be
written as
2B
1
cos ω
RF
t.
If we phase shift the RF by 90
◦
the ﬁeld will be modulated by a sine function
rather than a cosine:
2B
1
sin ω
RF
t.
How this comes about is illustrated in Fig. 5.3.
The difference between the cosine and sine modulations is that in the for
mer the ﬁeld starts at a maximum and then decays, whereas in the latter it
starts are zero and increases.
The resulting sine modulated ﬁeld can still be thought of as arising from
two counterrotating ﬁelds, B
+
1
and B
−
1
; this is shown in Fig. 5.4. The crucial
thing is that as the ﬁeld is zero at time zero the two counterrotating vectors
start out opposed. As they evolve from this position the component along x
grows, as required. Recall from section 3.4 that it is the component B
−
1
which
interacts with the magnetization. When we move to a rotating frame at ω
RF
this component will be along the y axis (the position it has at time zero). This
is in contrast to the case in Fig. 3.8 where B
−
1
starts out on the x axis.
x
x
y
y
2B
1
B
1

B
1
+
time
f
i
e
l
d
a
l
o
n
g
x
B
1

B
1

B
1
+
B
1
+
B
1

B
1
+
Fig. 5.4 Illustration of the decomposition of a ﬁeld oscillating along the x axis according to a sine wave into
two counter rotating components. This ﬁgure should be compared to Fig. 3.8. Note that the component
B
−
1
starts out along the y axis at time zero in contrast to the case of Fig. 3.8 where it starts out along x.
So, if the B
1
ﬁeld is modulated by a cosine wave the result is a pulse about
x, whereas if it is modulated by a sine wave the result is a pulse y. We see
that by phase shifting the RF we can apply pulses about any axis; there is no
need to install more than one coil in the probe.
5.4 The receiver 5–7
5.4 The receiver
The NMR signal emanating from the probe is very small (of the order of
µV) but for modern electronics there is no problem in amplifying this signal
to a level where it can be digitized. These ampliﬁers need to be designed
so that they introduce a minimum of extra noise (they should be lownoise
ampliﬁers).
The ﬁrst of these ampliﬁers, called the preampliﬁer or preamp is usually
placed as close to the probe as possible (you will often see it resting by the
foot of the magnet). This is so that the weak signal is boosted before being
sent down a cable to the spectrometer console.
One additional problemwhich needs to be solved comes about because the
coil in the probe is used for both exciting the spins and detecting the signal.
This means that at one moment hundreds of Watts of RF power are being
applied and the next we are trying to detect a signal of a few µV. We need
to ensure that the highpower pulse does not end up in the sensitive receiver,
thereby destroying it!
This separation of the receiver and transmitter is achieved by a gadget
known as a diplexer. There are various different ways of constructing such a
device, but at the simplest level it is just a fast acting switch. When the pulse
is on the high power RF is routed to the probe and the receiver is protected by
disconnecting it or shorting it to ground. When the pulse is off the receiver is
connected to the probe and the transmitter is disconnected.
Some diplexers are passive in the sense that they require no external power
to achieve the required switching. Other designs use fast electronic switches
(rather like the gate in the transmitter) and these are under the command of
the pulse programmer so that the receiver or transmitter are connected to the
probe at the right times.
5.5 Digitizing the signal
The analogue to digital converter
A device known as an analogue to digital converter or ADC is used to convert
the NMR signal from a voltage to a binary number which can be stored in
computer memory. The ADC samples the signal at regular intervals, resulting
in a representation of the FID as data points.
The output from the ADC is just a number, and the largest number that
the ADC can output is set by the number of binary “bits” that the ADC uses.
For example with only three bits the output of the ADC could take just 8
values: the binary numbers 000, 001, 010, 011, 100, 101, 110 and 111. The
smallest number is 0 and the largest number is decimal 7 (the total number of
possibilities is 8, which is 2 raised to the power of the number of bits). Such
an ADC would be described as a “3 bit ADC”.
The waveform which the ADC is digitizing is varying continuously, but
output of the 3 bit ADC only has 8 levels so what it has to do is simply
output the level which is closest to the current input level; this is illustrated in
Fig. 5.5. The numbers that the ADC outputs are therefore an approximation
5–8 How the spectrometer works
to the actual waveform.
(b)
(a)
Fig. 5.5 Digitization of a
waveform using an ADC with 8
levels (3 bits). The output of the
ADC can only be one of the 8
levels, so the smoothly varying
waveform has to be
approximated by data points at
one of the 8 levels. The data
points, indicated by ﬁlled circles,
are therefore an approximation
to the true waveform.
The approximation can be improved by increasing the number of bits; this
gives more output levels. At present, ADC with between 16 and 32 bits are
commonly in use in NMR spectrometers. The move to higher numbers of bits
is limited by technical considerations.
(a) (b)
Fig. 5.6 Spectrum (a) is from an FID which has been digitized using a 6 bit ADC (i.e. 64 levels); the vertical
scale has been expanded 10 fold so that the digitization sidebands are clearly visible. Spectrum (b) is
from an FID which has been digitized using an 8 bit ADC (256 levels); the improvement over (a) is evident.
The main consequence of the approximation process which the ADC uses
is the generation of a forest of small sidebands – called digitization sidebands
– around the base of the peaks in the spectrum. Usually these are not a prob
lem as they are likely to be swamped by noise. However, if the spectrum
contains a very strong peak the sidebands from it can swamp a nearby weak
peak. Improving the resolution of the ADC, i.e. the number of bits it uses,
reduces the digitization sidebands, as is shown in Fig. 5.6.
(a)
(b)
Fig. 5.7 Illustration of the effect
of sampling rate on the
representation of the FID. In (a)
the data points (shown by dots)
are quite a good representation
of the signal (shown by the
continuous line). In (b) the data
points are too widely separated
and so are a very poor
representation of the signal.
Sampling rates
Given that the ADC is only going to sample the signal at regular intervals
the question arises as to how frequently it is necessary to sample the FID i.e.
what should the time interval between the data points be. Clearly, if the time
interval is too long we will miss crucial features of the waveform and so the
digitized points will be a poor representation of the signal. This is illustrated
in Fig. 5.7.
It turns out that if the interval between the points is the highest fre
quency which can be represented correctly, f
max
, is given by
f
max
=
1
2
;
f
max
is called the Nyquist frequency. Usually we think of this relationship the
other way round i.e. if we wish to represent correctly frequencies up to f
max
the sampling interval is given by:
=
1
2 f
max
.
This sampling interval is often called the dwell time. A signal at f
max
will
have two data points per cycle.
We will see shortly that we are able to distinguish positive and negative
frequencies, so a dwell time of means that the range of frequencies from
−f
max
to +f
max
are represented correctly.
5.6 Quadrature detection 5–9
A signal at greater than f
max
will still appear in the spectrum but not at
the correct frequency; such a peak is said to be folded. For example a peak
at ( f
max
+ F) will appear in the spectrum at (−f
max
+ F) as is illustrated in
Fig. 5.8.
(a)
(b)
f
max
+f
max
0
f
max
+f
max
0
Fig. 5.8 Illustration of the
concept of folding. In spectrum
(a) the peak (shown in grey) is
at a higher frequency than the
maximum set by the Nyquist
condition. In practice, such a
peak would appear in the
position shown in (b).
This Nyquist condition quickly brings us to a problem. A typical NMR
frequency is of the order of hundreds of MHz but there simply are no ADCs
available which work fast enough to digitize such a waveform with the kind of
accuracy (i.e. number of bits) we need for NMR. The solution to this problem
is to mix down the signal to a lower frequency, as is described in the next
section.
Mixing down to a lower frequency
Luckily for us, although the NMR frequency is quite high, the range of fre
quencies that any typical spectrum covers is rather small. For example, the 10
ppm of a proton spectrum recorded at 400 MHz covers just 4000 Hz. Such a
small range of frequencies is easily within the capability of suitable ADCs.
The procedure is to take the NMR signals and then subtract a frequency
from them all such that we are left with a much lower frequency. Typically,
the frequency we subtract is set somewhere in the middle of the spectrum;
this frequency is called the receiver reference frequency or just the receiver
frequency.
For example, suppose that the 10 ppm range of proton shifts runs from
400.000 MHz to 400.004 MHz. If we set the receiver reference frequency at
400.002 MHz and then subtract this from the NMR frequencies we end up
with a range −2000 to +2000 Hz.
The subtraction process is carried out by a device called a mixer. There
are various ways of actually making a mixer – some are “active” and contain
the usual transistors and so on, some are passive and are mainly constructed
from transformers and diodes. The details need not concern us – all we need
to know is that a mixer takes two signal inputs at frequencies f
1
and f
2
and
produces an output signal which contains the signals at the sum of the two
inputs, ( f
1
+ f
2
) and the difference ( f
1
− f
2
). The process is visualized in
Fig. 5.9.
Usually one of the inputs is generated locally, typically by a synthesizer.
This signal is often called the local oscillator. The other input is the NMR
signal the probe. In the spectrometer the local oscillator is set to a required
receiver frequency so that the difference frequency will be quite small. The
other output of the mixer is the sum frequency, which will be at the order of
twice the Larmor frequency. This high frequency signal is easily separated
from the required low frequency signal by a passing the output of the mixer
though a low pass ﬁlter. The ﬁltered signal is then passed to the ADC.
5.6 Quadrature detection
In the previous section we saw that the signal which is passed to the ADC
contains positive and negative frequencies – essentially this corresponds to
positive and negative offsets. The question therefore arises as to whether or
5–10 How the spectrometer works
not we are able to discriminate between frequencies which only differ in their
sign; the answer is yes, but only provided we use a method known as quadra
ture detection.
x
y
time
M
y
M
x
x
y
time
M
y
M
x
positive frequency
negative frequency
Fig. 5.10 Illustration of the x and y components from a vector precessing as a positive and negative offset.
The only difference between the two is in the y component.
Figure 5.10 illustrates the problem. The upper part shows a vector precess
ing with a positive offset frequency (in the rotating frame); the x component
is a cosine wave and the y component is a sine wave. The lower part shows
the evolution of a vector with a negative frequency. The x component remains
the same but the y component has changed sign when compared with evolu
tion at a positive frequency. We thus see that to distinguish between positive
and negative frequencies we need to know both the x and y components of
the magnetization.
It is tempting to think that it would be sufﬁcient just to know the y compo
nent as positive frequencies give a positive sine wave and negative ones give
a negative sine wave. In the corresponding spectrum this would mean that
peaks at positive offsets would be positive and those at negative offsets would
be negative. If there were a few well separated peaks in the spectrum such a
spectrum might be interpretable, but if there are many peaks in the spectrum
5.6 Quadrature detection 5–11
the result would be an uninterpretable mess.
f
1
f
1
f
2
f
1
+f
2
f
2
Fig. 5.9 A radiofrequency mixer
takes inputs at two different
frequencies and produces an
output which contains the sum
and difference of the two inputs.
How then can we detect both the x and y components of the magnetiza
tion? One idea is that we should have two coils, one along x and one along
y: these would certainly detect the x and y components. In practice, it turns
out to be very hard to achieve such an arrangement, partly because of the con
ﬁned space in the probe and partly because of the difﬁculties in making the
two coils electrically isolated from one another.
Luckily, though, it turns out to be easy to achieve the same effect by phase
shifting the receiver reference frequency; we describe how this works in the
next section.
The mixing process
In section 5.5 we described the use of a mixer to subtract a locally gener
ated frequency from that of the NMR signal. The mixer achieves this by
multiplying together the two input signals, and it is instructive to look at the
consequence of this in more detail.
Suppose that the signal coming from the probe can be written as A cos ω
0
t
where A gives the overall intensity and as usual ω
0
is the Larmor frequency.
Let us write the local oscillator signal as cos ω
rx.
t , where ω
rx
is the receiver
frequency. Multiplying these two together gives:
A cos ω
0
t ×cos ω
rx
t =
1
2
A [cos(ω
0
+ω
rx
)t +cos(ω
0
−ω
rx
)t ]
where we have used the well known formula cos A cos B =
1
2
(cos(A + B) +
cos(A − B)).
After the low pass ﬁlter only the term cos(ω
0
− ω
rx
)t will survive. We
recognize the difference frequency (ω
0
− ω
rx
) as the separation between the
receiver frequency and the Larmor frequency. If the receiver frequency is
made to be the same as the transmitter frequency (as is usually the case) this
difference frequency is the offset we identiﬁed in section 3.4.
In summary, the output of the mixer is at the offset frequency in the rotat
ing frame and takes the form of a cosine modulation; it is thus equivalent to
detecting the x component of the magnetization in the rotating frame.
Now consider the case where the local oscillator signal is −sin ω
rx
t ; as
we saw in section 5.3 such a signal can be created by a phase shift of 90
◦
. The
output of the mixer is now given by
A cos ω
0
t ×−sin ω
rx
t =
1
2
A [−sin(ω
0
+ω
rx
)t +sin(ω
0
−ω
rx
)t ]
where we have used the well known formula cos A sin B =
1
2
(sin(A + B) −
sin(A−B)). We now see that the difference frequency term is sine modulated
at the offset frequency; it is the equivalent of the y component in the rotating
frame. Thus, simply by shifting the phase of the local oscillator signal we can
detect both the x and y components in the rotating frame.
The complete arrangement for quadrature detection is outlined in
Fig. 5.11. The NMR signal from the probe is split into two and fed to two
separate mixers. The local oscillator signals fed to the two mixers are phase
shifted by 90
◦
relative to one another; as a result, the outputs of the two mixers
5–12 How the spectrometer works
ADC
ADC
90˚
phase
shift
from
probe
local oscillator
low pass filter
low pass filter
mixer
mixer
M
x
real part
of FID
M
y
imag. part
of FID
Fig. 5.11 Schematic of the arrangement for quadrature detection; see text for details.
are the equivalents of the x and y components in the rotating frame. These two
outputs are digitized separately and become the real and imaginary parts of a
complex time domain signal (as described in section 4.1). Such a quadrature
detection scheme is the norm for all modern NMR spectrometers.
5.7 Time and frequency
We are now in a position to tidy up a number of points of terminology relating
to the FID and spectrum. It is important to understand these in order to report
correctly the parameters used to record a spectrum.
The range of frequencies correctly represented in the spectrum (i.e. not
folded) is called the spectral width, f
SW
. As quadrature detection is used, the
frequency scale on the spectrum runs from−
1
2
f
SW
to +
1
2
f
SW
i.e. positive and
negative frequencies are discriminated.
For a given spectral width, the sampling interval or dwell time, , is com
puted from:
=
1
f
SW
which means that the Nyquist frequency is
1
2
f
SW
– the edges of the spectrum.
If we record the FID for a time t
acq
, called the acquisition time, then the
number of data points, N, is
N =
t
acq
.
Each data point is complex, consisting of a real and an imaginary part.
To be clear about the conditions under which an FID is recorded it is
important to quote the acquisition time and the spectral width; quoting the
number of data points on its own is meaningless.
5.8 The pulse programmer 5–13
5.8 The pulse programmer
The pulse programmer has become an immensely sophisticated piece of com
puter hardware, controlling as it does all of the functions of the spectrometer.
As the pulse programmer needs to produce very precisely timed events, often
in rapid succession, it is usual for it to run independently of the main com
puter. Typically, the pulse program is speciﬁed in the main computer and
then, when the experiment is started, the instructions are loaded into the pulse
programmer and then executed there.
The acquisition of data is usually handled by the pulse programmer, again
separately from the main computer.
5–14 How the spectrometer works
5.9 Exercises
E 5–1
You have been offered a superconducting magnet which claims to have a ho
mogeneity of “1 part in 10
8
”. Your intention is to use it to record phosphorus
31 spectra at Larmor frequency of 180 MHz, and you know that your typical
linewidths are likely to be of the order of 25 Hz. Is the magnet sufﬁciently
homogeneous to be of use?
E 5–2
A careful pulse calibration experiment determines that the 180
◦
pulse is 24.8
µs. How much attenuation, in dB, would have to be introduced into the trans
mitter in order to give a ﬁeld strength, (ω
1
/2π), of 2 kHz?
E 5–3
A spectrometer is equipped with a transmitter capable of generating a maxi
mum of 100 W of RF power at the frequency of carbon13. Using this trans
mitter at full power the 90
◦
pulse width is found to be 20µs. What power
would be needed to reduce the 90
◦
pulse width to 7.5 µs? Would you have
any reservations about using this amount of power?
E 5–4
Explain what is meant by “a two bit ADC” and draw a diagram to illustrate
the outcome of such a ADC being used to digitize a sine wave.
Why is it generally desirable to improve the number of bits that the ADC
uses?
E 5–5
A spectrometer operates at 800 MHz for proton and it is desired to cover a
shift range of 15 ppm. Assuming that the receiver frequency is placed in the
middle of this range, what spectral width would be needed and what would
the sampling interval (dwell time) have to be?
If we recorded a FID for 2 s with the spectral width set as you have deter
mined, how many data points will have been collected?
E 5–6
Suppose that the spectral width is set to 100 Hz, but that a peak is present
whose offset from the receiver is +60 Hz. Where will the peak appear in the
spectrum? Can you explain why?
6–1
6 Product Operators
†
The vector model, introduced in Chapter 3, is very useful for describing basic
NMR experiments but unfortunately is not applicable to coupled spin systems.
When it comes to twodimensional NMR many of the experiments are only of
interest in coupled spin systems, so we really must have some way of
describing the behaviour of such systems under multiplepulse experiments.
The tools we need are provided by quantum mechanics, specifically in the
form of density matrix theory which is the best way to formulate quantum
mechanics for NMR. However, we do not want to get involved in a great deal
of complex quantum mechanics! Luckily, there is a way of proceeding which
we can use without a deep knowledge of quantum mechanics: this is the product
operator formalism.
The product operator formalism is a complete and rigorous quantum
mechanical description of NMR experiments and is well suited to calculating the
outcome of modern multiplepulse experiments. One particularly appealing
feature is the fact that the operators have a clear physical meaning and that the
effects of pulses and delays can be thought of as geometrical rotations, much in
the same way as we did for the vector model in Chapter 3.
6.1 A quick review of quantum mechanics
In this section we will review a few key concepts before moving on to a
description of the product operator formalism.
In quantum mechanics, two mathematical objects – wavefunctions and
operators – are of central importance. The wavefunction describes the system of
interest (such as a spin or an electron) completely; if the wavefunction is known
it is possible to calculate all the properties of the system. The simplest example
of this that is frequently encountered is when considering the wavefunctions
which describe electrons in atoms (atomic orbitals) or molecules (molecular
orbitals). One often used interpretation of such electronic wavefunctions is to
say that the square of the wavefunction gives the probability of finding the
electron at that point.
Wavefunctions are simply mathematical functions of position, time etc. For
example, the 1s electron in a hydrogen atom is described by the function
exp(–ar), where r is the distance from the nucleus and a is a constant.
In quantum mechanics, operators represent "observable quantities" such as
position, momentum and energy; each observable has an operator associated
with it.
Operators "operate on" functions to give new functions, hence their name
operator × function = (new function)
An example of an operator is d dx ( ); in words this operator says "differentiate
†
Chapter 6 "Product Operators" © James Keeler, 1998 & 2002
6–2
with respect to x". Its effect on the function sin x is
d
dx
x x sin cos ( ) =
the "new function" is cos x. Operators can also be simple functions, so for
example the operator x
2
just means "multiply by x
2
".
6.1.1 Spin operators
A mass going round a circular path (an orbit) possesses angular momentum; it
turns out that this is a vector quantity which points in a direction perpendicular to
the plane of the rotation. The x, y and zcomponents of this vector can be
specified, and these are the angular momenta in the x, y and zdirections. In
quantum mechanics, there are operators which represent these three components
of the angular momentum.
Nuclear spins also have angular momentum associated with them – called
spin angular momentum. The three components of this spin angular
momentum (along x, y and z) are represented by the operators I I I
x y z
, and .
6.1.2 Hamiltonians
The Hamiltonian, H, is the special name given to the operator for the energy of
the system. This operator is exceptionally important as its eigenvalues and
eigenfunctions are the "energy levels" of the system, and it is transitions between
these energy levels which are detected in spectroscopy. To understand the
spectrum, therefore, it is necessary to have a knowledge of the energy levels and
this in turn requires a knowledge of the Hamiltonian operator.
In NMR, the Hamiltonian is seen as having a more subtle effect than simply
determining the energy levels. This comes about because the Hamiltonian also
affects how the spin system evolves in time. By altering the Hamiltonian the
time evolution of the spins can be manipulated and it is precisely this that lies at
the heart of multiplepulse NMR.
The precise mathematical form of the Hamiltonian is found by first writing
down an expression for the energy of the system using classical mechanics and
then "translating" this into quantum mechanical form according to a set of rules.
In this chapter the form of the relevant Hamiltonians will simply be stated rather
than derived.
In NMR the Hamiltonian changes depending on the experimental situation.
There is one Hamiltonian for the spin or spins in the presence of the applied
magnetic field, but this Hamiltonian changes when a radiofrequency pulse is
applied.
6.2 Operators for one spin
6.2.1 Operators
In quantum mechanics operators represent observable quantities, such an
energy, angular momentum and magnetization. For a single spinhalf, the x y
p
A mass going round a circular
path possesses angular
moment, represented by a
vector which points
perpendicular to the plane of
rotation.
6–3
and zcomponents of the magnetization are represented by the spin angular
momentum operators I
x
, I
y
and I
z
respectively. Thus at any time the state of the
spin system, in quantum mechanics the density operator, σ, can be represented
as a sum of different amounts of these three operators
σ t a t I b t I c t I
x y z
( ) = ( ) + ( ) + ( )
The amounts of the three operators will vary with time during pulses and delays.
This expression of the density operator as a combination of the spin angular
momentum operators is exactly analogous to specifying the three components of
a magnetization vector.
At equilibrium the density operator is proportional to I
z
(there is only z
magnetization present). The constant of proportionality is usually unimportant,
so it is usual to write σ
eq
= I
z
6.1.2 Hamiltonians for pulses and delays
In order to work out how the density operator varies with time we need to know
the Hamiltonian (which is also an operator) which is acting during that time.
The free precession Hamiltonian (i.e. that for a delay), H
free
, is
H I
z free
= Ω
In the vector model free precession involves a rotation at frequency Ω about the
zaxis; in the quantum mechanical picture the Hamiltonian involves the zangular
momentum operator, I
z
– there is a direct correspondence.
The Hamiltonian for a pulse about the xaxis, H
pulse
, is
H I
x x pulse,
= ω
1
and for a pulse about the yaxis it is
H I
y y pulse,
= ω
1
Again there is a clear connection to the vector model where pulses result in
rotations about the x or yaxes.
6.1.3 Equation of motion
The density operator at time t, σ(t), is computed from that at time 0, σ(0) , using
the following relationship
σ σ t iHt iHt ( ) = − ( ) ( ) ( ) exp exp 0
where H is the relevant hamiltonian. If H and σ are expressed in terms of the
angular momentum operators if turns out that this equation can be solved easily
with the aid of a few rules.
Suppose that an xpulse, of duration t
p
, is applied to equilibrium
magnetization. In this situation H = ω
1
I
x
and σ(0) = I
z
; the equation to be solved
is
σ ω ω t i t I I i t I
x z x p p p
( )
= −
( ) ( )
exp exp
1 1
Such equations involving angular momentum operators are common in
6–4
quantum mechanics and the solution to them are already all know. The identity
required here to solve this equation is
exp exp cos sin − ( ) ( ) ≡ − i I I i I I I
x z x z y
β β β β [6.1]
This is interpreted as a rotation of I
z
by an angle β about the xaxis. By putting
β= ω
1
t
p
this identity can be used to solve Eqn. [6.1]
σ ω ω t t I t I
z y p p p
( )
= − cos sin
1 1
The result is exactly as expected from the vector model: a pulse about the xaxis
rotates zmagnetization towards the –yaxis, with a sinusoidal dependence on the
flip angle, β.
6.1.4 Standard rotations
Given that there are only three operators, there are a limited number of identities
of the type of Eqn. [6.1]. They all have the same form
exp exp
cos sin
− ( ) ( )
≡ +
i I i I
a a
θ θ
θ θ
{old operator}
{old operator} {new operator}
where {old operator}, {new operator} and I
a
are determined from the three
possible angular momentum operators according to the following diagrams; the
label in the centre indicates which axis the rotation is about
x
y
–x
–y
z
z
–y
–z
y
x
z
x
–z
–x
y
I II III
Angle of rotation = Ωt for offsets and ω
1
t
p
for pulses
First example: find the result of rotating the operator I
y
by θ about the xaxis,
that is
exp exp − ( ) ( ) i I I i I
x y x
θ θ
For rotations about x the middle diagram II is required. The diagram shows that
I
y
(the "old operator") is rotated to I
z
(the "new operator"). The required identity
is therefore
exp exp cos sin − ( ) ( ) ≡ + i I I i I I I
x y x y z
θ θ θ θ
Second example: find the result of
exp – exp −
( ) { } ( )
i I I i I
y z y
θ θ
This is a rotation about y, so diagram III is required. The diagram shows that –I
z
(the "old operator") is rotated to –I
x
(the "new operator"). The required identity
is therefore
6–5
exp exp cos sin –
cos sin
−
( )
− { } ( )
≡ − { } +
{ }
≡ − −
i I I i I I I
I I
y z y z x
z x
θ θ θ θ
θ θ
Finally, note that a rotation of an operator about its own axis has no effect e.g. a
rotation of I
x
about x leaves I
x
unaltered.
6.1.5 Shorthand notation
To save writing, the arrow notation is often used. In this, the term Ht is written
over an arrow which connects the old and new density operators. So, for
example, the following
σ ω σ ω t i t I i t I
x x p p p
( )
= −
( )
( )
( )
exp exp
1 1
0
is written
σ σ
ω
0
1
( ) →
( )
t I
x
t
p
p
For the case where σ(0) = I
z
I t I t I
z
t I
z y
x
ω
ω ω
1
1 1
p
p p
→ cos – sin
6.1.6 Example calculation: spin echo
90 180 °( ) °( ) x
a b
x
e f
delay delay acquire τ τ
At a the density operator is –I
y
. The transformation from a to b is free
precession, for which the Hamiltonian is ΩI
z
; the delay τ therefore corresponds
to a rotation about the zaxis at frequency Ω. In the shorthand notation this is
− → ( ) I b
y
I
z
Ωτ
σ
To solve this diagram I above is needed with the angle = Ωτ; the "new operator"
is I
x
− → − + I I I
y
I
y x
z
Ω
Ω Ω
τ
τ τ cos sin
In words this says that the magnetization precesses from –y towards +x.
The pulse about x has the Hamiltonian ω
1
I
x
; the pulse therefore corresponds
to a rotation about x for a time t
p
such that the angle, ω
1
t
p
, is π radians. In the
shorthand notation
− + → ( ) cos sin Ω Ω τ τ σ
ω
I I e
y x
t I
x 1 p
[6.2]
Each term on the left is dealt with separately. The first term is a rotation of y
about x; the relevant diagram is thus II
− → − − cos cos cos cos sin Ω Ω Ω τ τ ω τ ω
ω
I t I t I
y
t I
y z
x 1
1 1
p
p p
However, the flip angle of the pulse, ω
1
t
p
, is π so the second term on the right is
zero and the first term just changes sign (cos π = –1); overall the result is
− → cos cos Ω Ω τ τ
π
I I
y
I
y
x
The second term on the left of Eqn. [6.2] is easy to handle as it is unaffected by
6–6
a rotation about x. Overall, the effect of the 180° pulse is then
− + → + cos sin cos sin Ω Ω Ω Ω τ τ τ τ
π
I I I I
y x
I
y x
x
[6.3]
As was shown using the vector model, the ycomponent just changes sign. The
next stage is the evolution of the offset for time τ. Again, each term on the right
of Eqn. [6.3] is considered separately
cos cos cos sin cos
sin cos sin sin sin
Ω Ω Ω Ω Ω
Ω Ω Ω Ω Ω
Ω
Ω
τ τ τ τ τ
τ τ τ τ τ
τ
τ
I I I
I I I
y
I
y x
x
I
x y
z
z
→ −
→ +
Collecting together the terms in I
x
and I
y
the final result is
cos cos sin sin cos sin sin cos Ω Ω Ω Ω Ω Ω Ω Ω τ τ τ τ τ τ τ τ + ( ) + − ( ) I I
y x
The bracket multiplying I
x
is zero and the bracket multiplying I
y
is =1 because of
the identity cos sin
2 2
1 θ θ + = . Thus the overall result of the spin echo sequence
can be summarised
I I
z
x x
y
90 180 ° − ° − −
→
( )– ( ) τ τ
In words, the outcome is independent of the offset, Ω, and the delay τ, even
though there is evolution during the delays. The offset is said to be refocused by
the spin echo. This is exactly the result we found in section 3.8.
In general the sequence
– τ – 180°(x) – τ – [6.4]
refocuses any evolution due to offsets; this is a very useful feature which is
much used in multiplepulse NMR experiments.
One further point is that as far as the offset is concerned the spin echo
sequence of Eqn. [6.4] is just equivalent to 180°(x).
6.3 Operators for two spins
The product operator approach comes into its own when coupled spin systems
are considered; such systems cannot be treated by the vector model. However,
product operators provide a clean and simple description of the important
phenomena of coherence transfer and multiple quantum coherence.
6.1.1 Product operators for two spins
For a single spin the three operators needed for a complete description are I
x
, I
y
and I
z
. For two spins, three such operators are needed for each spin; an
additional subscript, 1 or 2, indicates which spin they refer to.
spin 1: spin 2 : I I I I I I
x y z x y z 1 1 1 2 2 2
I
1z
represents zmagnetization of spin 1, and I
2z
likewise for spin 2. I
1x
represents xmagnetization on spin 1. As spin 1 and 2 are coupled, the
spectrum consists of two doublets and the operator I
1x
can be further identified
with the two lines of the spin1 doublet. In the language of product operators I
1x
is said to represent inphase magnetization of spin 1; the description inphase
Ω
2
Ω
1
J
12
J
12
The spectrum from two coupled
spins, with offsets Ω
1
and Ω
2
(rad s
–1
) and mutual coupling
J
12
(Hz).
6–7
means that the two lines of the spin 1 doublet have the same sign and lineshape.
Following on in the same way I
2x
represents inphase magnetization on spin
2. I
1y
and I
2y
also represent inphase magnetization on spins 1 and 2,
respectively, but this magnetization is aligned along y and so will give rise to a
different lineshape. Arbitrarily, an absorption mode lineshape will be assigned
to magnetization aligned along x and a dispersion mode lineshape to
magnetization along y.
Ω
1
Ω
2
Ω
1
Ω
2
I
1x
I
1y
I
2x
I
2y
There are four additional operators which represent antiphase magnetization:
2I
1x
I
2z
, 2I
1y
I
2z
, 2I
1z
I
2x
, 2I
1z
I
2y
(the factors of 2 are needed for normalization
purposes). The operator 2I
1x
I
2z
is described as magnetization on spin 1 which is
antiphase with respect to the coupling to spin 2.
Ω
1
Ω
2
Ω
1
Ω
2
2I
1y
I
2z
2I
1z
I
2y
2I
1z
I
2x
2I
1x
I
2z
Note that the two lines of the spin1 multiplet are associated with different spin
states of spin2, and that in an antiphase multiplet these two lines have different
signs. Antiphase terms are thus sensitive to the spin states of the coupled spins.
There are four remaining product operators which contain two transverse (i.e.
x or yoperators) terms and correspond to multiplequantum coherences; they
are not observable
multiple quantum: 2 2 2 2 I I I I I I I I
x y y x x x y y 1 2 1 2 1 2 1 2
Finally there is the term 2I
1z
I
2z
which is also not observable and corresponds to a
particular kind of nonequilibrium population distribution.
6.1.2 Evolution under offsets and pulses
The operators for two spins evolve under offsets and pulses in the same way as
do those for a single spin. The rotations have to be applied separately to each
spin and it must be remembered that rotations of spin 1 do not affect spin 2, and
vice versa.
For example, consider I
1x
evolving under the offset of spin 1 and spin 2. The
relevant Hamiltonian is
H I I
z z free
= + Ω Ω
1 1 2 2
absorption dispersion
The absorption and dispersion
lineshapes. The absorption
lineshape is a maximum on
resonance, whereas the
dispersion goes through zero at
this point. The "cartoon" forms
of the lineshapes are shown in
the lower part of the diagram.
Ω
1
spin state
of spin 2
α β
The two lines of the spin1
doublet can be associated with
different spin states of spin 2.
6–8
where Ω
1
and Ω
2
are the offsets of spin 1 and spin 2 respectively. Evolution
under this Hamiltonian can be considered by applying the two terms sequentially
(the order is immaterial)
I
I
I
x
H t
x
tI tI
x
tI tI
z z
z z
1
1
1
1 1 2 2
1 1 2 2
free
→
→
→ →
+ Ω Ω
Ω Ω
The first "arrow" is a rotation about z
I t I t I
x
tI
x y
tI
z z
1 1 1 1 1
1 1 2 2
Ω Ω
Ω Ω → + → cos sin
The second arrow leaves the intermediate state unaltered as spin2 operators
have not effect on spin1 operators. Overall, therefore
I t I t I
x
tI tI
x y
z z
1 1 1 1 1
1 1 2 2
Ω Ω
Ω Ω
+
→ + cos sin
A second example is the term 2I
1x
I
2z
evolving under a 90° pulse about the y
axis applied to both spins. The relevant Hamiltonian is
H I I
y y
= + ω ω
1 1 1 2
The evolution can be separated into two successive rotations
2
1 2
1 1 1 2
I I
x z
tI tI
y y
ω ω
→ →
The first arrow affects only the spin1 operators; a 90° rotation of I
1x
about y
gives – I
1z
(remembering that ω
1
t = π/2 for a 90° pulse)
2 2 2
2 2
1 2 1 1 2 1 1 2
1 2
2
1 2
2
1 1 1 2
1 2
I I t I I t I I
I I I I
x z
tI
x z z z
tI
x z
I
z z
I
y y
y y
ω ω
π π
ω ω → − →
→ − →
cos sin
The second arrow only affects the spin 2 operators; a 90° rotation of z about y
takes it to x
2 2 2
1 2
2
1 2
2
1 2
1 2
I I I I I I
x z
I
z z
I
z x
y y
π π
→ − → −
The overall result is that antiphase magnetization of spin 1 has been transferred
into antiphase magnetization of spin 2. Such a process is called coherence
transfer and is exceptionally important in multiplepulse NMR.
6.1.3 Evolution under coupling
The new feature which arises when considering two spins is the effect of
coupling between them. The Hamiltonian representing this coupling is itself a
product of two operators:
H J I I
z z J
= 2
12 1 2
π
where J
12
is the coupling in Hz.
Evolution under coupling causes the interconversion of inphase and anti
phase magnetization according to the following diagrams
6–9
IV V
x
yz
–x
–yz zz
y
–xz
–y
xz zz
angle = πJt
For example, inphase magnetization along x becomes antiphase along y
according to the diagram d
I J t I J t I I
x
J t I I
x y z
z z
1
2
12 1 12 1 2
12 1 2
2
π
π π → + cos sin
note that the angle is πJ
12
t i.e. half the angle for the other rotations, I–III.
Antiphase magnetization along x becomes inphase magnetization along y;
using diagram V:
2 2
1 2
2
12 1 2 12 1
12 1 2
I I J t I I J t I
x z
J t I I
x z y
z z
π
π π → + cos sin
The diagrams apply equally well to spin2; for example
− → + 2 2
1 2
2
12 1 2 12 2
12 1 2
I I J t I I J t I
z y
J t I I
z y x
z z
π
π π – cos sin
Complete interconversion of inphase and antiphase magnetization requires a
delay such that π π J t
12
2 = i.e. a delay of 1/(2J
12
). A delay of 1/J
12
causes in
phase magnetization to change its sign:
I I I I I
x
J t I I t J
y z y
J t I I t J
y
z z z z
1
2 1 2
1 2 2
2 1
2
12 1 2 12 12 1 2 12
2
π π = =
→ → −
6.4 Spin echoes
It was shown in section 6.2.6 that the offset is refocused in a spin echo. In this
section it will be shown that the evolution of the scalar coupling is not
necessarily refocused.
6.4.1 Spin echoes in homonuclear spin system
In this kind of spin echo the 180° pulse affects both spins i.e. it is a nonselective
pulse:
– τ – 180°(x, to spin 1 and spin 2) – τ –
At the start of the sequence it will be assumed that only inphase xmagnetization
on spin 1 is present: I
1x
. In fact the starting state is not important to the overall
effect of the spin echo, so this choice is arbitrary.
It was shown in section 6.2.6 that the spin echo applied to one spin refocuses
the offset; this conclusion is not altered by the presence of a coupling so the
offset will be ignored in the present calculation. This greatly simplifies things.
For the first delay τ only the effect of evolution under coupling need be
considered therefore:
I J I J I I
x
J I I
x y z
z z
1
2
12 1 12 1 2
12 1 2
2
π τ
π τ π τ → + cos sin
The 180° pulse affects both spins, and this can be calculated by applying the
6–10
180° rotation to each in succession
cos sin π τ π τ
π π
J I J I I
x y z
I I
x x
12 1 12 1 2
2
1 2
+ → →
where it has already been written in that ω
1
t
p
= π, for a 180° pulse. The 180°
rotation about x for spin 1 has no effect on the operator I
1x
and I
2z
, and it simply
reverses the sign of the operator I
1y
cos sin cos sin π τ π τ π τ π τ
π π
J I J I I J I J I I
x y z
I
x y z
I
x x
12 1 12 1 2 12 1 12 1 2
2 2
1 2
+ → − →
The 180° rotation about x for spin 2 has no effect on the operators I
1x
and I
1y
, but
simply reverses the sign of the operator I
2z
. The final result is thus
cos sin cos sin
cos sin
π τ π τ π τ π τ
π τ π τ
π
π
J I J I I J I J I I
J I J I I
x y z
I
x y z
I
x y z
x
x
12 1 12 1 2 12 1 12 1 2
12 1 12 1 2
2 2
2
1
2
+ → −
→ +
Nothing has happened; the 180° pulse has left the operators unaffected! So, for
the purposes of the calculation it is permissible to ignore the 180° pulse and
simply allow the coupling to evolve for 2τ. The final result can therefore just be
written down:
I J I J I I
x
x
x y z 1
180
12 1 12 1 2
2 2 2
τ τ
π τ π τ
− ° −
→ +
( )
cos sin
From this it is easy to see that complete conversion to antiphase magnetization
requires 2πJ
12
τ = π/2 i.e. τ = 1/(4 J
12
).
The calculation is not quite as simple if the initial state is chosen as I
1y
, but the
final result is just the same – the coupling evolves for 2τ:
I J I J I I
y
x
y x z 1
180
12 1 12 1 2
2 2 2
τ τ
π τ π τ
− ° −
→ − +
( )
cos sin
In fact, the general result is that the sequence
– τ – 180°(x, to spin 1 and spin 2) – τ –
is equivalent to the sequence
– 2τ – 180°(x, to spin 1 and spin 2)
in which the offset is ignored and coupling is allowed to act for time 2τ.
6.1.2 Interconverting inphase and antiphase states
So far, spin echoes have been demonstrated as being useful for generating anti
phase terms, independent of offsets. For example, the sequence
90°(x) – 1/(4J
12
) – 180°(x) – 1/(4J
12
) –
generates pure antiphase magnetization.
Equally useful is the sequence
– 1/(4J
12
) – 180°(x) – 1/(4J
12
) –
which will convert pure antiphase magnetization, such as 2I
1x
I
2z
into inphase
magnetization, I
1y
.
6–11
6.1.3 Spin echoes in heteronuclear spin systems
If spin 1 and spin 2 are different nuclear species, such as
13
C and
1
H, it is
possible to choose to apply the 180° pulse to either or both spins; the outcome of
the sequence depends on the pattern of 180° pulses.
Sequence a has already been analysed: the result is that the offset is refocused
but that the coupling evolves for time 2τ. Sequence b still refocuses the offset of
spin 1, but it turns out that the coupling is also refocused. Sequence c refocuses
the coupling but leaves the evolution of the offset unaffected.
Sequence b
It will be assumed that the offset is refocused, and attention will therefore be
restricted to the effect of the coupling
I J I J I I
x
J I I
x y z
z z
1
2
12 1 12 1 2
12 1 2
2
π τ
π τ π τ → + cos sin
The 180°(x) pulse is only applied to spin 1
cos sin cos sin π τ π τ π τ π τ
π
J I J I I J I J I I
x y z
I
x y z
x
12 1 12 1 2 12 1 12 1 2
2 2
1
+ → − [6.5]
The two terms on the right each evolve under the coupling during the second
delay:
cos
cos cos sin cos
sin
cos sin sin sin
π τ
π τ π τ π τ π τ
π τ
π τ π τ π τ π τ
π τ
π τ
J I
J J I J J I I
J I I
J J I I J J I
x
J I I
x y z
y z
J I I
y z
z z
z z
12 1
2
12 12 1 12 12 1 2
12 1 2
2
12 12 1 2 12 12 1
12 1 2
12 1 2
2
2
2
→
+
− →
− +
xx
Collecting the terms together and noting that cos sin
2 2
1 θ θ + = the final result is
just I
1x
. In words, the effect of the coupling has been refocused.
Sequence c
As there is no 180° pulse applied to spin 1, the offset of spin 1 is not refocused,
but continues to evolve for time 2τ. The evolution of the coupling is easy to
calculate:
I J I J I I
x
J I I
x y z
z z
1
2
12 1 12 1 2
12 1 2
2
π τ
π τ π τ → + cos sin
This time the 180°(x) pulse is applied to spin 2
cos sin cos sin π τ π τ π τ π τ
π
J I J I I J I J I I
x y z
I
x y z
x
12 1 12 1 2 12 1 12 1 2
2 2
2
+ → −
The results is exactly as for sequence b (Eqn. [6.5]), so the final result is the
same i.e. the coupling is refocused.
Summary
In heteronuclear systems it is possible to choose whether or not to allow the
offset and the coupling to evolve; this gives great freedom in generating and
manipulating antiphase states which play a key role in multiple pulse NMR
experiments.
spin 1
spin 2
a
spin 1
spin 2
b
spin 1
spin 2
c
τ τ
τ τ
τ τ
Three different spin echo
sequences that can be applied
to heteronuclear spin systems.
The open rectangles represent
180° pulses.
6–12
6.5 Multiple quantum terms
6.5.1 Coherence order
In NMR the directly observable quantity is the transverse magnetization, which
in product operators is represented by terms such as I
1x
and 2I
1z
I
2y
. Such terms
are examples of single quantum coherences, or more generally coherences with
order, p = ±1. Other product operators can also be classified according to
coherence order e.g. 2I
1z
I
2z
has p = 0 and 2I
1x
I
2y
has both p = 0 (zeroquantum
coherence) and ±2 (double quantum coherence). Only single quantum
coherences are observable.
In heteronuclear systems it is sometimes useful to classify operators
according to their coherence orders with respect to each spin. So, for example,
2I
1z
I
2y
has p = 0 for spin 1 and p = ±1 for spin 2.
6.5.2 Raising and lowering operators
The classification of operators according to coherence order is best carried out
be reexpressing the Cartesian operators I
x
and I
y
in terms of the raising and
lowering operators, I
+
and I
–
, respectively. These are defined as follows
I I iI I I iI
x y x y + −
= + = − [6.6]
where i is the square root of –1. I
+
has coherence order +1 and I
–
has coherence
order –1; coherence order is a signed quantity.
Using the definitions of Eqn. [6.6] I
x
and I
y
can be expressed in terms of the
raising and lowering operators
I I I I I I
x y i
= + ( ) = ( )
+ − + −
1
2
1
2
– [6.7]
from which it is seen that I
x
and I
y
are both mixtures of coherences with p = +1
and –1.
The operator product 2I
1x
I
2x
can be expressed in terms of the raising and
lowering operators in the following way (note that separate operators are used
for each spin: I
1±
and I
2±
)
2 2
1 2
1
2 1 1
1
2 2 2
1
2 1 2 1 2
1
2 1 2 1 2
I I I I I I
I I I I I I I I
x x
= × + ( ) × + ( )
= + ( ) + + ( )
+ − + −
+ + − − + − − +
[6.8]
The first term on the right of Eqn. [6.8] has p = (+1+1) = 2 and the second term
has p = (–1–1) = –2; both are double quantum coherences. The third and fourth
terms both have p = (+1–1) = 0 and are zero quantum coherences. The value of
p can be found simply by noting the number of raising and lowering operators
in the product.
The pure double quantum part of 2I
1x
I
2x
is, from Eqn. [6.8],
double quantum part 2
1 2
1
2 1 2 1 2
I I I I I I
x x
[ ]
= + ( )
+ + − −
[6.9]
The raising and lowering operators on the right of Eqn. [6.9] can be re
expressed in terms of the Cartesian operators:
6–13
1
2 1 2 1 2
1
2 1 1 2 2 1 1 2 2
1
2 1 2 1 2
2 2
I I I I I iI I iI I iI I iI
I I I I
x y x y x y x y
x x y y
+ + − −
+ ( ) = +
( )
+
( )
+ −
( )
−
( )
[ ]
= +
[ ]
So, the pure double quantum part of 2I
1x
I
2x
is
1
2 1 2 1 2
2 2 I I I I
x x y y
+
( )
; by a similar
method the pure zero quantum part can be shown to be
1
2 1 2 1 2
2 2 I I I I
x x y y
−
( )
.
Some further useful relationships are given in section 6.9
6.6 Three spins
The product operator formalism can be extended to three or more spins. No
really new features arise, but some of the key ideas will be highlighted in this
section. The description will assume that spin 1 is coupled to spins 2 and 3 with
coupling constants J
12
and J
13
; in the diagrams it will be assumed that J
12
> J
13
.
6.6.1 Types of operators
I
1x
represents inphase magnetization on spin 1; 2I
1x
I
2z
represents magnetization
antiphase with respect to the coupling to spin 2 and 2I
1x
I
3z
represents
magnetization antiphase with respect to the coupling to spin 3. 4I
1x
I
2z
I
3z
represents magnetization which is doubly antiphase with respect to the
couplings to both spins 2 and 3.
As in the case of two spins, the presence of more than one transverse
operator in the product represents multiple quantum coherence. For example,
2I
1x
I
2x
is a mixture of double and zeroquantum coherence between spins 1 and
2. The product 4I
1x
I
2x
I
3z
is the same mixture, but antiphase with respect to the
coupling to spin 3. Products such as 4I
1x
I
2x
I
3x
contain, amongst other things,
triplequantum coherences.
6.6.2 Evolution
Evolution under offsets and pulses is simply a matter of applying sequentially
the relevant rotations for each spin, remembering that rotations of spin 1 do not
affect operators of spins 2 and 3. For example, the term 2I
1x
I
2z
evolves under
the offset in the following way:
2 2 2
1 2 1 1 2 1 1 2
1 1 2 2 3 3
I I t I I t I I
x z
tI tI tI
x z y z
z z z
Ω Ω Ω
Ω Ω → → → + cos sin
The first arrow, representing evolution under the offset of spin 1, affects only the
spin 1 operator I
1x
. The second arrow has no effect as the spin 2 operator I
2z
and
this is unaffected by a zrotation. The third arrow also has no effect as there are
no spin 3 operators present.
The evolution under coupling follows the same rules as for a twospin
system. For example, evolution of I
1x
under the influence of the coupling to spin
3 generates 2I
1y
I
3z
I J t I J t I I
x
J tI I
x y z
z z
1
2
13 1 13 1 3
13 1 3
2
π
π π → + cos sin
Further evolution of the term 2I
1y
I
3z
under the influence of the coupling to spin 2
generates a double antiphase term
spin 2
spin 3
Ω
1
J
13
J
12
α α β β
α β α β
The doublet of doublets from
spin 1 coupled to two other
spins. The spin states of t he
coupled spins are also
indicated.
I
1x
2I
1x
I
2z
2I
1x
I
3z
4I
1x
I
2z
I
3z
Representations of different
types of operators.
6–14
2 2 4
1 3
2
12 1 3 13 1 2 3
12 1 2
I I J t I I J t I I I
y z
J tI I
y z x z z
z z
π
π π → − cos sin
In this evolution the spin 3 operator is unaffected as the coupling does not
involve this spin. The connection with the evolution of I
1y
under a coupling can
be made more explicit by writing 2I
3z
as a "constant" γ
γ π γ π γ
π
I J t I J t I I
y
J tI I
y x z
z z
1
2
12 1 13 1 2
12 1 2
2 → − cos sin
which compares directly to
I J t I J t I I
y
J tI I
y x z
z z
1
2
12 1 13 1 2
12 1 2
2
π
π π → − cos sin
6.7 Alternative notation
In this chapter different spins have been designated with a subscript 1, 2, 3 ...
Another common notation is to distinguish the spins by using a different letter to
represent their operators; commonly I and S are used for two of the symbols
2 2
1 2
I I I S
x z x z
≡
Note that the order in which the operators are written is not important, although
it is often convenient (and tidy) always to write them in the same sequence.
In heteronuclear experiments a notation is sometimes used where the letter
represents the nucleus. So, for example, operators referring to protons are given
the letter H, carbon13 atoms the letter C and nitrogen15 atoms the letter N;
carbonyl carbons are sometimes denoted C'. For example, 4C
x
H
z
N
z
denotes
magnetization on carbon13 which is antiphase with respect to coupling to both
proton and nitrogen15.
6.8 Conclusion
The product operator method as described here only applies to spinhalf nuclei.
It can be extended to higher spins, but significant extra complexity is introduced;
details can be found in the article by Sørensen et al. (Prog. NMR Spectrosc. 16,
163 (1983)).
The main difficulty with the product operator method is that the more pulses
and delays that are introduced the greater becomes the number of operators and
the more complex the trigonometrical expressions multiplying them. If pulses
are either 90° or 180° then there is some simplification as such pulses do not
increase the number of terms. As will be seen in chapter 7, it is important to try
to simplify the calculation as much as possible, for example by recognizing
when offsets or couplings are refocused by spin echoes.
A number of computer programs are available for machine computation
using product operators within programs such as Mathematica or Maple. These
can be very labour saving.
6–15
6.9 Multiple quantum coherence
6.9.1 Multiplequantum terms
In the product operator representation of multiple quantum coherences it is usual
to distinguish between active and passive spins. Active spins contribute
transverse operators, such as I
x
, I
y
and I
+
, to the product; passive spins contribute
only zoperators, I
z
. In a sense the spins contributing transverse operators are
"involved" in the coherence, while those contributing zoperators are simply
spectators.
For double and zeroquantum coherence in which spins i and j are active it is
convenient to define the following set of operators which represent pure multiple
quantum states of given order. The operators can be expressed in terms of the
Cartesian or raising and lowering operators.
double quantum,
DQ
DQ
zero quantum,
ZQ
p
I I I I I I I I
I I I I I I I I
p
I I I I I
x
ij
ix jx iy jy i j i j
y
ij
ix jy iy jx i i j i j
x
ij
ix jx iy jy
= ±
≡ −
( )
≡ +
( )
≡ +
( )
≡ −
( )
=
≡ +
( )
≡
( )
+ + − −
( )
+ + − −
( )
2
2 2
2 2
0
2 2
1
2
1
2
1
2
1
2
1
2
1
2 ii j i j
y
ij
iy jx ix jy i i j i j
I I I
I I I I I I I I
+ − − +
( )
+ − − +
+
( )
≡ −
( )
≡ −
( )
ZQ
1
2
1
2
2 2
6.9.2 Evolution of multiple quantum terms
Evolution under offsets
The double and zeroquantum operators evolve under offsets in a way which is
entirely analogous to the evolution of I
x
and I
y
under free precession except that
the frequencies of evolution are (Ω
i
+ Ω
j
) and (Ω
i
– Ω
j
) respectively:
DQ DQ DQ
DQ DQ DQ
ZQ
x
ij tI tI
i j x
ij
i j y
ij
y
ij tI tI
i j y
ij
i j x
ij
x
ij tI tI
i iz j jz
i iz j jz
i iz j jz
t t
t t
( ) + ( ) ( )
( ) + ( ) ( )
( ) +
→ +
( )
+ +
( )
→ +
( )
− +
( )
→
Ω Ω
Ω Ω
Ω Ω
Ω Ω Ω Ω
Ω Ω Ω Ω
cos sin
cos sin
cos cos sin
cos sin
Ω Ω Ω Ω
Ω Ω Ω Ω
Ω Ω
i j x
ij
i j y
ij
y
ij tI tI
i j y
ij
i j x
ij
t t
t t
i iz j jz
−
( )
+ −
( )
→ −
( )
− −
( )
( ) ( )
( ) + ( ) ( )
ZQ ZQ
ZQ ZQ ZQ
Evolution under couplings
Multiple quantum coherence between spins i and j does not evolve under the
influence of the coupling between the two active spins, i and j.
Double and zeroquantum operators evolve under passive couplings in a
way which is entirely analogous to the evolution of I
x
and I
y
; the resulting
multiple quantum terms can be described as being antiphase with respect to the
effective couplings:
6–16
DQ DQ DQ
DQ DQ DQ
ZQ ZQ ZQ
ZQ
DQ,eff DQ,eff
DQ,eff DQ,eff
ZQ,eff ZQ,eff
x
ij
x
ij
kz y
ij
y
ij
y
ij
kz x
ij
x
ij
x
ij
kz y
ij
y
ij
J t J t I
J t J t I
J t J t I
J
( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( )
( )
→ +
→ −
→ +
→
cos cos
cos sin
cos sin
cos
π π
π π
π π
π
2
2
2
ZQ, ZQ,eff ZQ,eff
ZQ ZQ t J t I
y
ij
kz x
ij ( ) ( )
− sinπ 2
J
DQ,eff
is the sum of the couplings between spin i and all other spins plus the sum
of the couplings between spin j and all other spins. J
ZQ,eff
is the sum of the
couplings between spin i and all other spins minus the sum of the couplings
between spin j and all other spins.
For example in a threespin system the zeroquantum coherence between
spins 1 and 2, antiphase with respect to spin 3, evolves according to
2 2
3
12
3
12 12
13 23
I J t I J t
J J J
z y z y x
ZQ ZQ ZQ
where
ZQ,eff ZQ,eff
ZQ,eff
( ) ( ) ( )
→ −
= −
cos sin π π
Further details of multiplequantum evolution can be found in section 5.3 of
Ernst, Bodenhausen and Wokaun Principles of NMR in One and Two
Dimensions (Oxford University Press, 1987).
7–1
7 Twodimensional NMR
†
7.1 Introduction
The basic ideas of twodimensional NMR will be introduced by reference to the
appearance of a COSY spectrum; later in this chapter the product operator
formalism will be used to predict the form of the spectrum.
Conventional NMR spectra (onedimensional spectra) are plots of intensity
vs. frequency; in twodimensional spectroscopy intensity is plotted as a function
of two frequencies, usually called F
1
and F
2
. There are various ways of
representing such a spectrum on paper, but the one most usually used is to make
a contour plot in which the intensity of the peaks is represented by contour lines
drawn at suitable intervals, in the same way as a topographical map. The
position of each peak is specified by two frequency coordinates corresponding
to F
1
and F
2
. Twodimensional NMR spectra are always arranged so that the F
2
coordinates of the peaks correspond to those found in the normal one
dimensional spectrum, and this relation is often emphasized by plotting the one
dimensional spectrum alongside the F
2
axis.
The figure shows a schematic COSY spectrum of a hypothetical molecule
containing just two protons, A and X, which are coupled together. The one
dimensional spectrum is plotted alongside the F
2
axis, and consists of the
familiar pair of doublets centred on the chemical shifts of A and X, δ
A
and δ
X
respectively. In the COSY spectrum, the F
1
coordinates of the peaks in the
twodimensional spectrum also correspond to those found in the normal one
dimensional spectrum and to emphasize this point the onedimensional
spectrum has been plotted alongside the F
1
axis. It is immediately clear that
this COSY spectrum has some symmetry about the diagonal F
1
= F
2
which has
been indicated with a dashed line.
In a onedimensional spectrum scalar couplings give rise to multiplets in the
spectrum. In twodimensional spectra the idea of a multiplet has to be
expanded somewhat so that in such spectra a multiplet consists of an array of
individual peaks often giving the impression of a square or rectangular outline.
Several such arrays of peaks can be seen in the schematic COSY spectrum
shown above. These twodimensional multiplets come in two distinct types:
diagonalpeak multiplets which are centred around the same F
1
and F
2
frequency coordinates and crosspeak multiplets which are centred around
different F
1
and F
2
coordinates. Thus in the schematic COSY spectrum there
are two diagonalpeak multiplets centred at F
1
= F
2
= δ
A
and F
1
= F
2
= δ
X,
one
crosspeak multiplet centred at F
1
= δ
A
, F
2
= δ
X
and a second crosspeak
multiplet centred at F
1
= δ
X
, F
2
= δ
A
.
The appearance in a COSY spectrum of a crosspeak multiplet F
1
= δ
A
, F
2
=
δ
X
indicates that the two protons at shifts δ
A
and δ
X
have a scalar coupling
†
Chapter 7 "TwoDimensional NMR" © James Keeler 1998 and 2002
δ
A
δ
X
δ
A
δ
X
Schematic COSY spectrum for
two coupled spins, A and X
7–2
between them. This statement is all that is required for the analysis of a COSY
spectrum, and it is this simplicity which is the key to the great utility of such
spectra. From a single COSY spectrum it is possible to trace out the whole
coupling network in the molecule
7.1.1 General Scheme for twoDimensional NMR
In onedimensional pulsed Fourier transform NMR the signal is recorded as a
function of one time variable and then Fourier transformed to give a spectrum
which is a function of one frequency variable. In twodimensional NMR the
signal is recorded as a function of two time variables, t
1
and t
2
, and the resulting
data Fourier transformed twice to yield a spectrum which is a function of two
frequency variables. The general scheme for twodimensional spectroscopy is
evolution detection
t
1
t
2
mixing preparation
In the first period, called the preparation time, the sample is excited by one
or more pulses. The resulting magnetization is allowed to evolve for the first
time period, t
1
. Then another period follows, called the mixing time, which
consists of a further pulse or pulses. After the mixing period the signal is
recorded as a function of the second time variable, t
2
. This sequence of events
is called a pulse sequence and the exact nature of the preparation and mixing
periods determines the information found in the spectrum.
It is important to realize that the signal is not recorded during the time t
1
, but
only during the time t
2
at the end of the sequence. The data is recorded at
regularly spaced intervals in both t
1
and t
2
.
The twodimensional signal is recorded in the following way. First, t
1
is set
to zero, the pulse sequence is executed and the resulting free induction decay
recorded. Then the nuclear spins are allowed to return to equilibrium. t
1
is then
set to ∆
1
, the sampling interval in t
1
, the sequence is repeated and a free
induction decay is recorded and stored separately from the first. Again the
spins are allowed to equilibrate, t
1
is set to 2∆
1
, the pulse sequence repeated and
a free induction decay recorded and stored. The whole process is repeated
again for t
1
= 3∆
1
, 4∆
1
and so on until sufficient data is recorded, typically 50 to
500 increments of t
1
. Thus recording a twodimensional data set involves
repeating a pulse sequence for increasing values of t
1
and recording a free
induction decay as a function of t
2
for each value of t
1
.
7.1.2 Interpretation of peaks in a twodimensional spectrum
Within the general framework outlined in the previous section it is now
possible to interpret the appearance of a peak in a twodimensional spectrum at
particular frequency coordinates.
7–3
F
1
F
2
0,0 80
20
a b c
Suppose that in some unspecified twodimensional spectrum a peak appears at
F
1
= 20 Hz, F
2
= 80 Hz (spectrum a above) The interpretation of this peak is
that a signal was present during t
1
which evolved with a frequency of 20 Hz.
During the mixing time this same signal was transferred in some way to another
signal which evolved at 80 Hz during t
2
.
Likewise, if there is a peak at F
1
= 20 Hz, F
2
= 20 Hz (spectrum b) the
interpretation is that there was a signal evolving at 20 Hz during t
1
which was
unaffected by the mixing period and continued to evolve at 20 Hz during t
2
.
The processes by which these signals are transferred will be discussed in the
following sections.
Finally, consider the spectrum shown in c. Here there are two peaks, one at
F
1
= 20 Hz, F
2
= 80 Hz and one at F
1
= 20 Hz, F
2
= 20 Hz. The interpretation
of this is that some signal was present during t
1
which evolved at 20 Hz and that
during the mixing period part of it was transferred into another signal which
evolved at 80 Hz during t
2
. The other part remained unaffected and continued
to evolve at 20 Hz. On the basis of the previous discussion of COSY spectra,
the part that changes frequency during the mixing time is recognized as leading
to a crosspeak and the part that does not change frequency leads to a diagonal
peak. This kind of interpretation is a very useful way of thinking about the
origin of peaks in a twodimensional spectrum.
It is clear from the discussion in this section that the mixing time plays a
crucial role in forming the twodimensional spectrum. In the absence of a
mixing time, the frequencies that evolve during t
1
and t
2
would be the same and
only diagonalpeaks would appear in the spectrum. To obtain an interesting
and useful spectrum it is essential to arrange for some process during the
mixing time to transfer signals from one spin to another.
7.2 EXSY and NOESY spectra in detail
In this section the way in which the EXSY (EXchange SpectroscopY) sequence
works will be examined; the pulse sequence is shown opposite. This
experiment gives a spectrum in which a crosspeak at frequency coordinates F
1
= δ
A
, F
2
= δ
B
indicates that the spin resonating at δ
A
is chemically exchanging
with the spin resonating at δ
B
.
The pulse sequence for EXSY is shown opposite. The effect of the sequence
will be analysed for the case of two spins, 1 and 2, but without any coupling
between them. The initial state, before the first pulse, is equilibrium
t
1
t
2
τ
mix
The pulse sequence for EXSY
(and NOESY). All pulses have
90° flip angles.
7–4
magnetization, represented as I
1z
+ I
2z
; however, for simplicity only
magnetization from the first spin will be considered in the calculation.
The first 90° pulse (of phase x) rotates the magnetization onto –y
I I
z
I I
y
x x
1
2 2
1
1 2
π π
→ → −
(the second arrow has no effect as it involves operators of spin 2). Next follows
evolution for time t
1
− → → − + I t I t I
y
t I t I
y x
z z
1 1 1 1 1 1 1
1 1 1 2 1 2
Ω Ω
Ω Ω cos sin
again, the second arrow has no effect. The second 90° pulse turns the first term
onto the zaxis and leaves the second term unaffected
− → → −
→ →
cos cos
sin sin
Ω Ω
Ω Ω
1 1 1
2 2
1 1 1
1 1 1
2 2
1 1 1
1 2
1 2
t I t I
t I t I
y
I I
z
x
I I
x
x x
x x
π π
π π
Only the I
1z
term leads to crosspeaks by chemical exchange, so the other term
will be ignored (in an experiment this is achieved by appropriate coherence
pathway selection). The effect of the first part of the sequence is to generate, at
the start of the mixing time, τ
mix
, some zmagnetization on spin 1 whose size
depends, via the cosine term, on t
1
and the frequency, Ω
1
, with which the spin 1
evolves during t
1
. The magnetization is said to be frequency labelled.
During the mixing time, τ
mix
, spin 1 may undergo chemical exchange with
spin 2. If it does this, it carries with it the frequency label that it acquired
during t
1
. The extent to which this transfer takes place depends on the details of
the chemical kinetics; it will be assumed simply that during τ
mix
a fraction f of
the spins of type 1 chemically exchange with spins of type 2. The effect of the
mixing process can then be written
− → − − ( ) − cos cos cos Ω Ω Ω
1 1 1 1 1 1 1 1 2
1 t I f t I f t I
z z z
mixing
The final 90° pulse rotates this zmagnetization back onto the yaxis
− − ( ) → → − ( )
− → →
1 1
1 1 1
2 2
1 1 1
1 1 2
2 2
1 1 2
1 2
1 2
f t I f t I
f t I f t I
z
I I
y
z
I I
y
x x
x x
cos cos
cos cos
Ω Ω
Ω Ω
π π
π π
Although the magnetization started on spin 1, at the end of the sequence
there is magnetization present on spin 2 – a process called magnetization
transfer. The analysis of the experiment is completed by allowing the I
1y
and I
2y
operators to evolve for time t
2
.
1
1 1
1 1 1
1 2 1 1 1 1 2 1 1 1
1 1 2
2 2 1 1 2
1 2 1 2 2 2
1 2 1 2 2 2
− ( ) → →
− ( ) − − ( )
→ →
f t I
f t t I f t t I
f t I
f t t I
y
t I t I
y x
y
t I t I
z z
z z
cos
cos cos sin cos
cos
cos cos
Ω
Ω Ω Ω Ω
Ω
Ω Ω
Ω Ω
Ω Ω
yy x
f t t I − sin cos Ω Ω
2 2 1 1 2
If it is assumed that the ymagnetization is detected during t
2
(this is an arbitrary
choice, but a convenient one), the time domain signal has two terms:
7–5
1
1 2 1 1 2 2 1 1
− ( ) + f t t f t t cos cos cos cos Ω Ω Ω Ω
The crucial thing is that the amplitude of the signal recorded during t
2
is
modulated by the evolution during t
1
. This can be seen more clearly by
imagining the Fourier transform, with respect to t
2
, of the above function. The
cos Ω
1 2
t and cos Ω
2 2
t terms transform to give absorption mode signals centred
at Ω
1
and Ω
2
respectively in the F
2
dimension; these are denoted A
1
2 ( )
and A
2
2 ( )
(the subscript indicates which spin, and the superscript which dimension). The
time domain function becomes
1
1
2
1 1 2
2
1 1
− ( ) +
( ) ( )
f A t fA t cos cos Ω Ω
If a series of spectra recorded as t
1
progressively increases are inspected it
would be found that the cos Ω
1 2
t term causes a change in size of the peaks at Ω
1
and Ω
2
– this is the modulation referred to above.
Fourier transformation with respect to t
1
gives peaks with an absorption
lineshape, but this time in the F
1
dimension; an absorption mode signal at Ω
1
in
F
1
is denoted A
1
1 ( )
. The time domain signal becomes, after Fourier
transformation in each dimension
1
1
2
1
1
2
2
1
1
− ( ) +
( ) ( ) ( ) ( )
f A A fA A
Thus, the final twodimensional spectrum is predicted to have two peaks. One
is at (F
1
, F
2
) = (Ω
1
, Ω
1
) – this is a diagonal peak and arises from those spins of
type 1 which did not undergo chemical exchange during τ
mix
. The second is at
(F
1
, F
2
) = (Ω
1
, Ω
2
) – this is a cross peak which indicates that part of the
magnetization from spin 1 was transferred to spin 2 during the mixing time. It
is this peak that contains the useful information. If the calculation were
repeated starting with magnetization on spin 2 it would be found that there are
similar peaks at (Ω
2
, Ω
2
) and (Ω
2
, Ω
1
).
The NOESY (Nuclear Overhauser Effect SpectrocopY) spectrum is recorded
using the same basic sequence. The only difference is that during the mixing
time the crossrelaxation is responsible for the exchange of magnetization
between different spins. Thus, a crosspeak indicates that two spins are
experiencing mutual crossrelaxation and hence are close in space.
Having completed the analysis it can now be seen how the EXCSY/NOESY
sequence is put together. First, the 90° – t
1
– 90° sequence is used to generate
frequency labelled zmagnetization. Then, during τ
mix
, this magnetization is
allowed to migrate to other spins, carrying its label with it. Finally, the last
pulse renders the zmagnetization observable.
7.3 More about twodimensional transforms
From the above analysis it was seen that the signal observed during t
2
has an
amplitude proportional to cos(Ω
1
t
1
); the amplitude of the signal observed during
t
2
depends on the evolution during t
1
. For the first increment of t
1
(t
1
= 0), the
signal will be a maximum, the second increment will have size proportional to
cos(Ω
1
∆
1
), the third proportional to cos(Ω
1
2∆
1
), the fourth to cos(Ω
1
3∆
1
) and so
time
frequency
Ω
Fourier transform
The Fourier transform of a
decayi ng cosi ne functi on
cosΩt exp( –t/T
2
) i s an
absorption mode Lorentzian
centred at frequency Ω; the real
part of the spectrum has been
plotted.
7–6
on. This modulation of the amplitude of the observed signal by the t
1
evolution
is illustrated in the figure below.
In the figure the first column shows a series of free induction decays that
would be recorded for increasing values of t
1
and the second column shows the
Fourier transforms of these signals. The final step in constructing the two
dimensional spectrum is to Fourier transform the data along the t
1
dimension.
This process is also illustrated in the figure. Each of the spectra shown in the
second column are represented as a series of data points, where each point
corresponds to a different F
2
frequency. The data point corresponding to a
particular F
2
frequency is selected from the spectra for t
1
= 0, t
1
= ∆
1
, t
1
= 2∆
1
and so on for all the t
1
values. Such a process results in a function, called an
interferogram, which has t
1
as the running variable.
Illustration of how the modulation of a free induction decay by evolution during t
1
gives rise to a peak in the
7–7
twodimensional spectrum. In the left most column is shown a series of free induction decays that would be
recorded for successive values of t
1
; t
1
increases down the page. Note how the amplitude of these free
induction decays varies with t
1
, something that becomes even plainer when the time domain signals are
Fourier transformed, as shown in the second column. In practice, each of these F
2
spectra in column two
consist of a series of data points. The data point at the same frequency in each of these spectra is extracted
and assembled into an interferogram, in which the horizontal axis is the time t
1
. Several such
interferograms, labelled a to g, are shown in the third column. Note that as there were eight F
2
spectra in
column two corresponding to different t
1
values there are eight points in each interferogram. The F
2
frequencies at which the interferograms are taken are indicated on the lower spectrum of the second
column. Finally, a second Fourier transformation of these interferograms gives a series of F
1
spectra shown
in the right hand column. Note that in this column F
2
increases down the page, whereas in the first column t
1
increase down the page. The final result is a twodimensional spectrum containing a single peak.
Several interferograms, labelled a to g, computed for different F
2
frequencies
are shown in the third column of the figure. The particular F
2
frequency that
each interferogram corresponds to is indicated in the bottom spectrum of the
second column. The amplitude of the signal in each interferogram is different,
but in this case the modulation frequency is the same. The final stage in the
processing is to Fourier transform these interferograms to give the series of
spectra which are shown in the right most column of the figure. These spectra
have F
1
running horizontally and F
2
running down the page. The modulation of
the time domain signal has been transformed into a single twodimensional
peak. Note that the peak appears on several traces corresponding to different F
2
frequencies because of the width of the line in F
2
.
The time domain data in the t
1
dimension can be manipulated by multiplying
by weighting functions or zero filling, just as with conventional free induction
decays.
7.4 Twodimensional experiments using coherence transfer
through Jcoupling
Perhaps the most important set of twodimensional experiments are those which
transfer magnetization from one spin to another via the scalar coupling between
them. As was seen in section 6.3.3, this kind of transfer can be brought about
by the action of a pulse on an antiphase state. In outline the basic process is
I I I I I
x y z
x
z y 1 1 2 1 2
2 2
coupling 90 ( ) to both spins
spin 1 spin 2
→ →
°
7.4.1 COSY
The pulse sequence for this experiment is shown opposite. It will be assumed
in the analysis that all of the pulses are applied about the xaxis and for
simplicity the calculation will start with equilibrium magnetization only on spin
1. The effect of the first pulse is to generate ymagnetization, as has been
worked out previously many times
I I
z
I I
y
x x
1
2 2
1
1 2
π π
→ → −
This state then evolves for time t
1
, first under the influence of the offset of spin
1 (that of spin 2 has no effect on spin 1 operators):
− → − + I t I t I
y
t I
y x
z
1 1 1 1 1 1 1
1 1 1
Ω
Ω Ω cos sin
t
1
t
2
Pulse sequence for the two
dimensional COSY experiment
7–8
Both terms on the right then evolve under the coupling
− → − +
→ +
cos cos cos sin cos
sin cos sin sin sin
Ω Ω Ω
Ω Ω Ω
1 1 1
2
12 1 1 1 1 12 1 1 1 1 2
1 1 1
2
12 1 1 1 1 12 1 1 1 1
12 1 1 2
12 1 1 2
2
2
t I J t t I J t t I I
t I J t t I J t t I I
y
J t I I
y x z
x
J t I I
x y
z z
z z
π
π
π π
π π
22z
That completes the evolution under t
1
. Now all that remains is to consider the
effect of the final pulse, remembering that the effect of the pulse on both spins
needs to be computed. Taking the terms one by one:
− → → − { }
→ → − { }
cos cos cos cos
sin cos sin cos
cos sin
π π
π π
π
π π
π π
J t t I J t t I
J t t I I J t t I I
J t
y
I I
z
x z
I I
x y
x x
x x
12 1 1 1 1
2 2
12 1 1 1 1
12 1 1 1 1 2
2 2
12 1 1 1 1 2
12 1
1 2
1 2
1
2 2 2
Ω Ω
Ω Ω
ΩΩ Ω
Ω Ω
1 1 1
2 2
12 1 1 1 1
12 1 1 1 1 2
2 2
12 1 1 1 1 2
1 2
1 2
3
2 2 4
t I J t t I
J t t I I J t t I I
x
I I
x
y z
I I
z y
x x
x x
π π
π π
π
π π
→ → { }
→ → − { }
cos sin
sin sin sin sin
Terms {1} and {2} are unobservable. Term {3} corresponds to inphase
magnetization of spin 1, aligned along the xaxis. The t
1
modulation of this
term depends on the offset of spin 1, so a diagonal peak centred at (Ω
1
,Ω
1
) is
predicted. Term {4} is the really interesting one. It shows that antiphase
magnetization on spin 1, 2
1 2
I I
y z
, is transferred to antiphase magnetization on
spin 2, 2
1 2
I I
z y
; this is an example of coherence transfer. Term {4} appears as
observable magnetization on spin 2, but it is modulated in t
1
with the offset of
spin 1, thus it gives rise to a crosspeak centred at (Ω
1
,Ω
2
). It has been shown,
therefore, how cross and diagonalpeaks arise in a COSY spectrum.
Some more consideration should be give to the form of the cross and
diagonal peaks. Consider again term {3}: it will give rise to an inphase
multiplet in F
2
, and as it is along the xaxis, the lineshape will be dispersive.
The form of the modulation in t
1
can be expanded, using the formula,
cos sin sin sin A B B A B A = + ( ) + − ( ) { }
1
2
to give
cos sin sin sin π π π J t t t J t t J t
12 1 1 1
1
2 1 1 12 1 1 1 12
Ω Ω Ω = + ( ) + − ( ) { }
Two peaks in F
1
are expected at Ω
1 12
± πJ , these are just the two lines of the
spin 1 doublet. In addition, since these are sine modulated they will have the
dispersion lineshape. Note that both components in the spin 1 multiplet
observed in F
2
are modulated in this way, so the appearance of the two
dimensional multiplet can best be found by "multiplying together" the
multiplets in the two dimensions, as shown opposite. In addition, all four
components of the diagonalpeak multiplet have the same sign, and have the
double dispersion lineshape illustrated below
time
frequency
Ω
Fourier transform
The Fourier transform of a
decayi ng si ne f unct i on
sinΩt exp(–t/T
2
) is a dispersion
mode Lorentzian centred at
frequency Ω.
F
1
J
12
F
2
J
12
Schematic view of the diagonal
peak from a COSY spectrum.
The squares are supposed to
indicate the twodimensional
double dispersion lineshape
illustrated below
7–9
The double dispersion lineshape seen in pseudo 3D and as a contour plot; negative contours are indicated
by dashed lines.
Term {4} can be treated in the same way. In F
2
we know that this term gives
rise to an antiphase absorption multiplet on spin 2. Using the relationship
sin sin cos cos B A B A B A = − + ( ) + − ( ) { }
1
2
the modulation in t
1
can be expanded
sin sin cos cos π π π J t t t J t t J t
12 1 1
1
2 1 1 12 1 1 1 12
Ω Ω Ω = − + ( ) + − ( ) { }
Two peaks in F
1
, at Ω
1 12
± πJ , are expected; these are just the two lines of the
spin 1 doublet. Note that the two peaks have opposite signs – that is they are
antiphase in F
1
. In addition, since these are cosine modulated we expect the
absorption lineshape (see section 7.2). The form of the crosspeak multiplet
can be predicted by "multiplying together" the F
1
and F
2
multiplets, just as was
done for the diagonalpeak multiplet. The result is shown opposite. This
characteristic pattern of positive and negative peaks that constitutes the cross
peak is know as an antiphase square array.
The double absorption lineshape seen in pseudo 3D and as a contour plot.
COSY spectra are sometimes plotted in the absolute value mode, where all
the sign information is suppressed deliberately. Although such a display is
convenient, especially for routine applications, it is generally much more
desirable to retain the sign information. Spectra displayed in this way are said
to be phase sensitive; more details of this are given in section 7.6.
As the coupling constant becomes comparable with the linewidth, the
positive and negative peaks in the crosspeak multiplet begin to overlap and
cancel one another out. This leads to an overall reduction in the intensity of the
crosspeak multiplet, and ultimately the crosspeak disappears into the noise in
the spectrum. The smallest coupling which gives rise to a crosspeak is thus set
F
1
J
12
F
2
J
12
Schematic view of the cross
peak multiplet from a COSY
spectrum. The circles are
supposed to indicate the two
dimensional double absorption
lineshape illustrated below;
filled circles represent positive
i ntensi ty, open represent
negative intensity.
7–10
by the linewidth and the signaltonoise ratio of the spectrum.
7.4.2 Doublequantum filtered COSY (DQF COSY)
The conventional COSY experiment suffers from a disadvantage which arises
from the different phase properties of the cross and diagonalpeak multiplets.
The components of a diagonal peak multiplet are all inphase and so tend to
reinforce one another. In addition, the dispersive tails of these peaks spread far
into the spectrum. The result is a broad intense diagonal which can obscure
nearby crosspeaks. This effect is particularly troublesome when the coupling
is comparable with the linewidth as in such cases, as was described above,
cancellation of antiphase components in the crosspeak multiplet reduces the
overall intensity of these multiplets.
This difficulty is neatly sidestepped by a modification called double
quantum filtered COSY (DQF COSY). The pulse sequence is shown opposite.
Up to the second pulse the sequence is the same as COSY. However, it is
arranged that only doublequantum coherence present during the (very short)
delay between the second and third pulses is ultimately allowed to contribute to
the spectrum. Hence the name, "doublequantum filtered", as all the observed
signals are filtered through doublequantum coherence. The final pulse is
needed to convert the double quantum coherence back into observable
magnetization. This doublequantum derived signal is selected by the use of
coherence pathway selection using phase cycling or field gradient pulses.
In the analysis of the COSY experiment, it is seen that after the second 90°
pulse it is term {2} that contains doublequantum coherence; this can be
demonstrated explicitly by expanding this term in the raising and lowering
operators, as was done in section 6.5
2 2
1 2
1
2 1 1
1
2 2 2
1
2 1 2 1 2
1
2 1 2 1 2
I I I I I I
I I I I I I I I
x y i
i i
= × + ( ) × − ( )
= − ( ) + − + ( )
+ − + −
+ + − − + − − +
This term contains both double and zeroquantum coherence. The pure
doublequantum part is the term in the first bracket on the right; this term can
be reexpressed in Cartesian operators:
1
2 1 2 1 2
1
2 1 1 1 1 2 2 2 2
1
2 1 2 1 2
2 2
i i x y x y x y x y
x y y x
I I I I I iI I iI I iI I iI
I I I I
+ + − −
− ( ) = +
( )
+
( )
+ −
( )
−
( )
[ ]
= +
[ ]
The effect of the last 90°(x) pulse on the double quantum part of term {2} is
thus
− +
( )
→ →
− + ( )
1
2 12 1 1 1 1 2 1 2
2 2
1
2 12 1 1 1 1 2 1 2
2 2
2 2
1 2
sin cos
sin cos
π
π
π π
J t t I I I I
J t t I I I I
x y y x
I I
x z z x
x x
Ω
Ω
The first term on the right is antiphase magnetization of spin 1 aligned along
the xaxis; this gives rise to a diagonalpeak multiplet. The second term is anti
phase magnetization of spin 2, again aligned along x; this will give rise to a
t
1
t
2
The pulse sequence for DQF
COSY; the delay between the
last two pulses is usually just a
few microseconds.
7–11
crosspeak multiplet. Both of these terms have the same modulation in t
1
,
which can be shown, by a similar analysis to that used above, to lead to an anti
phase multiplet in F
1
. As these peaks all have the same lineshape the overall
phase of the spectrum can be adjusted so that they are all in absorption; see
section 7.6 for further details. In contrast to the case of a simple COSY
experiment both the diagonal and crosspeak multiplets are in antiphase in
both dimensions, thus avoiding the strong inphase diagonal peaks found in the
simple experiment. The DQF COSY experiment is the method of choice for
tracing out coupling networks in a molecule.
7.4.3 Heteronuclear correlation experiments
One particularly useful experiment is to record a twodimensional spectrum in
which the coordinate of a peak in one dimension is the chemical shift of one
type of nucleus (e.g. proton) and the coordinate in the other dimension is the
chemical shift of another nucleus (e.g. carbon13) which is coupled to the first
nucleus. Such spectra are often called shift correlation maps or shift
correlation spectra.
The onebond coupling between a carbon13 and the proton directly attached
to it is relatively constant (around 150 Hz), and much larger than any of the
longrange carbon13 proton couplings. By utilizing this large difference
experiments can be devised which give maps of carbon13 shifts vs the shifts of
directly attached protons. Such spectra are very useful as aids to assignment;
for example, if the proton spectrum has already been assigned, simply recording
a carbon13 proton correlation experiment will give the assignment of all the
protonated carbons.
Only one kind of nuclear species can be observed at a time, so there is a
choice as to whether to observe carbon13 or proton when recording a shift
correlation spectrum. For two reasons, it is very advantageous from the
sensitivity point of view to record protons. First, the proton magnetization is
larger than that of carbon13 because there is a larger separation between the
spin energy levels giving, by the Boltzmann distribution, a greater population
difference. Second, a given magnetization induces a larger voltage in the coil
the higher the NMR frequency becomes.
Trying to record a carbon13 proton shift correlation spectrum by proton
observation has one serious difficulty. Carbon13 has a natural abundance of
only 1%, thus 99% of the molecules in the sample do not have any carbon13 in
them and so will not give signals that can be used to correlate carbon13 and
proton. The 1% of molecules with carbon13 will give a perfectly satisfactory
spectrum, but the signals from these resonances will be swamped by the much
stronger signals from noncarbon13 containing molecules. However, these
unwanted signals can be suppressed using coherence selection in a way which
will be described below.
7–12
7.4.3.1 Heteronuclear multiplequantum correlation (HMQC)
The pulse sequence for this popular experiment is given opposite. The
sequence will be analysed for a coupled carbon13 proton pair, where spin 1
will be the carbon13 and spin 2 the proton.
The analysis will start with equilibrium magnetization on spin 1, I
1z
. The
whole analysis can be greatly simplified by noting that the 180° pulse is exactly
midway between the first 90° pulse and the start of data acquisition. As has
been shown in section 6.4, such a sequence forms a spin echo and so the
evolution of the offset of spin 1 over the entire period (t
1
+ 2∆) is refocused.
Thus the evolution of the offset of spin 1 can simply be ignored for the
purposes of the calculation.
At the end of the delay ∆ the state of the system is simply due to evolution of
the term –I
1y
under the influence of the scalar coupling:
− + cos sin π π J I J I I
y x z 12 1 12 1 2
2 ∆ ∆
It will be assumed that ∆ = 1/(2J
12
), so only the antiphase term is present.
The second 90° pulse is applied to carbon13 (spin 2) only
2 2
1 2
2
1 2
2
I I I I
x z
I
x y
x
π
→ −
This pulse generates a mixture of heteronuclear double and zeroquantum
coherence, which then evolves during t
1
. In principle this term evolves under
the influence of the offsets of spins 1 and 2 and the coupling between them.
However, it has already been noted that the offset of spin 1 is refocused by the
centrally placed 180° pulse, so it is not necessary to consider evolution due to
this term. In addition, it can be shown that multiplequantum coherence
involving spins i and j does not evolve under the influence of the coupling, J
ij
,
between these two spins. As a result of these two simplifications, the only
evolution that needs to be considered is that due to the offset of spin 2 (the
carbon13).
− → − + 2 2 2
1 2 2 1 1 2 2 1 1 2
2 1 2
I I t I I t I I
x y
t I
x y x x
z
Ω
Ω Ω cos sin
The second 90° pulse to spin 2 (carbon13) regenerates the first term on the
right into spin 1 (proton) observable magnetization; the other remains
unobservable
− → − cos cos Ω Ω
2 1 1 2
2
2 1 1 2
2 2
2
t I I t I I
x y
I
x z
x
π
This term then evolves under the coupling, again it is assumed that
∆ = 1/(2J
12
)
− → −
= ( )
cos cos
,
Ω Ω
∆ ∆
2 1 1 2
2 1 2
2 1 1
2
12 1 2 12
t I I t I
x z
J I I J
y
z z
π
1
H
13
C
t
1
∆ ∆
t
2
The pulse sequence for HMQC.
Filled rectangles represent 90°
pulses and open rectangles
represent 180° pulses. The
delay ∆ is set to 1/(2J
12
).
7–13
This is a very nice result; in F
2
there will be an inphase doublet centred at the
offset of spin 1 (proton) and these two peaks will have an F
1
coordinate simply
determined by the offset of spin 2 (carbon13); the peaks will be in absorption.
A schematic spectrum is shown opposite.
The problem of how to suppress the very strong signals from protons not
coupled to any carbon13 nuclei now has to be addressed. From the point of
view of these protons the carbon13 pulses might as well not even be there, and
the pulse sequence looks like a simple spin echo. This insensitivity to the
carbon13 pulses is the key to suppressing the unwanted signals. Suppose that
the phase of the first carbon13 90° pulse is altered from x to –x. Working
through the above calculation it is found that the wanted signal from the protons
coupled to carbon13 changes sign i.e. the observed spectrum will be inverted.
In contrast the signal from a proton not coupled to carbon13 will be unaffected
by this change. Thus, for each t
1
increment the free induction decay is recorded
twice: once with the first carbon13 90° pulse set to phase x and once with it set
to phase –x. The two free induction decays are then subtracted in the computer
memory thus cancelling the unwanted signals. This is an example of a very
simple phase cycle.
In the case of carbon13 and proton the one bond coupling is so much larger
than any of the long range couplings that a choice of ∆ = 1/(2J
one bond
) does not
give any correlations other than those through the onebond coupling. There is
simply insufficient time for the longrange couplings to become antiphase.
However, if ∆ is set to a much longer value (30 to 60 ms), longrange
correlations will be seen. Such spectra are very useful in assigning the
resonances due to quaternary carbon13 atoms. The experiment is often called
HMBC (heteronuclear multiplebond correlation).
Now that the analysis has been completed it can be seen what the function of
various elements in the pulse sequence is. The first pulse and delay generate
magnetization on proton which is antiphase with respect to the coupling to
carbon13. The carbon13 90° pulse turns this into multiple quantum
coherence. This forms a filter through which magnetization not bound to
carbon13 cannot pass and it is the basis of discrimination between signals from
protons bound and not bound to carbon13. The second carbon13 pulse returns
the multiple quantum coherence to observable antiphase magnetization on
proton. Finally, the second delay ∆ turns the antiphase state into an inphase
state. The centrally placed proton 180° pulse refocuses the proton shift
evolution for both the delays ∆ and t
1
.
7.4.3.2 Heteronuclear singlequantum correlation (HSQC)
This pulse sequence results in a spectrum identical to that found for HMQC.
Despite the pulse sequence being a little more complex than that for HMQC,
HSQC has certain advantages for recording the spectra of large molecules, such
a proteins. The HSQC pulse sequence is often embedded in much more
complex sequences which are used to record two and threedimensional
F
1
J
12
F
2
1
Ω
2
Ω
Schematic HMQC spectrum for
two coupled spins.
7–14
spectra of carbon13 and nitrogen15 labelled proteins.
1
H
13
C
t
1
y
A B C
t
2
∆
2
∆
2
∆
2
∆
2
The pulse sequence for HSQC. Filled rectangles represent 90° pulses and open rectangles represent 180°
pulses. The delay ∆ is set to 1/(2J
12
); all pulses have phase x unless otherwise indicated.
If this sequence were to be analysed by considering each delay and pulse in turn
the resulting calculation would be far too complex to be useful. A more
intelligent approach is needed where simplifications are used, for example by
recognizing the presence of spin echoes who refocus offsets or couplings.
Also, it is often the case that attention can be focused a particular terms, as
these are the ones which will ultimately lead to observable signals. This kind of
"intelligent" analysis will be illustrated here.
Periods A and C are spin echoes in which 180° pulses are applied to both
spins; it therefore follows that the offsets of spins 1 and 2 will be refocused, but
the coupling between them will evolve throughout the entire period. As the
total delay in the spin echo is 1/(2J
12
) the result will be the complete conversion
of inphase into antiphase magnetization.
Period B is a spin echo in which a 180° pulse is applied only to spin 1. Thus,
the offset of spin 1 is refocused, as is the coupling between spins 1 and 2; only
the offset of spin 2 affects the evolution.
With these simplifications the analysis is easy. The first pulse generates –I
1y
; during period A this then becomes –2I
1x
I
2z
. The 90°(y) pulse to spin 1 turns
this to 2I
1z
I
2z
and the 90°(x) pulse to spin 2 turns it to –2I
1z
I
2y
. The evolution
during period B is simply under the offset of spin 2
− → − + 2 2 2
1 2 2 1 1 2 2 1 1 2
2 1 2
I I t I I t I I
z y
t I
z y z x
z
Ω
Ω Ω cos sin
The next two 90° pulses transfer the first term to spin 1; the second term is
rotated into multiple quantum and is not observed
− + →
− −
+ ( )
cos sin
cos sin
Ω Ω
Ω Ω
2 1 1 2 2 1 1 2
2
2 1 1 2 2 1 1 2
2 2
2 2
1 2
t I I t I I
t I I t I I
z y z x
I I
y z y x
x x
π
The first term on the right evolves during period C into inphase magnetization
(the evolution of offsets is refocused). So the final observable term is
cos Ω
2 1 1
t I
x
. The resulting spectrum is therefore an inphase doublet in F
2
,
centred at the offset of spin 1, and these peaks will both have the same
frequency in F
1
, namely the offset of spin 2. The spectrum looks just like the
HMQC spectrum.
7.5 Advanced topic: Multiplequantum spectroscopy
A key feature of twodimensional NMR experiments is that no direct
7–15
observations are made during t
1
, it is thus possible to detect, indirectly, the
evolution of unobservable coherences. An example of the use of this feature is
in the indirect detection of multiplequantum spectra. A typical pulse sequence
for such an experiment is shown opposite
For a twospin system the optimum value for ∆ is 1/(2J
12
). The sequence can
be dissected as follows. The initial 90° – ∆/2 – 180° – ∆/2 – sequence is a spin
echo which, at time ∆, refocuses any evolution of offsets but allows the
coupling to evolve and generate antiphase magnetization. This antiphase
magnetization is turned into multiplequantum coherence by the second 90°
pulse. After evolving for time t
1
the multiple quantum is returned into
observable (antiphase) magnetization by the final 90° pulse. Thus the first
three pulses form the preparation period and the last pulse is the mixing period.
7.5.1 Doublequantum spectrum for a threespin system
The sequence will be analysed for a system of three spins. A complete
analysis would be rather lengthy, so attention will be focused on certain terms
as above, as many simplifying assumptions as possible will be made about the
sequence.
The starting point will be equilibrium magnetization on spin 1, I
1z
; after the
spin echo the magnetization has evolved due to the coupling between spin 1 and
spin 2, and the coupling between spin 1 and spin 3 (the 180° pulse causes an
overall sign change (see section 6.4.1) but this has no real effect here so it will
be ignored)
– cos sin
cos cos sin cos
cos sin sin sin
I J I J I I
J J I J J I I
J J I I J
y
J I I
y x z
J I I
y x z
x z
z z
z z
1
2
12 1 12 1 2
2
13 12 1 13 12 1 3
13 12 1 2 13
12 1 2
13 1 3
2
2
2
π
π
π π
π π π π
π π π
∆
∆
∆ ∆
∆ ∆ ∆ ∆
∆ ∆ ∆
→ − +
→ − +
+ + ππJ I I I
y z z 12 1 2 3
4 ∆
[3.1]
Of these four terms, all but the first are turned into multiplequantum by the
second 90° pulse. For example, the second term becomes a mixture of double
and zero quantum between spins 1 and 3
sin cos sin cos π π π π
π
J J I I J J I I
x z
I I I
x y
x x x
13 12 1 3
2
13 12 1 3
2 2
1 2 3
∆ ∆ ∆ ∆
+ + ( )
→ −
It will be assumed that appropriate coherence pathway selection has been used
so that ultimately only the doublequantum part contributes to the spectrum.
This part is
−
[ ]
+
( ) { }
≡
( )
sin cos π π J J I I I I B
x y y x y 13 12
1
2 1 3 1 3 13
2 2 ∆ ∆ DQ
13
The term in square brackets just gives the overall intensity, but does not affect
the frequencies of the peaks in the twodimensional spectrum as it does not
depend on t
1
or t
2
; this intensity term is denoted B
13
for brevity. The operators
in the curly brackets represent a pure double quantum state which can be
denoted DQ
13
y
( )
; the superscript (13) indicates that the double quantum is
between spins 1 and 3 (see section 6.9).
∆
2
∆
2
t
1
t
2
Pulse sequence for multiple
quantum spectroscopy.
7–16
As is shown in section 6.9, such a doublequantum term evolves under the
offset according to
B
B t B t
y
t I t I t I
y x
z z z
13
13 3 1 13 3 1
1 1 1 2 1 2 3 1 3
DQ
cos DQ DQ
13
1
13
1
13
( )
+ +
( ) ( )
→
+ ( ) − + ( )
Ω Ω Ω
Ω Ω Ω Ω sin
where DQ
x
13 ( )
≡ −
( )
1
2 1 3 1 3
2 2 I I I I
x x y y
. This evolution is analogous to that of a
single spin where y rotates towards –x.
As is also shown in section 6.9, DQ and DQ
13 13
y x
( ) ( )
do not evolve under the
coupling between spins 1 and 3, but they do evolve under the sum of the
couplings between these two and all other spins; in this case this is simply
(J
12
+J
23
). Taking each term in turn
B t
B t J J t
B t J J t I
B
y
J t I I J t I I
y
z x
z z z z
13 3 1
2 2
13 3 1 12 23 1
13 3 1 12 23 1 2
13
12 1 1 2 23 1 2 3
2
cos DQ
cos cos DQ
cos sin DQ
1
13
1
13
1
13
Ω Ω
Ω Ω
Ω Ω
+ ( ) →
+ ( ) + ( )
− + ( ) + ( )
−
( )
+
( )
( )
π π
π
π
sin ΩΩ Ω
Ω Ω
Ω Ω
1
13
1
13
1
13
DQ
cos DQ
DQ
+ ( ) →
− + ( ) + ( )
− + ( ) + ( )
( )
+
( )
( )
3 1
2 2
13 3 1 12 23 1
13 3 1 12 23 1 2
12 1 1 2 23 1 2 3
2
t
B t J J t
B t J J t I
x
J t I I J t I I
x
z y
z z z z
π π
π
π
sin
sin sin
Terms such as 2 2
2 2
I I
z y z x
DQ and DQ
13 13 ( ) ( )
can be thought of as doublequantum
coherence which has become "antiphase" with respect to the coupling to spin
2; such terms are directly analogous to singlequantum antiphase
magnetization.
Of all the terms present at the end of t
1
, only DQ
13
y
( )
is rendered observable
by the final pulse
cos cos DQ
cos cos
1
13
1
Ω Ω
Ω Ω
+ ( ) + ( ) →
+ ( ) + ( ) +
[ ]
( )
+ + ( )
3 1 12 23 1 13
2
3 1 12 23 1 13 1 3 1 3
1 2 3
2 2
t J J t B
t J J t B I I I I
y
I I I
x z z x
x x x
π
π
π
The calculation predicts that two twodimensional multiplets appear in the
spectrum. Both have the same structure in F
1
, namely an in–phase doublet,
split by (J
12
+ J
23
) and centred at (Ω
1
+ Ω
3
); this is analogous to a normal
multiplet. In F
2
one twodimensional multiplet is centred at the offset of spins
1, Ω
1
, and one at the offset of spin 3, Ω
3
; both multiplets are antiphase with
respect to the coupling J
13
. Finally, the overall amplitude, B
13
, depends on the
delay ∆ and all the couplings in the system. The schematic spectrum is shown
opposite. Similar multiplet structures are seen for the doublequantum between
spins 1 & 2 and spins 2 & 3.
F
1
F
2
1
Ω
3
Ω
+
1
Ω
3
Ω
Schematic twodimensional
double quantum spectrum
showing the multiplets arising
from evolution of double
quantum coherence between
spins 1 and 3. If has been
assumed that J
12
> J
13
> J
23
.
7–17
7.5.2 Interpretation of doublequantum spectra
The doublequantum spectrum shows the relationship between the frequencies
of the lines in the double quantum spectrum and those in the (conventional)
singlequantum spectrum. If two twodimensional multiplets appear at (F
1
, F
2
)
= (Ω
A
+ Ω
B
, Ω
A
) and (Ω
A
+ Ω
B
, Ω
B
) the implication is that the two spins A and
B are coupled, as it is only if there is a coupling present that doublequantum
coherence between the two spins can be generated (e.g. in the previous section,
if J
13
= 0 the term B
13
, goes to zero). The fact that the two twodimensional
multiplets share a common F
1
frequency and that this frequency is the sum of
the two F
2
frequencies constitute a double check as to whether or not the peaks
indicate that the spins are coupled.
Double quantum spectra give very similar information to that obtained from
COSY i.e. the identification of coupled spins. Each method has particular
advantages and disadvantages:
(1) In COSY the crosspeak multiplet is antiphase in both dimensions,
whereas in a doublequantum spectrum the multiplet is only antiphase in F
2
.
This may lead to stronger peaks in the doublequantum spectrum due to less
cancellation. However, during the two delays ∆ magnetization is lost by
relaxation, resulting in reduced peak intensities in the doublequantum
spectrum.
(2) The value of the delay ∆ in the doublequantum experiment affects the
amount of multiplequantum generated and hence the intensity in the spectrum.
All of the couplings present in the spin system affect the intensity and as
couplings cover a wide range, no single optimum value for ∆ can be given. An
unfortunate choice for ∆ will result in low intensity, and it is then possible that
correlations will be missed. No such problems occur with COSY.
(3) There are no diagonalpeak multiplets in a doublequantum spectrum, so
that correlations between spins with similar offsets are relatively easy to locate.
In contrast, in a COSY the crosspeaks from such a pair of spins could be
obscured by the diagonal.
(4) In more complex spin systems the interpretation of a COSY remains
unambiguous, but the doublequantum spectrum may show a peak with F
1
co
ordinate (Ω
A
+ Ω
B
) and F
2
coordinate Ω
A
(or Ω
B
) even when spins A and B are
not coupled. Such remote peaks, as they are called, appear when spins A and B
are both coupled to a third spin. There are various tests that can differentiate
these remote from the more useful direct peaks, but these require additional
experiments. The form of these remote peaks in considered in the next section.
On the whole, COSY is regarded as a more reliable and simple experiment,
although doublequantum spectroscopy is used in some special circumstances.
7.5.3 Remote peaks in doublequantum spectra
The origin of remote peaks can be illustrated by returning to the calculation of
section 7.5.1. and focusing on the doubly antiphase term which is present at
F
1
F
2
A
Ω
B
Ω
+
A
Ω
B
Ω
Schematic spectrum showing
the relationship between the
single and doublequantum
frequencies for coupled spins.
7–18
the end of the spin echo (the fourth term in Eqn. [3.1])
sin sin π π J J I I I
y z z 13 12 1 2 3
4 ∆ ∆
The 90° pulse rotates this into multiplequantum
sin sin sin sin π π π π
π
J J I I I J J I I I
y z z
I I I
z y y
x x x
13 12 1 2 3
2
13 12 1 2 3
4 4
1 2 3
∆ ∆ ∆ ∆
+ + ( )
→
The pure doublequantum part of this term is
− −
( )
≡
( )
1
2 13 12 1 2 3 1 2 3 23 1 1
23
4 4 2 sin sin
,
π π J J I I I I I I B I DQ
z x x z y y z
x
∆ ∆
In words, what has been generated in doublequantum between spins 2 and 3,
antiphase with respect to spin 1. The key thing is that no coupling between
spins 2 and 3 is required for the generation of this term – the intensity just
depends on J
12
and J
13
; all that is required is that both spins 2 and 3 have a
coupling to the third spin, spin 1.
During t
1
this term evolves under the influence of the offsets and the
couplings. Only two terms ultimately lead to observable signals; at the end of t
1
these two terms are
B t J J t I DQ
B t J J t DQ
z x
y
23 1 3 1 12 13 1 1
23
23 1 3 1 12 13 1
23
2
,
,
cos
cos
Ω Ω
Ω Ω
2
2
cos
sin
+ ( ) + ( )
+ ( ) + ( )
( )
( )
π
π
and after the final 90° pulse the observable parts are
B t J J t I I I
B t J J t I I I I
y z z
x z z x
23 1 3 1 12 13 1 1 2 3
23 1 3 1 12 13 1 2 3 2 3
4
2 2
,
,
cos
cos
Ω Ω
Ω Ω
2
2
cos
sin
+ ( ) + ( )
+ ( ) + ( ) + ( )
π
π
The first term results in a multiplet appearing at Ω
1
in F
2
and at (Ω
2
+ Ω
3
) in F
1
.
The multiplet is doubly antiphase (with respect to the couplings to spins 2 and
3) in F
2
; in F
1
it is inphase with respect to the sum of the couplings J
12
and J
13
.
This multiplet is a remote peak, as its frequency coordinates do not conform to
the simple pattern described in section 7.5.2. It is distinguished from direct
peaks not only by its frequency coordinates, but also by having a different
lineshape in F
2
to direct peaks and by being doubly antiphase in that
dimension.
The second and third terms are antiphase with respect to the coupling
between spins 2 and 3, and if this coupling is zero there will be cancellation
within the multiplet and no signals will be observed. This is despite the fact
that multiplequantum coherence between these two spins has been generated.
7.6 Advanced topic:
Lineshapes and frequency discrimination
This is a somewhat involved topic which will only be possible to cover in
outline here.
2I
2z
I
3x
Ω
3
J
23
J
23
decreasing
J
23
= 0
J
13
Illustration of how the intensity
of an antiphase multiplet
decreases as the coupling
which it is in antiphase with
respect to decreases. The in
phase multiplet is shown at the
top, and below are three
versions of the antiphase
mul ti pl et for successi vel y
decreasing values of J
23
.
7–19
7.6.1 Onedimensional spectra
Suppose that a 90°(y) pulse is applied to equilibrium magnetization resulting in
the generation of pure xmagnetization which then precesses in the transverse
plane with frequency Ω. NMR spectrometers are set up to detect the x and y
components of this magnetization. If it is assumed (arbitrarily) that these
components decay exponentially with time constant T
2
the resulting signals,
S t S t
x y
( ) ( ) and , from the two channels of the detector can be written
S t t t T S t t t T
x y
( ) = − ( ) ( ) = − ( ) γ γ cos exp sin exp Ω Ω
2 2
where γ is a factor which gives the absolute intensity of the signal.
Usually, these two components are combined in the computer to give a
complex timedomain signal, S(t)
S t S t iS t
t i t t T
i t t T
x y
( ) = ( ) + ( )
= + ( ) − ( )
= ( ) − ( )
γ
γ
cos sin exp
exp exp
Ω Ω
Ω
2
2
[7.2]
The Fourier transform of S(t) is also a complex function, S(ω):
S FT S t
A iD
ω
γ ω ω
( ) = ( )
[ ]
= ( ) + ( ) { }
where A(ω) and D(ω) are the absorption and dispersion Lorentzian lineshapes:
A
T
D
T
T
ω
ω
ω
ω
ω
( ) =
− ( ) +
( ) =
− ( )
− ( ) +
1
1 1
2
2
2
2
2
2
2
Ω
Ω
Ω
These lineshapes are illustrated opposite. For NMR it is usual to display the
spectrum with the absorption mode lineshape and in this case this corresponds
to displaying the real part of S(ω).
7.6.1.1 Phase
Due to instrumental factors it is almost never the case that the real and
imaginary parts of S(t) correspond exactly to the x and ycomponents of the
magnetization. Mathematically, this is expressed by multiplying the ideal
function by an instrumental phase factor, φ
instr
S t i i t t T ( ) = ( ) ( ) − ( ) γ φ exp exp exp
instr
Ω
2
The real and imaginary parts of S(t) are
Re cos cos sin sin exp
Im cos sin sin cos exp
S t t t t T
S t t t t T
( )
[ ]
= − ( ) − ( )
( )
[ ]
= + ( ) − ( )
γ φ φ
γ φ φ
instr instr
instr instr
Ω Ω
Ω Ω
2
2
Clearly, these do not correspond to the x– and ycomponents of the ideal time
domain function.
The Fourier transform of S(t) carries forward the phase term
S i A iD ω γ φ ω ω ( ) = ( ) ( ) + ( ) { } exp
instr
All modern spectrometers use a
method know as quadrature
det ect i on, which in effect
means that both the x and y
c o mp o n e n t s o f t h e
magnetization are detected
simultaneously.
Ω
ω
ω
Absorpt i on (above) and
dispersion (below) Lorentzian
l i neshapes, cent r ed at
frequency Ω.
7–20
The real and imaginary parts of S(ω) are no longer the absorption and
dispersion signals:
Re cos sin
Im cos sin
S A D
S D A
ω γ φ ω φ ω
ω γ φ ω φ ω
( )
[ ]
= ( ) − ( ) ( )
( )
[ ]
= ( ) + ( ) ( )
instr instr
instr instr
Thus, displaying the real part of S(ω) will not give the required absorption
mode spectrum; rather, the spectrum will show lines which have a mixture of
absorption and dispersion lineshapes.
Restoring the pure absorption lineshape is simple. S(ω) is multiplied, in the
computer, by a phase correction factor, φ
corr
:
S i i i A iD
i A iD
ω φ γ φ φ ω ω
γ φ φ ω ω
( ) ( ) = ( ) ( ) ( ) + ( ) { }
= + ( ) ( )
( ) + ( ) { }
exp exp exp
exp
corr corr instr
corr instr
By choosing φ
corr
such that (φ
corr
+ φ
inst
) = 0 (i.e. φ
corr
= – φ
instr
) the phase terms
disappear and the real part of the spectrum will have the required absorption
lineshape. In practice, the value of the phase correction is set "by eye" until the
spectrum "looks phased". NMR processing software also allows for an
additional phase correction which depends on frequency; such a correction is
needed to compensate for, amongst other things, imperfections in
radiofrequency pulses.
7.6.1.2 Phase is arbitrary
Suppose that the phase of the 90° pulse is changed from y to x. T h e
magnetization now starts along –y and precesses towards x; assuming that the
instrumental phase is zero, the output of the two channels of the detector are
S t t t T S t t t T
x y
( ) = − ( ) ( ) = − − ( ) γ γ sin exp cos exp Ω Ω
2 2
The complex timedomain signal can then be written
S t S t iS t
t i t t T
i t i t t T
i i t t T
i i t t T
x y
( ) = ( ) + ( )
= − ( ) − ( )
− ( ) + ( ) − ( )
= − ( ) ( ) − ( )
=
( )
( ) − ( )
γ
γ
γ
γ φ
sin cos exp
cos sin exp
exp exp
exp exp exp
Ω Ω
Ω Ω
Ω
Ω
2
2
2
2 exp
Where φ
e x p
, the "experimental" phase, is –π /2 (recall that
exp cos sin i i φ φ φ ( ) = + , so that exp(–i π/2) = –i).
It is clear from the form of S(t) that this phase introduced by altering the
experiment (in this case, by altering the phase of the pulse) takes exactly the
same form as the instrumental phase error. It can, therefore, be corrected by
applying a phase correction so as to return the real part of the spectrum to the
absorption mode lineshape. In this case the phase correction would be π/2.
The Fourier transform of the original signal is
7–21
S i A iD
S D S A
ω γ ω ω
ω γ ω ω γ ω
( ) = − ( ) ( ) + ( ) { }
( )
[ ]
= ( ) ( )
[ ]
= − ( ) Re Im
Thus the real part shows the dispersion mode lineshape, and the imaginary part
shows the absorption lineshape. The 90° phase shift simply swaps over the real
and imaginary parts.
7.6.1.3 Relative phase is important
The conclusion from the previous two sections is that the lineshape seen in the
spectrum is under the control of the spectroscopist. It does not matter, for
example, whether the pulse sequence results in magnetization appearing along
the x or y axis (or anywhere in between, for that matter). It is always possible
to phase correct the spectrum afterwards to achieve the desired lineshape.
However, if an experiment leads to magnetization from different processes
or spins appearing along different axes, there is no single phase correction
which will put the whole spectrum in the absorption mode. This is the case in
the COSY spectrum (section 7.4.1). The terms leading to diagonalpeaks
appear along the xaxis, whereas those leading to crosspeaks appear along y.
Either can be phased to absorption, but if one is in absorption, one will be in
dispersion; the two signals are fundamentally 90° out of phase with one
another.
7.6.1.4 Frequency discrimination
Suppose that a particular spectrometer is only capable of recording one, say the
x, component of the precessing magnetization. The time domain signal will
then just have a real part (compare Eqn. [7.2] in section 7.6.1)
S t t t T ( ) = − ( ) γ cos exp Ω
2
Using the identity cos exp exp θ θ θ = ( ) + − ( ) ( )
1
2
i i this can be written
S t i t i t t T
i t t T i t t T
( ) = ( ) + ( )
[ ]
− ( )
= ( ) − ( ) + ( ) − ( )
1
2 2
1
2 2
1
2 2
γ
γ γ
exp exp – exp
exp exp exp – exp
Ω Ω
Ω Ω
7–22
The Fourier transform of the first term gives, in the real part, an absorption
mode peak at ω = +Ω; the transform of the second term gives the same but at ω
= –Ω.
Re[ ]
–
S A A ω γ γ ( ) = +
+
1
2
1
2
where A
+
represents an absorption mode Lorentzian line at ω = +Ω and A
–
represents the same at ω = –Ω; likewise, D
+
and D
–
represent dispersion mode
peaks at +Ω and –Ω, respectively.
This spectrum is said to lack frequency discrimination, in the sense that it
does not matter if the magnetization went round at +Ω or –Ω, the spectrum still
shows peaks at both +Ω and –Ω. This is in contrast to the case where both the
x and ycomponents are measured where one peak appears at either positive or
negative ω depending on the sign of Ω.
The lack of frequency discrimination is associated with the signal being
modulated by a cosine wave, which has the property that cos(Ωt) = cos(–Ωt),
as opposed to a complex exponential, exp(iΩt) which is sensitive to the sign of
Ω. In onedimensional spectroscopy it is virtually always possible to arrange
for the signal to have this desirable complex phase modulation, but in the case
of twodimensional spectra it is almost always the case that the signal
modulation in the t
1
dimension is of the form cos(Ωt
1
) and so such spectra are
not naturally frequency discriminated in the F
1
dimension.
Suppose now that only the ycomponent of the precessing magnetization
could be detected. The time domain signal will then be (compare Eqn. [3.2] in
section 7.6.1)
S t i t t T ( ) = − ( ) γ sin exp Ω
2
Using the identity sin exp exp θ θ θ = ( ) − − ( ) ( )
1
2i
i i this can be written
S t i t i t t T
i t t T i t t T
( ) = ( ) − ( )
[ ]
− ( )
= ( ) − ( ) − ( ) − ( )
1
2 2
1
2 2
1
2 2
γ
γ γ
exp exp – exp
exp exp exp – exp
Ω Ω
Ω Ω
and so
Re[ ]
–
S A A ω γ γ ( ) = −
+
1
2
1
2
This spectrum again shows two peaks, at ±Ω, but the two peaks have opposite
signs; this is associated with the signal being modulated by a sine wave, which
has the property that sin(–Ωt) = – sin(Ωt). If the sign of Ω changes the two
peaks swap over, but there are still two peaks. In a sense the spectrum is
frequency discriminated, as positive and negative frequencies can be
distinguished, but in practice in a spectrum with many lines with a range of
positive and negative offsets the resulting set of possibly cancelling peaks
would be impossible to sort out satisfactorily.
ω + 0 –
ω + 0 –
ω + 0 –
a
b
c
Spectrum a has peaks at
posi t i ve and negat i ve
frequencies and is frequency
discriminated. Spectrum b
r esul t s f r om a cosi ne
modulated timedomain data
set; each peak appears at both
posi t i ve and negat i ve
frequency, regardl ess of
whether its real offset is
positive or negative. Spectrum
c results from a sine modulated
data set; like b each peak
appears twice, but with the
added complication that one
peak is inverted. Spectra b and
c lack frequency discrimination
and are quite uninterpretable as
a result.
7–23
7.6.2 Twodimensional spectra
7.6.2.1 Phase and amplitude modulation
There are two basic types of timedomain signal that are found in two
dimensional experiments. The first is phase modulation, in which the evolution
in t
1
is encoded as a phase, i.e. mathematically as a complex exponential
S t t i t t T i t t T
1 2 1 1 1 2
1
2 2 2 2
2
, exp exp exp exp ( ) = ( ) −
( ) ( ) −
( )
( ) ( )
phase
γ Ω Ω
where Ω
1
and Ω
2
are the modulation frequencies in t
1
and t
2
respectively, and
T
2
1 ( )
and T
2
2 ( )
are the decay time constants in t
1
and t
2
respectively.
The second type is amplitude modulation, in which the evolution in t
1
is
encoded as an amplitude, i.e. mathematically as sine or cosine
S t t t T i t t T
S t t t T i t t T
c
s
( ) = ( ) −
( ) ( ) −
( )
( ) = ( ) −
( ) ( ) −
( )
( ) ( )
( ) ( )
γ
γ
cos exp exp exp
sin exp exp exp
Ω Ω
Ω Ω
1 1 1 2
1
2 2 2 2
2
1 1 1 2
1
2 2 2 2
2
Generally, twodimensional experiments produce amplitude modulation, indeed
all of the experiments analysed in this chapter have produced either sine or
cosine modulated data. Therefore most twodimensional spectra are
fundamentally not frequency discriminated in the F
1
dimension. As explained
above for onedimensional spectra, the resulting confusion in the spectrum is
not acceptable and steps have to be taken to introduce frequency discrimination.
It will turn out that the key to obtaining frequency discrimination is the
ability to record, in separate experiments, both sine and cosine modulated data
sets. This can be achieved by simply altering the phase of the pulses in the
sequence.
For example, consider the EXSY sequence analysed in section 7.2 . The
observable signal, at time t
2
= 0, can be written
1
1 1 1 1 1 2
− ( ) + f t I f t I
y y
cos cos Ω Ω
If, however, the first pulse in the sequence is changed in phase from x to y the
corresponding signal will be
− − ( ) − 1
1 1 1 1 1 2
f t I f t I
y y
sin sin Ω Ω
i.e. the modulation has changed from the form of a cosine to sine. In COSY
and DQF COSY a similar change can be brought about by altering the phase of
the first 90° pulse. In fact there is a general procedure for effecting this change,
the details of which are given in a later chapter.
7.6.2.2 Twodimensional lineshapes
The spectra resulting from twodimensional Fourier transformation of phase
and amplitude modulated data sets can be determined by using the following
Fourier pair
FT i t t T A iD exp exp Ω ( ) − ( )
[ ]
= ( ) + ( ) { }
2
ω ω
7–24
where A and D are the dispersion Lorentzian lineshapes described in section
7.6.1
Phase modulation
For the phase modulated data set the transform with respect to t
2
gives
S t i t t T A iD
1 2 1 1 1
1 2 2
2
, exp exp ω γ ( ) = ( ) −
( )
+
[ ]
( )
+
( )
+
( )
phase
Ω
where A
+
( ) 2
indicates an absorption mode line in the F
2
dimension at ω
2
= +Ω
2
and with linewidth set by T
2
2 ( )
; similarly D
+
( ) 2
is the corresponding dispersion
line.
The second transform with respect to t
1
gives
S A iD A iD ω ω γ
1 2
1 1 2 2
, ( ) = +
[ ]
+
[ ] +
( )
+
( )
+
( )
+
( )
phase
where A
+
( ) 1
indicates an absorption mode line in the F
1
dimension at ω
1
= +Ω
1
and with linewidth set by T
2
1 ( )
; similarly D
+
( ) 1
is the corresponding dispersion
line.
The real part of the resulting twodimensional spectrum is
Re , S A A D D ω ω γ
1 2
1 2 1 2
( )
[ ]
= −
( )
+
( )
+
( )
+
( )
+
( )
phase
This is a single line at (ω
1
,ω
2
) = (+Ω
1
,+Ω
2
) with the phasetwist lineshape,
illustrated below.
Pseudo 3D view and contour plot of the phasetwist lineshape.
The phasetwist lineshape is an inextricable mixture of absorption and
dispersion; it is a superposition of the double absorption and double dispersion
lineshape (illustrated in section 7.4.1). No phase correction will restore it to
pure absorption mode. Generally the phase twist is not a very desirable
lineshape as it has both positive and negative parts, and the dispersion
component only dies off slowly.
Cosine amplitude modulation
For the cosine modulated data set the transform with respect to t
2
gives
S t t t T A iD
c
1 2 1 1 1 2
1 2 2
, cos exp ω γ ( ) = ( ) −
( )
+
[ ]
( )
+
( )
+
( )
Ω
The cosine is then rewritten in terms of complex exponentials to give
7–25
S t i t i t t T A iD
1 2
1
2 1 1 1 1 1 2
1 2 2
, exp exp exp ω γ ( ) = ( ) + − ( )
[ ]
−
( )
+
[ ]
( )
+
( )
+
( )
c
Ω Ω
The second transform with respect to t
1
gives
S A iD A iD A iD ω ω γ
1 2
1
2
1 1 1 1 2 2
, ( ) = +
{ }
+ +
{ }
[ ]
+
[ ] +
( )
+
( )
−
( )
−
( )
+
( )
+
( )
c
where A
−
( ) 1
indicates an absorption mode line in the F
1
dimension at ω
1
= –Ω
1
and with linewidth set by T
2
1 ( )
; similarly D
–
1 ( )
is the corresponding dispersion
line.
The real part of the resulting twodimensional spectrum is
Re , S A A D D A A D D ω ω γ γ
1 2
1
2
1 2 1 2
1
2
1 2 1 2
( )
[ ]
= −
( )
+ −
( )
+
( )
+
( )
+
( )
+
( )
−
( )
+
( )
−
( )
+
( )
c
This is a two lines, both with the phasetwist lineshape; one is located at
(+Ω
1
,+Ω
2
) and the other is at (–Ω
1
,+Ω
2
). As expected for a data set which is
cosine modulated in t
1
the spectrum is symmetrical about ω
1
= 0.
A spectrum with a pure absorption mode lineshape can be obtained by
discarding the imaginary part of the time domain data immediately after the
transform with respect to t
2
; i.e. taking the real part of S t
c
1 2
,ω ( )
S t S t
t t T A
c c
1 2 1 2
1 1 1 2
1 2
, Re ,
cos exp
ω ω
γ
( ) = ( )
[ ]
= ( ) −
( )
( )
+
( )
Re
Ω
Following through the same procedure as above:
S t i t i t t T A
c
1 2
1
2 1 1 1 1 1 2
1 2
, exp exp exp ω γ ( ) = ( ) + − ( )
[ ]
−
( )
( )
+
( )
Re
Ω Ω
S A iD A iD A
c
ω ω γ
1 2
1
2
1 1 1 1 2
, ( ) = +
{ }
+ +
{ }
[ ] +
( )
+
( )
−
( )
−
( )
+
( )
Re
The real part of the resulting twodimensional spectrum is
Re ,
Re
S A A A A
c
ω ω γ γ
1 2
1
2
1 2 1
2
1 2
( )
[ ]
= +
+
( )
+
( )
−
( )
+
( )
This is two lines, located at (+Ω
1
,+Ω
2
) and (–Ω
1
,+Ω
2
), but in contrast to the
above both have the double absorption lineshape. There is still lack of
frequency discrimination, but the undesirable phasetwist lineshape has been
avoided.
Sine amplitude modulation
For the sine modulated data set the transform with respect to t
2
gives
S t t t T A iD
1 2 1 1 1 2
1 2 2
, sin exp ω γ ( ) = ( ) −
( )
+
[ ]
( )
+
( )
+
( )
s
Ω
The cosine is then rewritten in terms of complex exponentials to give
S t i t i t t T A iD
i 1 2
1
2 1 1 1 1 1 2
1 2 2
, exp exp exp ω γ ( ) = ( ) − − ( )
[ ]
−
( )
+
[ ]
( )
+
( )
+
( )
s
Ω Ω
The second transform with respect to t
1
gives
S A iD A iD A iD
i
ω ω γ
1 2
1
2
1 1 1 1 2 2
, ( ) = +
{ }
− +
{ }
[ ]
+
[ ] +
( )
+
( )
−
( )
−
( )
+
( )
+
( )
s
The imaginary part of the resulting twodimensional spectrum is
7–26
Im , S A A D D A A D D ω ω γ γ
1 2
1
2
1 2 1 2
1
2
1 2 1 2
( )
[ ]
= − −
( )
+ −
( )
+
( )
+
( )
+
( )
+
( )
−
( )
+
( )
−
( )
+
( )
s
This is two lines, both with the phasetwist lineshape but with opposite signs;
one is located at (+Ω
1
,+Ω
2
) and the other is at (–Ω
1
,+Ω
2
). As expected for a
data set which is sine modulated in t
1
the spectrum is antisymmetric about ω
1
=
0.
As before, a spectrum with a pure absorption mode lineshape can be
obtained by discarding the imaginary part of the time domain data immediately
after the transform with respect to t
2
; i.e. taking the real part of S t
1 2
,ω ( )
s
S t S t
t t T A
s s
1 2 1 2
1 1 1
1 2
2
, Re ,
sin exp
Re
ω ω
γ
( ) = ( )
[ ]
= ( ) −
( )
( )
+
( )
Ω
Following through the same procedure as above:
S t i t i t t T A
s
i 1 2
1
2 1 1 1 1 1
1 2
2
, exp exp exp
Re
ω γ ( ) = ( ) − − ( )
[ ]
−
( )
( )
+
( )
Ω Ω
S A iD A iD A
s
i
ω ω γ
1 2
1
2
1 1 1 1 2
,
Re
( ) = +
{ }
− +
{ }
[ ] +
( )
+
( )
−
( )
−
( )
+
( )
The imaginary part of the resulting twodimensional spectrum is
Im , S A A A A ω ω γ γ
1 2
1
2
1 2 1
2
1 2
( )
[ ]
= − +
+
( )
+
( )
−
( )
+
( )
s
Re
The two lines now have the pure absorption lineshape.
7.6.2.3 Frequency discrimination with retention of absorption lineshapes
It is essential to be able to combine frequency discrimination in the F
1
dimension with retention of pure absorption lineshapes. Three different ways
of achieving this are commonly used; each will be analysed here.
StatesHaberkornRuben method
The essence of the StatesHaberkornRuben (SHR) method is the observation
that the cosine modulated data set, processed as described in section 7.6.1.2,
gives two positive absorption mode peaks at (+Ω
1
,+Ω
2
) and (–Ω
1
,+Ω
2
), whereas
the sine modulated data set processed in the same way gives a spectrum in
which one peak is negative and one positive. Subtracting these spectra from
one another gives the required absorption mode frequency discriminated
spectrum (see the diagram below):
Re , Im , S S
A A A A A A A A
A A
ω ω ω ω
γ γ γ γ
γ
1 2 1 2
1
2
1 2
1
2
1 2
1
2
1 2
1
2
1 2
1 2
( )
[ ]
− ( )
[ ]
= +
[ ]
− − +
[ ]
=
+
( )
+
( )
−
( )
+
( )
+
( )
+
( )
−
( )
+
( )
+
( )
+
( )
c
Re
s
Re
7–27
In practice it is usually more convenient to achieve this result in the following
way, which is mathematically identical.
The cosine and sine data sets are transformed with respect to t
2
and the real
parts of each are taken. Then a new complex data set is formed using the
cosine data for the real part and the sine data for the imaginary part:
S t S t iS t
t t T A i t t T A
i t t T A
1 2 1 2 1 2
1 1 1
1 2
1 1 1
1 2
1 1 1
1 2
2 2
2
, , ,
cos exp sin exp
exp exp
ω ω ω
γ γ
γ
( ) = ( ) + ( )
= ( ) −
( )
+ ( ) −
( )
= ( ) −
( )
( )
+
( ) ( )
+
( )
( )
+
( )
SHR
Re Re
c s
Ω Ω
Ω
Fourier transformation with respect to t
1
gives a spectrum whose real part
contains the required frequency discriminated absorption mode spectrum
S A iD A
A A iD A
ω ω γ
γ
1 2
1 1 2
1 2 1 2
, ( ) = +
[ ]
= +
+
( )
+
( )
+
( )
+
( )
+
( )
+
( )
+
( )
SHR
MarionWüthrich or TPPI method
The idea behind the TPPI (time proportional phase incrementation) or
Marion–Wüthrich (MW) method is to arrange things so that all of the peaks
have positive offsets. Then, frequency discrimination would not be required as
there would be no ambiguity.
One simple way to make all offsets positive is to set the receiver carrier
frequency deliberately at the edge of the spectrum. Simple though this is, it is
not really a very practical method as the resulting spectrum would be very
inefficient in its use of data space and in addition offresonance effects
associated with the pulses in the sequence will be accentuated.
In the TPPI method the carrier can still be set in the middle of the spectrum,
but it is made to appear that all the frequencies are positive by phase shifting
systematically some of the pulses in the sequence in concert with the
incrementation of t
1
.
In section 7.2 it was shown that in the EXSY sequence the cosine
modulation in t
1
, cos(Ω
1
t
1
), could be turned into sine modulation, – sin(Ω
1
t
1
), by
shifting the phase of the first pulse by 90°. The effect of such a phase shift can
be represented mathematically in the following way.
Recall that Ω is in units of radians s
–1
, and so if t is in seconds Ωt is in
radians; Ωt can therefore be described as a phase which depends on time. It is
also possible to consider phases which do not depend on time, as was the case
for the phase errors considered in section 7.6.1.1
The change from cosine to sine modulation in the EXSY experiment can be
though of as a phase shift of the signal in t
1
. Mathematically, such a phase
shifted cosine wave is written as cos Ω
1 1
t + ( ) φ , where φ is the phase shift in
radians. This expression can be expanded using the well known formula
cos cos cos sin sin A B A B A B + ( ) = − to give
1
ω
2
ω
0
cosine
sine
difference
– +
1
Ω +
1
Ω –
2
Ω
1
Ω +
1
Ω –
1
Ω +
Illustration of the way in which
the SHR method achieves
frequency discrimination by
combining cosine and sine
modulated spectra.
7–28
cos cos cos sin sin Ω Ω Ω
1 1 1 1
t t t + ( ) = − φ φ φ
If the phase shift, φ, is π/2 radians the result is
cos cos cos sin sin
sin
Ω Ω Ω
Ω
1 1 1 1
1
2 2 2 t t t
t
+ ( ) = −
= −
π π π
In words, a cosine wave, phase shifted by π/2 radians (90°) is the same thing as
a sine wave. Thus, in the EXSY experiment the effect of changing the phase of
the first pulse by 90° can be described as a phase shift of the signal by 90°.
Suppose that instead of a fixed phase shift, the phase shift is made
proportional to t
1
; what this means is that each time t
1
is incremented the phase
is also incremented in concert. The constant of proportion between the time
dependent phase, φ(t
1
), and t
1
will be written ω
additional
φ ω t t
1 1
( ) =
additional
Clearly the units of ω
additional
are radians s
–1
, that is ω
additional
is a frequency. The
new timedomain function with the inclusion of this incrementing phase is thus
cos cos
cos
Ω Ω
Ω
1 1 1 1 1 1
1 1
t t t t
t
+ ( ) ( )
= + ( )
= + ( )
φ ω
ω
additional
additional
In words, the effect of incrementing the phase in concert with t
1
is to add a
frequency ω
additional
to all of the offsets in the spectrum. The TPPI method
utilizes this option of shifting all the frequencies in the following way.
In onedimensional pulseFourier transform NMR the free induction signal is
sampled at regular intervals ∆. After transformation the resulting spectrum
displays correctly peaks with offsets in the range –(SW/2) to +(SW/2) where SW
is the spectral width which is given by 1/∆ (this comes about from the Nyquist
theorem of data sampling). Frequencies outside this range are not represented
correctly.
Suppose that the required frequency range in the F
1
dimension is from
–(SW
1
/2) to +(SW
1
/2) (in COSY and EXSY this will be the same as the range in
F
2
). To make it appear that all the peaks have a positive offset, it will be
necessary to add (SW
1
/2) to all the frequencies. Then the peaks will be in the
range 0 to (SW
1
).
As the maximum frequency is now (SW
1
) rather than (SW
1
/2) the sampling
interval, ∆
1
, will have to be halved i.e. ∆
1
= 1/(2SW
1
) in order that the range of
frequencies present are represented properly.
The phase increment is ω
additional
t
1
, but t
1
can be written as n∆
1
for the nth
increment of t
1
. The required value for ω
additional
is 2π(SW
1
/2) , where the 2π is
to convert from frequency (the units of SW
1
) to rad s
–1
, the units of ω
additional
.
Putting all of this together ω
additional
t
1
can be expressed, for the nth increment as
0 +SW
1
/2 –SW
1
/2
0 +SW
1
a
b
c
0 +SW
1
–SW
1
Illustration of the TPPI method.
The normal spectrum is shown
in a, with peaks in the range
–SW/2 to +SW/2. Adding a
frequency of SW/2 to all the
peaks gives them all positive
offsets, but some, shown
dotted) will then fall outside the
spectral window – spectrum b.
If the spectral width is doubled
all peaks are represented
correctly – spectrum c.
7–29
ω π
π
π
additional
t
SW
n
SW
n
SW
n
1
1
1
1
1
2
2
2
2
1
2
2
=
( )
=
=
∆
The way in which the phase incrementation increases the frequency of the
cosine wave is shown below:
time
The open circles lie on a cosine wave, cos(Ω × n∆ ), where ∆ is the sampling interval and n runs 0, 1, 2 ...
The closed circles lie on a cosine wave in which an additional phase is incremented on each point i.e. the
function is cos(Ω × n∆+ n φ); here φ = π/8. The way in which this phase increment increases the frequency
of the cosine wave is apparent.
In words this means that each time t
1
is incremented, the phase of the signal
should also be incremented by 90°, for example by incrementing the phase of
one of the pulses. The way in which it can be decided which pulse to increment
will be described in a later chapter.
A data set from an experiment to which TPPI has been applied is simply
amplitude modulated in t
1
and so can be processed according to the method
described for cosine modulated data so as to obtain absorption mode lineshapes.
As the spectrum is symmetrical about F
1
= 0 it is usual to use a modified
Fourier transform routine which saves effort and space by only calculating the
positive frequency part of the spectrum.
Echo antiecho method
Few twodimensional experiments naturally produce phase modulated data sets,
but if gradient pulses are used for coherence pathway selection it is then quite
often found that the data are phase modulated. In one way this is an advantage,
as it means that no special steps are required to obtain frequency discrimination.
However, phase modulated data sets give rise to spectra with phasetwist
lineshapes, which are very undesirable. So, it is usual to attempt to use some
method to eliminate the phasetwist lineshape, while at the same time retaining
frequency discrimination.
The key to how this can be done lies in the fact that two kinds of phase
modulated data sets can usually be recorded. The first is called the Ptype or
t
1
= 0
t
2
x
t
1
= ∆
t
2
y
t
1
= 2∆
t
2
–x
t
1
= 3∆
t
2
–y
t
1
= 4∆
t
1
t
2
x
TPPI phase incrementation
applied to a COSY sequence.
The phase of the first pulse is
incremented by 90° each time t
1
is incremented.
7–30
antiecho spectrum
S t t i t t T i t t T
1 2 1 1 1
1
2 2 2
2
2 2
, exp exp exp exp ( ) = ( ) −
( ) ( ) −
( )
( ) ( )
P
γ Ω Ω
the "P" indicates positive, meaning here that the sign of the frequencies in F
1
and F
2
are the same.
The second data set is called the echo or Ntype
S t t i t t T i t t T
1 2 1 1 1
1
2 2 2
2
2 2
, exp – exp exp exp ( ) = ( ) −
( ) ( ) −
( )
( ) ( )
N
γ Ω Ω
the "N" indicates negative, meaning here that the sign of the frequencies in F
1
and F
2
are opposite. As will be explained in a later chapter in gradient
experiments it is easy to arrange to record either the P or Ntype spectrum.
The simplest way to proceed is to compute two new data sets which are the
sum and difference of the P and Ntype data sets:
1
2 1 2 1 2
1
2 1 1 1 1 1 2
1
2 2 2 2
2
1 1 1 2
1
2 2 2 2
2
S t t S t t
i t i t t T i t t T
t t T i t t T
, ,
exp exp exp exp exp
cos exp exp exp
( ) ( )
( ) ( )
( ) + ( )
[ ]
=
( ) + − ( )
[ ]
−
( ) ( ) −
( )
= ( ) −
( ) ( ) −
( )
P N
γ
γ
Ω Ω Ω
Ω Ω
11
2 1 2 1 2
1
2 1 1 1 1 1 2
1
2 2 2 2
2
1 1 1 2
1
2 2 2 2
2
i
i
S t t S t t
i t i t t T i t t T
t t T i t t T
, ,
exp exp exp exp exp
sin exp exp exp
( ) ( )
( ) (
( ) − ( )
[ ]
=
( ) − − ( )
[ ]
−
( ) ( ) −
( )
= ( ) −
( ) ( ) −
P N
γ
γ
Ω Ω Ω
Ω Ω
))
( )
These two combinations are just the cosine and sine modulated data sets that
are the inputs needed for the SHR method. The pure absorption spectrum can
therefore be calculated in the same way starting with these combinations.
7.6.2.4 Phase in twodimensional spectra
In practice there will be instrumental and other phase shifts, possibly in both
dimensions, which mean that the timedomain functions are not the idealised
ones treated above. For example, the cosine modulated data set might be
S t t t T i t i t T
c
( ) = + ( ) −
( )
+ ( ) −
( )
( ) ( )
γ φ φ cos exp exp exp Ω Ω
1 1 1 1
1
2 2 2 2
2
2 2
where φ
1
and φ
2
are the phase errors in F
1
and F
1
, respectively. Processing this
data set in the manner described above will not give a pure absorption
spectrum. However, it is possible to recover the pure absorption spectrum by
software manipulations of the spectrum, just as was described for the case of
onedimensional spectra. Usually, NMR data processing software provides
options for making such phase corrections to twodimensional data sets.
8–1
8 Relaxation
†
Relaxation is the process by which the spins in the sample come to equilibrium
with the surroundings. At a practical level, the rate of relaxation determines
how fast an experiment can be repeated, so it is important to understand how
relaxation rates can be measured and the factors that influence their values.
The rate of relaxation is influenced by the physical properties of the molecule
and the sample, so a study of relaxation phenomena can lead to information on
these properties. Perhaps the most often used and important of these
phenomena in the nuclear Overhauser effect (NOE) which can be used to probe
internuclear distances in a molecule. Another example is the use of data on
relaxation rates to probe the internal motions of macromolecules.
In this chapter the language and concepts used to describe relaxation will be
introduced and illustrated. To begin with it will simply be taken for granted
that there are processes which give rise to relaxation and we will not concern
ourselves with the source of relaxation. Having described the experiments
which can be used to probe relaxation we will then go on to see what the source
of relaxation is and how it depends on molecular parameters and molecular
motion.
8.1 What is relaxation?
Relaxation is the process by which the spins return to equilibrium. Equilibrium
is the state in which (a) the populations of the energy levels are those predicted
by the Boltzmann distribution and (b) there is no transverse magnetization and,
more generally, no coherences present in the system.
In Chapter 3 we saw that when an NMR sample is placed in a static
magnetic field and allowed to come to equilibrium it is found that a net
magnetization of the sample along the direction of the applied field
(traditionally the zaxis) is developed. Magnetization parallel to the applied
field is termed longitudinal.
This equilibrium magnetization arises from the unequal population of the
two energy levels that correspond to the α and β spin states. In fact, the z
magnetization, M
z
, is proportional to the population difference
M n n
z
∝ −
( )
α β
where n
α
and n
β
are the populations of the two corresponding energy levels.
Ultimately, the constant of proportion just determines the absolute size of the
signal we will observe. As we are generally interested in the relative size of
magnetizations and signals we may just as well write
M n n
z
−
( )
α β
[2]
†
Chapter 8 "Relaxation" © James Keeler 1999 and 2002
α
β
x y
z
A population difference gives
rise to net magnetization along
the zaxis.
8–2
8.2 Rate equations and rate constants
The populations of energy levels are in many ways analogous to concentrations
in chemical kinetics, and many of the same techniques that are used to describe
the rates of chemical reactions can also be used to describe the dynamics of
populations. This will lead to a description of the dynamics of the z
magnetization.
Suppose that the populations of the α and β states at time t are n
α
and n
β
,
respectively. If these are not the equilibrium values, then for the system to
reach equilibrium the population of one level must increase and that of the other
must decrease. This implies that there must be transitions between the two
levels i.e. something must happen which causes a spin to move from the α state
to the β state or vice versa. It is this process which results in relaxation.
The simplest assumption that we can make about the rate of transitions from
α to β is that it is proportional to the population of state α, n
α
, and is a first
order process with rate constant W. With these assumptions the rate of loss of
population from state α is Wn
α
. In the same way, the rate of loss of population
of state β is Wn
β
.
However, the key thing to realize is that whereas a transition from α to β
causes a loss of population of level α, a transition from β to α causes the
population of state α to increase. So we can write
rate of change of population of state α = – Wn
α
+ Wn
β
The first term is negative as it represents a loss of population of state α and in
contrast the second term is negative as it represents a gain in the population of
state α. The rate of change of the population can be written using the language
of calculus as dn
α
/dt so we have
d
d
n
t
Wn Wn
α
α β
− + .
Similarly we can write for the population of state β:
d
d
n
t
Wn Wn
β
β α
− + .
These equations are almost correct, but we need to make one modification.
At equilibrium the populations are not changing so dn
α
/dt = 0; this immediately
implies that at equilibrium n
α
= n
β
, which simply is not correct. We know that
at equilibrium the population of state α exceeds that of state β. This defect is
easily remedied by replacing the population n
α
with the deviation of the
population from its equilibrium value n n
α α
−
( )
0
, where n
α
0
is the population of
state α at equilibrium. Doing the same with state β gives us the final, correct,
equations:
d
d
d
d
n
t
W n n W n n
n
t
W n n W n n
α
β β α α
β
α α β β
−
( )
− −
( )
−
( )
− −
( )
0 0 0 0
.
α
β
W
α
β
W
A transition from state α to
st at e β decreases t he
population of state α, but a
transition from state β to state α
increases the population of
state α.
8–3
You will recognise here that this kind of approach is exactly the same as that
used to analyse the kinetics of a reversible chemical reaction.
Using Eqn. [2], we can use these two equations to work out how the z
magnetization varies with time:
d
d
d
d
d
d
d
d
M
t
n n
t
n
t
n
t
W n n W n n W n n W n n
W n n W n n
W M M
z
z z
−
( )
−
−
( )
− −
( )
− −
( )
+ −
( )
− −
( )
+ −
( )
− −
( )
α β
α
β
β β α α α α β β
α β α β
0 0 0 0
0 0
0
2 2
2
where M n n
z
0 0 0
−
( )
α β
, the equilibrium zmagnetization.
8.2.1 Consequences of the rate equation
The discussion in the previous section led to a (differential) equation describing
the motion of the zmagnetization
d
d
M t
t
R M t M
z
z z z
( )
− ( ) −
( )
0
[7]
where the rate constant, R
z
= 2W and M
z
has been written as a function of time,
M
z
(t), to remind us that it may change.
What this equations says is that the rate of change of M
z
is proportional to
the deviation of M
z
from its equilibrium value, M
z
0
. If M
z
= M
z
0
, that is the
system is at equilibrium, the righthand side of Eqn. [7] is zero and hence so is
the rate of change of M
z
: nothing happens. On the other hand, if M
z
deviates
from M
z
0
there will be a rate of change of M
z
, and this rate will be proportional
to the deviation of M
z
from M
z
0
. The change will also be such as to return M
z
to
its equilibrium value, M
z
0
. In summary, Eqn. [7] predicts that over time M
z
will
return to M
z
0
; this is exactly what we expect. The rate at which this happens
will depend on R
z
.
This equation can easily be integrated:
d
d
const.
M t
M t M
R t
M t M R t
z
z z
z
z z z
( )
( ) −
( )
−
( ) −
( )
− +
∫ ∫ 0
0
ln
If, at time zero, the magnetization is M
z
(0), the constant of integration can be
determined as ln M M
z z
0
0
( ) −
( )
. Hence, with some rearrangement:
8–4
ln
M t M
M M
R t
z z
z z
z
( ) −
( ) −
]
]
]
−
0
0
0
[7A]
or
M t M M R t M
z z z z z
( ) ( ) −
[ ]
− ( ) + 0
0 0
exp [7B]
In words, this says that the zmagnetization returns from M
z
(0) to M
z
0
following
an exponential law. The time constant of the exponential is 1/R
z
, and this is
often called T
1
, the longitudinal or spinlattice relaxation time.
8.2.2 The inversion recovery experiment
We described this experiment in section 3.10. First, a 180° pulse is applied,
thereby inverting the magnetization. Then a delay t is left for the magnetization
to relax. Finally, a 90° pulse is applied so that the size of the zmagnetization
can be measured.
We can now analyse this experiment fully. The starting condition for M
z
is
M M
z z
0
0
( ) − , i.e. inversion. With this condition the predicted time evolution
can be found from Eqn. [7A] to be:
ln
M t M
M
R t
z z
z
z
( ) −
−
]
]
]
−
0
0
2
Recall that the amplitude of the signal we record in this experiment is
proportional to the zmagnetization. So, if this signal is S(t) it follows that
ln
S t S
S
R t
z
( ) −
−
]
]
]
−
0
0
2
[7C]
where S
0
is the signal intensity from equilibrium magnetization; we would find
this from a simple 90° – acquire experiment.
Equation [7C] implies that a plot of ln S t S S ( ) −
( )
−
[ ]
0 0
2 against t should be
a straight line of slope –R
z
. This, then, is the basis of a method of determining
the relaxation rate constant.
8.2.3 A quick estimate for R
z
(or T
1
)
Often we want to obtain a quick estimate for the relaxation rate constant (or,
equivalently, the relaxation time). One way to do this is to do an inversion
recovery experiment but rather than varying t systematically we look for the
value of t which results in no signal i.e. a null. If the time when S(t) is zero is
t
null
it follows immediately from Eqn. [7C] that:
ln
ln
ln
1
2
2
2
1
]
]
]
− R t R
t
T
t
z z null
null
null
or or
Probably the most useful relationship is the last, which is T
1
≈ 1.4 t
null
.
This method is rather crude, but it good enough for estimating T
1
. Armed
with this estimate we can then, for example, decide on the time to leave
between transients (typically three to five times T
1
).
t
M
z
M
z
(0)
M
z
0
t
S(t)–S
0
–2S
0
ln
[ ]
Plot used to extract a value of
R
z
from the data from an
inversion recovery experiment.
8–5
8.2.4 Writing relaxation in terms of operators
As we saw in Chapter 6, in quantum mechanics zmagnetization is represented
by the operator I
z
. It is therefore common to write Eqn. [7] in terms of
operators rather then magnetizations, to give:
d
d
I t
t
R I t I
z
z z z
( )
− ( ) −
( )
0
[8]
where I
z
(t) represents the zmagnetization at time t and I
z
0
represents the
equilibrium zmagnetization. As it stands this last equation seems to imply that
the operators change with time, which is not what is meant. What are changing
are the populations of the energy levels and these in turn lead to changes in the
zmagnetization represented by the operator. We will use this notation from
now on.
8.3 Solomon equations
The idea of writing differential equations for the populations, and then
transcribing these into magnetizations, is a particularly convenient way of
describing relaxation, especially in more complex system. This will be
illustrated in this section.
Consider a sample consisting of molecules which contain two spins, I and S;
the spins are not coupled. As was seen in section 2.4, the two spins have
between them four energy levels, which can be labelled according to the spin
states of the two spins.
αα>
ββ>
αβ> βα>
1
3 2
4
W
I
(1)
W
I
(2)
W
S
(1)
W
S
(2)
W
2
W
0
αα>
ββ>
αβ> βα>
1
3 2
4
(a) (b)
Diagram (a) shows the energy levels of a two spin system; the levels are labelled with the spin of I first and
the spin of S second. The dashed arrows indicate allowed transitions of the I spin, and the solid arrows
indicate allowed transitions of the S spin. Diagram (b) shows the relaxation induced transitions which are
possible amongst the same set of levels.
It turns out that in such a system it is possible to have relaxation induced
transitions between all possible pairs of energy levels, even those transitions
which are forbidden in normal spectroscopy; why this is so will be seen in
detail below. The rate constants for the two allowed I spin transitions will be
denoted W
I
1 ( )
and W
I
2 ( )
, and likewise for the spin S transitions. The rate constant
for the transition between the αα and ββ states is denoted W
2
, the "2" indicating
that it is a double quantum transition. Finally, the rate constant for the
transition between the αβ and βα states is denoted W
0
, the "0" indicating that it
is a zero quantum transition.
8–6
Just in the same was as was done in Section 8.2, rate equations can be
written for the flow of population from any of the levels. For example, for level
1
d
d
n
t
W n W n W n W n W n W n
S I S I
1
1
1
1
1 2 1
1
2
1
3 2 4
− − − + + +
( ) ( ) ( ) ( )
The negative terms are rates which lead to a loss of population of level 1 and
the positive terms are ones that lead to a gain in its population. As was
discussed in section 8.2 the populations ought to be written as deviations from
their equilibrium values, n n
i i
−
( )
0
. However, to do this results in unnecessary
complexity; rather, the calculation will be carried forward as written and then at
the last stage the populations will be replaced by their deviations from
equilibrium.
The corresponding equations for the other populations are
d
d
d
d
d
d
n
t
W n W n W n W n W n W n
n
t
W n W n W n W n W n W n
n
t
W n W n
S I S I
I S I S
S I
2
1
2
2
2 0 2
1
1
2
4 0 3
3
1
3
2
3 0 3
1
1
2
4 0 2
4
2
4
2
− − − + + +
− − − + + +
− −
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( )
44 2 4
2
3
2
2 2 1
− + + +
( ) ( )
W n W n W n W n
S I
All of this can be expressed in a more compact way if we introduce the I and S
spin zmagnetizations. The I spin magnetization is equal to the population
difference across the two I spin transitions, 1–3 and 2–4
I n n n n
z
− + −
1 3 2 4
[9]
As discussed above, the magnetization has been represented as the
corresponding operator, I
z
. Likewise for the Sspin magnetization
S n n n n
z
− + −
1 2 3 4
[10]
A third combination of populations will be needed, which is represented by the
operator 2I
z
S
z
2
1 3 2 4
I S n n n n
z z
− − + [11]
Comparing this with Eq. [9] reveals that 2I
z
S
z
represents the difference in
population differences across the two Ispin transitions; likewise, comparison
with Eq. [10] shows that the same operator also represents the difference in
population differences across the two Sspin transitions.
Taking the derivative of Eq. [9] and then substituting for the derivatives of
the populations gives
8–7
d
d
d
d
d
d
d
d
d
d
z
I
t
n
t
n
t
n
t
n
t
W n W n W n W n W n W n
W n W n W n W n W n W n
W
S I S I
I S I S
S
− + −
− − − + + +
+ + + − − −
−
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
1 3 2 4
1
1
1
1 2 1
1
2
1
3 2 4
1
3
2
3 0 3
1
1
2
4 0 2
11
2
2
2 0 2
1
1
2
4 0 3
2
4
2
4 2 4
2
3
2
2 2 1
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
− − + + +
+ + + − − −
n W n W n W n W n W n
W n W n W n W n W n W n
I S I
S I S I
[12]
This unpromising looking equation can be expressed in terms of I
z
, S
z
etc. by
first introducing one more operator E, which is essentially the identity or unit
operator
E n n n n + + +
1 2 3 4
[13]
and then realizing that the populations, n
i
, can be written in terms of E, I
z
, S
z
,
and 2I
z
S
z
:
n E I S I S
n E I S I S
n E I S I S
n E I S I S
z z z z
z z z z
z z z z
z z z z
1
1
4
2
1
4
3
1
4
4
1
4
2
2
2
2
+ + + ( )
+ − − ( )
− + − ( )
− − + ( )
where these relationships can easily be verified by substituting back in the
definitions of the operators in terms of populations, Eqs. [9] – [13].
After some tedious algebra, the following differential equation is found for I
z
d
d
z
z
I
t
W W W W I
W W S W W I S
I I
z I I z z
− + + +
( )
− − ( ) − −
( )
( ) ( )
( ) ( )
1 2
2 0
2 0
1 2
2
[14]
Similar algebra gives the following differential equations for the other operators
d
d
d2
d
z
z
z
z
S
t
W W I W W W W S W W I S
I S
t
W W I W W S
W W W
z S S S S z z
z
I I z S S
I I S
− − ( ) − + + +
( )
− −
( )
− −
( )
− −
( )
− + +
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( )
2 0
1 2
2 0
1 2
1 2 1 2
1 2 1
2
++
( )
( )
W I S
S z z
2
2
As expected, the total population, represented by E, does not change with time.
These three differential equations are known as the Solomon equations.
It must be remembered that the populations used to derive these equations
are really the deviation of the populations from their equilibrium values. As a
result, the I and S spin magnetizations should properly be their deviations from
their equilibrium values, I
z
0
and S
z
0
; the equilibrium value of 2I
z
S
z
is easily
shown, from its definition, to be zero. For example, Eq. [14] becomes
8–8
d
d
z
z
z
I I
t
W W W W I I
W W S S W W I S
z
I I z
z I I z z
−
( )
− + + +
( )
−
( )
− − ( ) −
( )
− −
( )
( ) ( )
( ) ( )
0
1 2
2 0
0
2 0
0 1 2
2
8.3.1 Interpreting the Solomon equations
What the Solomon equations predict is, for example, that the rate of change of
I
z
depends not only on I I
z z
−
0
, but also on S S
z z
−
0
and 2I
z
S
z
. In other words the
way in which the magnetization on the I spin varies with time depends on what
is happening to the S spin – the two magnetizations are connected. This
phenomena, by which the magnetizations of the two different spins are
connected, is called cross relaxation.
The rate at which S magnetization is transferred to I magnetization is given
by the term
W W S S
z 2 0
0
− ( ) −
( )
z
in Eq. [14]; (W
2
–W
0
) is called the crossrelaxation rate constant, and is
sometimes given the symbol σ
IS
. It is clear that in the absence of the relaxation
pathways between the αα and ββ states (W
2
), or between the αβ and βα states
(W
0
), there will be no cross relaxation. This term is described as giving rise to
transfer from S to I as it says that the rate of change of the I spin magnetization
is proportional to the deviation of the S spin magnetization from its equilibrium
value. Thus, if the S spin is not at equilibrium the I spin magnetization is
perturbed.
In Eq. [14] the term
W W W W I I
I I z
1 2
2 0
0 ( ) ( )
+ + +
( )
−
( )
z
describes the relaxation of I spin magnetization on its own; this is sometimes
called the self relaxation. Even if W
2
and W
0
are absent, self relaxation still
occurs. The self relaxation rate constant, given in the previous equation as a
sum of W values, is sometimes given the symbol R
I
or ρ
I
.
Finally, the term
W W I S
I I z z
1 2
2
( ) ( )
−
( )
in Eq. [14] describes the transfer of I
z
S
z
into I spin magnetization. Recall that
W
I
1 ( )
and W
I
2 ( )
are the relaxation induced rate constants for the two allowed
transitions of the I spin (1–3 and 2–4). Only if these two rate constants are
different will there be transfer from 2I
z
S
z
into I spin magnetization. This
situation arises when there is crosscorrelation between different relaxation
mechanisms; a further discussion of this is beyond the scope of these lectures.
The rate constants for this transfer will be written
∆ ∆
I I I S S S
W W W W −
( )
−
( )
( ) ( ) ( ) ( ) 1 2 1 2
According to the final Solomon equation, the operator 2I
z
S
z
shows self
relaxation with a rate constant
8–9
R W W W W
IS I I S S
+ + +
( )
( ) ( ) ( ) ( ) 1 2 1 2
Note that the W
2
and W
0
pathways do not contribute to this. This rate combined
constant will be denoted R
IS
.
Using these combined rate constants, the Solomon equations can be written
d
d
d
d
d2
d
z
z z
z
z z
z
z z
I I
t
R I I S S I S
S S
t
I I R S S I S
I S
t
I I S S R
z
I z IS z I z z
z
IS z S z S z z
z
I z S z IS
−
( )
− −
( )
− −
( )
−
−
( )
− −
( )
− −
( )
−
− −
( )
− −
( )
−
0
0 0
0
0 0
0 0
2
2
σ
σ
∆
∆
∆ ∆ 22I S
z z
[15]
The pathways between the different magnetization are visualized in the diagram
opposite. Note that as d d I t
z
0
0 (the equilibrium magnetization is a constant),
the derivatives on the lefthand side of these equations can equally well be
written d d I t
z
and d d S t
z
.
It is important to realize that in such a system I
z
and S
z
do not relax with a
simple exponentials. They only do this if the differential equation is of the
form
d
d
I
t
R I I
z
I z z
− −
( )
0
which is plainly not the case here. For such a twospin system, therefore, it is
not proper to talk of a "T
1
" relaxation time constant.
8.4 Nuclear Overhauser effect
The Solomon equations are an excellent way of understanding and analysing
experiments used to measure the nuclear Overhauser effect. Before embarking
on this discussion it is important to realize that although the states represented
by operators such as I
z
and S
z
cannot be observed directly, they can be made
observable by the application of a radiofrequency pulse, ideally a 90° pulse
aI aI
z
I
y
x
π 2 ( )
÷ → ÷÷ −
The subsequent recording of the free induction signal due to the evolution of
the operator I
y
will give, after Fourier transformation, a spectrum with a peak of
size –a at frequency Ω
I
. In effect, by computing the value of the coefficient a,
the appearance of the subsequently observed spectrum is predicted.
The basis of the nuclear Overhauser effect can readily be seen from the
Solomon equation (for simplicity, it is assumed in this section that ∆
I
= ∆
S
= 0)
d
d
z
z z
I I
t
R I I S S
z
I z IS z
−
( )
− −
( )
− −
( )
0
0 0
σ
What this says is that if the S spin magnetization deviates from equilibrium
there will be a change in the I spin magnetization at a rate proportional to (a)
the crossrelaxation rate, σ
IS
and (b) the extent of the deviation of the S spin
σ
IS
∆
I
∆
S
2I
z
S
z
I
z
S
z
8–10
from equilibrium. This change in the I spin magnetization will manifest itself
as a change in the intensity in the corresponding spectrum, and it is this change
in intensity of the I spin when the S spin is perturbed which is termed the
nuclear Overhauser effect.
Plainly, there will be no such effect unless σ
IS
is nonzero, which requires the
presence of the W
2
and W
0
relaxation pathways. It will be seen later on that
such pathways are only present when there is dipolar relaxation between the
two spins and that the resulting crossrelaxation rate constants have a strong
dependence on the distance between the two spins. The observation of a
nuclear Overhauser effect is therefore diagnostic of dipolar relaxation and
hence the proximity of pairs of spins. The effect is of enormous value,
therefore, in structure determination by NMR.
8.4.1 Transient experiments
A simple experiment which reveals the NOE is to invert just the S spin by
applying a selective 180° pulse to its resonance. The S spin is then not at
equilibrium so magnetization is transferred to the I spin by crossrelaxation.
After a suitable period, called the mixing time, τ
m
, a nonselective 90° pulse is
applied and the spectrum recorded.
After the selective pulse the situation is
I I S S
z z z z
0 0
0 0
( ) ( ) − [16]
where I
z
has been written as I
z
(t) to emphasize that it depends on time and
likewise for S. To work out what will happen during the mixing time the
differential equations
d
d
d
d
z
z z
z
z z
I t
t
R I t I S t S
S t
t
I t I R S t S
I z IS z
IS z S z
( )
− ( ) −
( )
− ( ) −
( )
( )
− ( ) −
( )
− ( ) −
( )
0 0
0 0
σ
σ
need to be solved (integrated) with this initial condition. One simple way to do
this is to use the initial rate approximation. This involves assuming that the
mixing time is sufficiently short that, on the righthand side of the equations, it
can be assumed that the initial conditions set out in Eq. [16] apply, so, for the
first equation
d
d
z
init
I t
t
R I I S S
S
I z z IS z z
IS z
( )
− −
( )
− − −
( )
0 0 0 0
0
2
σ
σ
This is now easy to integrate as the righthand side has no dependence on I
z
(t)
m
τ
90° 180°
S
90°
(a)
(b)
Pulse sequence for recording
transient NOE enhancements.
Sequence (a) involves selective
inversion of the S spin – shown
here using a shaped pulse.
Sequence (b) is used to record
the reference spectrum in
which the intensities are
unperturbed.
8–11
d d
z
z m z m
z m m
m m
I t S t
I I S
I S I
IS z
IS z
IS z z
( )
( ) − ( )
( ) +
∫ ∫
0
0
0
0
0 0
2
0 2
2
τ τ
σ
τ σ τ
τ σ τ
This says that for zero mixing time the I magnetization is equal to its
equilibrium value, but that as the mixing time increases the I magnetization has
an additional contribution which is proportional to the mixing time and the
crossrelaxation rate, σ
IS
. This latter term results in a change in the intensity of
the I spin signal, and this change is called an NOE enhancement.
The normal procedure for visualizing these enhancements is to record a
reference spectrum in which the intensities are unperturbed. In terms of z
magnetizations this means that I I
z z , ref
0
. The difference spectrum, defined as
(perturbed spectrum – unperturbed spectrum) corresponds to the difference
I I S I I
S
z IS z z z
IS z
z m ref m
m
τ σ τ
σ τ
( ) − + −
,
2
2
0 0 0
0
The NOE enhancement factor, η, is defined as
η
intensity in enhanced spectrum  intensity in reference spectrum
intensity in reference spectrum
so in this case η is
η τ
τ σ τ
m
z m ref
ref
m
( )
( ) −
I I
I
S
I
z
z
IS z
z
,
,
2
0
0
and if I and S are of the same nuclear species ( e.g. both proton), their
equilibrium magnetizations are equal so that
η τ σ τ
m m
( ) 2
IS
Hence a plot of η against mixing time will give a straight line of slope σ
IS
; this
is a method used for measuring the crossrelaxation rate constant. A single
experiment for one value of the mixing time will reveal the presence of NOE
enhancements.
This initial rate approximation is valid provided that
σ τ τ
IS S
R
m m
and << << 1 1
the first condition means that there is little transfer of magnetization from S to I,
and the second means that the S spin remains very close to complete inversion.
8.4.1.1 Advanced topic: longer mixing times
At longer mixing times the differential equations are a little more difficult to
solve, but they can be integrated using standard methods (symbolic
mathematical programmes such as Mathematica are particularly useful for this).
Using the initial conditions given in Eq. [16] and, assuming for simplicity that
I S
z z
0 0
the following solutions are found
enhanced
reference
difference
S I
Visualization of how an NOE
di f f er ence spect r um i s
recorded. The enhancement is
assumed to be positive.
8–12
I
I R
z
z
IS
τ σ
λ τ λ τ
m
m m
( )
− ( ) − − ( )
[ ]
+
0 2 1
2
1 exp exp
S
I
R R
R
z
z
I S
τ
λ τ λ τ
λ τ λ τ
m
m m
m m
( )
−
]
]
]
− ( ) − − ( )
[ ]
+ − − ( ) + − ( )
0 1 2
1 2
1
exp exp
exp exp
where
R R R R R
R R R R R R
I I S S IS
I S I S
− + +
+ +
[ ]
+ −
[ ]
2 2 2
1
1
2 2
1
2
2 4σ
λ λ
These definitions ensure that λ
1
> λ
2
. If R
I
and R
S
are not too dissimilar, R is of
the order of σ
IS
, and so the two rate constants λ
1
and λ
2
differ by a quantity of
the order of σ
IS
.
As expected for these two coupled differential equations, integration gives a
time dependence which is the sum of two exponentials with different time
constants.
The figure below shows the typical behaviour predicted by these equations
(the parameters are R
I
= R
S
= 5σ
IS
)
0
1.0
0.5
0.0
0.5
1.0
1.5
time
(S
z
/I
z
0
)
10 × (I
z
–I
z
0
)/I
z
0
(I
z
/I
z
0
)
The S spin magnetization returns to its equilibrium value with what appears to
be an exponential curve; in fact it is the sum of two exponentials but their time
constants are not sufficiently different for this to be discerned. The I spin
magnetization grows towards a maximum and then drops off back towards the
equilibrium value. The NOE enhancement is more easily visualized by plotting
the difference magnetization, (I
z
– I
z
0
)/I
z
0
, on an expanded scale; the plot now
shows the positive NOE enhancement reaching a maximum of about 15%.
Differentiation of the expression for I
z
as a function of τ
m
shows that the
maximum enhancement is reached at time
τ
λ λ
λ
λ
m, max
−
1
1 2
1
2
ln
and that the maximum enhancement is
8–13
I I
I R
z z
z
IS
R R
τ
σ λ
λ
λ
λ
λ λ
m, max
( )
−

(
'
`
J
J
−

(
'
`
J
J
]
]
]
]
]
− −
0
0
1
2
1
2
2
1 2
8.4.2 The DPFGSE NOE experiment
From the point of view of the relaxation behaviour the DPFGSE experiment is
essentially identical to the transient NOE experiment. The only difference is
that the I spin starts out saturated rather than at equilibrium. This does not
influence the build up of the NOE enhancement on I. It does, however, have
the advantage of reducing the size of the I spin signal which has to be removed
in the difference experiment. Further discussion of this experiment is deferred
to Chapter 9.
8.4.3 Steady state experiments
The steadystate NOE experiment involves irradiating the S spin with a
radiofrequency field which is sufficiently weak that the I spin is not affected.
The irradiation is applied for long enough that the S spin is saturated, meaning
S
z
= 0, and that the steady state has been reached, which means that none of the
magnetizations are changing, i.e. d d I t
z
( ) 0.
Under these conditions the first of Eqs. [15] can be written
d
d
z
SS
z,SS
I I
t
R I I S
z
I z IS z
−
( )
− −
( )
− −
( )
0
0 0
0 0 σ
therefore
I
R
S I
IS
I
z z z,SS
+
σ
0 0
As in the transient experiment, the NOE enhancement is revealed by
subtracting a reference spectrum which has equilibrium intensities. The NOE
enhancement, as defined above, will be
η
σ
SS
z,SS ref
ref
−
I I
I R
S
I
z
z
IS
I
z
z
,
,
0
0
90°
S
90°
(a)
(b)
Pulse sequence for recording
s t e a d y s t a t e NOE
enhancements. Sequence (a)
involves selective irradiation of
the S spin leading to saturation.
Sequence (b) is used to record
the reference spectrum in
which the intensities are
unperturbed.
8–14
In contrast to the transient experiment, the steady state enhancement only
depends on the relaxation of the receiving spin (here I); the relaxation rate of
the S spin does not enter into the relationship simply because this spin is held
saturated during the experiment.
It is important to realise that the value of the steadystate NOE enhancement
depends on the ratio of crossrelaxation rate constant to the self relaxation rate
constant for the spin which is receiving the enhancement. If this spin is
relaxing quickly, for example as a result of interaction with many other spins,
the size of the NOE enhancement will be reduced. So, although the size of the
enhancement does depend on the crossrelaxation rate constant the size of the
enhancement cannot be interpreted in terms of this rate constant alone. This
point is illustrated by the example in the margin.
8.4.4 Advanced topic: NOESY
The dynamics of the NOE in NOESY are very similar to those for the transient
NOE experiment. The key difference is that instead of the magnetization of the
S spin being inverted at the start of the mixing time, the magnetization has an
amplitude label which depends on the evolution during t
l
.
Starting with equilibrium magnetization on the I and S spins, the z
magnetizations present at the start of the mixing time are (other magnetization
will be rejected by appropriate phase cycling)
S t S I t I
z S z z I z
0 0
1
0
1
0
( ) − ( ) − cos cos Ω Ω
The equation of motion for S
z
is
d
d
z
z z
S t
t
I t I R S t S
IS z S z
( )
− ( ) −
( )
− ( ) −
( )
σ
0 0
As before, the initial rate approximation will be used:
d
d
z m
init
S
t
t I I R t S S
t I R t S
IS I z z S S z z
IS I z S S z
τ
σ
σ
( )
− − −
( )
− − −
( )
+ ( ) + + ( )
cos cos
cos cos
Ω Ω
Ω Ω
1
0 0
1
0 0
1
0
1
0
1 1
Integrating gives
d d
z
z m z m m
z m m
m m
S t t I R t S t
S S t I R t S
S t I
IS I z S S z
IS I z S S z
IS I z
( ) + ( ) + + ( )
[ ]
( ) − ( ) + ( ) + + ( )
( ) + ( )
∫ ∫
0
1
0
1
0
0
1
0
1
0
1
1 1
0 1 1
1
τ τ
σ
τ σ τ τ
τ σ τ
cos cos
cos cos
cos
Ω Ω
Ω Ω
Ω
00
1
0
1
0
0 0
1
0
1
0
1
1
+ + ( ) −
+
+
[ ]
+ −
[ ]
R t S t S
I R S
t I
t R S
S S z S z
IS z S z
I IS z
S S z
τ
σ τ τ
σ τ
τ
m
m m
m
m
{a}
{b}
{c}
cos cos
cos
cos
Ω Ω
Ω
Ω
After the end of the mixing time, this zmagnetization on spin S is rendered
H
A
H
B
H
C
H
D
X
Y
Irradiation of proton B gives a
much larger enhancement on
proton A than on C despite the
fact that the distances to the
two spins are equal. The
smaller enhancement on C is
due to the fact that it is relaxing
more quickly than A, due to the
interaction with proton D.
t
1
t
2
τ
mix
Pulse sequence for NOESY.
8–15
observable by the final 90° pulse; the magnetization is on spin S, and so will
precess at Ω
S
during t
2
.
The three terms {a}, {b} and {c} all represent different peaks in the NOESY
spectrum.
Term {a} has no evolution as a function of t
1
and so will appear at F
1
= 0; in
t
2
it evolves at Ω
S
. This is therefore an axial peak at {F
1
,F
2
} = {0, Ω
S
}. This
peak arises from zmagnetization which has recovered during the mixing time.
In this initial rate limit, it is seen that the axial peak is zero for zero mixing time
and then grows linearly depending on R
S
and σ
IS
.
Term {b} evolves at Ω
I
during t
1
and Ω
S
during t
2
; it is therefore a cross peak
at {Ω
I
, Ω
S
}. The intensity of the cross peak grows linearly with the mixing time
and also depends on σ
IS
; this is analogous to the transient NOE experiment.
Term {c} evolves at Ω
S
during t
1
and Ω
S
during t
2
; it is therefore a diagonal
peak at {Ω
S
, Ω
S
} and as R
s
τ
m
<< 1 in the initial rate, this peak is negative. The
intensity of the peak grows back towards zero linearly with the mixing time and
at a rate depending on R
S
. This peak arises from S spin magnetization which
remains on S during the mixing time, decaying during that time at a rate
determined by R
S
.
If the calculation is repeated using the differential equation for I
z
a
complimentary set of peaks at {0, Ω
I
}, {Ω
S
, Ω
I
} and {Ω
I
, Ω
I
} are found.
It will be seen later that whereas R
I
and R
S
are positive, σ
IS
can be either
positive or negative. If σ
IS
is positive, the diagonal and cross peaks will be of
opposite sign, whereas if σ
IS
is negative all the peaks will have the same sign.
8.4.5 Sign of the NOE enhancement
We see that the time dependence and size of the NOE enhancement depends on
the relative sizes of the crossrelaxation rate constant σ
IS
and the self relaxation
rate constants R
I
and R
S
. It turns out that these selfrates are always positive,
but the crossrelaxation rate constant can be positive or negative. The reason
for this is that σ
IS
= (W
2
– W
0
) and it is quite possible for W
0
to be greater or less
than W
2
.
A positive crossrelaxation rate constant means that if spin S deviates from
equilibrium crossrelaxation will increase the magnetization on spin I. This
leads to an increase in the signal from I, and hence a positive NOE
enhancement. This situation is typical for small molecules is nonviscous
solvents.
A negative crossrelaxation rate constant means that if spin S deviates from
equilibrium crossrelaxation will decrease the magnetization on spin I. This
leads to a negative NOE enhancement, a situation typical for large molecules in
viscous solvents. Under some conditions W
0
and W
2
can become equal and then
the NOE enhancement goes to zero.
I
Ω
I
Ω
S
Ω
S
Ω
F
1
F
2
{a}
{b}
{c}
0
8–16
8.5 Origins of relaxation
We now turn to the question as to what causes relaxation. Recall from section
8.1 that relaxation involves transitions between energy levels, so what we seek
is the origin of these transitions. We already know from Chapter 3 that
transitions are caused by transverse magnetic fields (i.e. in the xyplane) which
are oscillating close to the Larmor frequency. An RF pulse gives rise to just
such a field.
However, there is an important distinction between the kind of transitions
caused by RF pulses and those which lead to relaxation. When an RF pulse is
applied all of the spins experience the same oscillating field. The kind of
transitions which lead to relaxation are different in that the transverse fields are
local, meaning that they only affect a few spins and not the whole sample. In
addition, these fields vary randomly in direction and amplitude. In fact, it is
precisely their random nature which drives the sample to equilibrium.
The fields which are responsible for relaxation are generated within the
sample, often due to interactions of spins with one another or with their
environment in some way. They are made time varying by the random motions
(rotations, in particular) which result from the thermal agitation of the
molecules and the collisions between them. Thus we will see that NMR
relaxation rate constants are particularly sensitive to molecular motion.
If the spins need to lose energy to return to equilibrium they give this up to
the motion of the molecules. Of course, the amounts of energy given up by the
spins are tiny compared to the kinetic energies that molecules have, so they are
hardly affected. Likewise, if the spins need to increase their energy to go to
equilibrium, for example if the population of the β state has to be increased, this
energy comes from the motion of the molecules.
Relaxation is essentially the process by which energy is allowed to flow
between the spins and molecular motion. This is the origin of the original name
for longitudinal relaxation: spinlattice relaxation. The lattice does not refer to
a solid, but to the motion of the molecules with which energy can be
exchanged.
8.5.1 Factors influencing the relaxation rate constant
The detailed theory of the calculation of relaxation rate constants is beyond the
scope of this course. However, we are in a position to discuss the kinds of
factors which influence these rate constants.
Let us consider the rate constant W
ij
for transitions between levels i and j;
this turns out to depend on three factors:
W A Y J
ij ij ij
× ×
( )
ω
We will consider each in turn.
The spin factor, A
ij
This factor depends on the quantum mechanical details of the interaction. For
8–17
example, not all oscillating fields can cause transitions between all levels. In a
two spin system the transition between the αα and ββ cannot be brought about
by a simple oscillating field in the transverse plane; in fact it needs a more
complex interaction that is only present in the dipolar mechanism (section
8.6.2). We can think of A
ij
as representing a kind of selection rule for the
process – like a selection rule it may be zero for some transitions.
The size factor, Y
This is just a measure of how large the interaction causing the relaxation is. Its
size depends on the detailed origin of the random fields and often it is related to
molecular geometry.
The spectral density, J(ω
ij
)
This is a measure of the amount of molecular motion which is at the correct
frequency, ω
ij
, to cause the transitions. Recall that molecular motion is the
effect which makes the random fields vary with time. However, as we saw
with RF pulses, the field will only have an effect on the spins if it is oscillating
at the correct frequency. The spectral density is a measure of how much of the
motion is present at the correct frequency.
8.5.2 Spectral densities and correlation functions
The value of the spectral density, J(ω), has a large effect on relaxation rate
constants, so it is well worthwhile spending some time in understanding the
form that this function takes.
Correlation functions
To make the discussion concrete, suppose that a spin in a sample experiences
a magnetic field due to a dissolved paramagnetic species. The size of the
magnetic field will depend on the relative orientation of the spin and the
paramagnetic species, and as both are subject to random thermal motion, this
orientation will vary randomly with time (it is said to be a random function of
time), and so the magnetic field will be a random function of time. Let the field
experienced by this first spin be F
1
(t).
Now consider a second spin in the sample. This also experiences a random
magnetic field, F
2
(t), due to the interaction with the paramagnetic species. At
any instant, this random field will not be the same as that experienced by the
first spin.
For a macroscopic sample, each spin experiences a different random field,
F
i
(t). There is no way that a detailed knowledge of each of these random fields
can be obtained, but in some cases it is possible to characterise the overall
behaviour of the system quite simply.
The average field experienced by the spins is found by taking the ensemble
average – that is adding up the fields for all members of the ensemble (i.e. all
spins in the system)
Paramagnetic species have
unpaired electrons. These
generate magnetic fields which
can interact with nearby nuclei.
On account of the large
gyromagnetic ratio of the
electron (when compared to the
nucleus) such paramagnetic
species are often a significant
source of relaxation.
8–18
F t F t F t F t ( ) ( ) + ( ) + ( ) +
1 2 3
K
For random thermal motion, this ensemble average turns out to be independent
of the time; this is a property of stationary random functions. Typically, the
F
i
(t) are signed quantities, randomly distributed about zero, so this ensemble
average will be zero.
An important property of random functions is the correlation function,
G(t,τ), defined as
G t F t F t F t F t F t F t
F t F t
,
* * *
*
τ τ τ τ
τ
( ) ( ) + ( ) + ( ) + ( ) + ( ) + ( ) +
( ) + ( )
1 1 2 2 3 3
K
F
1
(t) is the field experienced by spin 1 at time t, and F
1
(t+τ) is the field
experienced at a time τ later; the star indicates the complex conjugate, which
allows for the possibility that F(t) may be complex. If the time τ is short the
spins will not have moved very much and so F
1
(t+τ) will be very little different
from F
1
(t). As a result, the product F t F t
1 1
( ) + ( )
*
τ will be positive. This is
illustrated in the figure below, plot (b).
5 1.2 5 1.2 5 1.2
F(t) F(t)F(t+1) F(t)F(t+15)
(a) (b) (c)
(a) is a plot of the random function F(t) against time; there are about 100 separate time points. (b) is a plot
of the value of F multiplied by its value one data point later – i.e. one data point to the right; all possible pairs
are plotted. (c) is the same as (b) but for a time interval of 15 data points. The two arrows indicate the
spacing over which the correlation is calculated.
The same is true for all of the other members of then ensemble, so when the
F t F t
i i
( ) + ( )
*
τ are added together for a particular time, t, – that is, the ensemble
average is taken – the result will be for them to reinforce one another and hence
give a finite value for G(t,τ).
As τ gets longer, the spin will have had more chance of moving and so
F
1
(t+τ) will differ more and more from F
1
(t); the product F t F t
1 1
( ) + ( )
*
τ need
not necessarily be positive. This is illustrated in plot (c) above. The ensemble
average of all these F t F t
i i
( ) + ( )
*
τ is thus less than it was when τ was shorter.
In the limit, once τ becomes sufficiently long, the F t F t
i i
( ) + ( )
*
τ are randomly
distributed and their ensemble average, G(t,τ), goes to zero. G(t,τ) thus has its
maximum value at τ = 0 and then decays to zero at long times. For stationary
random functions, the correlation function is independent of the time t; it will
therefore be written G(τ).
The correlation function, G(τ), is thus a function which characterises the
memory that the system has of a particular arrangement of spins in the sample.
For times τ which are much less than the time it takes for the system to
rearrange itself G(τ) will be close to its maximum value. As time proceeds, the
initial arrangement becomes more and more disturbed, and G(τ) falls. For
a
b
c
d
Visualization of the different
timescales for random motion.
(a) is the starting position: the
black dots are spins and the
open circle represents a
paramagnetic species. (b) is a
snap shot a very short time
after (a); hardly any of the spins
have moved. (c) is a snapshot
at a longer time; more spins
have moved, but part of the
ori gi nal pat t ern i s st i l l
discernible. (d) is after a long
time, all the spins have moved
and the original pattern is lost.
8–19
sufficiently long times, G(τ) tends to zero.
The simplest form for G(τ) is
G G τ τ τ ( ) ( ) − ( ) 0 exp
c
[18]
the variable τ appears as the modulus, resulting in the same value of G(τ) for
positive and negative values of τ. This means that the correlation is the same
with time τ before and time τ after the present time.
τ
c
is called the correlation time. For times much less than the correlation
time the spins have not moved much and the correlation function is close to its
original value; when the time is of the order of τ
c
significant rearrangements
have taken place and the correlation function has fallen to about half its initial
value. For times much longer than τ
c
the spins have moved to completely new
positions and the correlation function has fallen close to zero.
Spectral densities
The correlation function is a function of time, just like a free induction decay.
So, it can be Fourier transformed to give a function of frequency. The resulting
frequency domain function is called the spectral density; as the name implies,
the spectral density gives a measure of the amount of motion present at
different frequencies. The spectral density is usually denoted J(ω)
G J τ ω ( ) ÷ → ÷÷÷÷÷ ( )
Fourier Transform
If the spins were executing a well ordered motion, such as oscillating back
and forth about a mean position, the spectral density would show a peak at that
frequency. However, the spins are subject to random motions with a range of
different periods, so the spectral density shows a range of frequencies rather
than having peaks at discreet frequencies.
Generally, for random motion characterised by a correlation time τ
c
,
frequencies from zero up to about 1/τ
c
are present. The amount at frequencies
higher that 1/τ
c
tails off quite rapidly as the frequency increases.
For a simple exponential correlation function, given in Eq. [18], the
corresponding spectral density is a Lorentzian
exp − ( ) ÷ → ÷÷÷÷÷
+
τ τ
τ
ω τ
c
Fourier Transform
c
c
2
2
1
2
This function is plotted in the margin; note how it drops off significantly once
the product ωτ
c
begins to exceed ~1.
The plot opposite compares the spectral densities for three different
correlation times; curve a is the longest, b an intermediate value and c the
shortest. Note that as the correlation time decreases the spectral density moves
out to higher frequencies. However, the area under the plot remains the same,
so the contribution at lower frequencies is decreased. In particular, at the
frequency indicated by the dashed line the contribution at correlation time b is
greater than that for either correlation times a or c.
0
0
0
G( )
G(0)
τ τ
τ
c
0 1 2 3 4 5
0
J( ) ω
ωτ
c
Plot of the spectral density as a
function of the dimensionless
variable ωτ
c
. The curve is a
lorentzian
0 0
J( ) ω
ω
a
b
c
8–20
For this spectral density function, the maximum contribution at frequency ω
is found when τ
c
is 1/ω; this has important consequences which are described in
the next section.
8.5.3 The "T
1
minimum"
In the case of relaxation of a single spin by a random field (such as that
generated by a paramagnetic species), the only relevant spectral density is that
at the Larmor frequency, ω
0
. This is hardly surprising as to cause relaxation –
that is to cause transitions – the field needs to have components oscillating at
the Larmor frequency.
We have just seen that for a given frequency, ω
0
, the spectral density is a
maximum when τ
c
is 1/ω
0
, so to have the most rapid relaxation the correlation
time should be 1/ω
0
. This is illustrated in the plots below which show the
relaxation rate constant, W, and the corresponding relaxation time (T
1
= 1/W)
plotted as a function of the correlation time.
τ
c
1/ω
0
τ
c
W
T
1
1/ω
0
(a) (b)
Plot (a) shows how the relaxation rate constant, W, varies with the correlation time, τ
c
, for a given Larmor
frequency; there is a maximum in the rate constant when τ
c
= 1/ω
0
. Plot (b) shows the same effect, but here
we have plotted the relaxation time constant, T
1
; this shows a minimum.
At very short correlation times (τ
c
<< 1/ω
0
) there is some spectral density at
the Larmor frequency, but not that much as the energy of the motion is spread
over a very wide frequency range. As the correlation time increases the amount
of spectral density at the Larmor frequency increases and so the relaxation rate
constant increases, reaching a maximum when τ
c
= 1/ω
0
. After this point, the
spectra density at the Larmor frequency, and hence the rate constant, falls.
In terms of the relaxation time, T
1
, there is a minimum in T
1
which
corresponds to the maximum in W. We see that, like Goldilocks and the Three
Bears, efficient relaxation requires a correlation time which is neither too fast
nor too slow.
Motion which gives rise to correlation times which are much shorter than
1/ω
0
is described as being in the fast motion (or extreme narrowing) limit. Put
mathematically, fast motion means ω
0
τ
c
<< 1. Motion which gives rise to
correlation times which are much slower than 1/ω
0
is described as being in the
slow motion (or spin diffusion) limit; mathematically this limit is ω
0
τ
c
>> 1.
Clearly, which limit we are in depends on the Larmor frequency, which in turn
depends on the nucleus and the magnetic field.
For a Larmor frequency of 400 MHz we would expect the fastest relaxation
when the correlation time is 0.4 ns. Small molecules have correlation times
8–21
significantly shorter than this (say tens of ps), so such molecules are clearly in
the fast motion limit. Large molecules, such as proteins, can easily have
correlation times of the order of a few ns, and these clearly fall in the slow
motion limit.
Somewhat strangely, therefore, both very small and very large molecules
tend to relax more slowly than mediumsized molecules.
8.6 Relaxation mechanisms
So far, the source of the magnetic fields which give rise to relaxation and the
origin of their time dependence have not been considered. Each such source is
referred to as a relaxation mechanism. There are quite a range of different
mechanisms that can act, but of these only a few are really important for spin
half nuclei.
8.6.1 Paramagnetic species
We have already mentioned this source of varying fields several times. The
large magnetic moment of the electron means that paramagnetic species in
solution are particularly effective at promoting relaxation. Such species include
dissolved oxygen and certain transition metal compounds.
8.6.2 The dipolar mechanism
Each spin has associated with it a magnetic moment, and this is turn gives rise
to a magnetic field which can interact with other spins. Two spins are thus
required for this interaction, one to "create" the field and one to "experience" it.
However, their roles are reversible, in the sense that the second spin creates a
field which is experienced by the first. So, the overall interaction is a property
of the pair of nuclei.
The size of the interaction depends on the inverse cube of the distance
between the two nuclei and the direction of the vector joining the two nuclei,
measured relative to that of the applied magnetic field. As a molecule tumbles
in solution the direction of this vector changes and so the magnetic field
changes. Changes in the distance between the nuclei also result in a change in
the magnetic field. However, molecular vibrations, which do give such
changes, are generally at far too high frequencies to give significant spectral
density at the Larmor frequency. As a result, it is generally changes in
orientation which are responsible for relaxation.
The pair of interacting nuclei can be in the same or different molecules,
leading to intra and intermolecular relaxation. Generally, however, nuclei in
the same molecule can approach much more closely than those in different
molecules so that intramolecular relaxation is dominant.
The relaxation induced by the dipolar coupling is proportional to the square
of the coupling. Thus it goes as
8–22
γ γ
1
2
2
2
12
6
1
r
where γ
1
and γ
2
are the gyromagnetic ratios of the two nuclei involved and r
12
is
the distance between them.
As the size of the dipolar interaction depends on the product of the
gyromagnetic ratios of the two nuclei involved, and the resulting relaxation rate
constants depends on the square of this. Thus, pairs of nuclei with high
gyromagnetic ratios are most efficient at promoting relaxation. For example,
every thing else being equal, a protonproton pair will relax 16 times faster than
a carbon13 proton pair.
It is important to realize that in dipolar relaxation the effect is not primarily
to distribute the energy from one of the spins to the other. This would not, on
its own, bring the spins to equilibrium. Rather, the dipolar interaction provides
a path by which energy can be transferred between the lattice and the spins. In
this case, the lattice is the molecular motion. Essentially, the dipoledipole
interaction turns molecular motion into an oscillating magnetic field which can
cause transitions of the spins.
Relation to the NOE
The dipolar mechanism is the only common relaxation mechanism which can
cause transitions in which more than one spin flips. Specifically, with reference
to section 8.3, the dipolar mechanism gives rise to transitions between the αα
and ββ states (W
2
) and between the αβ and βα states (W
0
).
The rate constant W
2
corresponds to transitions which are at the sum of the
Larmor frequencies of the two spins, (ω
0,I
+ ω
0,S
) and so it is the spectral density
at this sum frequency which is relevant. In contrast, W
0
corresponds to
transitions at (ω
0,I
– ω
0,S
) and so for these it is the spectral density at this
difference frequency which is relevant.
In the case where the two spins are the same (e.g. two protons) the two
relevant spectral densities are J(2ω
0
) and J(0). In the fast motion limit (ω
0
τ
c
<<
1) J(2ω
0
) is somewhat less than J(0), but not by very much. A detailed
calculation shows that W
2
> W
0
and so we expect to see positive NOE
enhancements (section 8.4.5). In contrast, in the slow motion limit (ω
0
τ
c
>> 1)
J(2ω
0
) is all but zero and so J(0) >> J(2ω
0
); not surprisingly it follows that W
0
>
W
2
and a negative NOE enhancement is seen.
8.6.3 The chemical shift anisotropy mechanism
The chemical shift arises because, due to the effect of the electrons in a
molecule, the magnetic field experienced by a nucleus is different to that
applied to the sample. In liquids, all that is observable is the average chemical
shift, which results from the molecule rapidly experiencing all possible
orientations by rapid molecular tumbling.
At a more detailed level, the magnetic field experienced by the nucleus
8–23
depends on the orientation of the molecule relative to the applied magnetic
field. This is called chemical shift anisotropy (CSA). In addition, it is not only
the magnitude of the field which is altered but also its direction. The changes
are very small, but sufficient to be detectable in the spectrum and to give rise to
relaxation.
One convenient way of imagining the effect of CSA is to say that due to it
there are small additional fields created at the nucleus – in general in all three
directions. These fields vary in size as the molecule reorients, and so they have
the necessary time variation to cause relaxation. As has already been discussed,
it is the transverse fields which will give rise to changes in population.
The size of the CSA is specified by a tensor, which is a mathematical object
represented by a three by three matrix.
σ
σ σ σ
σ σ σ
σ σ σ

(
'
'
'
`
J
J
J
J
xx xy xz
yx yy yz
zx zy zz
The element σ
xz
gives the size of the extra field in the xdirection which results
from a field being applied in the zdirection; likewise, σ
yz
gives the extra field in
the ydirection and σ
zz
that in the zdirection. These elements depend on the
electronic properties of the molecule and the orientation of the molecule with
respect to the magnetic field.
Detailed calculations show that the relaxation induced by CSA goes as the
square of the field strength and is also proportional to the shift anisotropy. A
rough estimate of the size of this anisotropy is that it is equal to the typical shift
range. So, CSA relaxation is expected to be significant for nuclei with large
shift ranges observed at high fields. It is usually insignificant for protons.
8.7 Transverse relaxation
Right at the start of this section we mentioned that relaxation involved two
processes: the populations returning to equilibrium and the transverse
magnetization decaying to zero. So far, we have only discussed the fist of these
two. The second, in which the transverse magnetization decays, is called
transverse (or spinspin) relaxation.
8–24
(a)
(b)
(c)
Depiction of how the individual contributions from different spins (shown on the left) add up to give the net
transverse magnetization (on the right). See text for details.
Each spin in the sample can be thought of as giving rise to a small
contribution to the magnetization; these contributions can be in any direction,
and in general have a component along x, y and z. The individual contributions
along z add up to give the net zmagnetization of the sample.
The transverse contributions behave in a more complex way as, just like the
net transverse magnetization, these contributions are precessing at the Larmor
frequency in the transverse plane. We can represent each of these contributions
by a vector precessing in the transverse plane.
The direction in which these vectors point can be specified by giving each a
phase – arbitrarily the angle measured around from the xaxis. It is
immediately clear that if these phases are random the net transverse
magnetization of the sample will be zero as all the individual contributions will
cancel. This is the situation that pertains at equilibrium and is shown in (a) in
the figure above.
For there to be net magnetization, the phases must not be random, rather
there has to be a preference for one direction; this is shown in (b) in the figure
above. In quantum mechanics this is described as a coherence. An RF pulse
applied to equilibrium magnetization generates transverse magnetization, or in
other words the pulse generates a coherence. Transverse relaxation destroys
this coherence by destroying the alignment of the individual contributions, as
8–25
shown in (c) above.
Our picture indicates that there are two ways in which the coherence could
be destroyed. The first is to make the vectors jump to new positions, at
random. Drawing on our analogy between these vectors and the behaviour of
the bulk magnetization, we can see that these jumps could be brought about by
local oscillating fields which have the same effect as pulses.
This is exactly what causes longitudinal relaxation, in which we imagine the
local fields causing the spins to flip. So, anything that causes longitudinal
relaxation will also cause transverse relaxation.
The second way of destroying the coherence is to make the vectors get out of
step with one another as a result of them precessing at different Larmor
frequencies. Again, a local field plays the part we need but this time we do not
need it to oscillate; rather, all we need for it to do is to be different at different
locations in the sample.
This latter contribution is called the secular part of transverse relaxation; the
part which has the same origin as longitudinal relaxation is called the non
secular part.
It turns out that the secular part depends on the spectral density at zero
frequency, J(0). We can see that this makes sense as this part of transverse
relaxation requires no transitions, just a field to cause a local variation in the
magnetic field. Looking at the result from section 8.5.2 we see that J(0) = 2τ
c
,
and so as the correlation time gets longer and longer, so too does the relaxation
rate constant. Thus large molecules in the slow motion limit are characterised
by very rapid transverse relaxation; this is in contrast to longitudinal relaxation
is most rapid for a particular value of the correlation time.
The plot below compares the behaviour of the longitudinal and transverse
relaxation rate constants. As the correlation time increases the longitudinal rate
constant goes through a maximum. However, the transverse rate constant
carries on increasing and shows no such maximum. We can attribute this to the
secular part of transverse relaxation which depends on J(0) and which simply
goes on increasing as the correlation time increases. Detailed calculations show
that in the fast motion limit the two relaxation rate constants are equal.
τ
c
1/ω
0
W
longitudinal
transverse
Comparison of the longitudinal and transverse relaxation rate constants as a function of the correlation time
for the fixed Larmor frequency. The longitudinal rate constant shows a maximum, but the transverse rate
constant simply goes on increasing.
9–1
9 Coherence Selection:
Phase Cycling and Gradient Pulses
†
9.1 Introduction
The pulse sequence used in an NMR experiment is carefully designed to
produce a particular outcome. For example, we may wish to pass the spins
through a state of multiple quantum coherence at a particular point, or plan for
the magnetization to be aligned along the zaxis during a mixing period.
However, it is usually the case that the particular series of events we designed
the pulse sequence to cause is only one out of many possibilities. For example,
a pulse whose role is to generate doublequantum coherence from antiphase
magnetization may also generate zeroquantum coherence or transfer the
magnetization to another spin. Each time a radiofrequency pulse is applied
there is this possibility of branching into many different pathways. If no steps
are taken to suppress these unwanted pathways the resulting spectrum will be
hopelessly confused and uninterpretable.
There are two general ways in which one pathway can be isolated from the
many possible. The first is phase cycling. In this method the phases of the
pulses and the receiver are varied in a systematic way so that the signal from
the desired pathways adds and signal from all other pathways cancels. Phase
cycling requires that the experiment is repeated several times, something which
is probably required in any case in order to achieve the required signaltonoise
ratio.
The second method of selection is to use field gradient pulses. These are
short periods during which the applied magnetic field is made inhomogeneous.
As a result, any coherences present dephase and are apparently lost. However,
this dephasing can be undone, and the coherence restored, by application of a
subsequent gradient. We shall see that this dephasing and rephasing approach
can be used to select particular coherences. Unlike phase cycling, the use of
field gradient pulses does not require repetition of the experiment.
Both of these selection methods can be described in a unified framework
which classifies the coherences present at any particular point according to a
coherence order and then uses coherence transfer pathways to specify the
desired outcome of the experiment.
9.2 Phase in NMR
In NMR we have control over both the phase of the pulses and the receiver
phase. The effect of changing the phase of a pulse is easy to visualise in the
usual rotating frame. So, for example, a 90° pulse about the xaxis rotates
†
Chapter 9 "Coherence Selection: Phase Cycling and Gradient Pulses" © James Keeler 2001
and 2003.
9–2
magnetization from z onto –y, whereas the same pulse applied about the yaxis
rotates the magnetization onto the xaxis. The idea of the receiver phase is
slightly more complex and will be explored in this section.
The NMR signal – that is the free induction decay – which emerges from the
probe is a radiofrequency signal oscillating at close to the Larmor frequency
(usually hundreds of MHz). Within the spectrometer this signal is shifted down
to a much lower frequency in order that it can be digitized and then stored in
computer memory. The details of this downshifting process can be quite
complex, but the overall result is simply that a fixed frequency, called the
receiver reference or carrier, is subtracted from the frequency of the incoming
NMR signal. Frequently, this receiver reference frequency is the same as the
transmitter frequency used to generate the pulses applied to the observed
nucleus. We shall assume that this is the case from now on.
The rotating frame which we use to visualise the effect of pulses is set at the
transmitter frequency, ω
rf
, so that the field due to the radiofrequency pulse is
static. In this frame, a spin whose Larmor frequency is ω
0
precesses at (ω
0
–
ω
rf
), called the offset Ω. In the spectrometer the incoming signal at ω
0
is down
shifted by subtracting the receiver reference which, as we have already decided,
will be equal to the frequency of the radiofrequency pulses. So, in effect, the
frequencies of the signals which are digitized in the spectrometer are the offset
frequencies at which the spins evolve in the rotating frame. Often this whole
process is summarised by saying that the "signal is detected in the rotating
frame".
9.2.1 Detector phase
The quantity which is actually detected in an NMR experiment is the transverse
magnetization. Ultimately, this appears at the probe as an oscillating voltage,
which we will write as
S t
FID
= cosω
0
where ω
0
is the Larmor frequency. The downshifting process in the
spectrometer is achieved by an electronic device called a mixer; this effectively
multiplies the incoming signal by a locally generated reference signal, S
ref
,
which we will assume is oscillating at ω
rf
S t
ref rf
= cosω
The output of the mixer is the product S
FID
S
ref
S S A t t
A t t
FID ref rf
rf rf
=
= + ( ) + − ( )
[ ]
cos cos
cos cos
ω ω
ω ω ω ω
0
1
2 0 0
The first term is an oscillation at a high frequency (of the order of twice the
Larmor frequency as ω
0
≈ ω
rf
) and is easily removed by a lowpass filter. The
second term is an oscillation at the offset frequency, Ω. This is in line with the
previous comment that this downshifting process is equivalent to detecting the
precession in the rotating frame.
9–3
We can go further with this interpretation and say that the second term
represents the component of the magnetization along a particular axis (called
the reference axis) in the rotating frame. Such a component varies as cos Ωt,
assuming that at time zero the magnetization is aligned along the chosen axis;
this is illustrated below
θ
t = 0 time t
At time zero the magnetization is assumed to be aligned along the reference axis. After time t the
magnetization has precessed through an angle θ = Ωt. The projection of the magnetization onto the
reference axis is proportional to cos Ωt.
Suppose now that the phase of the reference signal is shifted by φ, something
which is easily achieved in the spectrometer. Effectively, this shifts the
reference axis by φ, as shown below
φ
t = 0 time t
θ φ –
Shifting the phase of the receiver reference by φ is equivalent to detecting the component along an axis
rotated by φ from its original position (the previous axis is shown in grey). Now the apparent angle of
precession is θ = Ωt – φ. and the projection of the magnetization onto the reference axis is proportional to
cos (Ωt – φ).
The component along the new reference axis is proportional to cos Ωt − ( ) φ .
How this is put to good effect is described in the next section.
9–4
9.2.2 Quadrature detection
We normally want to place the transmitter frequency in the centre of the
resonances of interest so as to minimise offresonance effects. If the receiver
reference frequency is the same as the transmitter frequency, it immediately
follows that the offset frequencies, Ω, may be both positive and negative.
However, as we have seen in the previous section the effect of the down
shifting scheme used is to generate a signal of the form cos Ωt. Since cos(θ) =
cos(–θ) such a signal does not discriminate between positive and negative
offset frequencies. The resulting spectrum, obtained by Fourier transformation,
would be confusing as each peak would appear at both +Ω and –Ω.
The way out of this problem is to detect the signal along two perpendicular
axes. As is illustrated opposite, the projection along one axis is proportional to
cos(Ωt) and to sin(Ωt) along the other. Knowledge of both of these projections
enables us to work out the sense of rotation of the vector i.e. the sign of the
offset.
The sin modulated component is detected by having a second mixer fed with
a reference whose phase is shifted by π/2. Following the above discussion the
output of the mixer is
cos cos cos sin sin
sin
Ω Ω Ω
Ω
t t t
t
− ( ) = +
=
π π π 2 2 2
The output of these two mixers can be regarded as being the components of the
magnetization along perpendicular axes in the rotating frame.
The final step in this whole process is regard the outputs of the two mixers as
being the real and imaginary parts of a complex number:
cos sin exp Ω Ω Ω t t t + = ( ) i i
The overall result is the generation of a complex signal whose phase varies
according to the offset frequency Ω.
9.2.3 Control of phase
In the previous section we supposed that the signal coming from the probe was
of the form cosω
0
t but it is more realistic to write the signal as
cos(ω
0
t + φ
sig
) in recognition of the fact that in addition to a frequency the signal
has a phase, φ
sig
. This phase is a combination of factors that are not under our
control (such as phase shifts produced in the amplifiers and filters through
which the signal passes) and a phase due to the pulse sequence, which certainly
is under our control.
The overall result of this phase is simply to multiply the final complex signal
by a phase factor, exp(iφ
sig
):
exp exp i i
sig
Ωt ( )
( )
φ
As we saw in the previous section, we can also introduce another phase shift by
altering the phase of the reference signal fed to the mixer, and indeed we saw
θ
x
y
The x and y projections of the
black vector are both positive.
If the vector had precessed in
the opposite direction (shown
shaded), and at the same
frequency, the projection along
x would be the same, but along
y it would be minus that of the
black vector.
9–5
that the cosine and sine modulated signals are generated by using two mixers
fed with reference signals which differ in phase by π/2. If the phase of each of
these reference signals is advanced by φ
rx
, usually called the receiver phase, the
output of the two mixers becomes cos Ωt − ( ) φ
rx
and sin Ωt − ( ) φ
rx
. In the
complex notation, the overall signal thus acquires another phase factor
exp exp exp i i i
sig rx
Ωt ( )
( )
− ( ) φ φ
Overall, then, the phase of the final signal depends on the difference between
the phase introduced by the pulse sequence and the phase introduced by the
receiver reference.
9.2.4 Lineshapes
Let us suppose that the signal can be written
S t B t t T ( ) = ( ) ( ) − ( ) exp exp exp i i Ω Φ
2
where Φ is the overall phase (= − φ φ
sig rx
) and B is the amplitude. The term,
exp(t/T
2
) has been added to impose a decay on the signal. Fourier
transformation of S(t) gives the spectrum S(ω):
S B A D ω ω ω ( ) = ( ) + ( )
[ ]
( ) i i exp Φ [1]
where A(ω) is an absorption mode lorentzian lineshape centred at ω = Ω and
D(ω) is the corresponding dispersion mode lorentzian:
A
T
T
D
T
T
ω
ω
ω
ω
ω
( ) =
+ − ( )
( ) =
− ( )
+ − ( )
2
2
2
2
2
2
2
2
2
1 1 Ω
Ω
Ω
Normally we display just the real part of S(ω) which is, in this case,
Re cos sin S B A D ω ω ω ( )
[ ]
= ( ) − ( )
[ ]
Φ Φ
In general this is a mixture of the absorption and dispersion lineshape. If we
want just the absorption lineshape we need to somehow set Φ to zero, which is
easily done by multiplying S(ω) by a phase factor exp(iΘ).
S B A D
B A D
ω ω ω
ω ω
( ) ( ) = ( ) + ( )
[ ]
( ) ( )
= ( ) + ( )
[ ]
+ [ ] ( )
exp exp exp
exp
i i i i
i i
Θ Φ Θ
Φ Θ
As this is a numerical operation which can be carried out in the computer we
are free to choose Θ to be the required value (here –Φ) in order to remove the
phase factor entirely and hence give an absorption mode spectrum in the real
part. This is what we do when we "phase the spectrum".
9.2.5 Relative phase and lineshape
We have seen that we can alter the phase of the spectrum by altering the phase
of the pulse or of the receiver, but that what really counts is the difference in
these two phases.
We will illustrate this with the simple vector diagrams shown below. Here,
the vector shows the position of the magnetization at time zero and its phase,
absorption
dispersion
Ω
9–6
φ
sig
, is measured anticlockwise from the xaxis. The dot shows the axis along
which the receiver is aligned; this phase, φ
rx
, is also measured anticlockwise
from the xaxis.
If the vector and receiver are aligned along the same axis, Φ = 0, and the real
part of the spectrum shows the absorption mode lineshape. If the receiver
phase is advanced by π/2, Φ = 0 – π/2 and, from Eq. [1]
S B A D
B A D
ω ω ω π
ω ω
( ) = ( ) + ( )
[ ]
− ( )
= − ( ) + ( )
[ ]
i i
i
exp 2
This means that the real part of the spectrum shows a dispersion lineshape. On
the other hand, if the magnetization is advanced by π/2, Φ = φ
sig
– φ
r x
= π/2 – 0 = π/2 and it can be shown from Eq. [1] that the real part of the
spectrum shows a negative dispersion lineshape. Finally, if either phase is
advanced by π, the result is a negative absorption lineshape.
x
y
φ
rx
φ
9.2.6 CYCLOPS
The CYCLOPS phase cycling scheme is commonly used in even the simplest
pulseacquire experiments. The sequence is designed to cancel some
imperfections associated with errors in the two phase detectors mentioned
above; a description of how this is achieved is beyond the scope of this
discussion. However, the cycle itself illustrates very well the points made in
the previous section.
There are four steps in the cycle, the pulse phase goes x, y, –x, –y i.e. it
advances by 90° on each step; likewise the receiver advances by 90° on each
step. The figure below shows how the magnetization and receiver phases are
related for the four steps of this cycle
9–7
x –x
y
–y
pulse x y –x –y
receiver x y –x –y
Although both the receiver and the magnetization shift phase on each step, the
phase difference between them remains constant. Each step in the cycle thus
gives the same lineshape and so the signal adds on all four steps, which is just
what is required.
Suppose that we forget to advance the pulse phase; the outcome is quite
different
x –x
y
–y
pulse x x x x
receiver x y –x –y
Now the phase difference between the receiver and the magnetization is no
longer constant. A different lineshape thus results from each step and it is clear
that adding all four together will lead to complete cancellation (steps 2 and 4
cancel, as do steps 1 and 3). For the signal to add up it is clearly essential for
the receiver to follow the magnetization.
9.2.7 EXORCYLE
EXORCYLE is perhaps the original phase cycle. It is a cycle used for 180°
pulses when they form part of a spin echo sequence. The 180° pulse cycles
through the phases x, y, –x, –y and the receiver phase goes x, –x, x, –x. The
diagram below illustrates the outcome of this sequence
9–8
x –x
y
y
–y
–y
90°(x) –
180°(±x)
180°(±y)
τ
τ
τ
If the phase of the 180° pulse is +x or –x the echo forms along the yaxis,
whereas if the phase is ±y the echo forms on the –y axis. Therefore, as the 180°
pulse is advanced by 90° (e.g. from x to y) the receiver must be advanced by
180° (e.g. from x to –x). Of course, we could just as well cycle the receiver
phases y, –y, y, –y; all that matters is that they advance in steps of 180°. We
will see later on how it is that this phase cycle cancels out the results of
imperfections in the 180° pulse.
9.2.8 Difference spectroscopy
Often a simple two step sequence suffices to cancel unwanted magnetization;
essentially this is a form of difference spectroscopy. The idea is well illustrated
by the INEPT sequence, shown opposite. The aim of the sequence is to transfer
magnetization from spin I to a coupled spin S.
With the phases and delays shown equilibrium magnetization of spin I, I
z
, is
transferred to spin S, appearing as the operator S
x
. Equilibrium magnetization
of S, S
z
, appears as S
y
. We wish to preserve only the signal that has been
transferred from I.
The procedure to achieve this is very simple. If we change the phase of the
second I spin 90° pulse from y to –y the magnetization arising from transfer of
the I spin magnetization to S becomes –S
x
i.e. it changes sign. In contrast, the
signal arising from equilibrium S spin magnetization is unaffected simply
because the S
z
operator is unaffected by the I spin pulses. By repeating the
experiment twice, once with the phase of the second I spin 90° pulse set to y
and once with it set to –y, and then subtracting the two resulting signals, the
undesired signal is cancelled and the desired signal adds. It is easily confirmed
that shifting the phase of the S spin 90° pulse does not achieve the desired
separation of the two signals as both are affected in the same way.
In practice the subtraction would be carried out by shifting the receiver by
180°, so the I spin pulse would go y, –y and the receiver phase go x, –x. This is
a two step phase cycle which is probably best viewed as difference
spectroscopy.
This simple two step cycle is the basic element used in constructing the
I
S
y
2J
1
2J
1
Pulse sequence for INEPT.
Filled rectangles represent 90°
pulses and open rectangles
represent 180° pulses. Unless
otherwise indicated, all pulses
are of phase x.
9–9
phase cycling of many two and threedimensional heteronuclear experiments.
9.3 Coherence transfer pathways
Although we can make some progress in writing simple phase cycles by
considering the vector picture, a more general framework is needed in order to
cope with experiments which involve multiplequantum coherence and related
phenomena. We also need a theory which enables us to predict the degree to
which a phase cycle can discriminate against different classes of unwanted
signals. A convenient and powerful way of doing both these things is to use the
coherence transfer pathway approach.
9.3.1 Coherence order
Coherences, of which transverse magnetization is one example, can be
classified according to a coherence order, p, which is an integer taking values 0,
± 1, ± 2 ... Single quantum coherence has p = ± 1, double has
p = ± 2 and so on; zmagnetization, "zz" terms and zeroquantum coherence
have p = 0. This classification comes about by considering the phase which
different coherences acquire is response to a rotation about the zaxis.
A coherence of order p, represented by the density operator σ
p ( )
, evolves
under a zrotation of angle φ according to
exp exp exp − ( ) ( ) = − ( )
( ) ( )
i i i φ σ φ φ σ F F p
z
p
z
p
[2]
where F
z
is the operator for the total zcomponent of the spin angular
momentum. In words, a coherence of order p experiences a phase shift of –pφ.
Equation [2] is the definition of coherence order.
To see how this definition can be applied, consider the effect of a zrotation
on transverse magnetization aligned along the xaxis. Such a rotation is
identical in nature to that due to evolution under an offset, and using product
operators it can be written
exp exp cos sin − ( ) ( ) = + i i φ φ φ φ I I I I I
z x z x y
[3]
The right hand sides of Eqs. [2] and [3] are not immediately comparable, but by
writing the sine and cosine terms as complex exponentials the comparison
becomes clearer. Using
cos exp exp exp exp φ φ φ φ φ φ = ( ) + − ( )
[ ]
= ( ) − − ( )
[ ]
1
2
1
2
i i in i i
i
s
Eq. [3] becomes
exp exp
exp exp exp exp
exp exp
− ( ) ( )
= ( ) + − ( )
[ ]
+ ( ) − − ( )
[ ]
= +
[ ]
( ) + −
[ ]
− ( )
i i
i i i i
i i
i
i i
φ φ
φ φ φ φ
φ φ
I I I
I I
I I I I
z x z
x y
x y x y
1
2
1
2
1
2
1 1
2
1
It is now clear that the first term corresponds to coherence order –1 and the
second to +1; in other words, I
x
is an equal mixture of coherence orders ±1.
The cartesian product operators do not correspond to a single coherence
9–10
order so it is more convenient to rewrite them in terms of the raising and
lowering operators, I
+
and I
–
, defined as
I I I I I I
x y x y +
= + = i i
–
–
from which it follows that
I I I I I I
x
= +
[ ]
= −
[ ]
+ +
1
2
1
2 – –
y i
[4]
Under zrotations the raising and lowering operators transform simply
exp exp exp − ( ) ( ) = ( )
± ±
i i i φ φ φ I I I I
z z
m
which, by comparison with Eq. [2] shows that I
+
corresponds to coherence
order +1 and I
–
to –1. So, from Eq. [4] we can see that I
x
and I
y
correspond to
mixtures of coherence orders +1 and –1.
As a second example consider the pure double quantum operator for two
coupled spins,
2 2
1 2 1 2
I I I I
x y y x
+
Rewriting this in terms of the raising and lowering operators gives
1
1 2 1 2 i
I I I I
+ + − −
−
( )
The effect of a zrotation on the term
I I
1 2
+ +
is found as follows:
exp exp exp exp
exp exp exp
exp exp exp
− ( ) − ( ) ( ) ( )
= − ( ) − ( ) ( )
= − ( ) − ( ) = − ( )
+ +
+ +
+ + + +
i i i i
i i i
i i i
φ φ φ φ
φ φ φ
φ φ φ
I I I I I I
I I I I
I I I I
z z z z
z z
1 2 1 2 2 1
1 1 2 1
1 2 1 2
2
Thus, as the coherence experiences a phase shift of –2φ the coherence is
classified according to Eq. [2] as having p = 2. It is easy to confirm that the
term
I I
1 2 − −
has p = –2. Thus the pure double quantum term, 2 2
1 2 1 2
I I I I
x y y x
+
, is
an equal mixture of coherence orders +2 and –2.
As this example indicates, it is possible to determine the order or orders of
any state by writing it in terms of raising and lowering operators and then
simply inspecting the number of such operators in each term. A raising
operator contributes +1 to the coherence order whereas a lowering operator
contributes –1. A zoperator, I
iz
, has coherence order 0 as it is invariant to z
rotations.
Coherences involving heteronuclei can be assigned both an overall order and
an order with respect to each nuclear species. For example the term
I S
1 1 + –
has
an overall order of 0, is order +1 for the I spins and –1 for the S spins. The term
I I S
z 1 2 1 + +
is overall of order 2, is order 2 for the I spins and is order 0 for the S
spins.
9.3.2 Evolution under offsets
The evolution under an offset, Ω, is simply a zrotation, so the raising and
lowering operators simply acquire a phase Ωt
9–11
exp exp exp − ( ) ( ) = ( )
± ±
i i i Ω Ω Ω tI I tI t I
z z
m
For products of these operators, the overall phase is the sum of the phases
acquired by each term
exp exp exp exp
exp
−
( )
− ( ) ( ) ( )
= −
( ) ( )
− +
− +
i i i i
i
Ω Ω Ω Ω
Ω Ω
j jz i iz i j i iz j jz
i j i j
tI tI I I tI tI
t I I
It also follows that coherences of opposite sign acquire phases of opposite signs
under free evolution. So the operator I
1+
I
2+
(with p = 2) acquires a phase –(Ω
1
+
Ω
2
)t i.e. it evolves at a frequency –(Ω
1
+ Ω
2
) whereas the operator I
1–
I
2–
(with p
= –2) acquires a phase (Ω
1
+ Ω
2
)t i.e. it evolves at a frequency (Ω
1
+ Ω
2
). We
will see later on that this observation has important consequences for the
lineshapes in twodimensional NMR.
The observation that coherences of different orders respond differently to
evolution under a zrotation (e.g. an offset) lies at the heart of the way in which
gradient pulses can be used to separate different coherence orders.
9.3.3 Phase shifted pulses
In general, a radiofrequency pulse causes coherences to be transferred from one
order to one or more different orders; it is this spreading out of the coherence
which makes it necessary to select one transfer among many possibilities. An
example of this spreading between coherence orders is the effect of a non
selective pulse on antiphase magnetization, such as 2I
1x
I
2z
, which corresponds to
coherence orders ±1. Some of the coherence may be transferred into double
and zeroquantum coherence, some may be transferred into twospin order and
some will remain unaffected. The precise outcome depends on the phase and
flip angle of the pulse, but in general we can see that there are many
possibilities.
If we consider just one coherence, of order p, being transferred to a
coherence of order p' by a radiofrequency pulse we can derive a very general
result for the way in which the phase of the pulse affects the phase of the
coherence. It is on this relationship that the phase cycling method is based.
We will write the initial state of order p as σ
p ( )
, and the final state of order p'
as σ
p' ( )
. The effect of the radiofrequency pulse causing the transfer is
represented by the (unitary) transformation U
φ
where φ is the phase of the
pulse. The initial and final states are related by the usual transformation
U U
p p
0 0
1
σ σ
( ) ( )
= +
– '
terms of other orders [5]
which has been written for phase 0; the other terms will be dropped as we are
only interested in the transfer from p to p'. The transformation brought about
by a radiofrequency pulse phase shifted by φ, U
φ
, is related to that with the
phase set to zero, U
0
, in the following way
U F U F
z z φ
φ φ = − ( ) ( ) exp exp i i
0
[6]
9–12
Using this, the effect of the phase shifted pulse on the initial state σ
p ( )
can be
written
U U
F U F F U F
p
z z
p
z z
φ φ
σ
φ φ σ φ φ
( )
( )
= − ( ) ( ) − ( ) ( )
–
–
exp exp exp exp
1
0 0
1
i i i i
[7]
The central three terms can be simplified by application of Eq. [2]
exp exp exp
–
i i i φ σ φ φ σ F F U p
z
p
z
p
( ) − ( ) = ( )
( ) ( )
0
1
giving
U U p F U U F
p
z
p
z φ φ
σ φ φ σ φ
( ) ( )
= ( ) − ( ) ( )
– –
exp exp exp
1
0 0
1
i i i
The central three terms can, from Eq. [5], be replaced by σ
p' ( )
to give
U U p F F
p
z
p
z φ φ
σ φ φ σ φ
( ) ( )
= ( ) − ( ) ( )
– '
exp exp exp
1
i i i
Finally, Eq. [5] is applied again to give
U U p p
p p
φ φ
σ φ φ σ
( ) ( )
= ( ) ( )
– '
exp exp '
1
i –i
Defining ∆p = (p' – p) as the change is coherence order, this simplifies to
U U p
p p
φ φ
σ φ σ
( ) ( )
= ( )
– '
exp –
1
i∆ [8]
Equation [8] says that if the phase of a pulse which is causing a change in
coherence order of ∆p is shifted by φ the coherence will acquire a phase label
(–∆p φ). It is this property which enables us to separate different changes in
coherence order from one another by altering the phase of the pulse.
In the discussion so far it has been assumed that U
φ
represents a single pulse.
However, any sequence of pulses and delays can be represented by a single
unitary transformation, so Eq. [8] applies equally well to the effect of phase
shifting all of the pulses in such a sequence. We will see that this property is
often of use in writing phase cycles.
If a series of phase shifted pulses (or pulse sandwiches) are applied a phase
(–∆p φ) is acquired from each. The total phase is found by adding up these
individual contributions. In an NMR experiment this total phase affects the
signal which is recorded at the end of the sequence, even though the phase shift
may have been acquired earlier in the pulse sequence. These phase shifts are,
so to speak, carried forward.
9.3.4 Coherence transfer pathways diagrams
In designing a multiplepulse NMR experiment the intention is to have specific
orders of coherence present at various points in the sequence. One way of
indicating this is to use a coherence transfer pathway (CTP) diagram along
with the timing diagram for the pulse sequence. An example of shown below,
which gives the pulse sequence and CTP for the DQF COSY experiment.
9–13
t
1
t
2
2
1
0
–1
–2
p
∆p=±1 ±1,±3
+1,–3
The solid lines under the sequence represent the coherence orders required
during each part of the sequence; note that it is only the pulses which cause a
change in the coherence order. In addition, the values of ∆p are shown for each
pulse. In this example, as is commonly the case, more than one order of
coherence is present at a particular time. Each pulse is required to cause
different changes to the coherence order – for example the second pulse is
required to bring about no less than four values of ∆p. Again, this is a common
feature of pulse sequences.
It is important to realise that the CTP specified with the pulse sequence is
just the desired pathway. We would need to establish separately (for example
using a product operator calculation) that the pulse sequence is indeed capable
of generating the coherences specified in the CTP. Also, the spin system which
we apply the sequence to has to be capable of supporting the coherences. For
example, if there are no couplings, then no double quantum will be generated
and thus selection of the above pathway will result in a null spectrum.
The coherence transfer pathway must start with p = 0 as this is the order to
which equilibrium magnetization (zmagnetization) belongs. In addition, the
pathway has to end with p = 1 as it is only single quantum coherence that is
observable. If we use quadrature detection (section 9.2.2) it turns out that only
one of p = ±1 is observable; we will follow the usual convention of assuming
that p = –1 is the detectable signal.
9.4 Lineshapes and frequency discrimination
9.4.1 Phase and amplitude modulation
The selection of a particular CTP has important consequences for lineshapes
and frequency discrimination in twodimensional NMR. These topics are
illustrated using the NOESY experiment as an example; the pulse sequence and
CTP is illustrated opposite.
If we imagine starting with I
z
, then at the end of t
1
the operators present are
− + cos sin Ω Ω t I t I
y x 1 1
The term in I
y
is rotated onto the zaxis and we will assume that only this term
survives. Finally, the zmagnetization is made observable by the last pulse (for
convenience set to phase –y) giving the observable term present at t
2
= 0 as
cosΩt I
x 1
As was noted in section 9.3.1, I
x
is in fact a mixture of coherence orders
p = ±1, something which is made evident by writing the operator in terms of I
+
t
1
t
2
1
0
–1
τ
m
–y
9–14
and I
–
1
2 1
cosΩt I I
+ −
+ ( )
Of these operators, only I
–
leads to an observable signal, as this corresponds to
p = –1. Allowing I
–
to evolve in t
2
gives
1
2 1 2
cos exp Ω Ω t t I i ( )
−
The final detected signal can be written as
S t t t t
C
i
1 2
1
2 1 2
, cos exp ( ) = ( ) Ω Ω
This signal is said to be amplitude modulated in t
1
; it is so called because the
evolution during t
1
gives rise, via the cosine term, to a modulation of the
amplitude of the observed signal.
The situation changes if we select a different pathway, as shown opposite.
Here, only coherence order –1 is preserved during t
1
. At the start of t
1
the
operator present is –I
y
which can be written
− − ( )
+ −
1
2i
I I
Now, in accordance with the CTP, we select only the I
–
term. During t
1
this
evolves to give
1
2 1 i
i exp Ωt I ( )
−
Following through the rest of the pulse sequence as before gives the following
observable signal
S t t t t
P
i i
1 2
1
4 1 2
, exp exp ( ) = ( ) ( ) Ω Ω
This signal is said to be phase modulated in t
1
; it is so called because the
evolution during t
1
gives rise, via exponential term, to a modulation of the
phase of the observed signal. If we had chosen to select p = +1 during t
1
the
signal would have been
S t t t t
N
–i i
1 2
1
4 1 2
, exp exp ( ) = ( ) ( ) Ω Ω
which is also phase modulated, except in the opposite sense. Note that in either
case the phase modulated signal is one half of the size of the amplitude
modulated signal, because only one of the two pathways has been selected.
Although these results have been derived for the NOESY sequence, they are
in fact general for any twodimensional experiment. Summarising, we find
• If a single coherence order is present during t
1
the result is phase
modulation in t
1
. The phase modulation can be of the form exp(iΩt
1
) or
exp(–iΩt
1
) depending on the sign of the coherence order present.
• If both coherence orders ±p are selected during t
1
, the result is amplitude
modulation in t
1
; selecting both orders in this way is called preserving
symmetrical pathways.
9.4.2 Frequency discrimination
The amplitude modulated signal contains no information about the sign of Ω,
t
1
t
2
1
0
–1
τ
m
–y
9–15
simply because cos(Ωt
1
) = cos(–Ωt
1
). As a consequence, Fourier
transformation of the time domain signal will result in each peak appearing
twice in the twodimensional spectrum, once at F
1
= +Ω and once at F
1
= –Ω.
As was commented on above, we usually place the transmitter in the middle of
the spectrum so that there are peaks with both positive and negative offsets. If,
as a result of recording an amplitude modulated signal, all of these appear
twice, the spectrum will hopelessly confused. A spectrum arising from an
amplitude modulated signal is said to lack frequency discrimination in F
1
.
On the other hand, the phase modulated signal is sensitive to the sign of the
offset and so information about the sign of Ω in the F
1
dimension is contained
in the signal. Fourier transformation of the signal S
P
(t
1
,t
2
) gives a peak at F
1
=
+Ω, F
2
= Ω, whereas Fourier transformation of the signal S
N
(t
1
,t
2
) gives a peak
at F
1
= –Ω, F
2
= Ω. Both spectra are said to be frequency discriminated as the
sign of the modulation frequency in t
1
is determined; in contrast to amplitude
modulated spectra, each peak will only appear once.
The spectrum from S
P
(t
1
,t
2
) is called the Ptype (P for positive) or echo
spectrum; a diagonal peak appears with the same sign of offset in each
dimension. The spectrum from S
N
(t
1
,t
2
) is called the Ntype (N for negative) or
antiecho spectrum; a diagonal peak appears with opposite signs in the two
dimensions.
It might appear that in order to achieve frequency discrimination we should
deliberately select a CTP which leads to a P– or an Ntype spectrum. However,
such spectra show a very unfavourable lineshape, as discussed in the next
section.
9.4.3 Lineshapes
In section 9.2.4 we saw that Fourier transformation of the signal
S t t t T ( ) = ( ) − ( ) exp exp iΩ
2
gave a spectrum whose real part is an absorption lorentzian and whose
imaginary part is a dispersion lorentzian:
S A D ω ω ω ( ) = ( ) + ( ) i
We will use the shorthand that A
2
represents an absorption mode lineshape at F
2
= Ω and D
2
represents a dispersion mode lineshape at the same frequency.
Likewise, A
1+
represents an absorption mode lineshape at F
1
= +Ω and D
1+
represents the corresponding dispersion lineshape. A
1–
and D
1–
represent the
corresponding lines at F
1
= –Ω.
The time domain signal for the Ptype spectrum can be written as
S t t t t t T t T
P
i i
1 2
1
4 1 2 1 2 2 2
, exp exp exp exp ( ) = ( ) ( ) − ( ) − ( ) Ω Ω
where the damping factors have been included as before. Fourier
transformation with respect to t
2
gives
S t F t t T A D
P
i i
1 2
1
4 1 1 2 2 2
, exp exp ( ) = ( ) − ( ) +
[ ]
Ω
9–16
and then further transformation with respect to t
1
gives
S F F A D A D
P
i i
1 2
1
4 1 1 2 2
, ( ) = +
[ ]
+
[ ]
+ +
The real part of this spectrum is
Re , S F F A A D D
P 1 2
1
4 1 2 1 2
( ) { }
= −
[ ]
+ +
The quantity in the square brackets on the right represents a phasetwist
lineshape at F
1
= +Ω, F
2
= Ω
Perspective view and contour plot of the phasetwist lineshape. Negative contours are shown dashed.
This lineshape is an inextricable mixture of absorption and dispersion, and it is
very undesirable for highresolution NMR. So, although a phase modulated
signal gives us frequency discrimination, which is desirable, it also results in a
phasetwist lineshape, which is not.
The time domain signal for the amplitude modulated data set can be written
as
S t t t t t T t T
C
i
1 2
1
2 1 2 1 2 2 2
, cos exp exp exp ( ) = ( ) ( ) − ( ) − ( ) Ω Ω
Fourier transformation with respect to t
2
gives
S t F t t T A D
C
i
1 2
1
2 1 1 2 2 2
, cos exp ( ) = ( ) − ( ) +
[ ]
Ω
which can be rewritten as
S t F t t t T A D
C
i i i
1 2
1
4 1 1 1 2 2 2
, exp exp exp ( ) = ( ) + − ( )
[ ]
− ( ) +
[ ]
Ω Ω
Fourier transformation with respect to t
1
gives, in the real part of the spectrum
Re , – –
–
S F F A A D D A A D D
C + 1 2
1
4 1 2 1 2
1
4 1 2 1 2
( ) { }
=
[ ]
+
[ ]
+ −
This corresponds to two phasetwist lineshapes, one at F
1
= +Ω, F
2
= Ω and the
other at F
1
= –Ω, F
2
= Ω; the lack of frequency discrimination is evident.
Further, the undesirable phasetwist lineshape is again present.
The lineshape can be restored to the absorption mode by discarding the
imaginary part of the time domain signal after the transformation with respect
to t
2
, i.e. by taking the real part
Re , cos exp S t F t t T A
C 1 2
1
2 1 1 2 2
( ) { }
= ( ) − ( ) Ω
Subsequent transformation with respect to t
1
gives, in the real part
9–17
1
4 1 2
1
4 1 2
A A A A
+ −
+
which is two double absorption mode lineshapes. Frequency discrimination is
lacking, but the lineshape is now much more desirable. The spectra with the
two phasetwist and two absorption mode lines are shown below on the left and
right, respectively.
0
F
1
F
2
0
F
1
F
2
9. 4. 4 Frequency discrimination with retention of absorption mode
lineshapes
For practical purposes it is essential to be able to achieve frequency
discrimination and at the same time retain the absorption mode lineshape.
There are a number of ways of doing this.
9.4.4.1 StatesHaberkornRuben (SHR) method
The key to this method is the ability to record a cosine modulated data set and a
sine modulated data set. The latter can be achieved simply by changing the
phase of appropriate pulses. For example, in the case of the NOESY
experiment, all that is required to generate the sine data set is to shift the phase
of the first 90° pulse by 90° (in fact in the NOESY sequence the pulse needs to
shift from x to –y). The two data sets have to kept separate.
The cosine data set is transformed with respect to t
1
and the imaginary part
discarded to give
Re , cos exp S t F t t T A
C 1 2
1
2 1 1 2 2
( ) { }
= ( ) − ( ) Ω [9]
The same operation is performed on the sine modulated data set
S t t t t t T t T
S
i
1 2
1
2 1 2 1 2 2 2
, sin exp exp exp ( ) = ( ) ( ) − ( ) − ( ) Ω Ω
Re , sin exp S t F t t T A
S 1 2
1
2 1 1 2 2
( ) { }
= ( ) − ( ) Ω [10]
A new complex data set is now formed by using the signal from Eq. [9] as the
real part and that from Eq. [10] as the imaginary part
S t F S t F S t F
t t T A
SHR C S
i
i
1 2 1 2 1 2
1
2 1 1 2 2
, Re , Re ,
exp exp
( ) = ( ) { }
+ ( ) { }
= ( ) − ( ) Ω
Fourier transformation with respect to t
1
gives, in the real part of the spectrum
Re , S F F A A
SHR 1 2
1
2 1 2
( ) { }
=
+
This is the desired frequency discriminated spectrum with a pure absorption
9–18
lineshape.
As commented on above, in NOESY all that is required to change from
cosine to sine modulation is to shift the phase of the first pulse by 90°. The
general recipe is to shift the phase of all the pulses that precede t
1
by 90°/p
1
,
where p
1
is the coherence order present during t
1
. So, for a double quantum
spectrum, the phase shift needs to be 45°. The origin of this rule is that, taken
together, the pulses which precede t
1
give rise to a pathway with ∆p = p
1
.
In heteronuclear experiments it is not usually necessary to shift the phase of
all the pulses which precede t
1
; an analysis of the sequence usually shows that
shifting the phase of the pulse which generates the transverse magnetization
which evolves during t
1
is sufficient.
9.4.4.2 Echo antiecho method
We will see in later sections that when we use gradient pulses for coherence
selection the natural outcome is P or Ntype data sets. Individually, each of
these gives a frequency discriminated spectrum, but with the phasetwist
lineshape. We will show in this section how an absorption mode lineshape can
be obtained provided both the P and the Ntype data sets are available.
As before, we write the two data sets as
S t t t t t T t T
S t t t t t T t T
P
N
i i
i i
1 2
1
4 1 2 1 2 2 2
1 2
1
4 1 2 1 2 2 2
, exp exp exp exp
, exp exp exp exp
( ) = ( ) ( ) − ( ) − ( )
( ) = − ( ) ( ) − ( ) − ( )
Ω Ω
Ω Ω
We then form the two combinations
S t t S t t S t t
t t t T t T
S t t S t t S t t
t t t T
C P N
S i P N
i
i
1 2 1 2 1 2
1
2 1 2 1 2 2 2
1 2
1
1 2 1 2
1
2 1 2 1 2
, , ,
cos exp exp exp
, , ,
sin exp exp
( ) = ( ) + ( )
= ( ) ( ) − ( ) − ( )
( ) = ( ) + ( )
[ ]
= ( ) ( ) −
Ω Ω
Ω Ω (( ) − ( ) exp t T
2 2
These cosine and sine modulated data sets can be used as inputs to the SHR
method described in the previous section.
An alternative is to Fourier transform the two data sets with respect to t
2
to
give
S t F t t T A D
S t F t t T A D
P
N
i i
i i
1 2
1
4 1 1 2 2 2
1 2
1
4 1 1 2 2 2
, exp exp
, exp – exp
( ) = ( ) − ( ) +
[ ]
( ) = ( ) − ( ) +
[ ]
Ω
Ω
We then take the complex conjugate of S
N
(t
1
,F
2
) and add it to S
P
(t
1
,F
2
)
S t F t t T A D
S t F S t F S t F
t t T A
N
N P
i i
i
1 2
1
4 1 1 2 2 2
1 2 1 2 1 2
1
2 1 1 2 2
, * exp exp
, , * ,
exp exp
( ) = ( ) − ( ) −
[ ]
( ) = ( ) + ( )
= ( ) − ( )
+
Ω
Ω
Transformation of this signal gives
9–19
S F F A D A
+ + +
( ) = +
[ ]
1 2
1
2 1 1 2
, i
which is frequency discriminated and has, in the real part, the required double
absorption lineshape.
9.4.4.3 MarionWüthrich or TPPI method
The idea behind the TPPI (time proportional phase incrementation) or
Marion–Wüthrich (MW) method is to arrange things so that all of the peaks
have positive offsets. Then, frequency discrimination is not required as there is
no ambiguity.
One simple way to make all offsets positive is to set the receiver carrier
frequency deliberately at the edge of the spectrum. Simple though this is, it is
not really a very practical method as the resulting spectrum would be very
inefficient in its use of data space and in addition offresonance effects
associated with the pulses in the sequence will be accentuated.
In the TPPI method the carrier can still be set in the middle of the spectrum,
but it is made to appear that all the frequencies are positive by phase shifting
some of the pulses in the sequence in concert with the incrementation of t
1
.
It was noted above that shifting the phase of the first pulse in the NOESY
sequence from x to –y caused the modulation to change from cos(Ωt
1
) to
sin(Ωt
1
). One way of expressing this is to say that shifting the pulse causes a
phase shift φ in the signal modulation, which can be written cos(Ωt
1
+ φ).
Using the usual trigonometric expansions this can be written
cos cos cos sin sin Ω Ω Ω t t t
1 1 1
+ ( ) = − φ φ φ
If the phase shift, φ, is –π/2 radians the result is
cos cos cos sin sin
sin
Ω Ω Ω
Ω
t t t
t
1 1 1
1
2 2 2 + ( ) = − ( ) − − ( )
=
π π π
This is exactly the result we found before.
In the TPPI procedure, the phase φ is made proportional to t
1
i.e. each time t
1
is incremented, so is the phase. We will suppose that
φ ω t t
1 1
( ) =
add
The constant of proportion between the time dependent phase, φ(t
1
), and t
1
has
been written ω
add
; ω
add
has the dimensions of rad s
–1
i.e. it is a frequency.
Following the same approach as before, the timedomain function with the
inclusion of this incrementing phase is thus
cos cos
cos
Ω Ω
Ω
t t t t
t
1 1 1 1
1
+ ( ) ( )
= + ( )
= + ( )
φ ω
ω
add
add
In words, the effect of incrementing the phase in concert with t
1
is to add a
frequency ω
add
to all of the offsets in the spectrum. The TPPI method utilizes
this in the following way.
In onedimensional pulseFourier transform NMR the free induction signal is
9–20
sampled at regular intervals ∆. After transformation the resulting spectrum
displays correctly peaks with offsets in the range –(SW/2) to +(SW/2) where SW
is the spectral width which is given by 1/∆ (this comes about from the Nyquist
theorem of data sampling). Frequencies outside this range are not represented
correctly.
Suppose that the required frequency range in the F
1
dimension is from
–(SW
1
/2) to +(SW
1
/2). To make it appear that all the peaks have a positive
offset, it will be necessary to add (SW
1
/2) to all the frequencies. Then the peaks
will be in the range 0 to (SW
1
).
As the maximum frequency is now (SW
1
) rather than (SW
1
/2) the sampling
interval, ∆
1
, will have to be halved i.e. ∆
1
= 1/(2SW
1
) in order that the range of
frequencies present are represented properly.
The phase increment is ω
add
t
1
, but t
1
can be written as n∆
1
for the nth
increment of t
1
. The required value for ω
add
is 2π(SW
1
/2) , where the 2π is to
convert from frequency (the units of SW
1
) to rad s
–1
, the units of ω
add
. Putting
all of this together ω
add
t
1
can be expressed, for the nth increment as
ω π
π
π
additional
t
SW
n
SW
n
SW
n
1
1
1
1
1
2
2
2
2
1
2
2
=
( )
=
=
∆
In words this means that each time t
1
is incremented, the phase of the signal
should also be incremented by 90°, for example by incrementing the phase of
one of the pulses.
A data set from an experiment to which TPPI has been applied is simply
amplitude modulated in t
1
and so can be processed according to the method
described above for cosine modulated data so as to obtain absorption mode
lineshapes. As the spectrum is symmetrical about F
1
= 0, it is usual to use a
modified Fourier transform routine which saves effort and space by only
calculating the positive frequency part of the spectrum.
9.4.4.4 StatesTPPI
When the SHR method is used, axial peaks (arising from magnetization which
has not evolved during t
1
) appear at F
1
= 0; such peaks can be a nuisance as
they may obscure other wanted peaks. We will see below (section 9.5.6) that
axial peaks can be suppressed with the aid of phase cycling, all be it at the cost
of doubling the length of the phase cycle.
The StatesTPPI method does not suppress these axial peaks, but moves
them to the edge of the spectrum so that they are less likely to obscure wanted
peaks. All that is involved is that, each time t
1
is incremented, both the phase of
the pulse which precedes t
1
and the receiver phase are advanced by 180° i.e. the
t
1
= 0
t
2
x
t
1
= ∆
t
2
y
t
1
= 2∆
t
2
–x
t
1
= 3∆
t
2
–y
t
1
= 4∆
t
1
t
2
x
Illustration of the TPPI method.
Each t i me t hat t
1
i s
incremented, so is the phase of
the pulse preceding t
1
.
9–21
pulse goes x, –x and the receiver goes x, –x.
For nonaxial peaks, the two phase shifts cancel one another out, and so have
no effect. However, magnetization which gives rise to axial peaks does not
experience the first phase shift, but does experience the receiver phase shift.
The sign alternation in concert with t
1
incrementation adds a frequency of SW
1
/2
to each peak, thus shifting it to the edge of the spectrum. Note that in States
TPPI the spectral range in the F
1
dimension is –(SW
1
/2) to +(SW
1
/2) and the
sampling interval is 1/2SW
1
, just as in the SHR method.
The nice feature of StatesTPPI is that is moves the axial peaks out of the
way without lengthening the phase cycle. It is therefore convenient to use in
complex three and fourdimensional spectra were phase cycling is at a
premium.
9.5 Phase cycling
In this section we will start out by considering in detail how to write a phase
cycle to select a particular value of ∆p and then use this discussion to lead on to
the formulation of general principles for constructing phase cycles. These will
then be used to construct appropriate cycles for a number of common
experiments.
9.5.1 Selection of a single pathway
To focus on the issue at hand let us consider the case of transferring from
coherence order +2 to order –1. Such a transfer has ∆p = (–1 – (2) ) = –3. Let
us imagine that the pulse causing this transformation is cycled around the four
cardinal phases (x, y, –x, –y, i.e. 0°, 90°, 180°, 270°) and draw up a table of the
phase shift that will be experienced by the transferred coherence. This is
simply computed as – ∆p φ, in this case = – (–3)φ = 3φ.
step pulse phase
phase shift experienced by
transfer with ∆p = –3
equivalent phase
1 0 0 0
2 90 270 270
3 180 540 180
4 270 810 90
The fourth column, labelled "equivalent phase", is just the phase shift
experienced by the coherence, column three, reduced to be in the range 0 to
360° by subtracting multiples of 360° (e.g. for step 3 we subtracted 360° and
for step 4 we subtracted 720°).
If we wished to select ∆p = –3 we would simply shift the phase of the
receiver in order to match the phase that the coherence has acquired; these are
the phases shown in the last column. If we did this, then each step of the cycle
would give an observed signal of the same phase and so they four contributions
would all add up. This is precisely the same thing as we did when considering
9–22
the CYCLOPS sequence in section 9.2.6; in both cases the receiver phase
follows the phase of the desired magnetization or coherence.
We now need to see if this four step phase cycle eliminates the signals from
other pathways. As an example, let us consider a pathway with
∆p = 2, which might arise from the transfer from coherence order –1 to +1.
Again we draw up a table to show the phase experienced by a pathway with ∆p
= 2, that is computed as – (2)φ
step
pulse
phase
phase shift experienced by
transfer with ∆p = 2
equivalent
phase
rx. phase to
select ∆p = –3
difference
1 0 0 0 0 0
2 90 –180 180 270 270 – 180 = 90
3 180 –360 0 180 180 – 0 = 180
4 270 –540 180 90 90 – 180 = –90
As before, the equivalent phase is simply the phase in column 3 reduced to the
range 0 to 360°. The fifth column shows the receiver (abbreviated to rx.)
phases determined above for selection of the transfer with ∆p = –3. The
question we have to ask is whether or not these phase shifts will lead to
cancellation of the transfer with ∆p = 2. To do this we compute the difference
between the receiver phase, column 5, and the phase shift experienced by the
transfer with ∆p = 2, column 4. The results are shown in column 6, labelled
"difference". Looking at this difference column we can see that step 1 will
cancel with step 3 as the 180° phase shift between them means that the two
signals have opposite sign. Likewise step 2 will cancel with step 4 as there is a
180° phase shift between them. We conclude, therefore, that this four step
cycl e cancel s t he si gnal ari si ng from a pat hway wi t h
∆p = 2.
An alternative way of viewing the cancellation is to represent the results of
the "difference" column by vectors pointing at the indicated angles. This is
shown below; it is clear that the opposed vectors cancel one another.
step 1 2 3 4
difference 0°
0°
90° 180° 270°
Next we consider the coherence transfer with ∆p = +1. Again, we draw up
the table and calculate the phase shifts experience by this transfer, which are
given by – (+1)φ = –φ.
9–23
step
pulse
phase
phase shift experienced
by transfer with ∆p = +1
equivalent
phase
rx. phase to
select ∆p = –3
difference
1 0 0 0 0 0
2 90 –90 270 270 270 – 270 = 0
3 180 –180 180 180 180 – 180 = 0
4 270 –270 90 90 90 – 90 = 0
Here we see quite different behaviour. The equivalent phases, that is the phase
shifts experienced by the transfer with ∆p = 1, match exactly the receiver
phases determined for ∆p = –3, thus the phases in the "difference" column are
all zero. We conclude that the four step cycle selects transfers both with ∆p =
–3 and +1.
Some more work with tables such as these will reveal that this four step
cycle suppresses contributions from changes in coherence order of –2, –1 and 0.
It selects ∆p = –3 and 1. It also selects changes in coherence order of 5, 9, 13
and so on. This latter sequence is easy to understand. A pathway with ∆p = 1
experiences a phase shift of –90° when the pulse is shifted in phase by 90°; the
equivalent phase is thus 270°. A pathway with ∆p = 5 would experience a
phase shift of –5 × 90° = –450° which corresponds to an equivalent phase of
270°. Thus the phase shifts experienced for ∆p = 1 and 5 are identical and it is
clear that a cycle which selects one will select the other. The same goes for the
series ∆p = 9, 13 ...
The extension to negative values of ∆p is also easy to see. A pathway with
∆p = –3 experiences a phase shift of 270° when the pulse is shifted in phase by
90°. A transfer with ∆p = +1 experiences a phase of –90° which corresponds to
an equivalent phase of 270°. Thus both pathways experience the same phase
shifts and a cycle which selects one will select the other. The pattern is clear,
this four step cycle will select a pathway with ∆p = –3, as it was designed to,
and also it will select any pathway with ∆p = –3 + 4n where n = ±1, ±2, ±3 ...
9.5.2 General Rules
The discussion in the previous section can be generalised in the following way.
Consider a phase cycle in which the phase of a pulse takes N evenly spaced
steps covering the range 0 to 2π radians. The phases, φ
k
, are
φ
k
= 2πk/N where k = 0, 1, 2 ... (N – 1).
To select a change in coherence order, ∆p, the receiver phase is set to
–∆p × φ
k
for each step and the resulting signals are summed. This cycle will, in
addition to selecting the specified change in coherence order, also select
pathways with changes in coherence order (∆ p ± nN) where
n = ±1, ±2 ..
The way in which phase cycling selects a series of values of ∆p which are
related by a harmonic condition is closely related to the phenomenon of aliasing
in Fourier transformation. Indeed, the whole process of phase cycling can be
9–24
seen as the computation of a discrete Fourier transformation with respect to the
pulse phase; in this case the Fourier codomains are phase and coherence order.
The fact that a phase cycle inevitably selects more than one change in
coherence order is not necessarily a problem. We may actually wish to select
more than one pathway, and examples of this will be given below in relation to
specific twodimensional experiments. Even if we only require one value of ∆p
we may be able to discount the values selected at the same time as being
improbable or insignificant. In a system of m coupled spins onehalf, the
maximum order of coherence that can be generated is m, thus in a two spin
system we need not worry about whether or not a phase cycle will discriminate
between double quantum and six quantum coherences as the latter simply
cannot be present. Even in more extended spin systems the likelihood of
generating highorder coherences is rather small and so we may be able to
discount them for all practical purposes. If a high level of discrimination
between orders is needed, then the solution is simply to use a phase cycle which
has more steps i.e. in which the phases move in smaller increments. For
example a six step cycle will discriminate between ∆p = +2 and +6, whereas a
four step cycle will not.
9.5.3 Refocusing Pulses
A 180° pulse simply changes the sign of the coherence order. This is easily
demonstrated by considering the effect of such a pulse on the operators I
+
and
I
–
. For example:
I I I I I I
x y
I
x y
x
+
≡ +
( )
→
( )
≡ i i
π
–
–
corresponds to p = +1 → p = –1. In a more complex product of operators, each
raising and lowering operator is affected in this way so overall the coherence
order changes sign.
We can now derive the EXORCYLE phase cycle using this property.
Consider a 180° pulse acting on single quantum coherence, for which the CTP
is shown opposite. For the pathway starting with p = 1 the effect of the 180°
pulse is to cause a change with ∆p = –2. The table shows a fourstep cycle to
select this change
Step
phase of
180° pulse
phase shift experienced by
transfer with ∆p = –2
Equivalent phase
= rx. phase
1 0 0 0
2 90 180 180
3 180 360 0
4 270 540 180
The phase cycle is thus 0, 90°, 180°, 270° for the 180° pulse and
0° 180° 0° 180° for the receiver; this is precisely the set of phases deduced
before for EXORCYCLE in section 9.2.7.
1
0
–1
180°
A 180° pulse simply changes
the sign of the coherence order.
The EXORCYLE phase cycling
selects both of the pathways
shown.
9–25
As the cycle has four steps, a pathway with ∆p = +2 is also selected; this is
the pathway which starts with p = –1 and is transferred to p = +1. Therefore,
the four steps of EXORCYLE select both of the pathways shown in the diagram
above.
A two step cycle, consisting of 0°, 180° for the 180° pulse and 0°, 0° for the
receiver, can easily be shown to select all even values of ∆p. This reduced form
of EXORCYCLE is sometimes used when it is necessary to minimise the
number of steps in a phase cycle. An eight step cycle, in which the 180° pulse
is advanced in steps of 45°, can be used to select the refocusing of double
quant um coher ence i n whi ch t he t r ansf er i s f r om
p = +2 to –2 (i.e. ∆p = –4) or vice versa.
9.5.4 Combining phase cycles
Suppose that we wish to select the pathway shown opposite; for the first pulse
∆p is 1 and for the second it is –2. We can construct a fourstep cycle for each
pulse, but to select the overall pathway shown these two cycles have to be
completed independently of one another. This means that there will be a total
of sixteen steps. The table shows how the appropriate receiver cycling can be
determined
Step
phase of 1st
pulse
phase for
∆p = 1
phase of 2nd
pulse
phase for
∆p = –2
total
phase
equivalent phase =
rx. phase
1 0 0 0 0 0 0
2 90 –90 0 0 –90 270
3 180 –180 0 0 –180 180
4 270 –270 0 0 –270 90
5 0 0 90 180 180 180
6 90 –90 90 180 90 90
7 180 –180 90 180 0 0
8 270 –270 90 180 –90 270
9 0 0 180 360 360 0
10 90 –90 180 360 270 270
11 180 –180 180 360 180 180
12 270 –270 180 360 90 90
13 0 0 270 540 540 180
14 90 –90 270 540 450 90
15 180 –180 270 540 360 0
16 270 –270 270 540 270 270
In the first four steps the phase of the second pulse is held constant and the
phase of the first pulse simply goes through the four steps 0° 90° 180° 270°. As
we are selecting ∆p = 1 for this pulse, the receiver phases are simply 0°, 270°,
180°, 90°.
Steps 5 to 8 are a repeat of steps 1–4 except that the phase of the second
pulse has been moved by 90°. As ∆p for the second pulse is –2, the required
1
0
–1
9–26
pathway experiences a phase shift of 180° and so the receiver phase must be
advanced by this much. So, the receiver phases for steps 5–8 are just 180°
ahead of those for steps 1–4.
In the same way for steps 9–12 the first pulse again goes through the same
four steps, and the phase of the second pulse is advanced to 180°. Therefore,
compared to steps 1–4 the receiver phases in steps 9–12 need to be advanced by
– (–2) × 180° = 360° = 0°. Likewise, the receiver phases for steps 13–16 are
advanced by – (–2) × 270° = 540° = 180°.
Another way of looking at this is to consider each step individually. For
example, compared to step 1, in step 14 the first pulse has been advanced by
90° so the phase from the first pulse is – (1) × 90° = –90°. The second pulse
has been advanced by 270° so the phase from this is – (–2) × 270° = 540°. The
total phase shift of the required pathway is thus –90 + 540 = 450° which is an
equivalent phase of 90°. This is the receiver phase shown in the final column.
The key to devising these sequences is to simply work out the two fourstep
cycles independently and then merge them together rather than trying to work
on the whole cycle. One writes down the first four steps, and then duplicates
this four times as the second pulse is shifted. We would find the same steps, in
a different sequence, if the phase of the second pulse is shifted in the first four
steps.
We can see that the total size of a phase cycle grows at an alarming rate.
With four phases for each pulse the number of steps grows as 4
l
where l is the
number of pulses in the sequence. A threepulse sequence such as NOESY or
DQF COSY would therefore involve a 64 step cycle. Such long cycles put a
lower limit on the total time of an experiment and we may end up having to run
an experiment for a long time not to achieve the desired signaltonoise ratio
but simply to complete the phase cycle.
Fortunately, there are several "tricks" which we can use in order to shorten
the length of a phase cycle. To appreciate whether or not one of these tricks
can be used in a particular sequence we need to understand in some detail what
the sequence is actually doing and what the likely problems are going to be.
9.5.5 Tricks
9.5.5.1 The first pulse
All pulse sequences start with equilibrium magnetization, which has coherence
order 0. It can easily be shown that when a pulse is applied to equilibrium
magnetization the only coherence orders that can be generated are ±1. If
retaining both of these orders is acceptable (which it often is), it is therefore not
necessary to phase cycle the first pulse in a sequence.
There are two additional points to make here. If the spins have not relaxed
completely by the start of the sequence the initial magnetization will not be at
equilibrium. Then, the above simplification does not apply. Secondly, the first
pulse of a sequence is often cycled in order to suppress axial peaks in two
9–27
dimensional spectra. This is considered in more detail in section 9.5.6.
9.5.5.2 Grouping pulses together
The sequence shown opposite can be used to generate multiple quantum
coherence from equilibrium magnetization; during the spin echo antiphase
magnetization develops and the final pulse transfers this into multiple quantum
coherence. Let us suppose that we wish to generate double quantum, with p =
±2, as show by the CTP opposite.
As has already been noted, the first pulse can only generate p = ±1 and the
180° pulse only causes a change in the sign of the coherence order. The only
pulse we need to be concerned with is the final one which we want to generate
only double quantum. We could try to devise a phase cycle for the last pulse
alone or we could simply group all three pulses together and imagine that, as a
group, they achieve the transformation p = 0 to p = ±2 i.e. ∆p = ±2. The phase
cycle would simply be for the three pulses together to go 0°, 90°, 180°, 270°,
with the receiver going 0°, 180°, 0°, 180°.
It has to be recognised that by cycling a group of pulses together we are only
selecting an overall transformation; the coherence orders present within the
group of pulses are not being selected. It is up to the designer of the experiment
to decide whether or not this degree of selection is sufficient.
The four step cycle mentioned above also selects ∆p = ±6; again, we would
have to decide whether or not such high orders of coherence were likely to be
present in the spin system. Finally, we note that the ∆p values for the final
pulse are ±1, ±3; it would not be possible to devise a four step cycle which
selects all of these pathways.
9.5.5.3 The last pulse
We noted above that only coherence order –1 is observable. So, although the
final pulse of a sequence may cause transfer to many different orders of
coherence, only transfers to p = –1 will result in observable signals. Thus, if
we have already selected, in an unambiguous way, a particular set of coherence
orders present just before the last pulse, no further cycling of this pulse is
needed.
9.5.5.4 Example – DQF COSY
A good example of the applications of these ideas is in devising a phase cycle
for DQF COSY, whose pulse sequence and CTP is shown below.
t
1
t
2
2
1
0
–1
–2
p
∆p=±1 ±1,±3
+1,–3
Note that we have retained symmetrical pathways in t
1
so that absorption mode
2
1
0
–1
–2
Pulse sequence for generating
doublequantum coherence.
Note that the 180° pulse simply
causes a change in the sign of
the coherence order.
9–28
lineshapes can be obtained. Also, both in generating the double quantum
coherence, and in reconverting it to observable magnetization, all possible
pathways have been retained. If we do not do this, signal intensity is lost.
One way of viewing this sequence is to group the first two pulses together
and view them as achieving the transformation 0 → ±2 i.e.
∆p = ±2. This is exactly the problem considered in section 9.5.5.2, where we
saw that a suitable four step cycle is for the first two pulses to go 0°, 90°, 180°,
270° and the receiver to go 0°, 180°, 0°, 180°. This unambiguously selects p =
±2 just before the last pulse, so phase cycling of the last pulse is not required
(see section 9.5.5.3).
An alternative view is to say that as only p = –1 is observable, selecting the
transformation ∆p = +1 and –3 on the last pulse will be equivalent to selecting p
= ±2 during the period just before the last pulse. Since the first pulse can only
generate p = ±1 (present during t
1
), the selection of ∆p = +1 and –3 on the last
pulse is sufficient to define the CTP completely.
A four step cycle to select ∆p = +1 involves the pulse going 0°, 90°, 180°,
270° and the receiver going 0°, 270°, 180°, 90°. As this cycle has four steps is
automatically also selects ∆p = –3, just as required.
The first of these cycles also selects ∆p = ±6 for the first two pulses i.e.
filtration through sixquantum coherence; normally, we can safely ignore the
possibility of such highorder coherences. The second of the cycles also selects
∆p = +5 and ∆p = –7 on the last pulse; again, these transfers involve such high
orders of multiple quantum that they can be ignored.
9.5.6 Axial peak suppression
Peaks are sometimes seen in twodimensional spectra at coordinates F
1
= 0 and
F
2
= frequencies corresponding to the usual peaks in the spectrum. The
interpretation of the appearance of these peaks is that they arise from
magnetization which has not evolved during t
1
and so has not acquired a
frequency label.
A common source of axial peaks is magnetization which recovers due to
longitudinal relaxation during t
1
. Subsequent pulses make this magnetization
observable, but it has no frequency label and so appears at
F
1
= 0. Another source of axial peaks is when, due to pulse imperfections, not
all of the original equilibrium magnetization is moved into the transverse plane
by the first pulse. The residual longitudinal magnetization can be made
observable by subsequent pulses and hence give rise to axial peaks.
A simple way of suppressing axial peaks is to select the pathway ∆p = ±1 on
the first pulse; this ensures that all signals arise from the first pulse. A twostep
cycle in which the first pulse goes 0°, 180° and the receiver goes 0°, 180°
selects ∆p = ±1. It may be that the other phase cycling used in the sequence
will also reject axial peaks so that it is not necessary to add an explicit axial
peak suppression steps. Adding a twostep cycle for axial peak suppression
9–29
doubles the length of the phase cycle.
9.5.7 Shifting the whole sequence – CYCLOPS
If we group all of the pulses in the sequence together and regard them as a unit
they simply achieve the transformation from equilibrium magnetization, p = 0,
to observable magnetization, p = –1. They could be cycled as a group to select
this pathway with ∆p = –1, that is the pulses going 0°, 90°, 180°, 270° and the
receiver going 0°, 90°, 180°, 270°. This is simple the CYCLOPS phase cycle
described in section 9.2.6.
If time permits we sometimes add CYCLOPSstyle cycling to all of the
pulses in the sequence so as to suppress some artefacts associated with
imperfections in the receiver. Adding such cycling does, of course, extend the
phase cycle by a factor of four.
This view of the whole sequence as causing the transformation ∆p = –1 also
enables us to interchange receiver and pulse phase shifts. For example, suppose
that a particular step in a phase cycle requires a receiver phase shift θ. The
same effect can be achieved by shifting all of the pulses by –θ and leaving the
receiver phase unaltered. The reason this works is that all of the pulses taken
together achieve the transformation ∆p = –1, so shifting their phases by –θ shift
the signal by – (–θ) = θ, which is exactly the effect of shifting the receiver by θ.
This kind of approach is sometimes helpful if hardware limitations mean that
small angle phaseshifts are only available for the pulses.
9.5.8 Equivalent cycles
For even a relatively simple sequence such as DQF COSY there are a number
of different ways of writing the phase cycle. Superficially these can look very
different, but it may be possible to show that they really are the same.
For example, consider the DQF COSY phase cycle proposed in section
9.5.5.4 where we cycle just the last pulse
step 1st pulse 2nd pulse 3rd pulse receiver
1 0 0 0 0
2 0 0 90 270
3 0 0 180 180
4 0 0 270 90
Suppose we decide that we do not want to shift the receiver phase, but want to
keep it fixed at phase zero. As described above, this means that we need to
subtract the receiver phase from all of the pulses. So, for example, in step 2 we
subtract 270° from the pulse phases to give –270°, –270° and –180° for the
phases of the first three pulses, respectively; reducing these to the usual range
gives phases 90°, 90° and 180°. Doing the same for the other steps gives a
rather strange looking phase cycle, but one which works in just the same way.
step 1st pulse 2nd pulse 3rd pulse receiver
9–30
1 0 0 0 0
2 90 90 180 0
3 180 180 0 0
4 270 270 180 0
We can play one more trick with this phase cycle. As the third pulse is required
to achieve the transformation ∆p = –3 or +1 we can alter its phase by 180° and
compensate for this by shifting the receiver by 180° also. Doing this for steps 2
and 4 only gives
step 1st pulse 2nd pulse 3rd pulse receiver
1 0 0 0 0
2 90 90 0 180
3 180 180 0 0
4 270 270 0 180
This is exactly the cycle proposed in section 9.5.5.4.
9.5.9 Further examples
In this section we will use a shorthand to indicate the phases of the pulses and
the receiver. Rather than specifying the phase in degrees, the phases are
expressed as multiples of 90°. So, EXORCYCLE becomes 0 1 2 3 for the
180° pulse and 0 2 0 2 for the receiver.
9.5.9.1 Double quantum spectroscopy
A simple sequence for double quantum spectroscopy is shown below
t
1
t
2
2
1
0
–1
–2
τ τ
Note that both pathways with p = ±1 during the spin echo and with p = ±2
during t
1
are retained. There are a number of possible phase cycles for this
experiment and, not surprisingly, they are essentially the same as those for DQF
COSY. If we regard the first three pulses as a unit, then they are required to
achieve the overall transformation ∆p = ±2, which is the same as that for the
first two pulses in the DQF COSY sequence. Thus the same cycle can be used
with these three pulses going 0 1 2 3 and the receiver going 0 2 0 2.
Alternatively the final pulse can be cycled 0 1 2 3 with the receiver going 0 3
2 1, as in section 9.5.5.4.
Both of these phase cycles can be extended by EXORCYCLE phase cycling
of the 180° pulse, resulting in a total of 16 steps.
9.5.9.2 NOESY
The pulse sequence for NOESY (with retention of absorption mode lineshapes)
is shown below
9–31
t
1
t
2
1
0
–1
τ
mix
If we group the first two pulses together they are required to achieve the
transformation ∆p = 0 and this leads to a four step cycle in which the pulses go
0 1 2 3 and the receiver remains fixed as 0 0 0 0. In this experiment axial
peaks arise due to zmagnetization recovering during the mixing time, and this
cycle will not suppress these contributions. Thus we need to add axial peak
suppression, which is conveniently done by adding the simple cycle 0 2 on the
first pulse and the receiver. The final 8 step cycle is 1st pulse: 0 1 2 3 2 3
0 1, 2nd pulse: 0 1 2 3 0 1 2 3, 3rd pulse fixed, receiver: 0 0 0 0 2 2
2 2.
An alternative is to cycle the last pulse to select the pathway ∆p = –1, giving
the cycle 0 1 2 3 for the pulse and 0 1 2 3 for the receiver. Once again, this
does not discriminate against zmagnetization which recovers during the mixing
time, so a two step phase cycle to select axial peaks needs to be added.
9.5.9.3 Heteronuclear Experiments
The phase cycling for most heteronuclear experiments tends to be rather trivial
in that the usual requirement is simply to select that component which has been
transferred from one nucleus to another. We have already seen in section 9.2.8
that this is achieved by a 0 2 phase cycle on one of the pulses causing the
transfer accompanied by the same on the receiver i.e. a difference experiment.
The choice of which pulse to cycle depends more on practical considerations
than with any fundamental theoretical considerations.
The pulse sequence for HMQC, along with the CTP, is shown below
t
2
1
0
–1
t
1
1
0
–1
∆ ∆
I
S
p
I
p
S
Note that separate coherence orders are assigned to the I and S spins.
Observable signals on the I spin must have p
I
= –1 and p
S
= 0 (any other value
of p
S
would correspond to a heteronuclear multiple quantum coherence). Given
this constraint, and the fact that the I spin 180° pulse simply inverts the sign of
p
I
, the only possible pathway on the I spins is that shown.
The S spin coherence order only changes when pulses are applied to those
spins. The first 90° S spin pulse generates p
S
= ±1, just as before. As by this
point p
I
= +1, the resulting coherences have p
S
= +1, p
I
= –1 (heteronuclear
zeroquantum) and p
S
= +1, p
I
= +1 (heteronuclear doublequantum). The I spin
9–32
180° pulse interconverts these midway during t
1
, and finally the last S spin
pulse returns both pathways to p
S
= 0. A detailed analysis of the sequence
shows that retention of both of these pathways results in amplitude modulation
in t
1
(provided that homonuclear couplings between I spins are not resolved in
the F
1
dimension).
Usually, the I spins are protons and the S spins some lowabundance
heteronucleus, such as
13
C. The key thing that we need to achieve is to suppress
the signals arising from vast majority of I spins which are not coupled to S
spins. This is achieved by cycling a pulse which affects the phase of the
required coherence but which does not affect that of the unwanted coherence.
The obvious targets are the two S spin 90° pulses, each of which is required to
give the transformation ∆p
S
= ±1. A two step cycle with either of these pulses
going 0 2 and the receiver doing the same will select this pathway and, by
difference, suppress any I spin magnetization which has not been passed into
multiple quantum coherence.
It is also common to add EXORCYCLE phase cycling to the I spin 180°
pulse, giving a cycle with eight steps overall.
9.5.10 General points about phase cycling
Phase cycling as a method suffers from two major practical problems. The first
is that the need to complete the cycle imposes a minimum time on the
experiment. In two and higherdimensional experiments this minimum time
can become excessively long, far longer than would be needed to achieve the
desired signaltonoise ratio. In such cases the only way of reducing the
experiment time is to record fewer increments which has the undesirable
consequence of reducing the limiting resolution in the indirect dimensions.
The second problem is that phase cycling always relies on recording all
possible contributions and then cancelling out the unwanted ones by combining
subsequent signals. If the spectrum has high dynamic range, or if spectrometer
stability is a problem, this cancellation is less than perfect. The result is
unwanted peaks and t
1
noise appearing in the spectrum. These problems
become acute when dealing with proton detected heteronuclear experiments on
natural abundance samples, or in trying to record spectra with intense solvent
resonances.
Both of these problems are alleviated to a large extent by moving to an
alternative method of selection, the use of field gradient pulses, which is the
subject of the next section. However, as we shall see, this alternative method is
not without its own difficulties.
9.6 Selection with field gradient pulses
9.6.1 Introduction
Like phase cycling, field gradient pulses can be used to select particular
coherence transfer pathways. During a pulsed field gradient the applied
9–33
magnetic field is made spatially inhomogeneous for a short time. As a result,
transverse magnetization and other coherences dephase across the sample and
are apparently lost. However, this loss can be reversed by the application of a
subsequent gradient which undoes the dephasing process and thus restores the
magnetization or coherence. The crucial property of the dephasing process is
that it proceeds at a different rate for different coherences. For example,
doublequantum coherence dephases twice as fast as singlequantum coherence.
Thus, by applying gradient pulses of different strengths or durations it is
possible to refocus coherences which have, for example, been changed from
single to doublequantum by a radiofrequency pulse.
Gradient pulses are introduced into the pulse sequence in such a way that
only the wanted signals are observed in each experiment. Thus, in contrast to
phase cycling, there is no reliance on subtraction of unwanted signals, and it
can thus be expected that the level of t
1
noise will be much reduced. Again in
contrast to phase cycling, no repetitions of the experiment are needed, enabling
the overall duration of the experiment to be set strictly in accord with the
required resolution and signaltonoise ratio.
The properties of gradient pulses and the way in which they can be used to
select coherence transfer pathways have been known since the earliest days of
multiplepulse NMR. However, in the past their wide application has been
limited by technical problems which made it difficult to use such pulses in
highresolution NMR. The problem is that switching on the gradient pulse
induces currents in any nearby conductors, such as the probe housing and
magnet bore tube. These induced currents, called eddy currents, themselves
generate magnetic fields which perturb the NMR spectrum. Typically, the eddy
currents are large enough to disrupt severely the spectrum and can last many
hundreds of milliseconds. It is thus impossible to observe a highresolution
spectrum immediately after the application of a gradient pulse. Similar
problems have beset NMR imaging experiments and have led to the
development of shielded gradient coils which do not produce significant
magnetic fields outside the sample volume and thus minimise the generation of
eddy currents. The use of this technology in highresolution NMR probes has
made it possible to observe spectra within tens of microseconds of applying a
gradient pulse. With such apparatus, the use of field gradient pulses in high
resolution NMR is quite straightforward, a fact first realised and demonstrated
by Hurd whose work has pioneered this whole area.
9.6.2 Dephasing caused by gradients
A field gradient pulse is a period during which the B
0
field is made spatially
inhomogeneous; for example an extra coil can be introduced into the sample
probe and a current passed through the coil in order to produce a field which
varies linearly in the zdirection. We can imagine the sample being divided into
thin discs which, as a consequence of the gradient, all experience different
magnetic fields and thus have different Larmor frequencies. At the beginning
9–34
of the gradient pulse the vectors representing transverse magnetization in all
these discs are aligned, but after some time each vector has precessed through a
different angle because of the variation in Larmor frequency. After sufficient
time the vectors are disposed in such a way that the net magnetization of the
sample (obtained by adding together all the vectors) is zero. The gradient pulse
is said to have dephased the magnetization.
It is most convenient to view this dephasing process as being due to the
generation by the gradient pulse of a spatially dependent phase. Suppose that
the magnetic field produced by the gradient pulse, B
g
, varies linearly along the
zaxis according to
B Gz
g
=
where G is the gradient strength expressed in, for example, T m
–1
or G cm
–1
; the
origin of the zaxis is taken to be in the centre of the sample. At any particular
position in the sample the Larmor frequency, ω
L
(z), depends on the applied
magnetic field, B
0
, and B
g
ω γ γ
L 0 g 0
= +
( )
= + ( ) B B B Gz
,
where γ is the gyromagnetic ratio. After the gradient has been applied for time
t, the phase at any position in the sample, Φ(z), is given by Φ z B Gz t ( ) = + ( ) γ
0
.
The first part of this phase is just that due to the usual Larmor precession in the
absence of a field gradient. Since this is constant across the sample it will be
ignored from now on (which is formally the same result as viewing the
magnetization in a frame of reference rotating at γB
0
). The remaining term γGzt
is the spatially dependent phase induced by the gradient pulse.
If a gradient pulse is applied to pure xmagnetization, the following
evolution takes place at a particular position in the sample
I Gzt I Gzt I
x
GztI
x y
z
γ
γ γ → ( ) + ( ) cos sin .
The total xmagnetization in the sample, M
x
, is found by adding up the
magnetization from each of the thin discs, which is equivalent to the integral
M
x
t ( ) =
1
r
max
cos γGzt ( ) dz
–
1
2
r
max
1
2
r
max
∫
where it has been assumed that the sample extends over a region ±
1
2
r
max
.
Evaluating the integral gives an expression for the decay of xmagnetization
during a gradient pulse
M t
Gr t
Gr t
x
( ) =
( ) sin
1
2
1
2
γ
γ
max
max
[11]
The plot below shows M
x
(t) as a function of time
9–35
10 20 30 40 50
0.5
0.0
0.5
1.0
M
x
Gr
max
t
γ
The black line shows the decay of magnetization due to the action of a gradient pulse. The grey line is an
approximation, valid at long times, for the envelope of the decay.
Note that the oscillations in the decaying magnetization are imposed on an
overall decay which, for long times, is given by 2/(γGtr
max
). Equation [11]
embodies the obvious points that the stronger the gradient (the larger G) the
faster the magnetization decays and that magnetization from nuclei with higher
gyromagnetic ratios decays faster. It also allows a quantitative assessment of
the gradient strengths required: the magnetization will have decayed to a
fraction α of its initial value after a time of the order of 2 γ α G r
max
( ) (the
relation is strictly valid for α << 1). For example, if it is assumed that r
max
is 1
cm, then a 2 ms gradient pulse of strength 0.37 T m
–1
(37 G cm
–1
) will reduce
proton magnetization by a factor of 1000. Gradients of such strength are
readily obtainable using modern shielded gradient coils that can be built into
high resolution NMR probes
This discussion now needs to be generalised for the case of a field gradient
pulse whose amplitude is not constant in time, and for the case of dephasing a
general coherence of order p. The former modification is of importance as for
instrumental reasons the amplitude envelope of the gradient is often shaped to a
smooth function. In general after applying a gradient pulse of duration τ the
spatially dependent phase, Φ(r,τ) is given by
Φ r sp B r ,τ γ τ ( ) = ( )
g
[12]
The proportionality to the coherence order comes about due to the fact that the
phase acquired as a result of a zrotation of a coherence of order p through an
angle φ is pφ, (see Eqn. [2] in section 9.3.1). In Eqn. [12] s is a shape factor: if
the envelope of the gradient pulse is defined by the function A(t), where
A t ( ) ≤1, s is defined as the area under A(t)
s A t t = ( )
∫
1
0
τ
τ
d
The shape factor takes a particular value for a certain shape of gradient,
regardless of its duration. A gradient applied in the opposite sense, that is with
the magnetic field decreasing as the zcoordinate increases rather than vice
9–36
versa, is described by reversing the sign of s. The overall amplitude of the
gradient is encoded within B
g
.
In the case that the coherence involves more than one nuclear species, Eqn.
[12] is modified to take account of the different gyromagnetic ratio for each
spin, γ
i
, and the (possibly) different order of coherence with respect to each
nuclear species, p
i
:
Φ r sB r p
i i
i
,τ τ γ ( ) = ( )
∑ g
From now on we take the dependence of Φ on r and τ, and of B
g
on r as being
implicit.
9.6.3 Selection by refocusing
The method by which a particular coherence transfer pathway is selected using
field gradients is illustrated opposite. The first gradient pulse encodes a
spatially dependent phase, Φ
1
and the second a phase Φ
2
where
Φ Φ
1 1 1 1 2 2 2 2
= = s p B s p B γ τ γ τ
g,1 g,2
and
.
After the second gradient the net phase is (Φ
1
+ Φ
2
). To select the pathway
involving transfer from coherence order p
1
to coherence order p
2
, this net phase
should be zero; in other words the dephasing induced by the first gradient pulse
is undone by the second. The condition (Φ
1
+ Φ
2
) = 0 can be rearranged to
s B
s B
p
p
1 1
2 2
2
1
g,1
g,2
τ
τ
=
–
. [13]
For example, if p
1
= +2 and p
2
= –1, refocusing can be achieved by making the
second gradient either twice as long (τ
2
= 2τ
1
), or twice as strong
(B
g,2
= 2B
g,1
) as the first; this assumes that the two gradients have identical
shape factors. Other pathways remain dephased; for example, assuming that we
have chosen to make the second gradient twice as strong and the same duration
as the first, a pathway with p
1
= +3 to p
2
= –1 experiences a net phase
Φ Φ
1 2 1 1 1
3 + = = sB sB sB
g,1 g,2 g,1
τ τ τ –
.
Provided that this spatially dependent phase is sufficiently large, according the
criteria set out in the previous section, the coherence arising from this pathway
remains dephased and is not observed. To refocus a pathway in which there is
no sign change in the coherence orders, for example, p
1
= –2 to p
2
= –1, the
second gradient needs to be applied in the opposite sense to the first; in terms of
Eqn. [13] this is expressed by having s
2
= –s
1
.
The procedure can easily be extended to select a more complex coherence
transfer pathway by applying further gradient pulses as the coherence is
transferred by further pulses, as illustrated opposite. The condition for
refocusing is again that the net phase acquired by the required pathway be zero,
which can be written formally as
p
1
p
2
RF
g
1
τ
2
τ
Illustration of the use of a pair
of gradients to select a single
pathway. The radiofrequency
pulses are on the line marked
"RF" and the field gradient
pulses are denoted by shaded
rectangles on the line marked
"g".
RF
g
1
τ
3
τ
4
τ
2
τ
9–37
s p B
i i i i
i
γ τ
g, i ∑
= 0 .
With more than two gradients in the sequence, there are many ways in which a
given pathway can be selected. For example, the second gradient may be used
to refocus the first part of the required pathway, leaving the third and fourth to
refocus another part. Alternatively, the pathway may be consistently dephased
and the magnetization only refocused by the final gradient, just before
acquisition.
At this point it is useful to contrast the selection achieved using gradient
pulses with that achieved using phase cycling. From Eqn. [13] it is clear that a
particular pair of gradient pulses selects a particular ratio of coherence orders;
in the above example any two coherence orders in the ratio –2 : 1 or 2 : –1 will
be refocused. This selection according to ratio of coherence orders is in
contrast to the case of phase cycling in which a phase cycle consisting of N
steps of 2π/N radians selects a particular change in coherence order ∆p = p
2
–
p
1
, and further pathways which have ∆p = (p
2
– p
1
) ± mN, where m = 0, 1, 2 ...
It is straightforward to devise a series of gradient pulses which will select a
single coherence transfer pathway. It cannot be assumed, however, that such a
sequence of gradient pulses will reject all other pathways i.e. leave coherence
from all other pathways dephased at the end of the sequence. Such assurance
can only be given be analysing the fate of all other possible coherence transfer
pathways under the particular gradient sequence proposed. In complex pulse
sequences there may also be several different ways in which gradient pulses can
be included in order to achieve selection of the desired pathway. Assessing
which of these alternatives is the best, in the light of the requirement of
suppression of unwanted pathways and the effects of pulse imperfections may
be a complex task.
9.6.3.1 Selection of multiple pathways
As we have seen earlier, it is not unusual to want to select two or more
pathways simultaneously, for example either to maximise the signal intensity or
to retain absorptionmode lineshapes. A good example of this is the double
quantum filter pulse sequence element, shown opposite.
9–38
The ideal pathway, shown in (a), preserves coherence orders p = ± 2 during the
interpulse delay. Gradients can be used to select the pathway –2 to –1 or +2 to
–1, shown in (b) and (c) respectively. However, no combination of gradients
can be found which will select simultaneously both of these pathways. In
contrast, it is easy to devise a phase cycle which selects both of these pathways
(section 9.5.5.4). Thus, selection with gradients will in this case result in a loss
of half of the available signal when compared to an experiment of equal length
which uses selection by phase cycling. Such a loss in signal is, unfortunately, a
very common feature when gradients are used for pathway selection.
9.6.3.2 Selection versus suppression
Coherence order zero, comprising zmagnetization, zzterms and homonuclear
zeroquantum coherence, does not accrue any phase during a gradient pulse.
Thus, it can be separated from all other orders simply by applying a single
gradient. In a sense, however, this is not a gradient selection process; rather it
is a suppression of all other coherences. A gradient used in this way is often
called a purge gradient. In contrast to experiments where selection is achieved,
there is no inherent sensitivity loss when gradients are used for suppression.
We will see examples below of cases where this suppression approach is very
useful.
9.6.3.3 Gradients on other axes
The simplest experimental arrangement generates a gradient in which the
magnetic field varies in the z direction, however it is also possible to generate
gradients in which the field varies along x or y. Clearly, the spatially dependent
phase generated by a gradient applied in one direction cannot be refocused by a
gradient applied in a different direction. In sequences where more than one pair
of gradients are used, it may be convenient to apply further gradients in
different directions to the first pair, so as to avoid the possibility of accidentally
refocusing unwanted coherence transfer pathways. Likewise, a gradient which
is used to destroy all coherences can be applied in a different direction to
gradients used for pathway selection.
9.6.4 Refocusing and inversion pulses
Refocusing and inversion pulses play an important role in multiplepulse NMR
experiments and so the interaction between such pulses and field gradient
pulses will be explored in some detail. As has been noted above in section
9.5.3, a perfect refocusing pulse simply changes the sign of the order of any
coherences present, p → –p. If the pulse is imperfect, there will be transfer to
coherence orders other than –p.
A perfect inversion pulse simply inverts zmagnetization or, more generally,
all zoperators: I
z
→ –I
z
. If the pulse is imperfect, it will generate transverse
magnetization or other coherences. Inversion pulses are used extensively in
heteronuclear experiments to control the evolution of heteronuclear couplings.
RF
g
2
1
0
–1
–2
p
RF
2
1
0
–1
–2
p
g
RF
2
1
0
–1
–2
p
(a)
(b)
(c)
9–39
We start out the discussion by considering the refocusing of coherences,
illustrated opposite. The net phase, Φ, at the end of such a sequence is
Φ Ω Ω = + + +
( ) ( ' )
'
p p
sp B sp B δ γ τ δ γ τ
g g
where Ω
p ( )
is the frequency with which coherence of order p evolves in the
absence of a gradient; note that Ω Ω
− ( ) ( )
= −
p p
. The net phase is zero if, and
only if, p' = –p. With sufficiently strong gradients all other pathways remain
dephased and the gradient sequence thus selects the refocused component. As
is expected for a spin echo, the underlying evolution of the coherence (as would
occur in the absence of a gradient) is also refocused by the selection of the
pathway shown. Any transverse magnetization which an imperfect refocusing
pulse might create is also dephased.
Placing equal gradients either side of a refocusing pulse therefore selects the
coherence transfer pathway associated with a perfect refocusing pulse. This
selection works for all coherence orders so, in contrast to the discussion in
section 9.6.3.1, there is no loss of signal. Such a pair of gradients are often
described as being used to "clean up" a refocusing pulse, referring to their role
in eliminating unwanted pathways.
We cannot use gradients to select the pathway associated with an inversion
pulse as p = 0 both before and after the pulse. However, we can apply a
gradient after the pulse to dephase any magnetization which might be created
by an imperfect pulse. Taking the process a step further, we can apply a
gradient both before and after the pulse, with the two gradients in opposite
directions. The argument here is that this results in the maximum dephasing of
unwanted coherences – both those present before the pulse and those that might
be generated by the pulse. Again, this sequence is often described as being
used to "clean up" an inversion pulse.
In heteronuclear experiments an inversion pulse applied to one nucleus is
used to refocus the evolution of a coupling to another nucleus. For example, in
the sequence shown opposite the centrally placed S spin 180° pulse refocuses
the IS coupling over the period 2δ. The pair of gradients shown have no net
effect on I spin coherences as the dephasing due to the first gradient is exactly
reversed by the second. The gradient sequence can be thought of as "cleaning
up" the S spin inversion pulse.
9.6.4.1 Phase errors due to gradient pulses
For the desired pathway, the spatially dependent phase created by a gradient
pulse is refocused by a subsequent gradient pulse. However, the underlying
evolution of offsets (chemical shifts) and couplings is not refocused, and phase
errors will accumulate due to the evolution of these terms. Since gradient
pulses are typically of a few milliseconds duration, these phase errors are
substantial.
p
p'
RF
g
τ τ
180°
δ δ
Gradient sequence used to
"clean up" a refocusing pulse.
Note that the two gradients are
of equal area. The refocused
pathway has p' = –p.
RF
g
τ
τ
180°
Gradient sequence used to
"clean up" an inversion pulse.
I
S
g
τ
δ δ
τ
180°
In a heteronuclear experiment
(separate pulses shown on the
I and S spins) a 180° pulse on
the S spin refocuses the I S
coupling over the period 2δ.
The gradient pulses shown are
used to "clean up" the inversion
pulse.
9–40
In multidimensional NMR the uncompensated evolution of offsets during
gradient pulses has disastrous effects on the spectra. This is illustrated here for
the DQF COSY experiment, for which a pulse sequence using the gradient
pulses is shown opposite. The problem with this sequence is that the double
quantum coherence will evolve during delay τ
1
. Normally, the delay between
the last two pulses would be a few microseconds at most, during which the
evolution is negligible. However, once we need to apply a gradient the delay
will need to be of the order of milliseconds and significant evolution takes
place. The same considerations apply to the second gradient.
We can investigate the effect of this evolution by analysing the result for a
twospin system. In the calculation, it will be assumed that only the indicated
pathway survives and that the spatially dependent part of the evolution due to
the gradients can be ignored as ultimately it is refocused. The coherence with
order of + 2 present during τ
1
evolves according to
I I I I i
I I
z z
1 2 1 2 1 2 1
1 1 1 2 1 2
+ +
+
+ +
→ + ( ) ( )
Ω Ω
Ω Ω
τ τ
τ exp – ,
where Ω
1
and Ω
2
are the offsets of spins 1 and 2, respectively. After the final
90° pulse and the second gradient the observable terms on spin 1 are
i
2 1 2 1 1 2 1 2 1 2 1 2
2 2 exp – cos sin i I I I I
x z y z
Ω Ω Ω Ω + ( ) ( )
+
[ ]
τ τ τ [14]
where it has been assumed that τ
2
is sufficiently short that evolution of the
coupling can be ignored. It is clearly seen from Eqn. [14] that, due to the
evolution during τ
2
, the multiplet observed in the F
2
dimension will be a
mixture of dispersion and absorption antiphase contributions. In addition,
there is an overall phase shift due to the evolution during τ
1
. The phase
correction needed to restore this multiplet to absorption depends on both the
frequency in F
2
and the doublequantum frequency during the first gradient.
Thus, no single linear frequency dependent phase correction could phase
correct a spectrum containing many multiplets. The need to control these phase
errors is plain.
The general way to minimise these problems is to associate each gradient
with a refocusing pulse, as shown opposite. In sequence (a) the gradient is
placed in one of the delays of a spin echo; the evolution of the offset during the
first delay τ is refocused during the second delay τ. So, overall there are no
phase errors due to the evolution of the offset.
An alternative is the sequence (b). Here, as in (a), the offset is refocused
over the whole sequence. The first gradient results in the usually spatially
dependent phase and then the 180° pulse changes the sign of the coherence
order. As the second gradient is opposite to the first, it causes further
dephasing; effectively, it is as if a gradient of length 2τ is applied. Sequence
(b) will give the same dephasing effect as (a) if each gradient in (b) is of
duration τ/2; the overall sequence will then be of duration τ. If losses due to
relaxation are a problem, then clearly sequence (b) is to be preferred as it takes
less time than (a).
g
RF
2
1
0
–1
–2
t
1
t
2
1
τ
2
τ
Simple pulse sequence for
DQF COSY with gradient
selection and retention of
symmetrical pathways in t
1
.
The second gradient is twice
the area of the first.
RF
180°
g
τ
τ τ
(a)
(b)
RF
180°
g
τ
τ
τ τ
9–41
The sequence below shows the gradientselected DQF COSY pulse sequence
modified by the inclusion of extra 180° pulses to remove phase errors. Note
that although the extra 180° pulses are effective at refocusing offsets, they do
not refocus the evolution of homonuclear couplings. It is essential, therefore, to
keep the gradient pulses as short as is feasible.
g
RF
t
1
1
τ
1
τ
1
τ
2
τ
2
τ
2
τ
In many pulse sequences there are already periods during which the
evolution of offsets is refocused. The evolution of offsets during a gradient
pulse placed within such a period will therefore also be refocused, making it
unnecessary to include extra refocusing pulses. Likewise, a gradient may be
placed during a "constant time" evolution period of a multidimensional pulse
sequence without introducing phase errors in the corresponding dimension; the
gradient simply becomes part of the constant time period. This approach is
especially useful in three and fourdimensional experiments used to record
spectra of
15
N,
13
C labelled proteins.
9.6.5 Sensitivity
The use of gradients for coherence selection has consequences for the signalto
noise ratio of the spectrum when it is compared to a similar spectrum recorded
using phase cycling. If a gradient is used to suppress all coherences other than
p = 0, i . e. it is used simply to remove all coherences, leaving just z
magnetization or zz terms, there is no inherent loss of sensitivity when
compared to a corresponding phase cycled experiment. If, however, the
gradient is used to select a particular order of coherence the signal which is
subsequently refocused will almost always be half the intensity of that which
can be observed in a phase cycled experiment. This factor comes about simply
because it is likely that the phase cycled experiment will be able to retain two
symmetrical pathways, whereas the gradient selection method will only be able
to refocus one of these.
The foregoing discussion applies to the case of a selection gradient placed in
a fixed delay of a pulse sequence. The matter is different if the gradient is
placed within the incrementable time of a multidimensional experiment, e.g. in
t
1
of a twodimensional experiment. To understand the effect that such a
gradient has on the sensitivity of the experiment it is necessary to be rather
careful in making the comparison between the gradient selected and phase
cycled experiments. In the case of the latter experiments we need to include the
SHR or TPPI method in order to achieve frequency discrimination with
absorption mode lineshapes. If a gradient is used in t
1
we will need to record
separate P and Ntype spectra so that they can be recombined to give an
9–42
absorption mode spectrum. We must also ensure that the two spectra we are
comparing have the same limiting resolution in the t
1
dimension, that is they
achieve the same maximum value of t
1
and, of course, the total experiment time
must be the same.
The detailed argument which is needed to analyse this problem is beyond the
scope of this lecture; it is given in detail in J. Magn. Reson Ser. A, 111, 7076
(1994)
†
. The conclusion is that the signaltonoise ratio of an absorption mode
spectrum generated by recombining P and Ntype gradient selected spectra is
lower, by a factor 1 2 , than the corresponding phase cycled spectrum with
SHR or TPPI data processing.
The potential reduction in sensitivity which results from selection with
gradients may be more than compensated for by an improvement in the quality
of the spectra obtained in this way. Often, the factor which limits whether or
not a cross peak can be seen is not the thermal noise level by the presence of
other kinds of "noise" associated with imperfect cancellation etc.
9.6.6 Diffusion
The process of refocusing a coherence which has been dephased by a gradient
pulse is inhibited if the spins move either during or between the defocusing and
refocusing gradients. Such movement alters the magnetic field experienced by
the spins so that the phase acquired during the refocusing gradient is not exactly
opposite to that acquired during the defocusing gradient.
In liquids there is a translational diffusion of both solute and solvent which
causes such movement at a rate which is fast enough to cause significant effects
on NMR experiments using gradient pulses. As diffusion is a random process
we expect to see a smooth attenuation of the intensity of the refocused signal as
the diffusion contribution increases. These effects have been known and
exploited to measure diffusion constants since the very earliest days of NMR.
The effect of diffusion on the signal intensity from the simple echo sequence
shown opposite is relatively simple to analyse and captures all of the essential
points. Note that the two gradient pulses can be placed anywhere in the
intervals δ either side of the 180° pulse. For a single uncoupled resonance, the
intensity of the observed signal, S, expressed as a fraction of the signal intensity
in the absence of a gradient, S
0
is given by
S
S
G D
0
2 2 2
3
= − −
exp γ τ
τ
∆ [15]
where D is the diffusion constant, ∆ is the time between the start of the two
gradient pulses and τ is the duration of the gradient pulses; relaxation has been
ignored. For a given pair of gradient pulses it is diffusion during the interval
between the two pulses, ∆, which determines the attenuation of the echo. The
†
There is an error in this paper: in Fig. 1(b) the penultimate S spin 90° pulse should be phase y
and the final S spin 90° pulse is not required.
RF
g
τ τ
δ
∆
δ
180°
9–43
stronger the gradient the more rapidly the phase varies across the sample and
thus the more rapidly the echo will be attenuated. This is the physical
interpretation of the term γ τ
2 2 2
G in Eqn. [15].
Diffusion constants generally decrease as the molecular mass increases. A
small molecule, such as water, will diffuse up to twenty times faster than a
protein with molecular weight 20,000. The table shows the loss in intensity due
to diffusion for typical gradient pulse pair of 2 ms duration and of strength 10 G
cm
–1
for a small, medium and large sized molecule; data is given for ∆ = 2 ms
and ∆ = 100 ms. It is seen that even for the most rapidly diffusing molecules
the loss of intensity is rather small for ∆ = 2 ms, but becomes significant for
longer delays. For large molecules, the effect is small in all cases.
Fraction of transverse magnetization refocused after a spin echo with gradient refocusing
a
∆/ms small molecule
b
medium sized molecule
c
macro molecule
d
2 0.99 1.00 1.00
100 0.55 0.88 0.97
a
Calculated for the pulse sequence shown above for two gradients of strength 10 G cm
–1
and
duration, τ, 2 ms; relaxation is ignored.
b
Diffusion constant, D, taken as that for water, which
is 2.1 × 10
–9
m
2
s
–1
at ambient temperatures.
c
Diffusion constant taken as 0.46 × 10
–9
m
2
s
–1
.
d
Diffusion constant taken as 0.12 × 10
–9
m
2
s
–1
.
9.6.6.1 Minimisation of Diffusion Losses
The foregoing discussion makes it clear that in order to minimise intensity
losses due to diffusion the product of the strength and durations of the gradient
pulses, G
2 2
τ , should be kept as small as is consistent with achieving the
required level of suppression. In addition, a gradient pulse pair should be
separated by the shortest time, ∆, within the limits imposed by the pulse
sequence. This condition applies to gradient pairs the first of which is
responsible for dephasing, and the second for rephasing. Once the coherence is
rephased the time that elapses before further gradient pairs is irrelevant from
the point of view of diffusion losses.
In twodimensional NMR, diffusion can lead to line broadening in the F
1
dimension if t
1
intervenes between a gradient pair. Consider the two alternative
pulse sequences for recording a simple COSY spectrum shown opposite. In (a)
the gradient pair are separated by the very short time of the final pulse, thus
keeping the diffusion induced losses to an absolute minimum. In (b) the two
gradients are separated by the incrementable time t
1
; as this increases the losses
due to diffusion will also increase, resulting in an extra decay of the signal in t
1
.
The extra line broadening due to this decay can be estimated from Eqn. [15],
with ∆ = t
1
, as γ τ π
2 2 2
G D
Hz. For a pair of 2 ms gradients of strength 10 G
cm
–1
this amounts ≈ 2 Hz in the case of a small molecule.
This effect by which diffusion causes an extra line broadening in the F
1
dimension is usually described as diffusion weighting. Generally it is possible
to avoid it by careful placing of the gradients. For example, the sequences (a)
g
RF
t
1
t
2
g
RF
t
1
t
2
(a)
(b)
9–44
and (b) are in every other respect equivalent, thus there is no reason not to
chose (a). It should be emphasised that diffusion weighting occurs only when t
1
intervenes between the dephasing and refocusing gradients.
9.6.7 Some examples of gradient selection
9.6.7.1 Introduction
Reference has already been made to the two general advantages of using
gradient pulses for coherence selection, namely the possibility of a general
improvement in the quality of spectra and the removal of the requirement of
completing a phase cycle for each increment of a multidimensional
experiment. This latter point is particularly significant when dealing with
three and fourdimensional experiments.
The use of gradients results in very significant improvement in the quality of
protondetected heteronuclear experiments, especially when unlabelled samples
are used. In such experiments, gradient selection results in much lower
dynamic range in the free induction decay as compared to phase cycled
experiments.
As has been discussed above, special care needs to be taken in experiments
which use gradient selection in order to retain absorption mode lineshapes.
In the following sections the use of gradient selection in several different
experiments will be described. The gradient pulses used in these sequences will
be denoted G
1
, G
2
etc. where G
i
implies a gradient of duration τ
i
, strength B
g,i
and shape factor s
i
. There is always the choice of altering the duration, strength
or, conceivably, shape factor in order to establish refocusing. Thus, for brevity
we shall from now on write the spatially dependent phase produced by gradient
G
i
acting on coherence of order p as γpG
i
in the homonuclear case or
γ
j j i
j
p G
∑
in the heteronuclear case; the sum is over all types of nucleus.
9.6.7.2 Doublequantum Filtered COSY
g
RF
(a) (b)
t
1
G
1
G
2
1
τ
2
τ
2
1
0
–1
–2
g
RF
t
1
G
1
G
2
1
τ
2
1
0
–1
–2
This experiment has already been discussed in detail in previous sections;
sequence (a) is essentially that described already and is suitable for recording
absorption mode spectra. The refocusing condition is G
2
= 2G
1
; frequency
9–45
discrimination in the F
1
dimension is achieved by the SHR or TPPI procedures.
Multiple quantum filters through higher orders can be implemented in the same
manner.
In sequence (b) the final spin echo is not required as data acquisition is
started immediately after the final radiofrequency pulse; phase errors which
would accumulate during the second gradient pulse are thus avoided. Of
course, the signal only rephases towards the end of the final gradient, so there is
little signal to be observed. However, the crucial point is that, as the
magnetization is all in antiphase at the start of t
2
, the signal grows from zero at
a rate determined by the size the couplings on the spectrum. Provided that the
gradient pulse is much shorter that 1/J, where J is a typical protonproton
coupling constant, the part of the signal missed during the gradient pulse is not
significant and the spectrum is not perturbed greatly. An alternative procedure
is to start to acquire the data after the final gradient, and then to right shift the
free induction decay, bringing in zeroes from the left, by a time equal to the
duration of the gradient.
9.6.7.3 HMQC
t
2
1
0
–1
t
1
1
0
–1
∆ ∆
I
S
p
I
p
S
g
G
1
G
1
G
2
1
τ
1
τ
There are several ways of implementing gradient selection into the HMQC
experiment, one of which, which leads to absorption mode spectra, is shown
above. The centrally placed I spin 180° pulse results in no net dephasing of the
I spin part of the heteronuclear multiple quantum coherence by the two
gradients G
1
i.e. the dephasing of the I spin coherence caused by the first is
undone by the second. However, the S spin coherence experiences a net
dephasing due to these two gradients and this coherence is subsequently
refocused by G
2
. Two 180° S spin pulses together with the delays τ
1
refocus
shift evolution during the two gradients G
1
. The centrally placed 180° I spin
pulse refocuses chemical shift evolution of the I spins during the delays ∆ and
all of the gradient pulses (the last gradient is contained within the final delay,
∆). The refocusing condition is
m2 0
1 2
γ γ
s I
G G − =
where the + and – signs refer to the P and Ntype spectra respectively. The
switch between recording these two types of spectra is made simply by
reversing the sense of G
2
. The P and Ntype spectra are recorded separately
9–46
and then combined in the manner described in section 9.4.4.2 to give a
frequency discriminated absorption mode spectrum.
In the case that I and S are proton and carbon13 respectively, the gradients
G
1
and G
2
are in the ratio 2:±1. Proton magnetization not involved in
heteronuclear multiple quantum coherence, i.e. magnetization from protons not
coupled to carbon13, is refocused after the second gradient G
1
but is then
dephased by the final gradient G
2
. Provided that the gradient is strong enough
these unwanted signals, and the t
1
noise associated with them, will be
suppressed.
9.6.7.4 HSQC
t
2
1
0
–1
t
1
1
0
–1
I
a
b
y
S
p
I
p
S
g
G
∆
2
∆
2
∆
2
∆
2
The sequence above shows the simplest way of implementing gradients into the
HSQC experiment. An analysis using product operators shows that at point a
the required signal is present as the operator 2I
z
S
z
whereas the undesired signal
(from I spins not coupled to S spins) is present as I
y
. Thus, a field gradient
applied at point a will dephase the unwanted magnetization and leave the
wanted term unaffected. This is an example of using gradients not for
selection, but for suppression of unwanted magnetization (see section 9.6.3.2).
The main practical difficulty with this approach is that the unwanted
magnetization is only along y at point a provided all of the pulses are perfect; if
the pulses are imperfect there will be some zmagnetization present which will
not be eliminated by the gradient. In the case of observing proton
13
C or
proton
15
N HSQC spectra from natural abundance samples, the magnetization
from uncoupled protons is very much larger than the wanted magnetization, so
even very small imperfections in the pulses can give rise to unacceptably large
residual signals. However, for globally labelled samples the degree of
suppression is often sufficient and such an approach is used successfully in
many three and fourdimensional experiments applied to globally
13
C and
15
N
labelled proteins.
The key to obtaining the best suppression of the uncoupled magnetization is
to apply a gradient when transverse magnetization is present on the S spin. An
example of the HSQC experiment utilising such a principle is shown below
9–47
t
2
1
0
–1
t
1
1
0
–1
I
S
p
I
p
S
g
G
1
G
2
1
τ
∆
2
∆
2
∆
2
∆
2
HSQC pulse sequence with gradient selection. The CTP for an Ntype spectrum is shown by the full line
and for the Ptype spectrum by the dashed line.
Here, G
1
dephases the S spin magnetization present at the end of t
1
and, after
transfer to the I spins, G
2
refocuses the signal. An extra 180° pulse to S in
conjunction with the extra delay τ
1
ensures that phase errors which accumulate
during G
1
are refocused; G
2
is contained within the final spin echo which is part
of the usual HSQC sequence. The refocusing condition is
mγ γ
S I
G G
1 2
0 − =
where the – and + signs refer to the N and Ptype spectra respectively. As
before, an absorption mode spectrum is obtained by combining the N and P
type spectra, which can be selected simply by reversing the sense of G
2
.
9.7 Zeroquantum dephasing and purge pulses
Both zmagnetization and homonuclear zeroquantum coherence have
coherence order 0, and thus neither are dephased by the application of a
gradient pulse. Selection of coherence order zero is achieved simply by
applying a gradient pulse which is long enough to dephase all other coherences;
no refocusing is used. In the vast majority of experiments it is the z
magnetization which is required and the zeroquantum coherence that is
selected at the same time is something of a nuisance.
A number of methods have been developed to suppress contributions to the
spectrum from zeroquantum coherence. Most of these utilise the property that
zeroquantum coherence evolves in time, whereas zmagnetization does not.
Thus, if several experiments in which the zeroquantum has been allowed to
evolve for different times are coadded, cancellation of zeroquantum
contributions to the spectrum will occur. Like phase cycling, such a method is
time consuming and relies on a difference procedure. However, it has been
shown that if a field gradient is combined with a period of spinlocking the
coherences which give rise to these zeroquantum coherences can be dephased.
Such a process is conveniently considered as a modified purging pulse.
9.7.1 Purging pulses
A purging pulse consists of a relatively long period of spinlocking, taken here
to be applied along the xaxis. Magnetization not aligned along x will precess
9–48
about the spinlocking field and, because this field is inevitably
inhomogeneous, such magnetization will dephase. The effect is thus to purge
all magnetization except that aligned along x. However, in a coupled spin
system certain antiphase states aligned perpendicular to the spinlock axis are
also preserved. For a two spin system (with spins k and l), the operators
preserved under spinlocking are I
kx
, I
lx
and the antiphase state 2 2 I I I I
ky lz kz ly
− .
Thus, in a coupled spin system, the purging effect of the spinlocking pulse is
less than perfect.
The reason why these antiphase terms are preserved can best be seen by
transforming to a tilted coordinate system whose zaxis is aligned with the
effective field seen by each spin. For the case of a strong B
1
field placed close
to resonance the effective field seen by each spin is along x, and so the
operators are transformed to the tilted frame simply by rotating them by –90°
about y
I I I I
I I I I I I I I
kx
I
kz lx
I
lz
ky lz kz ly
I I
ky lx kx ly
ky ly
ky ly
− −
− +
( )
→ →
− → −
π π
π
2 2
2
2 2 2 2
T T
T T T T
Operators in the tilted frame are denoted with a superscript T. In this frame the
xmagnetization has become z, and as this is parallel with the effective field, it
clearly does not dephase. The antiphase magnetization along y has become
2 2 I I I I
ky lx kx ly
T T T T
−
which is recognised as zeroquantum coherence in the tilted frame. Like zero
quantum coherence in the normal frame, this coherence does not dephase in a
strong spinlocking field. There is thus a connection between the inability of a
field gradient to dephase zeroquantum coherence and the preservation of
certain antiphase terms during a purging pulse.
Zeroquantum coherence in the tilted frame evolves with time at a frequency,
Ω
ZQ
T
, given by
Ω Ω Ω
ZQ
T
= +
( )
− +
( )
k l
2
1
2 2
1
2
ω ω
where Ω
i
is the offset from the transmitter of spin i and ω
1
is the B
1
field
strength. If a field gradient is applied during the spinlocking period the zero
quantum frequency is modified to
Ω Ω Ω
ZQ
T
g g
r B r B r
k l
( ) = + ( )
{ }
+
( )
− + ( )
{ }
+
( )
γ ω γ ω
2
1
2
2
1
2
This frequency can, under certain circumstances, become spatially dependent
and thus the zeroquantum coherence in the tilted frame will dephase. This is in
contrast to the case of zeroquantum coherence in the laboratory frame which is
not dephased by a gradient pulse.
The principles of this dephasing procedure are discussed in detail elsewhere
(J. Magn. Reson. Ser. A 105, 167183 (1993) ). Here, we note the following
features. (a) The optimum dephasing is obtained when the extra offset induced
9–49
by the gradient at the edges of the sample, γB
g
(r
max
), is of the order of ω
1
. (b)
The rate of dephasing is proportional to the zeroquantum frequency in the
absence of a gradient, (Ω
k
– Ω
l
). (c) The gradient must be switched on and off
adiabatically. (d) The zeroquantum coherences may also be dephased using
the inherent inhomogeneity of the radiofrequency field produced by typical
NMR probes, but in such a case the optimum dephasing rate is obtained by spin
locking offresonance so that
tan
–
,
1
1
54 ω Ω
k l
( )
≈ °.
(e) Dephasing in an inhomogeneous B
1
field can be accelerated by the use of
special composite pulse sequences.
The combination of spinlocking with a gradient pulse allows the
implementation of essentially perfect purging pulses. Such a pulse could be
used in a twodimensional TOCSY experiment whose pulse sequence is shown
below as (a).
t
1
τ
m
g
G
A
G
A
G
A
G
2
y
t
2 t
1
t
2
DIPSI
(a) (b)
Pulse sequences using purging pulses which comprise a period of spin locking with a magnetic field
gradient. The field gradient must be switched on and off in an adiabatic manner.
In this experiment, the period of isotropic mixing transfers inphase
magnetization (say along x) between coupled spins, giving rise to crosspeaks
which are absorptive and inphase in both dimensions. However, the mixing
sequence also both transfers and generates antiphase magnetization along y,
which gives rise to undesirable dispersive antiphase contributions in the
spectrum. In sequence (a) these antiphase contributions are eliminated by the
use of a purging pulse as described here. Of course, at the same time all
magnetization other than x is also eliminated, giving a near perfect TOCSY
spectrum without the need for phase cycling or other difference measures.
These purging pulses can be used to generate pure zmagnetization without
contamination from zeroquantum coherence by following them with a 90°(y)
pulse, as is shown in the NOESY sequence (b). Zeroquantum coherences
present during the mixing time of a NOESY experiment give rise to
troublesome dispersive contributions in the spectra, which can be eliminated by
the use of this sequence.
1
Exercises for Chapters 6, 7, 8 & 9
The more difficult and challenging problems are marked with an asterisk, *
Chapter 6: Product operators
E61 Using the standard rotations from section 6.1.4, express the following
rotations in terms of sines and cosines:
exp exp exp exp exp exp
exp exp exp exp exp exp
− ( ) ( ) − ( ) −
( ) ( ) − −
( ) ( )
−
( ) ( )
− ( ) ( ) ( ) − ( ) − ( )
i I I i I i I I i I i I I i I
i I I i I i I I i I i I I i I
x y x z y z y z y
y x y x x x x z x
θ θ θ θ θ θ
θ θ θ θ θ θ
1
2
1
2
Express all of the above transformations in the shorthand notation of section
6.1.5.
E62. Repeat the calculation in section 6.1.6 for a spin echo with the 180°
pulse about the yaxis. You should find that the magnetization refocuses onto
the –y axis.
E63. Assuming that magnetization along the yaxis gives rise to an
absorption mode lineshape, draw sketches of the spectra which arise from the
following operators
I
1y
I
2x
2I
1y
I
2z
2I
1z
I
2x
E64. Describe the following terms in words:
I
1y
I
2z
2I
1y
I
2z
2I
1x
I
2x
E65. Give the outcome of the following rotations
I
I I
I I
I
I
I I
x
t I
x z
I I
x z
I
x
tI tI
x
J tI I
z y
J tI I
y
y y
y
z z
z z
z z
1
1 2
2
1 2
1
1
2
1 2
2
1 1
1 2
2
1 1 2 2
12 1 2
12 1 2
2
2
2
ω
π
π
π
π
p
→
→
→
→ →
− →
− →
( ) +
( )
−
Ω Ω
Describe the outcome in words in each case.
E66. Consider the spin echo sequence
– τ – 180°(x, to spin 1 and spin 2) – τ –
applied to a twospin system. Starting with magnetization along y, represented
by I
1y
, show that overall effect of the sequence is
2
I J I J I I
y y x z 1 12 1 12 1 2
2 2 2
spin echo
→ − ( ) + ( ) cos sin π τ π τ
You should ignore the effect of offsets, which are refocused, are just consider
evolution due to coupling.
Is your result consistent with the idea that this echo sequence is
equivalent to
– 2τ – 180°(x, to spin 1 and spin 2)
[This calculation is rather more complex than that in section 6.4.1. You will
need the identities
cos cos sin sin cos sin 2 2 2
2 2
θ θ θ θ θ θ = − = and ]
E67. For a twospin system, what delay, τ, in a spin echo sequence would
you use to achieve the following overall transformations (do not worry about
signs)? [cos π/4 = sin π/4 = 1/√2]
I I I
I I I
I I I I
I I
y z x
z x x
x x y z
x x
2 1 2
1 2 2
1
1
2 1
1
2 1 2
1 1
2
2
2
→
→
→
( )
+
( )
→ −
E68. Confirm by a calculation that spin echo sequence c shown on page 6–11
does not refocus the evolution of the offset of spin 1. [Start with a state I
1x
or I
1y
;
you may ignore the evolution due to coupling].
*E69. Express 2
1 2
I I
x y
in terms of raising and lowering operators: see section
6.5.2. Take the zeroquantum part of your expression and then rewrite this in
terms of Cartesian operators using the procedure shown in section 6.5.2.
E610. Consider three coupled spins in which J
23
> J
12
. Following section 6.6,
draw a sketch of the doublet and doublets expected for the multiplet on spin 2
and label each line with the spin states of the coupled spins, 1 and 3. Lable the
splittings, too.
Assuming that magnetization along x gives an absorption mode
lineshape, sketch the spectra from the following operators:
I I I I I I I I
x z x y z z x z 2 1 2 2 3 1 2 3
2 2 4
E611. Complete the following rotations.
3
I
I I
I I I
I I
x
tI tI tI
y z
I I I
x z z
I I I
z x
J tI I
z z z
y y y
y y y
z z
2
1 2
2
1 2 3
2
1 2
2
1 1 2 2 3 3
1 2 3
1 2 3
12 1 2
2
2
2
Ω Ω Ω
→ → →
→
→
→
( ) + +
( )
( ) + +
( )
π
π
π
22
4
1 2
2
1 2 3
2
1
2 2
23 2 3
23 2 3
12 1 2 13 1 3
I I
I I I
I
z x
J tI I
z y z
J tI I
x
J tI I tJ I I
z z
z z
z z z z
π
π
π π
→
→
→ →
Chapter 7: Twodimensional NMR
E71. Sketch the COSY spectra you would expect from the following
arrangements of spins. In the diagrams, a line represents a coupling. Assume
that the spins have well separated shifts; do not concern yourself with the details
of the multiplet structures of the cross and diagonalpeaks.
A B C A B C
A B C
E72. Sketch labelled twodimensional spectra which have peaks arising from
the following transfer processes
frequencies in F
1
/ Hz frequencies in F
2
/ Hz
a 30 Transferred to 30
b 30 and 60 transferred to 30
c 60 transferred to 30 and 60
d 30 transferred to 20, 30 and 60
e 30 and 60 transferred to 30 and 60
E73. What would the diagonalpeak multiplet of a COSY spectrum of two
spins look like if we assigned the absorption mode lineshape in F
2
to
magnetization along x and the absorption mode lineshape in F
1
to sine
modulated data in t
1
?
What would the crosspeak multiplet look like with these assignments?
E74. The smallest coupling that will gives rise to a discernible crosspeak in a
COSY spectra depends on both the linewidth and the signaltonoise ratio of the
spectrum. Explain this observation.
4
*E75. Complete the analysis of the DQF COSY spectrum by showing in
detail that both the cross and diagonalpeak multiplets have the same lineshape
and are in antiphase in both dimensions. Start from the expression in the
middle of page 7–10, section 7.4.21. [You will need the identity
cos sin sin sin A B B A B A = + ( ) + − ( )
[ ]
1
2
].
*E76. Consider the COSY spectrum for a threespin system. Start with
magnetization just on spin 1. The effect of the first pulse is
I I
z
I I I
y
x x x
1
2
1
1 2 3
π ( ) + + ( )
→ −
Then, only the offset of spin 1 has an effect
− → − + I t I t I
y
t I
y x
z
1 1 1 1 1 1 1
1 1 1
Ω
Ω Ω cos sin
Only the term in I
1x
leads to cross and diagonal peaks, so consider this term
only from now on.
First allow it to evolve under the coupling to spin 2 and then the coupling to spin
3
sinΩ
1 1 1
2 2
12 1 1 2 13 1 1 3
t I
x
J t I I J t I I
z z z z
π π
→ →
Then, consider the effect of a 90°(x) pulse applied to all three spins. After this
pulse, you should find one term which represents a diagonalpeak multiplet, one
which represents a crosspeak multiplet between spin 1 and spin 2, and one
which represents a crosspeak multiplet between spin 1 and spin 3. What does
the fourth term represent?
[More difficult] Determine the form of the crosspeak multiplets, using the
approach adopted in section 6.4.1. Sketch the multiplets for the case J
12
≈ J
23
>
J
13
. [You will need the identity
sin sin cos cos A B A B A B = + ( ) − − ( )
[ ]
1
2
]
*E77. The pulse sequence for twodimensional TOCSY (total correlation
spectroscopy) is shown below
t
1
isotropic mixing
t
2
τ
The mixing time, of length τ, is a period of isotropic mixing. This is a multiple
pulse sequence which results in the transfer of inphase magnetization from one
spin to another. In a two spin system the mixing goes as follows:
I J I J I
x x x 1
2
12 1
2
12 2
isotropic mixing for time τ
π τ π τ → + cos sin
We can assume that all terms other than I
1x
do not survive the isotropic mixing
sequence, and so can be ignored.
Predict the form of the twodimensional TOCSY spectrum for a twospin
system. What is the value of τ which gives the strongest cross peaks? For this
optimum value of τ, what happens to the diagonal peaks? Can you think of any
5
advantages that TOCSY might have over COSY?
E78. Repeat the analysis for the HMQC experiment , section 7.4.3.1, with
the phase of the first spin2 (carbon13) pulse set to –x rather than +x. Confirm
that the observable signals present at the end of the sequence do indeed change
sign.
E79. Why must the phase of the second spin1 (proton) 90° pulse in the
HSQC sequence, section 7.4.3.2, be y rather than x?
E710. Below is shown the pulse sequence for the HETCOR (heteronuclear
correlation) experiment
1
H
13
C
t
1
A C
t
2
∆
2
∆
2
∆
2
∆
2
B
This sequence is closely related to HSQC, but differs in that the signal is
observed on carbon13, rather than being transferred back to proton for
observation. Like HSQC and HMQC the resulting spectrum shows cross peaks
whose coordinates are the shifts of directly attached carbon13 proton pairs.
However, in contrast to these sequences, in HETCOR the proton shift
is in F
1
and the carbon13 shift is in F
2
. In the early days of twodimensional
NMR this was a popular sequence for shift correlation as it is less demanding of
the spectrometer; there are no strong signals from protons not coupled to
carbon13 to suppress.
We shall assume that spin 1 is proton, and spin 2 is carbon13. During period
A, t
1
, the offset of spin 1 evolves but the coupling between spins 1 and 2 is
refocused by the centrally placed 180° pulse. During period B the coupling
evolves, but the offset is refocused. The optimum value for the time ∆ is
1/(2J
12
), as this leads to complete conversion into antiphase. The two 90°
pulses transfer the antiphase magnetization to spin 2.
During period C the antiphase magnetization rephases (the offset is refocused)
and if ∆ is 1/(2J
12
) the signal is purely inphase at the start of t
2
.
Make an informal analysis of this sequence, along the lines of that given in
section 7.4.3.2, and hence predict the form of the spectrum. In the first instance
assume that ∆ is set to its optimum value. Then, make the analysis slightly
more complex and show that for an arbitrary value of ∆ the signal intensity goes
as sin
2
πJ
12
∆.
Does altering the phase of the second spin1 (proton) 90° pulse from x to y
make any difference to the spectrum?
[Harder] What happens to carbon13 magnetization, I
2z
, present at the beginning
of the sequence? How could the contribution from this be removed?
6
Chapter 8: Relaxation
E81. In an inversionrecovery experiment the following peak heights (S,
arbitrary units) were measured as a function of the delay, t, in the sequence:
t / s 0.1 0.5 0.9 1.3 1.7 2.1 2.5 2.9
S –98.8 –3.4 52.2 82.5 102.7 115.2 120.7 125.1
The peak height after a single 90° pulse was measured as 130.0 Use a graphical
method to analyse these data and hence determine a value for the longitudinal
relaxation rate constant and the corresponding value of the relaxation time, T
1
.
E82. In an experiment to estimate T
1
using the sequence [180° – τ – 90°]
acquire three peaks in the spectrum were observed to go through a null at 0.5,
0.6 and 0.8 s respectively. Estimate T
1
for each of these resonances.
A solvent resonance was still inverted after a delay of 1.5 s; what does this tell
you about the relaxation time of the solvent?
*E83. Using the diagram at the top of page 8–5, write down expressions for
dn
1
/dt, dn
2
/dt etc. in terms of the rate constants W and the populations n
i
. [Do
this without looking at the expressions given on page 8–6 and then check
carefully to see that you have the correct expressions].
*E84. Imagine a modified experiment, designed to record a transient NOE
enhancement, in which rather than spin S being inverted at the beginning of the
experiment, it is saturated. The initial conditions are thus
I I S
z z z
0 0 0
0
( ) = ( ) =
Using these starting conditions rather than those of Eq. [16] on page 810, show
that in the initial rate limit the NOE enhancement builds up at a rate proportional
to σ
IS
rather than 2σ
IS
. You should use the method given in Section 8.4.1 for
your analysis.
Without detailed calculation, sketch a graph, analogous to that given on page 8
12, for the behaviour of I
z
and S
z
for these new initial conditions as a function of
mixing time.
E85. Why is it that in a two spin system the size of transient NOE
enhancements depends on R
I
, R
S
and σ
IS
, whereas in a steady state experiment
the enhancement only depends on R
I
and σ
IS
? [Spin S is the target].
In a particular twospin system, S relaxes quickly and I relaxes slowly. Which
experiment would you choose in order to measure the NOE enhancement
between these two spins? Include in your answer an explanation of which spin
you would irradiate.
7
E86. For the molecule shown on the right, a transient
NOE experiment in which H
B
is inverted gave equal
initial NOE buildup rates on H
A
and H
C
. If H
A
was
inverted the initial buildup rate on H
B
was the same as in
the first experiment; no enhancement is seen of H
C
. In
steady state experiments, irradiation of H
B
gave equal
enhancements on H
A
and H
B
. However, irradiation of
H
A
gave a much smaller enhancement on H
B
than for the
case where H
B
was the irradiated spin and the
enhancement was observed on H
A
. Explain.
Z
H
A
H
B
H
C
X
Y
E87. What do you understand by the terms correlation time and spectral
density? Why are these quantities important in determining NMR relaxation rate
constants?
E88. The simplest form of the spectral density, J(ω), is the Lorentzian:
J ω
τ
ω τ
( ) =
+
2
1
2
c
c
2
Describe how this spectral density varies with both ω and τ
c
. For a given
frequency, ω
0
, at what correlation time is the spectral density a maximum?
Show how this form of the spectral density leads to the expectation that, for a
given Larmor frequency, T
1
will have a minimum value at a certain value of the
correlation time, τ
c
.
E89. Suppose that he Larmor frequency (for proton) is 800 MHz. What
correlation time will give the minimum value for T
1
? What kind of molecule
might have such a correlation time?
E810. Explain why the NOE enhancements observed in small molecules are
positive whereas those observed for large molecules are negative.
E811. Explain how it is possible for the sign of an NOE enhancement to
change when the magnetic field strength used by the spectrometer is changed.
E812. What is transverse relaxation and how it is different from longitudinal
relaxation? Explain why it is that the rate constant for transverse relaxation
increases with increasing correlation times, whereas that for longitudinal
relaxation goes through a maximum.
8
Chapter 9: Coherence selection: phase cycling and gradient pulses
E91. (a) Show, using vector diagrams like those of section 9.1.6, that in a
pulseacquire experiment a phase cycle in which the pulse goes x, y, –x, –y and
in which the receiver phase is fixed leads to no signal after four transients have
been coadded.
(b) In a simple spin echo sequence
90° – τ – 180° – τ –
the EXORCYCLE sequence involves cycling the 180° pulse x, y, –x, –y and the
receiver x, –x, x, –x. Suppose that, by accident, the 180° pulse has been omitted.
Use vector pictures to show that the four step phase cycle cancels all the signal.
(c) In the simple echo sequence, suppose that there is some zmagnetization
present at the end of the first τ delay; also suppose that the 180° pulse is
imperfect so that some of the zmagnetization is made transverse. Show that the
four steps of EXORCYCLE cancels the signal arising from this magnetization.
*E92. (a) For the INEPT pulse sequence of section 9.1.8, confirm with
product operator calculations that: [You should ignore the evolution of offsets
as this is refocused by the spin echo; assume that the spin echo delay is 1/(2J
IS
)].
(i) the sign of the signal transferred from I to S is altered by changing the phase
of the second I spin 90° pulse from y to –y;
(ii) the signs of both the transferred signal and the signal originating from
equilibrium S spin magnetization, S
z
, are altered by changing the phase of the
first S spin 90° pulse by 180°.
On the basis of your answers to (i) and (ii), suggest a suitable phase cycle,
different to that given in the notes, for eliminating the contribution from the
equilibrium S spin magnetization.
(b) Imagine that in the INEPT sequence the first I spin 180° pulse is cycled x, y,
–x, –y. Without detailed calculations, deduce the effect of this cycle on the
transferred signal and hence determine a suitable phase cycle for the receiver
[hint  this 180° pulse is just forming a spin echo]. Does your cycle eliminate
the contribution from the equilibrium S spin magnetization?
(c) Suppose now that the first S spin 180° pulse is cycled x, y, –x, –y; what
effect does this have on the signal transferred from I to S?
E93. Determine the coherence order or orders of each of the following
operators [you will need to express I
x
and I
y
in terms of the raising and lowering
operators, see section 9.3.1]
I I I I I I I I I I I I I
z x y x z x x y y 1 2 1 2 3 1 1 1 2 1 2 1 2
4 2 2 2
+ − + +
+
( )
In a heteronuclear system a coherence order can be assigned to each spin
9
separately. If I and S represent different nuclei, assign separate coherence orders
for the I and S spins to the following operators
I S I S I S
x y x z x x
2 2
E94. (a) Consider the phase cycle devised in section 9.5.1 which was
designed to select ∆p = –3: the pulse phase goes 0, 90, 180, 270 and the receiver
phase goes 0, 270, 180, 90. Complete the following table and use it to show that
such a cycle cancels signals arising from a pathway with ∆p = 0.
step
pulse
phase
phase shift experienced
by pathway with ∆p = 0
equivalent
phase
rx. phase for
∆p = –3
difference
1 0 0
2 90 270
3 180 180
4 270 90
Construct a similar table to show that a pathway with ∆p = –1 is cancelled, but
that one with ∆p = +5 is selected by this cycle.
(b) Bodenhausen et al. have introduced a notation in which the sequence of
possible ∆p values is written out in a line; the values of ∆p which are selected by
the cycle are put into bold print, and those that are rejected are put into
parenthesis, viz (1). Use this notation to describe the pathways selected and
rejected by the cycle given above for pathways with ∆p between –5 and +5 [the
fate of several pathways is given in section 9.5.1, you have worked out two
more in part (a) and you may also assume that the pathways with ∆p = –5, –4,
–2, 3 and 4 are rejected]. Confirm that, as expected for this fourstep cycle, the
selected values of ∆p are separated by 4.
(c) Complete the following table for a threestep cycle designed to select ∆p =
+1.
step
pulse
phase
phase shift experienced by
pathway with ∆p = +1
equivalent
phase
rx. phase
1 0
2 120°
4 240°
(d) Without drawing up further tables, use the general rules of section 9.5.2 to
show that, in Bodenhausen's notation, the selectivity of the cycle devised in (c)
can be written:
–2 (–1) (0) 1 (2)
(e) Use Bodenhausen's notation to describe the selectivity of a 6 step cycle
designed to select ∆p = +1; consider ∆p values in the range –6 to +6.
E95. Draw coherence transfer pathways for (a) fourquantum filtered COSY
10
[use the sequence in section 9.5.5.4 as a model]; (b) a 180° pulse used to refocus
doublequantum coherence; (c) Ntype NOESY.
E96. Write down fourstep phase cycles to select (a) ∆p = –1 and (b) ∆p =
+2. Suppose that the cycles are applied to different pulses. Combine them to
give a 16step cycle as was done in section 9.5.4; give the required receiver
phase shifts.
*E97. The CTP and pulse sequence of triplequantum filtered COSY are
t
1
t
2
0
–1
–2
–3
1
2
3
(a) Write down the values of ∆p brought about by (i) the first pulse, (iii) the
second pulse, (iii) the first and second pulses acting together and considered as a
group, (iv) the last pulse.
(b) Imagine the spin system under consideration is able to support multiple
quantum coherence up to and including p = ±4. Consider a phase cycle in which
the first two pulses are cycled as a group and, having in mind your answer to (a)
(iii), use the notation of Bodenhausen to indicate which pathways need to be
selected and which blocked. Hence argue that the required phase cycle must
have 6 steps. Draw up such a sixstep phase cycle and confirm that the
proposed sequence of pulse and receiver phases does indeed discriminate against
filtration through doublequantum coherence.
(c) Consider an alternative phase cycle in which just the phase of the last pulse
is cycled. As in (b), use Bodenhausen's notation to describe the wanted and
unwanted values of ∆p. Devise a suitable phase cycle to select the required
values of ∆p.
(d) Try to devise a cycle which involves shifting just the phase of the second
pulse.
E98. (a) Write down a 16 step cycle which selects the pathways shown in
the double quantum spectroscopy pulse sequence of section 9.5.9.1; include in
your cycle doublequantum selection and EXORCYCLE phase cycling of the
180° pulse.
(b) Write down an 8 step cycle which selects the pathway for NOESY shown
in section 9.5.9.2; include in your cycle explicit axial peak suppression steps.
*E99. The following sequence is one designed to measure the relaxation
induced decay of doublequantum coherence as a function of the time 2τ
11
0
–1
–2
1
2
τ τ
The 180° pulse placed in the middle of the doublequantum period is used to
refocus evolution due to offsets and inhomogeneous line broadening.
Devise a suitable phase cycle for the second 180° pulse bearing in mind that the
180° pulse may be imperfect. [hint: are four steps sufficient?]
E910. Use the formula given in section 9.6.2 for the overall decay of
magnetization during a gradient
2
γGtr
max
to calculate how long a gradient is needed to dephase magnetization to (a) 10%
and (b) 1% of its initial value assuming that: G = 0.1 T m
–1
(10 G
cm
–1
), r
max
= 0.005 m (0.5 cm) and γ = 2.8 × 10
8
rad s
–1
. [Put all the quantities
in SI units].
E911. Imagine that two gradients, G
1
and G
2
, are placed before and after a
radiofrequency pulse. For each of the following ratios between the two
gradients, identify two coherence transfer pathways which will be refocused:
(a) G
1
:G
2
= 1:1 (b) G
1
:G
2
= –1:1 (c) G
1
:G
2
= 1:–2 (d) G
1
:G
2
= 0.5:1
E912. Determine the ratio G
1
:G
2
needed to select the following pathways:
(a) p = 2 → 1 (b) p = 3 → 1 (c) p = –3 → 1
(d) p = –1 → 1 and p = 1 → –1 (e) p = 0 → 1
Comment on the way in which case (e) differs from all the others.
E913. Consider a gradient selected Ntype DQF COSY
t
1
G
1
G
2
G
3
t
2
0
–1
–2
1
2
We use three gradients; G
1
in t
1
so as to select p = +1 during t
1
; G
2
to select
double quantum during the filter delay and G
3
to refocus prior to acquisition.
(a) Show that the pathway shown, which can de denoted 0→1→2→–1, is
12
selected by gradients in the ratio G
1
:G
2
:G
3
= 1:1:3.
(b) Show that this set of gradients also selects the pathway
0→–1→4→–1. What kind of spectrum does such a pathway give rise to?
(c) Consider gradients in the ratios G
1
:G
2
:G
3
(i) 1:
1
2
:2 and (ii) 1:
1
3
:
5
3
. Show
that these combination select the DQ filtered pathway desired. In each case, give
another possible pathway which has p = ±1 during t
1
and p = –1 during t
2
that
these gradient combinations select.
(d) In the light of (b) and (c), consider the utility of the gradient ratio
1.0 : 0.8 : 2.6
*E914.
Devise a gradient selected version of the triplequantum filtered COSY
experiment, whose basic pulse sequence and CTP was given in E97. Your
sequence should include recommendations for the relative size of the gradients
used. The resulting spectrum must have pure phase (i.e. p = ±1 must be
preserved in t
1
) and phase errors due to the evolution of offsets during the
gradients must be removed.
How would you expect the sensitivity of your sequence to compare with its
phasecycled counterpart?
*E91. A possible sequence for P/N selected HMQC is
t
2
1
0
–1
t
1
1
0
–1
a
b
13
C
1
H
p
I
p
S
g
G
1
G
2
∆ ∆ τ τ
The intention is to recombine P and Ntype spectra so as to obtain absorption
mode spectra.
(a) Draw CTPs for the Ptype and Ntype versions of this experiment [refer to
section 9.6.7.3 for some hints].
(b) What is the purpose of the two 180° pulses a and b, and why are such
pulses needed for both proton and carbon13?
(c) Given that γ
H
/γ
C
= 4, what ratios of gradients are needed for the Ptype and
the Ntype spectra?
(d) Compare this sequence with that given in section 9.6.7.3, pointing out any
advantages and disadvantages that each has.
Understanding NMR Spectroscopy
James Keeler
University of Cambridge, Department of Chemistry
c James Keeler, 2002
1
What this course is about
This course is aimed at those who are already familiar with using NMR on a daytoday basis, but who wish to deepen their understanding of how NMR experiments work and the theory behind them. It will be assumed that you are familiar with the concepts of chemical shifts and couplings, and are used to interpreting proton and 13 C spectra. It will also be assumed that you have at least come across simple twodimensional spectra such as COSY and HMQC and perhaps may have used such spectra in the course of your work. Similarly, some familiarity with the nuclear Overhauser effect (NOE) will be assumed. That NMR is a useful for chemists will be taken as self evident. This course will always use the same approach. We will ﬁrst start with something familiar – such as multiplets we commonly see in proton NMR spectra – and then go deeper into the explanation behind this, introducing along the way new ideas and new concepts. In this way the new things that we are learning are always rooted in the familiar, and we should always be able to see why we are doing something. In NMR there is no escape from the plain fact that to understand all but the simplest experiments we need to use quantum mechanics. Luckily for us, the quantum mechanics we need for NMR is really rather simple, and if we are prepared to take it on trust, we will ﬁnd that we can make quantum mechanical calculations simply by applying a set of rules. Also, the quantum mechanical tools we will use are quite intuitive and many of the calculations can be imagined in a very physical way. So, although we will be using quantum mechanical ideas, we will not be using any heavyduty theory. It is not necessary to have studied quantum mechanics at anything more than the most elementary level. Inevitably, we will have to use some mathematics in our description of NMR. However, the level of mathematics we need is quite low and should not present any problems for a science graduate. Occasionally we will use a few ideas from calculus, but even then it is not essential to understand this in great detail. Course structure The course is accompanied by a detailed set of handouts, which for convenience is divided up into “chapters”. You will notice an inconsistency in the style of these chapters; this comes about because they have been prepared (or at least the early versions of them) over a number of years for a variety of purposes. The notes are sufﬁciently complete that you should not need to take many extra notes during the lectures. Each chapter has associated with it some exercises which are intended to illustrate the course material; unless you do the exercises you will not understand the material. In addition, there will be some practical exercises which
We can also use the vector model to understand the basic. Chapter 5 is concerned with how the spectrometer works. Chapter 6 introduces the product operator formalism for analysing NMR experiments. We then go on to consider the sources of relaxation and how it is related to molecular properties. analysing these in some detail. and that despite its quantum mechanical rigour it retains a relatively intuitive approach. We will concentrate on the most important types of spectra. There are quite a lot of useful manipulations that we can carry out on the data to enhance the sensitivity or resolution. Chapter 4 is concerned with data processing. which will be useful to us later on in the course. Using product operators we can describe important phenomena such as the evolution of couplings during spin echoes. These manipulations are described and their limitations discussed. We start out by considering the effects of relaxation. depending on what we require. and are responsible for the explosive growth of NMR over the past 20 years or so. In this chapter we also introduce some important ideas about how the NMR signal is turned into a digital form. Chapter 3 introduces the vector model of NMR. but very important. Chapter 8 considers the important topic of relaxation in NMR. such as COSY and HMQC. coherence transfer and the generation of multiple quantum coherences. This model has its limitations. inversion recovery and most importantly the spin echo. Such spectra have proved to be enormously useful in structure determination. It is not necessary to understand this is great detail. and we have to convert this to the usual representation (intensity as a function of frequency) using Fourier transformation. but it is very useful for understanding how pulses excite NMR signals. but it does help to have some basic understanding of what is going on when we “shim the magnet” or “tune the probe”. This approach is quantum mechanical. These exercises will give you a feel for what you can do with NMR data and how what you see relates to the theory you have studied. The chapter closes with a discussion of strongly coupled spectra and how they can be analysed. concentrating in particular on the very important nuclear Overhauser effect. We will see that the formalism is well adapted to describing pulsed NMR experiments. The signal we actually record in an NMR experiment is a function of time. Chapter 7 puts the tools from Chapter 6 to immediate use in analysing and understanding twodimensional spectra. Although this approach is more complex than the familiar “successive splitting” method for constructing multiplets it does help us understand how to think about multiplets in terms of “active” and “passive” spins. Quite a lot of the exercises will involve processing experimental data. and the consequences that this has. in contrast to the semiclassical approach taken by the vector model. .1–2 What this course is about involve mainly data processing on a PC. NMR experiments such as pulseacquire. This approach also makes it possible to understand the form of multiple quantum spectra. Chapter 2 considers how we can understand the form of the NMR spectrum in terms of the underlying nuclear spin energy levels.
The University of Cambridge is acknowledged for a period of study leave to enable me to come to UC Irvine. at a number of different levels. Hunter Modern NMR Spectroscopy (OUP). Hore Nuclear Magnetic Resonance (OUP).ch. Once again I am grateful to the organizers and sponsors of these meetings for the opportunity to present this material. Irvine. K. An understanding of how these work is helpful in getting to grips with the details of how experiments are actually run. Texts There are innumerable books written about NMR. There are also a number of texts which take a more theorybased approach. The chapter is concerned with the two methods used in multiple pulse NMR to select a particular outcome in an NMR experiment: phase cycling and ﬁeld gradient pulses.uk wwwkeeler.Keeler@ch. Probably the best of the more elementary books if P. Department of Chemistry March 2002 James. Sanders and B. for the invitation to give this course.ac. It has been modiﬁed subsequently for the courses in Japan mentioned above and for another EMBO course held in Lucca in 2000. but is an optional advanced topic. Chapter 9 was originally prepared (in a somewhat different form) for an EMBO course held in Turin (Italy) in 1995. I wish to express my thanks to Professor AJ Shaka and to the Department of Chemistry. Chapters 6.ac.1–3 Chapter 9 does not form a part of the course. An excellent example of this kind of book is J. Acknowledgements Chapters 2 to 5 have been prepared especially for this course. Levitt Spin Dynamics (Wiley). H.cam. 7 and 8 are modiﬁed from notes prepared for summer schools held in Mishima and Sapporo (Japan) in 1998 and 1999. J.cam.uk . thanks are due to Professor F Inagaki for the opportunity to present this material. James Keeler University of Cambridge. K. University of California. Finally. For a deeper understanding you can do no better that the recently published M. M. but rather concentrate on the interpretation of various kinds of spectra. Many of these avoid any serious attempt to describe how the experiments work.
if the line from TMS comes out on one spectrometer at 400 MHz. The most familiar effect of strong coupling is the “rooﬁng” or “tilting” of multiplets. we will see that in principle there are other transitions. 2. which can take place. there are. If we wanted to quote the NMR frequency it would be inconvenient to have to specify the exact magnetic ﬁeld strength as well. In NMR spectroscopy we tend not to use this approach of thinking about energy levels and the transitions between them. For the case shown in Fig. especially for understanding more complex experiments. 2. to think about how the familiar NMR spectra we see are related to energy levels. Finally. it is useful. These are spectra in which the simple rules used to construct multiplets no longer apply because the shift differences between the spins have become small compared to the couplings. ν. the numbers we would have Chapter 2 “NMR and energy levels” c James Keeler. called multiple quantum transitions. Such a transition can be caused by a photon of light whose frequency. In such spin systems. Chemical shifts It is found to a very good approximation that the frequencies at which NMR absorptions (lines) occur scale linearly with the magnetic ﬁeld strength. 2. In addition.1 A line in the spectrum is associated with a transition between two energy levels. doubling the magnetic ﬁeld will result in it coming out at 800 MHz. we will look at strongly coupled spectra. E = E 2 − E 1 . we use different rules for working out the appearance of multiplets and so on. as they are known. as we shall see. To start with we will look at the energy levels of just one spin and them move on quickly to look at two and three coupled spins. However. E. NMR spectroscopists tend to use some rather unusual units. Rather.1 Frequency and energy: sorting out the units E2 hν = E2–E1 ν spectrum E1 energy levels Fig. but we can detect them indirectly using twodimensional experiments. 2002 . We will see how such spectra can be analysed in some simple cases. is related to the energy gap. important applications of such multiple quantum transitions.1. So.2 NMR and energy levels The picture that we use to understand most kinds of spectroscopy is that molecules have a set of energy levels and that the lines we see in spectra are due to transitions between these energy levels. Such transitions are not observed in simple NMR spectra. and so we need to know about these and how to convert from one to another if we are not to get into a muddle. between the two levels according to: E = hν where h is a universal constant known as Planck’s constant.
in the case of proton NMR the reference compound is TMS. νTMS (2.2–2 NMR and energy levels to quote would not be very memorable. Angular frequency Frequencies are most commonly quoted in Hz. τ . about which we will explain later. Suppose that there are two peaks in the spectrum at shifts δ1 and δ2 in ppm. the chemical shift is rather small so it is common to multiply the value for δ by 106 and then quote its value in parts per million. the point returns to the same position at regular intervals each time it has competed 360◦ of rotation. With this deﬁnition the chemical shift of the reference compound is 0 ppm. If the frequency of the line we are interested in is ν (in Hz) and the frequency of the line from TMS is νTMS (also in Hz). The time taken for the point to return to its original position is called the period.2) Sometimes we want to convert from shifts in ppm to frequencies. but it is more convenient to use the chemical shift scale in which frequencies are expressed relative to that of an agreed reference compound. Writing this difference out in terms of the deﬁnition of chemical shift given in Eq. What is the frequency separation between the two peaks? It is easy enough to work out what it is in ppm. νTMS Multiplying both sides by νTMS now gives us what we want: (ν2 − ν1 ) = 10−6 × νTMS × (δ2 − δ1 ). . such as TMS in the case of proton spectra. 2. would you like to quote the shift of the protons in benzene as 400. However. or ppm. the chemical shift of the line is computed as: ν frequency νTMS δ= ν − νTMS .2 An NMR spectrum can be plotted as a function of frequency. it is just (δ2 − δ2 ). 2. Think about a point on the edge of a disc which is rotating about its centre. For example. As all the frequencies scale with the magnetic ﬁeld. δppm = 106 × ν − νTMS .001234 MHz? We neatly sidestep both of these problems by quoting the chemical shift relative to an agreed reference compound. there are plenty of other possibilities so it is as well to regard the zero point on the frequency scale as arbitrary. νTMS (2. It is often sufﬁciently accurate to replace νTMS with the spectrometer reference frequency. For example. One choice for the zero frequency point is the line from the reference compound.2 we have: (δ2 − δ1 ) = 106 × ν2 − νTMS ν1 − νTMS − 106 × νTMS νTMS ν2 − ν1 = 106 × .1) δ δTMS = 0 chemical shift Fig. which is the same as “per second” or s−1 . If the disc is moving at a constant speed. Typically. If we want to change the ppm scale of a spectrum into a frequency scale we need to decide where zero is going to be. this ratio is independent of the magnetic ﬁeld strength.
Recall that 360◦ is 2π radians. ν.2. we can say that it has rotated though 2π radians in τ seconds. During each period the point moves through 2π radians or 360◦ . if the period is 0. ω. 2. is given by ω= 2π . All we have to remember is that there is a factor of h or h needed to convert to Joules if we need to. For all the nuclei that we are going .2 Nuclear spin and spin states The frequency.2 Nuclear spin and spin states NMR spectroscopy arises from the fact that nuclei have a property known as spin. If we want to express the frequency in angular units then the relationship with the energy is E =h ω 2π = hω ¯ Fig. In SI units the frequency is in Hz and h is in J s−1 . there is really nothing wrong with quoting energies in frequency units. τ one period The units of this frequency are “radians per second” or rad s−1 . I . we will not concern ourselves with where this comes from. The point to notice here is that frequency. is directly proportional to energy. There is another way of expressing the frequency. if the point completes a rotation in τ seconds. but just take it as a fact. 2π We will ﬁnd that angular frequencies are often the most natural units to use in NMR calculations. τ time start 2–3 For example. do not be concerned to see an energy expressed in Hz or rad s−1 . Quantum mechanics tells us that this nuclear spin is characterised by a nuclear spin quantum number. Angular frequencies will be denoted by the symbols ω or whereas frequencies in Hz will be denoted ν.001 s.3 A point at the edge of a circle which is moving at a constant speed returns to its original position after a time called the period. Energies A photon of frequency ν has energy E given by E = hν where h is Planck’s constant. and so this is what we will do. ν and ω are related via ω ν= or ω = 2π ν. So. which is in angular units. So. the frequency is 1/0. in either Hz or rad s−1 .001 = 1000 Hz. is simply the inverse of the period: ν= 1 . 2. where h (pronounced “h bar” or “h cross”) is Planck’s constant divided by ¯ 2π . So. The angular frequency. It turns out to be much ¯ more convenient to work in frequency units throughout.
If we have two spins in our molecule. E α and E β . It should be noted here that there is only a onetoone correspondence between these spin state combinations and the energy levels in the case of weak coupling. α is the low energy state for nuclei with a positive gyromagnetic ratio. in 1 the case of a spinhalf. and these can be labelled with either the m value of the labels α and β.1 . . but it is a bit superﬂuous here as we only have one spin). where again we have used the subscript 1 to distinguish the of nucleus. 2. So. From quantum mechanics we can show that the energies of these two levels. α1 β2 . B0 : ν0.1 Strictly. more of which below. The gyromagnetic ratio is characteristic of a particular kind of nucleus. although other values are possible.1 2 and 1 E β = − 2 ν0.β1 α2 and β1 β2 . We will see in a later chapter that all this negative frequency really means is that the spin precesses in a particular sense. You will note here that. These four possibilities correspond to four energy levels. but to be consistent we need to stick with this convention. We can continue the same process for three or more spins.2–4 NMR and energy levels to be concerned with. as explained in section 2. For such nuclei we therefore see that the Larmor frequency is negative. there are only two values of m. For the nuclei we are interested in. these are characterised by another quantum number m which quantum mechanics tells us is restricted to the values −I to I in integer steps. for three spins there are 8 combinations leading to 8 energy levels. and so there are four possibilities: α1 α2 . such as proton or carbon13. So αβ implies α1 β2 etc.1 = − 1 γ1 (1 + δ1 )B0 2π (2. are: E α = + 1 ν0. Further details are to be found in section 2. then each spin can be in the α or β state. the chemical shift δ. The state 2 1 with m = − 2 is denoted β and is sometimes described as “spin down”.1 we have written the energies in frequency units. In fact it is easy to see that the energies just depend on m: E m = mν0. Note that we have added a subscript 1 or 2 to differentiate the two nuclei. and as each spin is added the number of possible combinations increases. for example. A spinhalf 2 nucleus has an interaction with a magnetic ﬁeld which gives rise to two energy levels.1 is the Larmor frequency of spin 1 (we will need the 1 later on. and the strength of the applied magnetic ﬁeld. for the commonest nuclei (protons and carbon13) it is positive. as it is sometimes called) with m = 1 is denoted α and is sometimes described as “spin up”. its units are normally rad s−1 T−1 . The Larmor frequency depends on a quantity known as the gyromagnetic ratio. So. The magnetic ﬁeld is normally given in units of Tesla (symbol T). In fact. the α state is the one with the lowest energy. I = 1 .6. − 2 and + 1 . although often we will dispense with these and simply take it that the ﬁrst spin state is for spin 1 and the second for spin 2. 2 By tradition in NMR the energy level (or state. which we will assume from now on. where ν0.3) This negative Larmor frequency for nuclei with a positive γ sometimes seems a bit unnatural and awkward.3 One spin There are just two energy levels for a single spin. γ . γ can be positive or negative.
2.4 Two spins To take a speciﬁc example, for protons γ = +2.67 × 108 rad s−1 T−1 , so in a magnetic ﬁeld of 4.7 T the Larmor frequency of a spin with chemical shift zero is ν0 = − 1 γ (1 + δ)B0 2π 1 × 2.67 × 108 × 4.7 = −200 × 106 Hz. =− 2π
2–5
In other words, the Larmor frequency is −200 MHz. We can also calculate the Larmor frequency in angular units, ω0 , in which case the factor of 1/2π is not needed: ω0 = −γ (1 + δ)B0 which gives a value of 1.255 × 109 rad s−1 . Spectrum As you may know from other kinds of spectroscopy you have met, only certain transitions are allowed i.e. only certain ones actually take place. There are usually rules – called selection rules – about which transitions can take place; these rules normally relate to the quantum numbers which are characteristic of each state or energy level. In the case of NMR, the selection rule refers to the quantum number m: only transitions in which m changes by one (up or down) are allowed. This is sometimes expressed as m = m(initial state) − m(ﬁnal state) = ±1. Another way as saying this is that one spin can ﬂip between “up” and “down” or vice versa. In the case of a single spinhalf, the change in m between the two states is (+ 1 − (− 1 )) = 1 so the transition is allowed. We can now simply work out 2 2 the frequency of the allowed transition: ναβ = E β − E α = − 1 ν0,1 − (+ 1 ν0,1 ) 2 2 = −ν0,1 . Note that we have taken the energy of the upper state minus that of the lower state. In words, therefore, we see one transition at the minus the Larmor frequency, −ν0,1 . You would be forgiven for thinking that this is all an enormous amount of effort to come up with something very simple! However, the techniques and ideas developed in this section will enable us to make faster progress with the case of two and three coupled spins, which we consider next. 2.4 Two spins
Eβ
m = –1/2
ναβ = Eβ–Eα
ναβ = ν0,1 spectrum
Eα
m = +1/2
energy levels
Fig. 2.4 The transition between the two energy levels of a spinhalf is allowed, and results in a single line at the Larmor frequency of the spin.
2–6
NMR and energy levels We know that the spectrum of two coupled spins consists of two doublets, each split by the same amount, one centred at the chemical shift of the ﬁrst spin and one at the shift of the second. The splitting of the doublets is the scalar coupling, J12, quoted in Hz; the subscripts indicate which spins are involved. We will write the shifts of the two spins as δ1 and δ2 , and give the corresponding Larmor frequencies, ν0,1 and ν0,2 as: ν0,1 = − ν0,2 1 γ1 (1 + δ1 )B0 2π 1 = − γ2 (1 + δ2 )B0. 2π
J12
J12
ν0,2
ν0,1 frequency
Fig. 2.5 Schematic spectrum of two coupled spins showing two doublets with equal splittings. As indicated by the dashed lines, the separation of the Larmor frequencies is much larger than the coupling between the spins.
If the two nuclei are of the same type, such a proton, then the two gyromagnetic ratios are equal; such a two spin system would be described as homonuclear. The opposite case is where the two nuclei are of different types, such as proton and carbon13; such a spin system is described as heteronuclear. Energy levels As was already described in section 2.2, there are four possible combinations of the spin states of two spins and these combinations correspond to four energy levels. Their energies are given in the following table: number spin states 1 αα 2 αβ 3 βα 4 ββ + + − −
1 2 1 2 1 2 1 2
energy ν0,1 + 1 ν0,2 + 2 ν0,1 − 1 ν0,2 − 2 ν0,1 + 1 ν0,2 − 2 ν0,1 − 1 ν0,2 + 2
1 4 1 4 1 4 1 4
J12 J12 J12 J12
The second column gives the spin states of spins 1 and 2, in that order. It is easy to see that these energies have the general form: E m 1 m 2 = m 1 ν0,1 + m 2 ν0,2 + m 1 m 2 J12 where m 1 and m 2 are the m values for spins 1 and 2, respectively. For a homonuclear system ν0,1 ≈ ν0,2 ; also both Larmor frequencies are much greater in magnitude than the coupling (the Larmor frequencies are of the order of hundreds of MHz, while couplings are at most a few tens of Hz). Therefore, under these circumstances, the energies of the αβ and βα states are rather similar, but very different from the other two states. For a heteronuclear system, in which the Larmor frequencies differ signiﬁcantly, the four levels are all at markedly different energies. These points are illustrated in Fig. 2.6. Spectrum The selection rule is the same as before, but this time it applies to the quantum number M which is found by adding up the m values for each of the spins. In this case: M = m 1 + m 2. The resulting M values for the four levels are:
2.4 Two spins
ββ 4
2–7
ββ αβ αβ βα
4 2
2
3 βα αα
3 1
αα
1H –1H
1
13C –1H
Fig. 2.6 Energy levels, drawn approximately to scale, for two spin systems. On the left is shown a homonuclear system (two protons); on this scale the αβ and βα states have the same energy. On the right is the case for a carbon13 – proton pair. The Larmor frequency of proton is about four times that of carbon13, and this is clear reﬂected in the diagram. The αβ and βα states now have substantially different energies.
number spin states M 1 αα 1 2 αβ 0 3 βα 0 4 ββ −1 The selection rule is that M = ±1, i.e. the value of M can change up or down by one unit. This means that the allowed transitions are between levels 1 & 2, 3 & 4, 1 & 3 and 2 & 4. The resulting frequencies are easily worked out; for example, the 1–2 transition: ν12 = E 2 − E 1 1 = + 2 ν0,1 − 1 ν0,2 − 1 J12 − ( 1 ν0,1 + 1 ν0,2 + 1 J12 ) 2 4 2 2 4 = −ν0,2 − 1 J12. 2 The complete set of transitions are: transition spin states 1→2 αα → αβ 3→4 βα → ββ 1→3 αα → βα 2→4 αβ → ββ frequency −ν0,2 − 1 J12 2 −ν0,2 + 1 J12 2 −ν0,1 − 1 J12 2 −ν0,1 + 1 J12 2
Throughout we will use the convention that when computing the transition frequency we will take the energy of the upper state minus the energy of the lower: E = E upper − E lower .
The energy levels and corresponding schematic spectrum are shown in Fig. 2.7. There is a lot we can say about this spectrum. Firstly, each allowed transition corresponds to one of the spins ﬂipping from one spin state to the other, while the spin state of the other spin remains ﬁxed. For example, transition 1–2 involves a spin 2 going from α to β whilst spin 1 remains in the α state. In this transition we say that spin 2 is active and spin 1 is passive. As
2–8
ββ 4 flips α β 13 24
NMR and energy levels
24
34
α β flips 12 34
spin 1 spin 2
αβ
2
βα
3
12
13 −ν0,1 −ν0,2 frequency
αα
1
Fig. 2.7 On the left, the energy levels of a twospin system; the arrows show the allowed transitions: solid lines for transitions in which spin 1 ﬂips and dotted for those in which spin 2 ﬂips. On the right, the corresponding spectrum; it is assumed that the Larmor frequency of spin 2 is greater in magnitude than that of spin 1 and that the coupling J12 is positive.
spin 2 ﬂips in this transition, it is not surprising that the transition forms one part of the doublet for spin 2. Transition 3–4 is similar to 1–2 except that the passive spin (spin 1) is in the β state; this transition forms the second line of the doublet for spin 2. This discussion illustrates a very important point, which is that the lines of a multiplet can be associated with different spin states of the coupled (passive) spins. We will use this kind of interpretation very often, especially when considering twodimensional spectra. The two transitions in which spin 1 ﬂips are 1–3 and 2–4, and these are associated with spin 2 being in the α and β spin states, respectively. Which spin ﬂips and the spins states of the passive spins are shown in Fig. 2.7. What happens is the coupling is negative? If you work through the table you will see that there are still four lines at the same frequencies as before. All that changes is the labels of the lines. So, for example, transition 1–2 is now the right line of the doublet, rather than the left line. From the point of view of the spectrum, what swaps over is the spin state of the passive spin associated with each line of the multiplet. The overall appearance of the spectrum is therefore independent of the sign of the coupling constant. Multiple quantum transitions There are two more transitions in our twospin system which are not allowed by the usual selection rule. The ﬁrst is between states 1 and 4 (αα → ββ) in which both spins ﬂip. The M value is 2, so this is called a doublequantum transition. Using the same terminology, all of the allowed transitions described above, which have M = 1, are singlequantum transitions. From the table of energy levels it is easy to work out that its frequency is (−ν0,1 − ν0,2 ) i.e. the sum of the Larmor frequencies. Note that the coupling has no effect on the frequency of this line. The second transition is between states 2 and 3 (αβ → βα); again, both
In a two spin system the double. but spin 3 will still show a doublet of doublets. The appearance of these multiplets will depend on the relative sizes of the couplings.1 + ν0. However. The energies are given by: E m 1 m 2 m 3 = m 1 ν0. Energy levels Each of the three spins can be in the α or β spin state. As with the doublequantum transition. if J12 = 0.3 − ν0.2 + ν0.1 − ν0. The resulting spectrum consists of three doublets of doublets. and its frequency is (−ν0.1 + m 2 ν0. 2.10.3 − ν0.2 + m 3 ν0. On the other hand.1 + ν0.3 − 1 J12 + 4 ν0.1 − 2 J12 − J12 − J12 + J12 + We have grouped the energy levels into two groups of four. one centred at the shift of each of the three spins.1 + 2 − 1 ν0. The energy levels (for a homonuclear system) are shown schematically in Fig.5 Three spins αα 1 If we have three spins.3 + 1 J12 − 4 ν0.2 + ν0.3 − 1 J12 − 4 ν0.2.1 + 2 + 1 ν0. βα 2 2–9 ββ 14 23 4 αβ 3 2. The M value is 0. then the spectrum consists of three doublets of doublets.9.1 + ν0.1 Fig.and zeroquantum spectra are not especially interesting. a typical example of which is shown for spin 1 with the assumption that J12 is greater than J13 . the spin topology is shown in Fig.2 ) i.3 + 1 J12 + 4 ν0. The energies and corresponding M values (= m 1 + m 2 + m 3 ) are shown in the table: number spin states M 3 1 ααα 2 1 2 αβα 2 1 3 βαα 2 4 ββα −1 2 5 6 7 8 ααβ αββ βαβ βββ − − − 1 2 1 2 1 2 3 2 Fig. 1 J13 3 J12 J23 J12 2 J13 J13 + + − − 1 2 1 2 1 2 1 2 ν0.2 − ν0. there is the additional constraint that only . so this is called a zeroquantum transition.3 + ν0. + 1 ν0. each of which is coupled to the other two. We will also see later on that in twodimensional spectra we can exploit to our advantage the special properties of these multiplequantum spectra.2 + ν0. 2. 2.8 In a twospin system there is one double quantum transition (1–4) and one zeroquantum transition (2–3). if J12 = J13 the doublet of doublets from spin 1 will collapse to a 1:2:1 triplet. the coupling has no effect on the frequency of this line.9 The topology – that is the number of spins and the couplings between them – for a threespin system in which each spin is coupled to both of the others.2 − ν0.2 + ν0. the ﬁrst group all have spin 3 in the α state and the second have spin 3 in the β state. the frequency of neither of these transitions are affected by the size of the coupling between the two spins.e. only doublets will be seen for spin 1 and spin 2.5 Three spins spins ﬂip.2 − ν0. so there are a total of 8 possible combinations corresponding to 8 energy levels. Spectrum The selection rule is as before. the difference of the Larmor frequencies. in the case of more than two spins.1 − 2 − 1 ν0. For example. that is M can only change by 1.2 − 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 energy ν0.3 + m 1 m 2 J12 + m 1 m 3 J13 + m 2 m 3 J23 where m i is the value of the quantum number m for the i th spin.1 − 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 ν0. 2.3 + 1 4 1 4 1 4 1 4 1 4 1 4 1 4 1 4 1 4 1 4 1 4 1 4 J13 + J13 − J13 + J13 − J13 − J13 + J13 − J13 + 1 4 1 4 1 4 1 4 1 4 1 4 1 4 1 4 J23 J23 J23 J23 J23 J23 J23 J23 ν0. but we will see in a threespin system that the situation is rather different.
2.1 + 1 J12 − 2 −ν0. The results are shown in the table.10 Energy levels for a homonuclear threespin system. where it has been assuming that ν0. The levels can be grouped into two sets of four: those with spin 3 in the α state (shown on the left with solid lines) and those with spin 3 in the β state.11.1 = −100 Hz.2–10 βββ NMR and energy levels 8 ββα 4 αββ 6 βαβ 7 αβα 2 βαα 3 ααβ 5 ααα 1 Fig. one spin can ﬂip. The multiplet is labelled with the spin states of the passive spins. 2. As in the case of a twospin system.1 104 108 ααα 1 Fig. along with the spin states of the passive spins (2 and 3 in this case). The schematic spectrum is illustrated in Fig. shown on the right (dashed lines).1 − 1 J12 − 2 −ν0. J12 = 10 Hz and J13 = 2 Hz. The frequencies of these lines can easily be worked out from the table of energy levels on page 2– 9. Applying these rules we see that there are four allowed transitions in which spin 1 slips: 1–3. transition state of spin 2 1–3 α 2–4 β 5–7 α 6–8 β state of spin 3 α α β β frequency −ν0. 2–4. These four transitions form the four lines of the multiplet (a doublet of doublets) at the shift of spin 1.11 Energy levels for a threespin system showing by the arrows the four allowed transitions which result in the doublet of doublets at the shift of spin 1.1 − 1 J12 + 2 −ν0. 5–7 and 6–8.1 + 1 J12 + 2 α α βββ 8 α β β α 1 2 1 2 1 2 1 2 J13 J13 J13 J13 β spin 2 β spin 3 13 57 24 68 J12 J13 ββα 4 αββ 6 βαβ 7 J12 J13 αβα 2 βαα 3 ααβ 5 92 96 100 −ν0. The schematic multiplet is shown on the right. we can label each line of the . 2.
the two doublets are not centred at −ν0.2 .1 and −ν0.2 = ν0. 2–6 and 4–8). as 3 ν0. The same is true of those transitions associated with spin 3 being in the β state (dashed lines).1 . 3–4.1 and −ν0. The ﬁrst group consists of the lines which have spin 3 in the α state and the second group consists of the lines which have spin 3 in the α state.12. 3–7. this means the spin states of spins 2 and 3. but at (−ν0. we can deﬁne effective Larmor frequencies for spin 3 being in the β state: 3 ν0.1 − 1 J13 2 β 3 ν0. Similarly. We can deﬁne an effective Larmor frequency for α3 spin 1 with spin 3 in the α spin state.1 J12 αβα 2 βαα 3 ααβ 5 α3 −ν0.2 − 1 J23 .1 − 1 1 2 J13 ) and (−ν0. 2 β .1 + 1 J13 2 α and likewise for spin 2: 3 ν0.12 Illustration of the division of the two multiplets from spins 1 and 2 into subspectra according to the spin state of spin 3. 5–6 and 7–8) and the four which contribute to that from spin 3 (1–5.2 + 1 J23. ν0.2 ββα 4 αββ 6 βαβ 7 J12 spin 3 in α state α3 −ν0.1 = ν0.2 Fig.5 Three spins multiplet with the spin states of the passive spins – in the case of the multiplet from spin 1. except that the shift is in the opposite direction.2 . Subspectra J13 J23 2–11 βββ 8 −ν0.1 −ν0.2. In the same way. However. 2.2 = ν0. There are four lines which have spin3 in the α state. we can identify the four transitions which contribute to the multiplet from spin 2 (1–2.1 = ν0. and as can be seen from the spectrum these form two doublets with a common separation of J12. 2 α The two doublets in the subspectrum corresponding to spin 3 being in the α α3 α3 state are thus centred at −ν0. The transitions associated with spin 3 in the α state (indicated by the full lines on the energy level diagram) give rise to a pair of doublets.2 − 2 J23 ). One was of thinking about the spectrum from the threespin system is to divide up the lines in the multiplets for spins 1 and 2 into two groups or subspectra. 2.2 J12 ααα 1 spin 3 in β state J12 β3 −ν0.1 β3 −ν0. but with their centres shifted from the Larmor frequencies by half the coupling to spin 3. This separation is illustrated in Fig.
1 − ν0.2 − ν0. We can think of the spectrum of spin 1 and 2 as being composed of two subspectra.2–12 NMR and energy levels 3 3 The two doublets in the β subspectrum are centred at −ν0.2 ααα 1 frequency Fig.3 + 1 J12 + 1 J13 2 2 These transitions come in three pairs. The two resulting lines form a doublet which is centred at the sum of the Larmor frequencies of spins 1 and 2 and which is split by the sum of the couplings to spin 3.6 Note that the separation of the spectra according to the spin state of spin 3 is arbitrary.1 − ν0.13 There are two double quantum transitions in which spins 1 and 2 both ﬂip (1–4 and 5–8). We could just as well separate the two multiplets from spins 1 and 3 according to the spin state of spin 2. We can describe these transitions as a kind of double quantum doublet. It has been assumed that both couplings are positive. section 2.2 . Transitions 1–4 and 5–8 are centred at the sum of the Larmor frequencies of spins 1 and 2.2 − 1 J13 − 1 J23 2 2 −ν0.3 + 1 J12 + 1 J23 2 2 −ν0. βββ 8 α 14 ββα 4 αββ 6 7 βαβ β 58 spin 3 J13 + J23 αβα 2 3 βαα ααβ 5 −ν0. This kind of approach is particularly useful when it comes to dealing with strongly coupled spin systems. β β Multiple quantum transitions There are six transitions in which M changes by 2.1 and −ν0.2 − ν0. and spin 3 is passive. this is not surprising as we note that in these transitions it is the spin states of both spins 1 and 2 which ﬂip. transition initial state 1–4 ααα 5–8 ααβ 1–7 2–8 1–6 3–8 ααα αβα ααα βαα ﬁnal state ββα βββ βαβ βββ αββ βββ frequency −ν0. but they are unaffected by the coupling J12 which is between the two spins which ﬂip.1 −ν0.2 + 1 J13 + 1 J23 2 2 −ν0.1 − ν0. each from a two spin system but in which the Larmor frequencies are effectively shifted one way of the other by half the coupling to the third spin.1 − ν0. Their frequencies are given in the table. 2. As with the singlequantum spectra. we can associate the two lines of the doublet with different spin states of the third spin.3 − 1 J12 − 1 J23 2 2 −ν0.3 − 1 J12 − 1 J13 2 2 −ν0. The two transitions are separated by the sum of the couplings to spin 3 (J13 + J23 ). Just as . Spins 1 and 2 are both active in these transitions.
in this limit all of the lines in a multiplet have the same intensity.e.2. These are spin systems in which the differences between the Larmor frequencies (in Hz) of the spins are much greater in magnitude than the magnitude of the couplings between the spins.3 2–13 Notice that the frequencies of these lines are not affected by any of the couplings. In this transition all of the spins ﬂip. in which M changes by 3.2 − ν0. Finally.13. This illustrates the idea that as we move to higher orders of multiple quantum. like multiple quantum transitions. The table gives the frequencies of these three lines: transition initial state ﬁnal state 2–7 αβα βαβ 3–6 βαα αββ 4–5 ββα ααβ frequency −ν0. they are called combination lines. in the weak coupling limit it is possible to work out the energies of the levels present simply by making all possible combinations of spin states. Combination lines There are three more transitions which we have not yet described.3 +ν0. but this time centred around the difference in the Larmor frequencies of two of the spins. We will also see in section 2.2 − ν0. the differences between Larmor frequencies and the values of couplings can be positive or negative! In deciding whether or not a spectrum will be strongly coupled we need to compare the magnitude of the difference in the Larmor frequencies with the magnitude of the coupling. So far all we have said about energy levels and spectra applies to what are called weakly coupled spin systems. the corresponding spectra become simpler. There are also six zeroquantum transitions in which M does not change. A schematic representation of the spectrum is shown in Fig.2 − ν0. the sum of the Larmor frequencies. Such lines are not seen in normal spectra but.1 − ν0. 2. This feature has been used in the analysis of some complex spin systems. just as we have done above.1 + ν0. between levels 1 (ααα) and 8 (βββ). There are thus many similarities between the double. the doublequantum spectrum of three doublets and the triplequantum spectrum of a single line.2 − ν0.and zeroquantum spectra. i.6 Strong coupling We need to be careful here as Larmor frequencies. Under these circumstances the rules from predicting spectra are very simple – they are the ones you are already familiar with which you use for constructing multiplets. we can associate one line with spin 3 being in the α state (1–4) and one with it being in the β state (5–8).6 that these lines may be observable in strongly coupled spectra. they can be detected indirectly using twodimensional spectra. 2. Like the double quantum transitions these group in three pairs.3 +ν0. In addition.6 Strong coupling we did before. These zeroquantum doublets are split by the difference of the couplings to the spin which does not ﬂip in the transitions. In a three spin system there is one triplequantum transition. We see that the singlequantum spectrum consists of three doublets of doublets.3 . M changes by 1 but all three spins ﬂip. .1 − ν0.1 + ν0. and from the table of energies we can easily work out that its frequency is −ν0. For these.
intensity (1 + sin 2θ ) (1 − sin 2θ ) (1 − sin 2θ ) (1 + sin 2θ ) is the sum of the Larmor frequencies: (2. However. Under these circumstances all of the lines have unit intensity. and really the only practical approach is to use computer simulation. In this limit it is clear that sin 2θ = 2 D 2 = (ν0. This makes the spectrum more and more strongly coupled. This is sloppy usage.2 )2 and so D = (ν0. J A Jones and S Wimperis (Oxford University Press.1 − ν0.2 )2 + J12 . You will often ﬁnd that people talk of two spins being strongly coupled when what they really mean is the coupling between the two spins is large. we will always use the term strong coupling in the sense described in this section. the peaks are just about all the same intensity and where we expect them to be. It is a relatively simple exercise in quantum mechanics to work out the energy levels and hence frequencies and intensities of the lines from a strongly coupled twospin system1 . When D is very much larger than J12 the fraction J12/D becomes small. D The ﬁrst thing to do is to verify that these formulae give us the expected result when we impose the condition that the separation of the Larmor frequencies is large compared to the coupling. for example.2–14 NMR and energy levels If the separation of the Larmor frequencies is not sufﬁcient to satisfy the weak coupling criterion. .1 − ν0.2 and D is the positive quantity deﬁned as 2 D 2 = (ν0. Chapter 10 of NMR: The Toolkit.1 − ν0.2 )2 + J12 ≈ (ν0. it is useful to look at the spectrum from two strongly coupled spins as the spectrum is simple enough to analyse by hand and will reveal most of the of the crucial features of strong coupling.1 − ν0. This makes predicting or analysing the spectra much more difﬁcult than in the case of weak coupling. the system is said to be strongly coupled. 2000) 1 See.2 ). Putting this value into the table above gives us exactly the frequencies we had before on page 2–7.14 shows a series of spectra computed using the above formulae in which the Larmor frequency of spin 1 is held constant while the Larmor frequency of spin 2 is progressively moved towards that of spin 1.1 + ν0. Figure 2. the weak coupling limit is regained. and so sin 2θ ≈ 0 (sin φ goes to zero as φ goes to zero). In this limit none of the rules outlined in the previous paragraph apply. These are given in the following table.4) The angle θ is called the strong coupling parameter and is deﬁned via J12 . by P J Hore. transition frequency 1 1 1–2 − 1 J12 2D − 2 2 1 D − 1 + 1 J12 3–4 2 2 2 1–3 − 1 D − 1 − 1 J12 2 2 2 2–4 − 1 D − 1 + 1 J12 2 2 2 In this table = ν0. So. The spectrum at the bottom is almost weakly coupled.
but the Larmor frequencies (the shifts) are no longer midway between the two lines of the doublet. 2. The parameters used for these spectra were ν0. lines 1–3 and 3–4 appear elsewhere but have intensity zero.14 A series of spectra of a two spin system in which the Larmor frequency of spin 1 is help constant and that of spin 2 is moved in closer to spin 1.2 = –10 –ν0. These spectra illustrate the socalled roof effect in which the intensities of the lines in a strongly coupled multiplet tilt upwards towards the multiplet from the coupled spin. In fact it is easy Fig.2 = –90 Fig. as the Larmor frequencies of the two spins get closer and closer together we notice two things: (1) the “outer” two lines get weaker and the “inner” two lines get stronger. The spectra become more and more strongly coupled showing a pronounced roof effect until in the limit that the two Larmor frequencies are equal only one line is observed. (2) the two lines which originally formed the doublet are no longer symmetrically spaced about the Larmor frequency.14 also illustrate the point that when the two Larmor frequencies are identical there is only one line seen in the spectrum and this is at this Larmor frequency. There is one more thing to notice which is not so clear from the spectra but is clear if one looks at the frequencies in the table. 2. this is called the “roof” effect. However.6 Strong coupling ~ 2–15 ν0. . the peak in the top most spectrum has been truncated. Note that as the “outer” lines get weaker the “inner” lines get proportionately stronger. The “take home message” is that from such strongly coupled spectra we can easily measure the coupling.1 20 40 60 frequency (Hz) 80 –ν0.1 = −10 Hz and J12 = 5 Hz. the same is true for the other doublet. The spectra in Fig 2.2 100 ν0.15 The intensity distributions in multiplets from stronglycoupled spectra are such that the multiplets “tilt” towards one another.15 illustrates the idea. 2. In this limit lines 1–2 and 2–4 both appear at the Larmor frequency and with intensity 2.2 J12 ν0.2 = –20 –ν0. making a kind of roof.2 = –50 –ν0.2 13 24 12 34 0 –ν0.2 J12 12 34 ν0. in fact the stronger of the two lines moves progressively closer to the Larmor frequency. Fig.2. This is that the two lines that originally formed the spin 1 doublet are always separated by J12 .
it is clear from the table that the frequency separation of the lefthand lines of the two multiplets (3–4 and 2–4) is D J12 D D J12 ν34 − ν24 = ( 1 D − 2 = D. 2. whereas three weakly coupled spins would be denoted AMX.1 + ν0.2 )) ν0.1 = 1 ( 2 + (ν0.5 that we could think about the spectrum of three coupled spins in terms of subspectra in which the Larmor frequencies were replaced by effective Larmor frequencies.2 ) we can ﬁnd ν0. Eqn.2 )2 + J12 Fig. 1 2 + 1 J12) − (− 1 D − 2 2 1 2 + 1 J12) 2 34 12 24 13 The separation of the other two lines is also D.1 − ν0. So. the idea is illustrated in Fig. Each spin is given a letter (rather than a number). therefore (ν0. then the convention is to used letters which are close in the alphabet. 1 2 − 1 J12) + (− 1 D − 2 2 1 2 + 1 J12) 2 Now we have a values for = (ν0.2–16 NMR and energy levels enough to work out the Larmor frequencies using the following method.1 − ν0. So.16.1 − ν0.2 ) from its deﬁnition. with a different letter for spins which have different Larmor frequencies (chemical shifts). and so once we know D it is easy to compute (ν0. 2. The ABX spin system We noted in section 2.1 − ν0.16 The quantities J12 and D are readily measurable from the spectrum of two strongly coupled spins. . The notation ABX implies that two of the spins (A and B) are strongly coupled but the Larmor frequency of the third spin is widely separated from that of A and B. 2 D 2 = (ν0.2 separately: ν0.1 − ν0. a strongly coupled twospin system would be denoted AB and a strongly coupled threespin system ABC.4.2 = 1 ( 2 − (ν0. Notation for spin systems There is a traditional notation for spin systems which it is sometimes useful to use. If the Larmor frequencies of two spins are widely separated they are given letters which are widely separated in the alphabet. If spins are strongly coupled. Remember we can easily measure J12 directly from the splitting.1 and ν0. This kind of approach is very useful for understanding the AB part of the ABX spectrum.1 − ν0.2 )). two weakly coupled spins are usually denoted as A and X.2 ) = 2 D 2 − J12 . If we denote the frequency of transition 1–2 as ν12 and so on. In this way we can extract the Larmor frequencies of the two spins (the shifts) and the coupling from the strongly coupled spectrum. Now we notice from the table on 2–14 that the sum of the frequencies of the two stronger lines (1–2 and 2–4) or the two weaker lines (3–4 and 2–4) gives us − : ν12 + ν24 = ( 1 D − 2 =− .2 ) and a value for (ν0. 2.
Once we have done this. rather than the four which would be expected in the weak coupling limit.A–1/2JAX 20 30 –ν0. JAX = 15 Hz and JBX = 3 Hz. and so the degree of strong coupling (and hence the intensity patterns) in the two subspectra will be different. JAB = 5 Hz. the two subspectra can be analysed in exactly the way described above for an AB system.B = −30 Hz. Figure 2. 2 2 The other has effective Larmor frequencies ν0.B − 1 JBX.A + 1 JAX and ν0. ν0. 2. The two extra lines are combination lines which become observable when strong coupling is present.17 AB parts of an ABX spectrum illustrating the decomposition into two subspectra with different effective Larmor frequencies (indicated by the arrows). respectively.B+1/2JBX α subspectrum 0 10 –ν0.2. one is associated with the X spin being in the α state and the other with the spin being in the β state. All we can measure is the complete spectrum (the sum of the two subspectra) but once we know that it is in fact the sum of two ABtype spectra it is usually possible to disentangle these two contributions.A+1/2JAX –ν0.17 illustrates this decomposition. In general it contains 6 lines.B .A = −20 Hz. If spins A and B have Larmor frequencies ν0. then one subspectrum has effective Larmor frequencies ν0. .A –ν0.B–1/2JBX 40 50 frequency (Hz) Fig.B β subspectrum –ν0. 2 2 The separation between the two effective Larmor frequencies in the two subspectra can easily be different.A − 1 JAX and ν0. The parameters used in the simulation were ν0.A and ν0. The form of the X part of the ABX spectrum cannot be deduced from this simple analysis.B + 1 JBX. As spin X is weakly coupled to the others we can think of the AB part of the spectrum as two superimposed AB subspectra.6 Strong coupling 2–17 full spectrum –ν0.
2. and which four comprise the spin 3 multiplet.34 ppm. and let the coupling between the two spins be −5 Hz. Make a sketch of the multiplet (roughly to scale) and label the lines in the same way as is done in Fig. however.136305 MHz? Recalculate the shift using the spectrometer frequency. νspec quoted by the manufacturer as 400. and label each line with the energy levels involved (i. Also indicate for each line which spin ﬂips and the spin state of the passive spin. How would these labels change if J23 = −4 Hz? On an energy level diagram.11.135705 MHz.54 and 5. compute the frequencies of the lines which comprise the spin 2 multiplet.2–18 2. E 2–4 Consider a system of two weakly coupled spins. indicate the four transitions which comprise the spin 2 multiplet. in ppm. 1–2 etc. 2. draw up a table similar to that on page 2–10 showing the frequencies of the four lines of the multiplet from spin 2.2 = −200 Hz. Make a sketch of the spectrum. 2. 2. E 2–6 For a three spin system. What is the shift. . Compare your sketch with Fig. Then. roughly to scale. of a peak which has a frequency of 400.). E 2–5 For a three spin system. taking ν0.13 MHz rather than νTMS in the denominator of Eq. What is the separation of these two lines in Hz and in rad s−1 ? E 2–3 Calculate the Larmor frequency (in Hz and in rad s−1 ) of a carbon13 resonance with chemical shift 48 ppm when recorded in a spectrometer with a magnetic ﬁeld strength of 9.7 Exercises NMR and energy levels E 2–1 In a proton spectrum the peak from TMS is found to be at 400.7283 × 107 rad s−1 T−1 .11. the principles remain the same.e.4 T. νspec Does this make a signiﬁcant difference to the value of the shift? E 2–2 Two peaks in a proton spectrum are found at 1. compute the frequencies of the six zeroquantum Of course in reality the Larmor frequencies out to be tens or hundreds of MHz. Compute the frequencies of the lines in the normal (single quantum) spectrum. it makes the numbers easier to handle if we use these unrealistic small values. J23 = 4 Hz and the rest of the parameters as in Fig. The spectrometer frequency is quoted as 400. The gyromagnetic ratio of carbon13 is +6.2: δppm = 106 × ν − νTMS . not 100 Hz! However. Let the Larmor frequency of the ﬁrst spin be −100 Hz and that of the second spin be −200 Hz.7 and comment on any differences.13 MHz.
0 (0. in Hz. Also.7). Show that the values you ﬁnd are consistent with the observed intensities. J12 = 5 Hz.3). (intensities are given in brackets): 32. 10 15 20 25 30 35 frequency (Hz) 40 Fig.18 The AB part of an ABX spectrum E 2–9 Figure 2. 6. 39.7 Exercises transitions and also mark these on an energy level diagram. give the value of the AB coupling. 2.2 = −20 Hz. . Determine the values of the coupling constant and the two Larmor frequencies. Make a sketch of the spectrum (roughly to scale) indicating which transition is which and the position of the Larmor frequencies. 2–19 Make sure that you have your calculator set to “radians” when you compute sin 2θ .0 (0. 13. Disentangle the two subspectra.7).1 = −10 Hz.3).18 shows the AB part of an ABX spectrum.2. Do these six transitions fall into natural groups? How would you describe the spectrum? E 2–7 Calculate the line frequencies and intensities of the spectrum for a system of two spins with the following parameters: ν0. mark in the rough positions of the effective Larmor frequencies and hence estimate the size of the AX and BX couplings.0 (1.0 (1. E 2–8 The spectrum from a stronglycoupled two spin system showed lines at the following frequencies. ν0.
This would be a good point to comment on the axis system we are going to use – it is called a righthanded set. 1 It z magnetic field magnetization vector x y Fig. so what we need to be concerned with is the net effect of all these nuclei. The magnetization can be represented by a vector – called the magnetization vector – pointing along the direction of the applied ﬁeld (z).3 The vector model For most kinds of spectroscopy it is sufﬁcient to think about energy levels and selection rules. In fact. This is called the bulk magnetization of the sample. which is what we will use throughout these lectures. The name righthanded comes about from the fact that if you imagine grasping is a common misconception to state that the magnetic moment must either be aligned with or against the magnetic ﬁeld. To make any progress in understanding NMR experiments we need some more tools. let alone the large number of subtle twodimensional experiments which have been developed. to be more precise. and that it is this which gives rise the energy levels and ultimately an NMR spectrum. 3. In an NMR experiment. 3. as shown in Fig. In many ways. and the ﬁrst of these we are going to explore is the vector model. a magnetic moment. and not surprisingly the language and ideas which ﬂow from the model have become the language of NMR to a large extent. the ideas that ﬂow from even this rather restricted area in which the model can be applied are carried over into more sophisticated treatments. In fact. However.1 Bulk magnetization We commented before that the nuclear spin has an interaction with an applied magnetic ﬁeld. the vector model can only be applied to a surprisingly small number of situations. 3. it is permissible to think of the nucleus as behaving like a small bar magnet or. quantum mechanics says no such thing (see Levitt Chapter 9 for a very lucid discussion of this point). we do not observe just one nucleus but a very large number of them (say 1020 ).1. and such a set of axes is show in Fig. but note that the quantum mechanics tells us that the magnetic moment can be aligned in any direction1. If the magnetic moments were all to point in random directions.1 At equilibrium. The axis set in this diagram is a righthanded one. This model has been around as long as NMR itself. as this is what we will observe. in the strictest sense. However. Chapter 3 “The vector model” c James Keeler. a sample has a net magnetization along the magnetic ﬁeld direction (the z axis) which can be represented by a magnetization vector. 2002 . 3. From now on we will only be concerned with what happens to this vector. It is therefore essential to have a good grasp of the vector model and how to apply it. it turns out that at equilibrium the magnetic moments are not aligned randomly but in such a way that when their contributions are all added up there is a net magnetic ﬁeld along the direction of the applied ﬁeld (B0). then the small magnetic ﬁeld that each generates will cancel one another out and there will be no net effect. We will not go into the details here. using this energy level approach we cannot even describe how the most basic pulsed NMR experiment works.1. For example. this is not true for NMR.
The ﬁngers then curl in the sense of a positive precession. 3. the Larmor frequency is a signed quantity and is negative for nuclei with a positive gyromagnetic ratio. The whole process is analogous to the way in which electric current can be generated by a magnet rotating inside a coil.3.3 The precessing magnetization will cut a coil wound round the x axis.3. 3. and this corresponds to a negative Larmor frequency. such that it makes an angle β to that axis. The vector is said to precesses about the ﬁeld and this particular motion is called Larmor precession. 3. 3. this is no accident. to tip the magnetization vector away from the z axis. As the magnetization vector “cuts” the coil a current is induced which we can amplify and then record – this is the socalled free induction signal which is detected in a pulse NMR experiment. with the axis of the coil aligned in the x yplane.2. For clarity. . z x y where γ is the gyromagnetic ratio. Imagine grasping the z axis with your right hand. it is given by ν0 = − 1 γ B0 2π z magnetic field x y Fig. 3. 3. some how. thereby inducing a current in the coil. In words. The precession of the magnetization vector is what we actually detect in an NMR experiment. which is precisely what is shown in Fig.2 will show that the magnetization vector is rotating in the opposite sense to your ﬁngers. with the thumb pointing along the +z direction.3. then the frequency of the Larmor precession is ω0 (in rad s−1 ) ω0 = −γ B0 or if we want the frequency in Hz. We can sort out positive and negative frequencies in the following way. the frequency at which the magnetization precesses around the B0 ﬁeld is exactly the same as the frequency of the line we see from the spectrum on one spin. These are of course exactly the same frequencies that we encountered in section 2. This means that for such spins the precession frequency is negative.2 Larmor precession Suppose that we have managed. your ﬁngers curl from the x to the y axes. All we have to do is to mount a small coil of wire round the sample. Once tilted away from the z axis we ﬁnd is that the magnetization vector rotates about the direction of the magnetic ﬁeld sweeping out a cone with a constant angle. it is this that forms the free induction signal. We will see later on that such a tilt can be brought about by a radiofrequency pulse. 3. The direction of precession shown is for a nucleus with a positive gyromagnetic ratio and hence a negative Larmor frequency.3–2 The vector model the z axis with your right hand. Inspection of Fig. the coil has only been shown on one side of the x axis. If the magnetic ﬁeld strength is B0. This current can be ampliﬁed and detected.3 Detection Fig. this is illustrated in Fig.2 If the magnetization vector is tilted away from the z axis it executes a precessional motion in which the vector sweeps out a cone of constant angle to the magnetic ﬁeld direction. 3. see Fig. As was discussed in section 2.2.
At a later time. say τ2 . the xcomponent is M0 sin β. we can deduce that the x. Fig. so 1 = ω0 τ1 . We will also see in a later section that in practice we can easily detect both the x. we can visualize what happens to the x. 3.5. We now turn to the important question as to how we can rotate the magnetization away from its equilibrium position along the z axis. 3.4 Pulses Mx My time Fig. We can now easily work out the x. 3. 3.4 Tilting the magnetization through an angle θ gives an x component of size M 0 sin β. we can see that after time t the angle is = ω0 t.and ycomponents. the coil detects the xcomponent of the magnetization. 3. After a time τ1 the vector has rotated through a certain angle. in time τ1 the vector has moved through (ω0 × τ1 ) radians. 3. Mx and M y respectively. In general. Plots of these signals are shown in Fig. Although the magnetization vector precesses on a cone. Suppose that the equilibrium magnetization vector is of size M0 . At time zero. are: Mx = M0 sin β cos(ω0 t) M y = −M0 sin β sin(ω0 t). if this has been tilted through an angle β towards the x axis.and ycomponents of the magnetization. The angle through which the vector has precessed is given by ω0 t.and ycomponents much more simply by just thinking about the projection onto the x yplane. 3. We see that they are both simple oscillations at the Larmor frequency.and y components of the magnetization predicted using the approach of Fig.6.3. 3. On the righthand diagram we see the geometry for working out the x and y components of the vector. this is illustrated in Fig. The xcomponent is proportional to cos and the ycomponent is negative (along −y) and proportional to sin .4 illustrates the geometry. 3. This is shown in Fig. which we will call 1 . the vector has had longer to precess and the angle 2 will be (ω0 τ2 ). Recalling that the initial size of the vector is M0 sin β.5. Conceptually it is . As the vector is rotating at ω0 radians per second.4 Pulses Essentially.and ycomponents of the magnetization using simple geometry. Fourier transformation of these signals will give rise to the usual spectrum. we will assume that there is only an xcomponent.5. the details of how this works will be covered in a later chapter.6 Plots of the x .5 Illustration of the precession of the magnetization vector in the xy plane. x ε1 ε2 ε τ1 τ2 Fig. y 3–3 z β M0 sin β x Fig. Fourier transformation of these signals gives us the familiar spectrum – in this case a single line at ω0 . We can easily work out what this will be.
Rotating frame When RF power is applied to the coil wound along the x axis the result is a magnetic ﬁeld which oscillates along the x axis. as illustrated in Fig. .7 If the magnetic ﬁeld along the z axis is replaced quickly by one along x. and it turns out that the key is to use the idea of resonance. but this time along the −x axis. As time proceeds. 3. The magnetization would then precess about the new magnetic ﬁeld which would bring the vector down away from the z axis. this linearly oscillating ﬁeld with two counterrotating ﬁelds.7. The magnetic ﬁeld moves back and forth from +x to −x passing through zero along the way. rotates in the negative sense. The ﬁelds continue to rotate. Conveniently. we can use the same coil to generate this oscillating magnetic ﬁeld as the one we used to detect the magnetization (Fig. Thus we see that the two z x magnetic field y z x y Fig. However. reaching a maximum when the angle of rotation is π . Unfortunately it is all but impossible to switch the magnetic ﬁeld suddenly in this way. We will take the frequency of this oscillation to be ωRF (in rad s−1 ) and the size of the magnetic ﬁeld to be 2B1 (in T). the reason for the 2 will become apparent later. denoted B1 . the xcomponents shrink towards zero as the angle through which the vectors have rotated approaches 1 π radians or 90◦ .3). 3. both are rotating at the transmitter frequency ωRF . we will need to ﬁnd another approach. rotates in the positive sense (from x to y) − and the other. B0 . the vectors move away from x. The idea is to apply a very small magnetic ﬁeld along the x axis but one which is oscillating at or near to the Larmor frequency – that is resonant with the Larmor frequency. The two counter rotating ﬁelds have the same + magnitude B1. denoted B1 . To understand how this weak RF ﬁeld can rotate the magnetization we need to introduce the idea of the rotating frame. even in the presence of the very strong applied ﬁeld. The resulting ﬁeld is called the radiofrequency or RF ﬁeld. in opposite directions. 3. All we do is feed some radiofrequency (RF) power to the coil and the resulting oscillating current creates an oscillating magnetic ﬁeld along the xdirection. Fig. Remember that the main magnetic ﬁeld is supplied by a powerful superconducting magnet. As the 2 angle increases beyond this point the xcomponent grows once more. We will show that this small magnetic ﬁeld is able to rotate the magnetization away from the z axis.8 illustrates the idea. they are both aligned along the x axis and so add up to give a total ﬁeld of 2B1 along the x axis. All that is required is to (suddenly) replace the magnetic ﬁeld along the z axis with one in the x yplane (say along the x axis). and there is no way that this can be switched off. 3. At time zero. One. This frequency is also called the transmitter frequency for the reason that a radiofrequency transmitter is used to produce the power. in our minds. the magnetization will then precess about the x axis and so move towards the transverse plane. As the two vectors have the same magnitude and are rotating at the same frequency the ycomponents always cancel one another out.3–4 The vector model easy to see what we have to do. causing the xcomponent to drop back to zero and rise again to a value 2B1 along the +x axis. It turns out to be a lot easier to work out what is going on if we replace.
which is rotating in the opposite sense to the Larmor precession. or rotating frame. 3. The 1 1 graph at the bottom shows how the ﬁeld along x varies with time. B1 appears to be static .4 Pulses y + B1 B1 3–5 x y x 2B1 field along x time Fig.8 Illustration of how two counterrotating ﬁelds (shown in the upper part of the diagram and marked B + and B − ) add together to give a ﬁeld which is oscillating along the x axis (shown in the lower part). at − −ωRF ).e. We now employ a mathematical trick which is to move to a coordinate system which. This is the same direction as the rotation − of B1 . has no signiﬁcant interaction with the magnetization and so from now on we will ignore it.ωRF Fig. recall that this means the Larmor frequency is negative so that the sense of precession is from x towards −y. y x . counterrotating ﬁelds add up to the linearly oscillating one.ωRF y B1 x .3. It turns out that the other ﬁeld. The same ﬁeld viewed in a set of axes rotating at −ωRF appears to be static. rather than being static (called the laboratory frame) is rotating − about the z axis in the same direction and at the same rate as B1 (i. Suppose now that we think about a nucleus with a positive gyromagnetic ratio.9 The top row shows a ﬁeld rotating at −ωRF when viewed in a ﬁxed axis system. 3. In this rotating set of axes.
In the rotating frame this ﬁeld along the z axis appears to shrink. γ B This apparent magnetic ﬁeld in the rotating frame is usually called the reduced ﬁeld. If we choose the rotating frame to be at the Larmor frequency. given the symbol B. fram. fram. fram. fram. (3. In this discussion we will assume that the gyromagnetic ratio is positive so that the Larmor frequency is negative. First. In the rotating frame the magnetization will appear not to move i. ).3–6 The vector model and directed along the x axis of the rotating frame. This difference frequency is called the offset and is given the symbol : = ω0 − ωrot. 3.2) From this it follows that if the apparent Larmor frequency in the rotating frame is different from that in ﬁxed frame it must also be the case that the apparent magnetic ﬁeld in the rotating frame must be different from the actual applied magnetic ﬁeld. We can use Eq. . The effective ﬁeld From the discussion so far we can see that when an RF ﬁeld is being applied there are two magnetic ﬁelds in the rotating frame. = −ωRF it follows that the offset is = ω0 − (−ωRF ) = ω0 + ωRF . in such a frame the Larmor precession will appear to be at (ω0 − ωrot. fram. . there is the reduced ﬁeld. In the ﬁxed frame the precession is at ω0 . there is the RF ﬁeld (or B1 ﬁeld) of magnitude B1. the apparent Larmor frequency will be zero! It is clear that moving to a rotating frame has an effect on the apparent Larmor frequency. and under the right conditions can become small enough that the RF ﬁeld is dominant. This is the key to how the very weak RF ﬁeld can affect the magnetization in the presence of the much stronger B0 ﬁeld. The general case is when the rotating frame is at frequency ωrot. but suppose that we choose the rotating frame to be at the same frequency.9.e. the offset will be zero and so too will the reduced ﬁeld. Second. This is a very nice result as the time dependence has been removed from the problem. (3. as is shown in Fig. from the apparent Larmor frequency. we will make this ﬁeld static by choosing the rotating frame frequency to be equal to −ωRF . ) and ωrot. Larmor precession in the rotating frame We need to consider what happens to the Larmor precession of the magnetization when this is viewed in this rotating frame.1) We have used several times the relationship between the magnetic ﬁeld and the precession frequency: ω = −γ B. B.2 to compute the apparent magnetic ﬁeld. Since = (ω0 − ωrot. : = −γ hence B=− . B. . given by (− /γ ). 3.
Using these deﬁnitions in Eq.10 In the rotating frame the effective ﬁeld B eff is the vector sum of the reduced ﬁeld B and the B 1 ﬁeld. 3. For each ﬁeld the precession frequency is proportional to the magnetic ﬁeld. 3. ωeff . By making the offset small. (3.10. is deﬁned as the angle between B and B eff . The tilt angle.3.10 we can see that: B1 B cos θ = Beff Beff All three deﬁnitions are equivalent. which is exactly what we want to achieve. It is therefore convenient to think about the construction of the effective ﬁeld not in terms of magnetic ﬁelds but in terms of the precession frequencies that they cause.10.3 ωeff can be written as ωeff = 2 ω1 + 2. The angle between B and Beff is called the tilt angle and is usually given the symbol θ . z Ω θ ωeff ω1 x Fig.10 can be redrawn in terms of frequencies.4 Pulses This looks rather strange. Similarly. For example. From Fig. the constant of proportion is γ . and so the magnetization will be rotated from z down to the plane. the tilt angle can be expressed in terms of these frequencies: sin θ = ω1 ωeff cos θ = ωeff tan θ = ω1 . we have already seen that in the rotating frame the apparent Larmor precession frequency. the gyromagnetic ratio. 3. B Fig. the reduced ﬁeld (which is along z) and the RF or B1 ﬁeld (which is along x) add vectorially to give an effective ﬁeld.3) Beff x B1 The magnetization precesses about this effective ﬁeld at frequency ωeff given by ωeff = γ Beff just in the same way that the Larmor precession frequency is related to B0 . θ . 3. but recall that ω0 is negative. Beff as illustrated in Fig.11. sin θ = tan θ = B1 . . The trick is that although B0 is much larger than B1 we can affect the magnetization with B1 by making it oscillate close to the Larmor frequency. or zero. 3. In the rotating frame. 3. depends on the reduced ﬁeld: = −γ B. . the effective ﬁeld lies close to the x yplane. so if the transmitter frequency and the Larmor frequency are comparable the offset will be small. This is the phenomena of resonance. We deﬁne ω1 as the precession frequency about the B1 ﬁeld (the positive sign is intentional): ω1 = γ B1 and we already have ωeff = γ Beff . as shown in Fig. The effective ﬁeld in frequency units For practical purposes the thing that is important is the precession frequency about the effective ﬁeld. Figure 3.11 The effective ﬁeld can be thought of in terms of frequencies instead of the ﬁelds used in Fig 3. The size of this effective ﬁeld is given by: Beff = 2 B1 + 3–7 z ∆B θ B 2.
through which the magnetization has been rotated will be given by β = ω1 tp . . your ﬁngers then curl in the sense of a positive rotation. of the effective ﬁeld is π/2 or 90◦ . β is called the ﬂip angle of the pulse. this is called an inversion pulse. The effective ﬁeld lies along the x axis and so the magnetization precesses in the yz plane. For completeness we also note that the tilt angle. In general. In (b) the pulse ﬂip angle is 180◦ and so the magnetization ends up along −z. is zero and so the reduced ﬁeld. The motion of the magnetization vector during onresonance 90◦ and 180◦ pulses are shown in Fig. Under these circumstances the offset. By altering the time for which the pulse has been applied we can alter than angle through which the magnetization is rotated. is also zero. In many experiments the commonly used ﬂip angles are π/2 (90◦ ) and π (180◦ ). In this situation the motion of the magnetization vector is very simple. this is because the rotation is in the positive sense.3–8 The vector model Onresonance pulses The simplest case to deal with is where the transmitter frequency is exactly the same as the Larmor frequency – it is said that the pulse is exactly on resonance. Imagine grasping the axis about which the rotation is taking place (the x axis) with your right hand. thereby rotating in the zyplane. Just as in Fig. 3. θ . As we have seen above the precession frequency is ω1 . (a) (b) z z B1 x B1 x y y Fig. The magnetization is assumed to start out along +z.12 A “grapefruit” diagram in which the thick line shows the motion of a magnetization vector during an onresonance pulse. for a ﬂip angle β . The rotation is in the positive sense about x so the magnetization moves toward the −y axis. If the RF ﬁeld is applied for a time tp . If the pulse ﬂip angle is set to 180◦ the magnetization is taken all the way from +z to −z. 3. 3. 3.10 we see that the effective ﬁeld is therefore the same as the B1 ﬁeld and lies along the x axis. β. Referring to Fig.7 the magnetization precesses about the ﬁeld.12. The 90◦ pulse rotates the magnetization from the equilibrium position to the −y axis. B. the angle. In (a) the pulse ﬂip angle is 90◦ .
It turns out that because of the way the spectrometer works the signal that we detect appears to be that in the rotating frame. By “hard” enough we mean that the B1 ﬁeld has to be large enough that it is much greater than the size of the reduced ﬁeld. if we put the transmitter frequency at about 5 ppm. 3. this is almost always the case. rather than detecting an oscillation at the Larmor frequency. tp We know that for a 90◦ pulse β = π/2 and the duration.5 Detection in the rotating frame simple geometry tells us that the z.3.13 If the pulse ﬂip angle is β we can use simple geometry to work out the y . If this condition holds. 3. 3. .3 × 105 rad s−1 . For example. However. and so the pulse can be regarded as strong over the whole width of the spectrum.5 Detection in the rotating frame To work out what is happening during an RF pulse we need to work in the rotating frame. B. From this we can work out the value of ω1 . If the spectrometer frequency is 500 MHz the maximum offset is 5 × 500 = 2500 Hz. 2 Strictly . consider a proton spectrum covering about 10 ppm. 10 5 transmitter frequency 0 ppm Fig. this is illustrated in Fig. in the rotating frame the precession is at the offset frequency. if we make the RF ﬁeld strong enough we can effectively achieve this condition.14 Illustration of the range of offsets one might see in a typical proton spectrum.14. A typical spectrometer might have a 90◦ pulse lasting 12 µs. More details will be given in Chapter 5. each of which has a different Larmor frequency. . tp is 12 × 10−6 s. ωRF . 3. It is often relatively easy to achieve this condition. The maximum offset is 2500 Hz. Larmor precession can be viewed just as easily in the laboratory and rotating frames.13. the effective ﬁeld lies along B1 and so the situation is identical to the case of an onresonance pulse. either positive or negative. we cannot therefore be on resonance with all of the lines in the spectrum.and ycomponents are Mz = M0 cos β this is illustrated in Fig.and z components. the maximum offset of a resonance will be 5 ppm. . M y = −M0 sin β. Thinking in terms of frequencies this condition translates to ω1 being greater in magnitude than the offset.2 this is only true if we set the receiver reference frequency to be equal to the transmitter frequency. Hard pulses In practical NMR spectroscopy we usually have several resonances in the spectrum. If the transmitter frequency is placed as shown. we see an oscillation at the offset. therefore ω1 = π/2 12 × 10−6 = 1. which is 2π × 2500 = 1. 3. So. 3–9 z M0 cos β β M0 sin β y Fig. and we have seen that to get this simpliﬁcation the frequency of the rotating frame must match the transmitter frequency. We see that the RF ﬁeld is about eight times the offset.6 × 104 rad s−1 . the maximum offset is 5 ppm. We start from β β = ω1 tp hence ω1 = .
In full the signals are: M y = −M0 cos( t) Mx = M0 sin( t).15 Timing diagram or pulse sequence for the simple pulse–acquire experiment. Some simple geometry. The magne(here assumed to be positive). 3. 3. with a peak appearing at frequency . During period 3 the magnetization precesses in the transverse plane at the offset . resulting x . 3. From now on we will work exclusively in the rotating frame. enables us to deduce how the x. As was described above. The experiment comes in three periods: 1. As we commented on before. a 90◦ pulse is applied. x y 90˚ RF acq { { 1 2 Fig.3–10 The vector model It also turns out that we can detect both the x. and the line marked “acq” shows when the signal is recorded or acquired. 3. this takes us to the end of period 2. The line marked “RF” shows the location of the pulses. 2. During period 1 equilibrium magnetization builds up along the z axis.6 The basic pulse–acquire experiment At last we are in a position to describe how the simplest NMR experiment works – the one we use every day to record spectra. The timing diagram – or pulse sequence as it is usually known – is shown in Fig.e. The ycomponent is therefore proportional to cos t and the xcomponent to sin t. After the RF power is switched off we start to detect the signal which arises from the magnetization as it rotates in the transverse plane.17. . RF power is switched on for long enough to rotate the magnetization through 90◦ i. shown in Fig. 3. the 90◦ pulse rotates this magnetization onto the −y axis.and y magnetizations are shown below. The sample is allowed to come to equilibrium. this is illustrated in Fig. 3. The offset is so after time t the vector has precessed through an angle ( × t).and ycomponents of the magnetization in the rotating frame.16 Evolution during the acquisition time (period 3) of the pulse–acquire experiment. Fourier transformation of these signals will give the usual spectrum.16.and ymagnetizations vary with time. 3.15. The tization starts out along −y and evolves at the offset frequency. { 3 My time Mx Fig.
the ﬂip angle is then 180◦ . For example. It is crucial that the pulses we use in NMR experiments have the correct ﬂip angles. A typical outcome of such an experiment is shown in Fig. to obtain the maximum intensity in the pulse–acquire experiment we must use a 90◦ pulse.17 The magnetization starts out along the −y axis and rotates through an angle t during time t.13 that the ycomponent of magnetization after a pulse of ﬂip angle β is proportional to sin β. signal intensity 90º π/2 180º π flip angle 270º 3π/2 Fig.7 Pulse calibration Spectrum with several lines If the spectrum has more than one line. During the acquisition time each precesses at its own offset.1 cos 1t 3–11 x Ωt y − M0.18. 2 etc. Once the length of a 180◦ pulse is found. 3.7 Pulse calibration Fig. where M0.1 is the equilibrium magnetization of spin 1. 1 is its offset and so on for the other spins.3.6) and vary the ﬂip angle of the pulse. and if we wish to invert magnetization we must use a 180◦ pulse. 3. Usually we wish to observe all the lines at once. 3. simply halving the time gives a 90◦ pulse. the signal goes negative. we should see that the intensity of the signal varies as sin β. The reason for doing this is that the null is sharper than the maximum.. The normal practice is to increase the ﬂip angle until a null is found. so we choose the B1 ﬁeld to be strong enough that for the range of offsets that these lines cover all the associated magnetization vectors will be rotated onto the −y axis. The signal is a maximum for a ﬂip angle of 90◦ and goes through a null at 180◦ .2 cos 2t − M0. If we therefore do a pulseacquire experiment (section 3. If we imagine an onresonance or hard pulse we have already determined from Fig.3 cos 3t .. then to a good approximation we can associate a magnetization vector with each. Pulse calibration is therefore an important preliminary to any experiment. so the detected signal will be: M y = −M0.18 Illustration of how pulse calibration is achieved. . 3. Fourier transformation of the free induction signal will produce a spectrum with lines at 1 . Along the top are the spectra which would be expected for various different ﬂip angles (indicated by the dashed lines). The signal intensity varies as (sin β) as shown by the curve at the bottom of the picture. after that. 3.
usually called the RF ﬁeld strength or the B1 ﬁeld strength. .5 × 10−6 In frequency units the calculation is (ω1 /2π ) = 1 1 = = 32.5 × 10−6 In normal NMR parlance we would say “the B1 ﬁeld is 32.8 The spin echo x 180 pulse y Ωτ Ωτ π−Ωτ 0 τ/2 τ τ 3τ/2 2τ π phase. see text for details. This is mixing the units up rather strangely. t180 In this way we can determine ω1 .3 kHz”. in which case we need to divide by 2π : (ω1 /2π ) = 1 Hz. but the meaning is clear once you know what is going on! 3.3 kHz. 2t180 2 × 15.5 µs. Since the ﬂip angle is given by β = ω1 tp we can see that for a 180◦ pulse in which the ﬂip angle is π π = ω1 t180 π hence ω1 = . let us suppose that we found the null condition at 15. Also shown is a phase evolution diagram for two different offsets (the solid and the dashed line). It is usual to quote the ﬁeld strength not in rad s−1 but in Hz. thus ω1 = π t180 = π = 2. 3. φ π/2 0 τ time 2τ Fig.03 × 105 rad s−1 .3–12 The vector model Suppose that the 180◦ pulse was found to be of duration t180 .19 Vector diagrams showing how a spin echo refocuses the evolution of the offset. 15. 2t180 For example.
but all the vectors end up in mirror image positions. During the ﬁrst τ delay the phase increases linearly with time. . 3. which is a component of a very large number of more complex experiments. The 180◦ pulse is called a refocusing pulse because of the property that the evolution due to the offset during the ﬁrst delay τ is refocused during the second delay.20. The signal is acquired after the second delay τ . The durations of the pulses are in practice very much shorter than the delays τ but for clarity the length of the pulses has been exaggerated. This conclusion is independent of the value of the offset . During the second delay τ the vector continues to evolve. The effect of the 180◦ pulse is to move the vector to a mirror image position. Here we plot the phase.3. the vector is now at an angle ( τ ) to the y axis rather than being at ( τ ) to the −y axis. the vector will therefore always end up along the y axis regardless of the offset. with the mirror in question being in the x zplane.19. The arc through which the vectors are moved is different for each. which we can describe as a phase. Figure 3. φ. during this time it will rotate through a further angle of ( τ ) and therefore at the end of the second delay the vector will be aligned along the y axis. The effect of the 180◦ pulse is to change the phase from ( τ ) to (π − τ ).21. A pulse about y is said to be “phase shifted by 90◦ ” (we take an . or put another way. . As has already been mentioned. The 180◦ pulse (indicated by an open rectangle as opposed to the closed one for a 90◦ pulse) is in the centre of a delay of duration 2τ . So. The angle through which the vector rotates is simply ( t). thus separating the sequence into two equal periods. this does not have to be so. meaning that at the end of the sequence it is just as if the offset had been zero and hence there had been no evolution of the magnetization. We describe this outcome by saying that “the offset has been refocused”. the diagram illustrates this by the dashed line which represents the evolution of vector with a smaller offset. this is the jump on the diagram at time τ . when the time from the beginning of the sequence is 2τ . τ . The special thing about the spin echo sequence is that at the end of the second τ delay the magnetization ends up along the same axis. Further evolution for time τ causes the phase to increase by ( τ ) leading to a ﬁnal phase at the end of the second τ delay of π . 3.20 Pulse sequence for the spin echo experiment. regardless of the values of τ and the offset. and is shown in Fig. Another way of thinking about the spin echo is to plot a phase evolution diagram. A few moments thought will reveal that as the angle through which the vector rotates during the ﬁrst τ delay must be equal to that through which it rotates during the second τ delay. the spin echo. 3. φ. the effect of the 180◦ pulse is to reﬂect the vectors in the x zplane. It is interesting to note that the spin echo sequence gives exactly the same result as the sequence 90◦ – 180◦ with the delays omitted. and we can just as easily apply it along the y axis. The way this works is illustrated in Fig. During the ﬁrst delay τ the vector precesses from −y towards the x axis. The pulse sequence is quite simple.9 Pulses of different phases 3–13 90˚ RF τ 180˚ τ 2τ acq Fig.9 Pulses of different phases We are now able to analyse the most famous pulsed NMR experiment. 3. as a function of time.19 illustrates how the sequence works after the initial 90◦ pulse has placed the magnetization along the −y axis. 3. this is done at the bottom of Fig. So far we have assume that the B1 ﬁeld is applied along the x axis. for example.
As a result. as is also shown in Fig. the vectors end up in mirror image positions with respect to the xz plane. 3. As before.22 Grapefruit plots showing the effect on equilibrium magnetization of (a) a 90◦ pulse about the y axis and (b) a 90◦ pulse about the −x axis. xpulse to have a phase shift of zero). All three are rotated by 180◦ about the x axis on the trajectories indicated by the thick lines which dip into the southern hemisphere. In a similar way. We have seen that a 180◦ pulse about the x axis causes the vectors to move to mirror image positions with respect to the x zplane.3–14 z The vector model y x y Fig. . If we apply a 90◦ pulse about the y axis to equilibrium magnetization we ﬁnd that the vector rotates in the yzplane such that the magnetization ends up along x. grey and light grey). this is illustrated in Fig.22. 3.21 Illustration of the effect of a 180◦ pulse on three vectors which start out at different angles from the −y axis (coloured in black.22. we can determine the effect of such a pulse by thinking of it as causing a positive rotation about the y axis. 3. likewise a pulse about −x would be said to be phase shifted by 180◦. 3. A 90◦ pulse about −x causes the magnetization to appear along y. Note the position of the B 1 ﬁeld in each case. On modern spectrometers it is possible to produce pulses with arbitrary phase shifts. (a) (b) z z B1 B1 x y x y Fig.
This loss of x. The size and sign of the z magnetization is reﬂected in the spectra (shown underneath). but one whose size is decreasing.3. So.10 Relaxation a 180◦ pulse about the y axis causes the vectors to be reﬂected in the yzplane. what we see for short values of τ is a negative line. Once perturbed. by recording spectra with different values of the delay τ we can map out the recovery of the zmagnetization from the intensity of the observed lines. So. which results from the magnetization precessing in the x y plane will therefore decay away in amplitude. the zmagnetization will try to return to its equilibrium position. The . 3.24 Visualization of the outcome of an inversion recovery experiment. Suppose that the delay τ is very short so that at the end of this delay the magnetization has not changed. and this process is called longitudinal relaxation. Now the 90◦ pulse will rotate the magnetization onto the +y axis. If this gives a positive line in the spectrum. As τ gets longer more relaxation takes place and the magnetization shrinks towards zero.23 The pulse sequence for the inversion recovery experiment used to measure longitudinal relaxation. 180º RF acq τ 90º Fig.and ymagnetization is called transverse relaxation. The 180◦ pulse rotates the equilibrium magnetization to the −z axis. 3.10 Relaxation 3–15 We will have a lot more to say about relaxation later on. but at this point we will just note that the magnetization has a tendency to return to its equilibrium position (and size) – a process known as relaxation. 3. 3. Thus. Recall that the equilibrium situation has magnetization of size M0 along z and no transverse (x or y) magnetization. then having the magnetization along +y will give a negative line. Eventually the magnetization goes through zero and then starts to increase along +z – this gives a positive line in the spectrum. We can measure the rate of this process using the inversion recovery experiment whose pulse sequence is shown in Fig.23. increasing τ Fig. this result is a negative line in the spectrum. The free induction signal. if we have created some transverse magnetization (for example by applying a 90◦ pulse) over time relaxation will cause this magnetization to decay away to zero. By analysing the peak heights as a function of the delay τ it is possible to ﬁnd the rate of recovery of the z magnetization. note that this is in contrast to the case where the magnetization starts out along +z and it is rotated onto −y.
as expected for a 90◦ pulse.5 times the RF ﬁeld strength.24. The consequences of this are sometimes a problem to us. there is a signiﬁcant component of magnetization generated along the x direction. As a consequence.25. . 3. the effective ﬁeld lies along x and is indicated by the dashed line A. this is illustrated in Fig.3–16 whole process is visualized in Fig. We now turn to the case where the offset is comparable to the RF ﬁeld strength. The plot of the xmagnetization. as the offset increases the amount of ymagnetization generally decreases but imposed on this overall decrease there is an oscillation.11 Offresonance effects and soft pulses z D C B d c a y b The vector model A x Fig. 3. (b). on resonance the equilibrium magnetization ends up entirely along −y. The further off resonance we go. rather than the magnetization moving in the yz plane from z to −y. these are shown in Fig. Also.26. The effective ﬁeld directions are labelled C and D. 3. respectively. at some offsets the magnetization is zero and at others it is positive. the magnetization starts to follow a more complex curved path. the effective ﬁeld is marked B. but again with a strong oscillation.25 Grapefruit diagram showing the path followed during a pulse for various different resonance offsets. which is given by Mabs = 2 2 Mx + M y . Plot (c) is of the magnitude of the magnetization. the further the vector ends up from the x y plane. So far we have only dealt with the case where the pulse is either on resonance or where the RF ﬁeld strength is large compared to the offset (a hard pulse) which is in effect the same situation. In (a) we see the ymagnetization and. something which does not occur in the onresonance case. 3. shows a similar story with the magnetization generally falling off as the offset increases. We can see more clearly what is going on if we plot the x and y magnetization as a function of the offset. but they can also be turned to our advantage for selective excitation. Path b is for the case where the offset is half the RF ﬁeld strength. Paths c and d are for offsets equal to and 1. However. 3. Path a is for the onresonance case. the effective ﬁeld begins to move up from the x axis towards the z axis. As the offset becomes comparable to the RF ﬁeld.
always positive. the equilibrium magnetization has been assumed to be of size 1. translates to 20 kHz. if we accept a reduction to 90% of the full intensity at the edges of the spectrum we would need an RF ﬁeld strength of 20/1.5 20 Mx 0. the maximum offset that a peak can have is 100 ppm which.5 20 10 0 10 20 Ω / ω1 Fig.3. the region over which it does so efﬁciently is really rather small. at this Larmor frequency. Excitation of a range of shifts There are some immediate practical consequences of this observation. We can achieve this by putting the transmitter in the centre of the region we wish to excite and then reducing the RF ﬁeld .6 times the RF ﬁeld strength. What plot (c) tells us is that although a pulse can excite magnetization over a wide range of offsets.and ymagnetization leads to phase errors in the spectrum which can be difﬁcult to correct.26 Plots of the magnetization produced by a pulse as a function of the offset. We see from this plot the characteristic nulls and subsidiary maxima as the offset increases. 3. Suppose that we are trying to record the full range of carbon13 shifts (200 ppm) on an spectrometer whose magnetic ﬁeld gives a proton Larmor frequency of 800 MHz and hence a carbon13 Larmor frequency of 200 MHz. of course. If the manufacturer failed to provide sufﬁcient power to produce this pulse width we can see that the excitation of the spectrum will fall below our (arbitrary) 90% mark. If we want at least 90% of the full intensity the offset must be less than about 1.5 10 Ω / ω1 20 Ω / ω1 1 1 (c) 1 Mabs 0. This gives the total transverse magnetization in any direction.5 kHz. The pulse length has been adjusted so that on resonance the ﬂip angle is 90◦ . According to our criterion above.5 20 20 10 0. for example just the lines of a single multiplet. ω1 . This would correspond to a 90◦ pulse width of 20 µs.5 10 10 0. it is.6 ≈ 12. If we place the transmitter frequency at 100 ppm. We will see in a later section that the presence of a mixture of x. The horizontal axes of the plots is the offset expressed as a ratio of the RF ﬁeld strength.11 Offresonance effects and soft pulses 1 1 3–17 (b) (a) My 0. Selective excitation Sometimes we want to excite just a portion of the spectrum.
Much effort has been put into getting round both of these problems. 3. or unlucky and have an unwanted peak fall on a maximum. The simplest of these are basically bellshaped (like a gaussian function. Figure 3. The key feature of all of the approaches is to “shape” the envelope of the RF pulses i. however. To attack this part of the problem requires an altogether more sophisticated approach. 3. In the end a balance has to be made between making the pulse too long (and hence losing signal due to relaxation) and allowing a small amount of excitation of the unwanted signals. they do not. 3. Such pulses are called shaped pulses. The closer in the unwanted peaks are the weaker the RF ﬁeld must become and hence the longer the 90◦ pulse.27 does not portray one problem with this approach. We might be lucky and have an unwanted peak fall on a null. improve the phase properties. more of which in a later section. This is described as a phase error. . these are sometimes called “wiggles”. The whole process is visualized in Fig.3–18 The vector model strong pulse excitation selective pulse offset transmitter Fig. the peaks at nonzero offsets are attenuated and the spectrum which is excited will be as shown at the bottom. Of course.and ymagnetization is generated (as shown in Fig. for example).27. not just switch it on and off abruptly. reducing the RF ﬁeld strength lengthens the duration of a 90◦ pulse. If the transmitter is placed on resonance with one line and the strength of the RF ﬁeld reduced then the pattern of excitation we expect is as shown in the middle. The value to which we need to reduce the RF ﬁeld depends on the separation of the peak we want to excite from those we do not want to excite.27 Visualization of the use of selective excitation to excite just one line in the spectrum. As a result.e. These suppress the “wiggles” at large offsets and give just a smooth decay. which is that for peaks away from the transmitter a mixture of x. Such pulses which are designed to affect only part of the spectrum are called selective pulses or soft pulses (as opposed to nonselective or hard pulses). At the top is shown the spectrum that would be excited using a hard pulse. The second problem that the ﬁgure does show is that the excitation only falls off rather slowly and “bounces” through a series of maximum and nulls. strength until the degree of excitation of the rest of the spectrum is essentially negligible.26). but with a smooth variation.
28 shows the zmagnetization generated as a function of offset for a pulse. . The good consequence is that a selective 180◦ pulse is.3. This observation has two consequences: one “good” and one “bad”. Thus.11 Offresonance effects and soft pulses Selective inversion Sometimes we want to invert the magnetization associated with just one resonance while leaving all the others in the spectrum unaffected. that the 180◦ pulse is longer. Plot (b) covers a narrower range of offsets than plot (a). We see that the range over which there is signiﬁcant inversion is rather small. In particular. such a pulse would be called a selective inversion pulse.26. 1 1 3–19 (a) 0.28 with Figure 3. Figure 3.5 Mz Ω / ω1 0. These plots should be compared with those in Fig. 3. more selective than a corresponding 90◦ pulse. the weaker “bouncing” sidelobes are a useful feature. and that the oscillations are smaller in amplitude than for the excitation pulse.26.28 Plots of the z magnetization produced by a pulse as a function of the offset. 3. Just as for selective excitation. so some of the improvement may be illusory! The bad consequence is that when it comes to hard pulses the range of offsets over which there is anything like complete inversion is much more limited than the range of offsets over which there is signiﬁcant excitation. all we need to do is to place the transmitter on the line we wish to invert and reduce the RF ﬁeld until the other resonances in the spectrum are not affected signiﬁcantly. the ﬂip angle onresonance has been set to 180◦ . This can be seen clearly by comparing Fig. 180◦ pulses are often the source of problems in spectra with large offset ranges.5 5 1 1 Fig. Of course we need to adjust the pulse duration so that the onresonance ﬂip angle is 180◦ .5 20 10 10 Mz Ω / ω1 20 5 0.5 (b) 0. Do not forget. for a given ﬁeld strength. though. 3.
Suppose that the transmitter is placed at 5 ppm.08 × 108 rad s−1 T−1 . it follows that these nulls in the excitation . as the pulse was lengthened further the intensity decreased going through a null at 20.26 we see that there are some offsets at which the transverse magnetization goes to zero. to what do you attribute this? E 3–3 Use vector diagrams to describe what happens during a spin echo sequence in which the 180◦ pulse is applied about the y axis. of a spin with this offset. what is the offset (in Hz) of a peak at 10 ppm? Compute the tilt angle. Explain what is happening in this experiment and use the data to determine the RF ﬁeld strength in Hz and the length of a 90◦ pulse.5 µs and then turning negative.3–20 3. Does this have a signiﬁcant effect of the carbon13 nuclei? Explain your answer carefully. E 3–6 Referring to the plots of Fig. ωeff . Also. Estimate the minimum 90◦ pulse width you would need to excite peaks in this complete range to within 90% of the their theoretical maximum for a spectrometer with a B0 ﬁeld strength of 9. is this 25 kHz ﬁeld strong enough to give what could be classed as hard pulses? E 3–2 In an experiment to determine the pulse length an operator observed a positive signal for pulse widths of 5 and 10 µs. θ . For the normal range of proton shifts (0 – 10 ppm). Suppose that a 90◦ pulse of length 10 µs is applied to the protons.4 T. ω1 /(2π ) has been determined to be 25 kHz. covering some 700 ppm.12 Exercises The vector model E 3–1 A spectrometer operates with a Larmor frequency of 600 MHz for protons. For a particular set up the RF ﬁeld strength. draw a phase evolution diagram appropriate for this pulse sequence. This nucleus shows a wide range of shifts. 3. In what way is the outcome different from the case where the refocusing pulse is applied about the x axis? What would the effect of applying the refocusing pulse about the −x axis be? E 3–4 The gyromagnetic ratio of phosphorus31 is 1. A further null in the signal was seen at 41. E 3–5 A spectrometer operates at a Larmor frequency of 400 MHz for protons and hence 100 MHz for carbon13.0 µs. Recall that the magnetization is rotating about the effective ﬁeld.
E 3–9 Consider the spin echo sequence to which a 90◦ pulse has been added at the end: 90◦ (x) − delay τ − 180◦ (x) − delay τ − 90◦ (φ).3. is give from ω1 τp = π/2 thus τp = π . explain how the x. 3. Does this agree with Fig. E 3–7 When calibrating a pulse by looking for the null produced by a 180◦ rotation. when ( /ω1 ) = 15. one with a small offset)? E 3–8 Use vector diagrams to predict the outcome of the sequence: 90◦ − delay τ − 90◦ applied to equilibrium magnetization.e. Show that with this values of the effective ﬁeld is given by: ωeff = ω1 1 + κ 2 . 2ω1 The angle of rotation about the effective ﬁeld for a pulse of duration τp is (ωeff τp ). a complete rota√ √ tion.e. Make a sketch of a “grapefruit” showing this. βeff = 2 The null in the excitation will occur when βeff is 2π i. . (The reason for doing this is to reduce the number of variables. In your answer.12 Exercises come about when the magnetization executes complete 360◦ rotations about the effective ﬁeld. y and z magnetizations depend on the delay τ and the offset .) Let us assume that onresonance the pulse ﬂip angle is π/2. τp . so the duration of the pulse. 3–21 Suppose that we express the offset as a multiple κ of the RF ﬁeld strength: = κω1 . The effective ﬁeld is given by ωeff = 2 ω1 + 2. why is it important to choose a line which is close to the transmitter frequency (i.26? Predict other values of κ at which there will be nulls in the excitation. βeff is given by π 1 + κ 2. both pulses are about the x axis. Show that for the effective ﬁeld given above this angle.e. Show that this occurs when κ = 15 i. In such a rotation the magnetization is returned to the z axis.
3–22 The vector model The axis about which the pulse is applied is given in brackets after the ﬂip angle. How does one go about choosing the value for τ ? E 3–11 The socalled “1–1” sequence is: 90◦ (x) − delay τ − 90◦ (y). τ is chosen so that the peaks of interest are excited. if the delay is such that τ = π no transverse magnetization is excited. Describe the excitation that this sequence produces as a function of offset. −x or −y. The sequence has been used for suppressing strong solvent signals which might otherwise overwhelm the spectrum. y. for a peak with an offset such that τ = π/2 the sequence results in all of the equilibrium magnetization appearing along the x axis. Explain in what way the outcome is different depending on whether the phase φ of the pulse is chosen to be x. However. Further. Explain these observations and make a sketch graph of the amount of transverse magnetization generated as a function of the offset for a ﬁxed delay τ. and so is not excited. The solvent is placed on resonance. How it could be used for observing spectra in the presence of strong solvent signals? . E 3–10 The socalled “1–1” sequence is: 90◦ (x) − delay τ − 90◦ (−x) For a peak which is on resonance the sequence does not excite any observable magnetization.
we will see later that this choice of starting position is essentially arbitrary. 4. the signal is therefore often called the free induction decay or FID. 4. For the purposes of this chapter we are going to assume that the magnetization starts out along x. y x frequency Fig. This process takes the time domain function (the FID) and converts it into a frequency domain function (the spectrum).2 Evolution of the magnetization over time. 4. into the a spectrum. such as phase and lineshapes. Mx time My Fig.1 Fourier transformation is the mathematical process which takes us from a function of time (the time domain) – such as a FID – to a function of frequency – the spectrum.6 we saw that the x and y components of the free induction signal could be computed by thinking about the evolution of the magnetization during the acquisition time. this is shown in Fig. Chapter 4 “Fourier transformation and data processing” c James Keeler. the offset is assumed to be positive and the magnetization starts out along the x axis. The question is how do we turn this signal. in which the horizontal axis is frequency. In that discussion we assumed that the magnetization started out along the −y axis as this is where it would be rotated to by a 90◦ pulse.1 The FID time Fourier transformation In section 3. In this chapter we will start out by exploring some features of the spectrum.1. 2002 . 4. which depends on time. This conversion is made using a mathematical process known as Fourier transformation. which are closely associated with the Fourier transform and then go on to explore some useful manipulations of NMR data such as sensitivity and resolution enhancement.4 Fourier transformation and data processing In the previous chapter we have seen how the precessing magnetization can be detected to give a signal which oscillates at the Larmor frequency – the free induction signal. We also commented that this signal will eventually decay away due to the action of relaxation.
4.4 Illustration of a typical FID.2. both decay over time. 4. part Sx(t) or real part time Fig.1) Fig. as is illustrated in Fig. We mentioned at the start of this section that the transverse magnetization decays over time. Another way of writing this is to deﬁne a (ﬁrst order) rate constant R2 = 1/T2 and so S(t) becomes S(t) = S0 exp(i t) exp(−R2 t). From this we can easily see that the x and y components of the magnetization are: Sx Sy S0 Ωt Mx = M0 cos t M y = M0 sin t.4–2 Fourier transformation and data processing If the magnetization does indeed start along x then Fig. (4. 4. Sy(t) or imag. (4. but just note that the timedomain signal is complex. 4.3. 3.3 The x and y components of the signal can be thought of as arising from the rotation of a vector S 0 at frequency . The signal that we detect is proportional to these magnetizations. It is convenient to think of this signal as arising from a vector of length S0 rotating at frequency . with the real and imaginary parts corresponding to the x and y components of the signal. and this is most simply represented by an exponential decay with a time constant T2 . 4. 4. showing the real and imaginary parts of the signal.16 needs to be redrawn. 4.2) The shorter the time T2 (or the larger the rate constant R2 ) the more rapidly the signal decays.1 gives the frequency domain signal which we know as the spectrum. The signal then becomes S(t) = S0 exp(i t) exp −t T2 . A typical example is illustrated in Fig. we will simply write the detected x and y signals. the x and y components of the vector give Sx and S y . Sx (t) and S y (t) as Sx (t) = S0 cos t and S y (t) = S0 sin t where S0 gives is the overall size of the signal and we have reminded ourselves that the signal is a function of time by writing it as Sx (t) etc. The constant of proportion depends on all sorts of instrumental factors which need not concern us here. Like the time domain signal the frequency domain signal has a real and an imaginary part. The real . as is shown in Fig. it is also convenient to regard Sx (t) and S y (t) as the real and imaginary parts of a complex signal S(t): S(t) = Sx (t) + i S y (t) = S0 cos t + i S0 sin t = S0 exp(i t). 4. We need not concern ourselves too much with the mathematical details here. As a consequence of the way the Fourier transform works.2 Fourier transformation Fourier transformation of a signal such as that given in Eq.
they are not in fact related to one another. these points are illustrated in Fig. but it also has positive and negative parts. Whereas the absorption lineshape is always positive. This offset can be positive or negative and. In a complex spectrum these might cancel one another out. Positive and negative frequencies As we discussed in section 3. The imaginary part of the spectrum gives a lineshape known as the dispersion mode Lorentzian.6 Illustration of the fact that the more rapidly the FID decays the broader the line in the corresponding spectrum.2 Fourier transformation part of the spectrum shows what we call an absorption mode line. in fact in the case of the exponentially decaying signal of Eq.4. Not only is it broader than the absorption mode. . Although these two lineshapes do look roughly the same. So. the evolution we observe is at frequency i. This means that the faster the decay of the FID the broader the line becomes. A series of FIDs are shown at the top of the ﬁgure and below are the corresponding spectra. as we will see later. it turns out to be possible to determine the sign of the frequency. it also extends further. the dispersion lineshape has positive and negative parts. If you are familiar with ESR spectra you might recognize the dispersion mode lineshape as looking like the derivative lineshape which is traditionally used to plot ESR spectra. The dispersion line shape is not one that we would choose to use. for example by increasing S0 the height of the peak increases in direct proportion. 4. the area under the line – that is the integral – remains constant so as it gets broader so the peak height reduces.5. 4.5 Illustration of the absorption and dispersion mode Lorentzian lineshapes.1 the line has a shape known as a Lorentzian. 4. 4. so as they get broader the peak height decreases. If the size of the time domain signal increases. 4.6. all plotted on the same vertical scale. the apparent Larmor frequency in the rotating frame. in our spectrum we have positive and negative frequencies. These observations lead to the very important consequence that by integrating the lines in the spectrum we can determine the relative number of protons (typically) which contribute to each.e. leading to a great deal of confusion. The integral of the peaks remains constant. absorption dispersion frequency Ω Fig. Both lineshapes are illustrated in Fig.5. However. or to be precise the absorption mode Lorentzian. and it is usual to plot these with zero in the middle. This absorption lineshape has a width at half of its maximum height of 1/(π T2 ) Hz or (R/π ) Hz. 4–3 time frequency Fig.
φ. that is on the phase of the signal at time zero. In (b) we see the effect of a phase shift.3 exp(i 3 t) exp −t T2(3) . inclusion of this phase shift means that the (complex) signal becomes: S(t) = S0 exp(i φ) exp(i t) exp −t T2 . and relaxation time constant. (i) i . as we saw in section 3.2 exp(i 2 t) exp −t T2(2) + S0.7 (a) we see the situation we had before.1 exp(i 1 t) exp −t T2 (1) + S0. The situation is portrayed in Fig. just as we require it.i . the FID will be the sum of contributions from each line. rather than at zero. where we have allowed each line to have a separate intensity. frequency. of 45◦ .4–4 Fourier transformation and data processing Several lines What happens if we have more than one line in the spectrum? In this case. Now it is S y which takes the form of a damped cosine wave. The Fourier transform gives a spectrum in which the absorption mode signal now appears in the imaginary part. 4. both are a mixture of absorption and dispersion. What we see is that in general the appearance of the spectrum depends on the position of the signal at time zero. For example. 4.3) . if there are three lines S(t) will be: S(t) =S0. Fourier transformation gives a spectrum in which the real part contains the absorption mode lineshape and the imaginary part the dispersion mode. in general this need not be the case – it might just as well be the other way round or anywhere in between. S y now starts out at a ﬁnite value. (4. the Fourier transformation of S(t) will be the sum of these lines – which is the complete spectrum. whereas Sx is a sine wave. S0. as Fourier transformation of each of the terms in S(t) gives a line of appropriate width and frequency. In (c) the phase shift is 90◦ . at the start of the FID) Sx (t) is a maximum and S y (t) is zero. The real part of the FID (corresponding to Sx ) is a damped cosine wave and the imaginary part (corresponding to S y ) is a damped sine wave. So.5. In Fig. T2 .7. with the signal starting out along x and precessing towards y. We describe this general situation be saying that the signal is phase shifted or that it has a phase error. As a result neither the real nor the imaginary part of the spectrum has the absorption mode lineshape. Finally in (d) the phase shift is 180◦ and this gives a negative absorption mode signal in the real part of the spectrum.e. 4.3 Phase So far we have assumed that at time zero (i. However. The Fourier transform is a linear process which means that if the time domain is a sum of functions the frequency domain will be a sum of Fourier transforms of those functions. Mathematically.
so the . and multiply it by exp(i φcorr ): exp(i φcorr )S(t) = exp(i φcorr ) × S0 exp(i φ) exp(i t) exp −t T2 . of 45◦ .7 Illustration of the effect of a phase shift of the time domain signal on the spectrum. 4.4. now the absorption mode appears in the imaginary part of the spectrum. This is undesirable as for the best resolution we require an absorption mode lineshape. this leads to a situation in which the real part of the spectrum (which is normally the part we display) does not show a pure absorption lineshape. Suppose with take the FID. the real and imaginary parts of the spectrum are now mixtures of absorption and dispersion. In (b) there is a phase shift. Luckily. This is easy to do as by now the FID is stored in computer memory.3 Phase Sy Sy φ 4–5 (a) Sx (b) Sx Sx Sy Sx Sy real imag real imag Sy φ (c) Sy φ Sx (d) Sx Sx Sy Sx Sy real imag real imag Fig. Finally in (d) the phase shift is 180◦ giving a negative absorption line in the real part of the spectrum. Phase correction It turns out that for instrumental reasons the axis along which the signal appears cannot be predicted. In (a) the signal starts out along x and so the spectrum is the absorption mode in the real part and the dispersion mode in the imaginary part. so in any practical situation there is an unknown phase shift. φ. The vector diagrams illustrate the position of the signal at time zero. restoring the spectrum to the absorption mode is easy.3. 4. In general. In (c) the phase shift is 90◦ . represented by Eq.
if the spectrum is represented by S(ω) (a function of frequency. If there were only one line in the spectrum it would be possible to ensure that the line appeared in the absorption mode simply by adjusting the phase Attempts have been made over the years to automate this phasing process. regardless of their offset. So. We can now describe this mixture of x and y magnetization as resulting in a phase shift or phase error of the spectrum. where the magnetization appears only along −y. as exp(0) = 1 the time domain signal becomes: −t . Figure 3. Frequency dependent phase errors We saw in section 3.4–6 Fourier transformation and data processing multiplication is just a mathematical operation on some numbers. In addition to the phase shifts introduced by the spectrometer we can of course deliberately introduce a shift of phase by. We then adjust the phase correction (i. In practice what happens is that we Fourier transform the FID and display the real part of the spectrum. resulting in a phase error which also increases with offset. This is illustrated schematically in the upper spectrum shown in Fig. exp(i φcorr )S(t) = S0 exp(i t) exp T2 The signal now has no phase shift and so will give us a spectrum in which the real part shows the absorption mode lineshape – which is exactly what we want. but these automatic algorithms tend to have more trouble with poorlyresolved spectra. Exponentials have the property that exp(A) exp(B) = exp(A + B) so we can rewrite the time domain signal as exp(i φcorr )S(t) = exp(i (φcorr + φ)) S0 exp(i t) exp −t T2 . This is in contrast to the case of a hard pulse. It turns out that the phase correction can just as easily be applied to the spectrum as it can to the FID.25 illustrates very clearly how the x component increases as the offset increases. In a sense it does not matter what the phase of the signal is – we can always obtain an absorption spectrum by phase correcting the spectrum later on. In any case. Therefore lines at different offsets in the spectrum will have different phase errors. Such a correction is called a frequency independent or zero order phase correction as it is the same for all peaks in the spectrum. ω) the phase correction is applied by computing exp(i φcorr )S(ω).8. altering the phase of a pulse. for example. what constitutes a correctly phased spectrum is rather subjective. 4. . the error increasing as the offset increases. The whole process is called phasing the spectrum and is something we have to do each time we record a spectrum. All we need to do is ﬁnd the correct φcorr . Now suppose that we set φcorr = −φ. the value of φcorr ) until the spectrum appears to be in the absorption mode – usually this adjustment is made by turning a knob or by a “click and drag” operation with the mouse.11 that if the offset becomes comparable with the RF ﬁeld strength a 90◦ pulse about x results in the generation of magnetization along both the x and y axes. on well resolved spectra the results are usually good.e.
Of course. and so it will be impossible to phase all of the lines at once. All we have to do is to vary the rate of change of phase with frequency (the slope of the line) until the spectrum appears to be phased. where tp is the length of the pulse. it is often the case that the phase correction needed is directly proportional to the offset – called a linear or ﬁrst order phase correction. In (a) the line which is on resonance (at zero frequency) is in pure absorption.8 (b). Such an frequency dependent phase error would result from the use of a pulse whose RF ﬁeld strength was not much larger than the range of offsets.3 Phase (a) 4–7 0 (b) phase frequency (c) 0 frequency Fig. in the way described above. In practice. if there is more than one line present in the spectrum the phase correction for each will be different. the maximum offset is about 100 ppm which translates to 12500 Hz. but as the offset increases the phase error increases. Luckily. by applying a phase correction which varies with the offset in a linear manner. However. The spectrum can be returned to the absorption mode. For reasons which we cannot go into here it turns out that the linear phase . For a carbon13 spectrum recorded at a Larmor frequency of 125 MHz. Such a variation in phase with offset is shown in Fig. For a pulse the phase error due to these offresonance effects for a peak with offset is of the order of ( tp ). Let us suppose that the 90◦ pulse width is 15 µs. then the phase error is 2π × 12500 × 15 × 10−6 ≈ 1. 4. as with the zeroorder phase correction the computer software usually makes it easy for us to do this by turning a knob or pushing the mouse.2 radians which is about 68◦ . The usual convention is to express the frequency dependent phase correction as the value that the phase takes at the extreme edges of the spectrum.4. note that in the calculation we had to convert the offset from Hz to rad s−1 by multiplying by 2π . this is a signiﬁcant effect.8 Illustration of the appearance of a frequency dependent phase error in the spectrum. such a correction by 100◦ means that the phase correction is zero in the middle (at zero offset) and rises linearly to +100◦ at on edge and falls linearly to −100◦ at the opposite edge. as shown in (b). So. to obtain a correctly phased spectrum we have to choose the correct slope of the graph of phase against offset. we expect the frequency dependent phase error to vary from zero in the middle of the spectrum (where the offset is zero) to 68◦ at the edges. (c).and ﬁrstorder phase corrections. for example. 4. So. to phase the spectrum correctly usually requires some iteration of the zero.
but for broad lines it is not so good. the FID decays over time but in contrast the noise just goes on and on. that the early parts of the FID are . but the major contributor is the thermal noise from the coil used to detect the signal. if we carry on recording data for long after the FID has decayed we will just measure noise and no signal. As time progresses.4 Sensitivity enhancement Inevitably when we record a FID we also record noise at the same time. Provided that the lines in the spectrum are sharp a linear correction works very well. In (a) the FID has decayed to next to nothing within the ﬁrst quarter of the time. The resulting spectrum will therefore have a poor signaltonoise ratio. Attempting to use a ﬁrstorder phase correction on such spectra often results in distortions of the baseline. but the noise carries on unabated for the whole time. Therefore. the SNR improves signiﬁcantly. NMR is not a sensitive technique. 4. Of course. By its very nature. Just shortening the time spent recording the signal (called the acquisition time) will improve the SNR since more or less all the signal is contained in the early part of the FID.4–8 Fourier transformation and data processing correction is sometimes only a ﬁrst approximation to the actual correction needed.9 Illustration of the effect of the time spent acquiring the FID on the signaltonoise ratio (SNR) in the spectrum. Sensitivity enhancement Looking at the FID we can see that at the start the signal is strongest. Some of the noise is contributed by the ampliﬁers and other electronics in the spectrometer. therefore. 4. we must not shorten the acquisition time too much or we will start to miss the FID. This point is illustrated in Fig. 4. (a) (b) (c) Fig. so we need to take any steps we can to improve the signaltonoise ratio in the spectrum.9 where we see that by recording the FID for long after it has decayed all we end up doing is recording more noise and no signal. the signal decays and so gets weaker but the noise remains at the same level. Reducing the noise contributed by these two sources is largely a technical matter which will not concern use here. Shown in(b) is the effect of halving the time spent acquiring the data. We will see that there are some manipulations we can perform on the FID which will give us some improvement in the signaltonoise ratio (SNR). The idea arises. which would result in a reduction in SNR. In (c) we see that taking the ﬁrst quarter of the data gives a further improvement in the SNR.
The original FID is shown in (a) and the corresponding spectrum is (f). Using a more rapidly decaying weighting function. 4. A typical choice for this function – called a weighting function – is an exponential: (4.4. The improvement in SNR is less marked. causing it to decay more quickly. Multiplying (a) by the more rapidly decaying weighting function (d) gives (e). Multiplying the FID by a weighting function (b) gives (c). As was explained in section 4. the use .10 Illustration of how multiplying a FID by a decaying function (a weighting function) can improve the SNR. The idea is that this function will cut off the later parts of the FID where the signal is weakest. Fourier transformation of (c) gives the spectrum (g).10 illustrates the effect on the SNR of different choices of this decay constant.4 Sensitivity enhancement “more important” as it is here where the signal is the strongest. This effect can be exploited by deliberately multiplying the FID by a function which starts at 1 and then steadily tails away to zero. It should not be forgotten that the weighting function also acts on the signal. the corresponding spectrum is (h). this shows most clearly the improvement in the SNR. (d). The same spectra are plotted in (i) – (k) but this time normalized so that the peak height is the same. So. gives a further small improvement in the SNR (h). Note the improvement in SNR of (g) compared to (f). Spectrum (g) shows a large improvement in the SNR when compared to spectrum (f) simply because the long tail of noise is suppressed. Figure 4. but leave the early parts unaffected.2 a more rapidly decaying signal leads to a broader line.4) W (t) = exp(−RLB t) where RLB is a rate constant which we are free to choose. Spectra (f) – (h) are all plotted on the same vertical scale so that the decrease in peak height can be seen. 4–9 (a) (b) (c) original FID weighting function weighting function (d) (e) weighted FID weighted FID (f) (g) (h) (i) (j) (k) Fig.
For example if 5 Hz of linebroadening is required then (RLB /π ) = 5 giving RLB = 15.4–10 Fourier transformation and data processing of the weighting function will not only attenuate the noise it will also broaden the lines. So. applying an additional line broadening of 2 Hz will give the optimum SNR. and this will reduce the SNR. a “linebroadening of 1 Hz” is a function which will increase the linewidth in the spectrum by 1 Hz. . called the matched ﬁlter. if the linewidth is 2 Hz in the original spectrum. It can be shown that the best SNR is obtained by applying a weighting function which matches the linewidth in the original spectrum – such a weighting function is called a matched ﬁlter.7 s−1 . However. Matched ﬁlter Suppose that the FID can be represented by the exponentially decaying function introduced in Eq. this reduction in peak height will reduce the SNR. 4. the linewidth in the spectrum is the sum of the linewidths in the original spectrum and the additional linebroadening imposed by the weighting function.2: S(t) = S0 exp(i t) exp(−R2 t). for example. 4.e. Hence there is a tradeoff we have to make: the more rapidly decaying the weighting function the more the noise in the tail of the FID is attenuated thus reducing the noise level in the spectrum. Broadening the lines also reduces the peak height (remember that the integral remains constant) – something that is again evident from Fig. 4. Under these circumstances we may choose to use sufﬁcient line broadening to cut off the excess noise in the tail of the FID. So. In words. but still less than the matched ﬁlter.10. Also. The line in the corresponding spectrum is of width (R2 /π ) Hz. Combining the two terms describing the exponential decays gives W (t) × S(t) = S0 exp(i t) exp (−(RLB + R2 )t) .4 i. It turns out that there is an optimum weighting function. It is usual to specify the weighting function in terms of the extra line broadening it will cause. the extra line broadening caused by the matched ﬁlter may not be acceptable on the grounds of the decrease in resolution it causes. Let us apply a weighting function of the kind described by Eq. a decaying exponential: W (t) × S(t) = exp(−RLB t) S0 exp(i t) exp(−R2 t) . We can see easily from this argument that if there is a range of linewidths in the spectrum we cannot ﬁnd a value of the linebroadening which is the optimum for all the peaks. From this we see that the weighted FID will give a linewidth of (RLB + R2 )/π. 4. this is clear from Fig. The decay due to the weighting function on its own would give a linewidth of (RLB /π ) Hz.10. Clearly. at the same time a more rapidly decaying function will cause greater line broadening.
This function starts at 1 when t = 0 and then rises indeﬁnitely. Note that the scales of the plots have been altered to make the relevant features clear. To get round this it is. for example a rising exponential: W (t) = exp(+RRE t) RRE > 0. The problem with multiplying the FID with such a function is that the noise in the tail of the FID is ampliﬁed. The basic idea is simple. All we need to do is to multiply the FID by a weighting function which increases with time. after applying the positive exponential we multiply by a second decaying function to “clip” the noise at the .11 Illustration of the use of weighting functions to enhance the resolution in the spectrum. (a) (c) (d) × = (b) (e) (f) (a) (h) × × (g) = (i) (j) Fig. In this section we will consider the opposite case.5 Resolution enhancement 4–11 We noted in section 4. where the weighting function is designed to narrow the lines in the spectrum and so increase the resolution.2 that the more rapidly the time domain signal decays the broader the lines become. A weighting function designed to improve the SNR inevitably leads to a broadening of the lines as such a function hastens the decay of the signal. See text for details.4.5 Resolution enhancement 4. 4. thus making the SNR in the spectrum very poor indeed.
The larger α. 4.4–12 Fourier transformation and data processing tail of the FID. We can see clearly from this plot how the two functions combine together to ﬁrst increase the timedomain function and then to attenuate it at longer times. the two lineshapes are compared in Fig. this gives the timedomain signal (h). Usually.12. The original FID. Parameters for the LorentztoGauss transformation The combined weighting function for this transformation is W (t) = exp(RRE t) exp(−αt 2). the faster the decay rate. the resolution has clearly been improved when compared to (b). which is often considered to be superior to the Lorentzian as it is narrower at the base. Careful choice of the parameters RRE and α are needed to obtain the optimum result. plot (j) shows the product of the two weighting functions (f) and (g). This transformation to a Gaussian lineshape is often called the LorentztoGauss transformation. Usually. Fourier transformation of (h) gives the spectrum (i). Fourier transformation of (d) gives the spectrum (e). (a). is given by RRE /π . The weighting function can therefore be rewritten W (t) = exp(−π Lt) exp(−αt 2 ).11. but the noise in the tail of the FID has been greatly magniﬁed. but the noise in the tail of the FID is not as large as it is in (d). If (a) is multiplied by the rising exponential function plotted in (c). The resulting peak in the spectrum will have a Gaussian lineshape. contains a signiﬁcant amount of noise and has been recorded well beyond the point where the signal decays into the noise. Fourier transformation of (a) gives the spectrum (b). . the second function is chosen to be one that decays relatively slowly over most of the FID and then quite rapidly at the end – a common choice is the Gaussian function: W (t) = exp(−αt 2 ). L.11 we can see the effect of introducing a Gaussian weighting function as well. the two peaks have been adjusted so that their peak heights and widths at half height are equal. but at the expense of a large reduction in the SNR. For compatibility with the linebroadening role of a decaying exponential it is usual to deﬁne the exponential weighting function as exp(−π Lt) so that Lorentzian Gaussian Fig. 4. The Gaussian is a more “compact” lineshape. Referring now to the bottom part of Fig. the resolution has clearly been improved. The whole process is illustrated in Fig. but without too great a loss of SNR. If we can set RRE to cancel exactly the original decay of the FID then the result of this process is to generate a timedomain function which only has a Gaussian decay. Note that once again the signal decay has been slowed. where α is a parameter which sets the decay rate. Finally. a process of trial and error is adopted.12 Comparison of the Lorentzian and Gaussian lineshapes. note how the decay of the signal has been slowed. The original FID (a) is multiplied by the rising exponential (f) and the decaying Gaussian (g). Recall that in such a situation the linewidth. Thus RRE = π L. the result is the FID (d). The usual approach is to specify RRE in terms of the linewidth that it would create on its own if it were used to specify a decaying exponential. 4. 4.
A little mathematics shows that this maximum occurs at time tmax given by: Lπ tmax = − 2α (recall that L is negative) so that α=− Lπ . 4.6 Other weighting functions a positive value for L leads to line broadening and a negative value leads to resolution enhancement. f . The basic sine bell is just the ﬁrst part of a sin θ for θ = 0 to θ = π .4. tacq : tmax = f tacq . tmax is expressed as a fraction. The weighting function is chosen so that the sine bell ﬁts exactly across the acquisition time.11 (j) we can see that the effect of the rising exponential and the Gaussian is to give an overall weighting function which has a maximum in it. 2tmax 4–13 We simply select a value of tmax and use this to deﬁne the value of α. It is usual to deﬁne the Gaussian parameter α from the position of this maximum.13. 4. In this form the function will give resolution enhancement rather like the combination of a rising exponential and a Gaussian function (compare Fig. see text for details. which are illustrated in Fig. phase 0 π/8 π/4 π/2 0 tacq /2 tacq Fig. tacq . mathematically the required function is: W (t) = sin πt . From Fig. this is illustrated in the top lefthand plot of Fig. 4. α=− 2 f tacq 4.13. of the acquisition time. 4.13 The top row shows sine bell and the bottom row shows sine bell squared weighting functions for different choices of the phase parameter. On some spectrometers. 4.6 Other weighting functions Many other weighting functions have been used for sensitivity enhancement and resolution enhancement. In this case Lπ .11 (j)). Perhaps the most popular are the sine bell are variants on it.
So. This is illustrated in Fig. 4. 4.7 Zero ﬁlling Before being processed the FID must be converted into a digital form so that it can be stored in computer memory.13. The FID is therefore represented by a series of data points. We will have more to say about this in Chapter 5 but for now we will just note that in this process the signal is sampled at regular intervals. tacq Much of the popularity of these functions probably rests of the fact that there is only one parameter to adjust. 4.4–14 Fourier transformation and data processing The sine bell can be modiﬁed by shifting it the left. The shape of all of these weighting functions are altered subtly by squaring them to give the sine bell squared functions. it is in fact a series of closely spaced points.14 Illustration of the results of zero ﬁlling. 4. and in the limit that the shift is by π/2 or 90◦ the function is simply a decaying one and so will broaden the lines. The weighting function is then W (t) = sin2 (π − φ)t +φ . tacq tacq tacq (a) (b) (c) Fig. the resulting weighting function is: (π − φ)t +φ . no extra data has been acquired. 4. We can see that there are only a few data points which deﬁne the line in the spectrum.13. W (t) = sin tacq Note that this deﬁnition of the function ensures that it goes to zero at tacq . . The FIDs along the top row have been supplemented with increasing numbers of zeroes and so contain more and more data points. these are also shown in Fig. although we plot the spectrum as a smooth line.14 (a) which shows the FID and the corresponding spectrum. as is shown in Fig. rather than joining up the points which make up the spectrum we have just plotted the points. When the FID is Fourier transformed the spectrum is also represented by a series of data points. The further the shift to the left the smaller the resolution enhancement effect will be. rather than two in the case of the LorentztoGauss transformation. Fourier transformation preserves the number of data points so the line in the spectrum is represented by more points as zeroes are added to the end of the FID. The shift is usually expressed in terms of a phase φ (in radians). Note that the FID remains the same for all three cases.
the corresponding spectrum has double the number of points and so the line is represented by more data points. to zero ﬁll beyond this point. In practice. 4.14 (b). So. If we stop recording the signal before it has fully decayed the FID is said to be “truncated”. the corresponding spectrum shows the expected lineshape.15 Illustration of how truncation leads to artefacts (called sinc wiggles) in the spectrum.4.14 (c). having added one set of zeroes. at least for aesthetic reasons if nothing else! It turns out that the Fourier transform algorithm used by computer programs is most suited to a number of data points which is a power of 2. .8 Truncation If we take the original FID and add an equal number of zeroes to it. of course. 4. This is desirable. 4. We can carry on with this zero ﬁlling process. Fig.8 Truncation 4–15 In conventional NMR it is virtually always possible to record the FID until it has decayed almost to zero (or into the noise). This results in an even larger number of data points deﬁning the line. Zero ﬁlling costs nothing in the sense that no extra data is required. for example. it is usual to zero ﬁll the time domain data so that the total number of points is a power of 2. as is shown in Fig. Adding a set of zeroes equal to the number of data points is called “one zero ﬁlling”. For example. if data recording is stopped before the signal has fully decayed the corresponding spectra show oscillations around the base of the peak. in multidimensional NMR this may not be the case. it does not improve the resolution as the measured signal remains the same. it is always an option. particularly in the “indirect dimension” (see Chapter X for further details). we can add another to double the total number of data points (“two zero ﬁllings”). 4. However. this is illustrated in Fig. Of course. it is just a manipulation in the computer.15. 4. simply because of the restrictions on the amount of data which can be recorded. but 15000 is not. 214 = 16384 is a suitable number of data points to transform. This is illustrated in Fig. However. The FID on the left has been recorded for sufﬁcient time that it has decayed almost the zero. but the lines will be better deﬁned in the spectrum. therefore.
and higherdimensional NMR experiments. the larger the sinc wiggles. these are usually called sinc wiggles or truncation artefacts – the name arises as the peak shape is related to a sinc function. a truncated FID leads to oscillations around the base of the peak. Much effort has therefore been put into ﬁnding alternatives to the Fourier transform which will generate spectra without these truncation artefacts.4–16 Fourier transformation and data processing As is shown clearly in the ﬁgure. this will have the side effects of broadening the lines and reducing the SNR. the way forward is to apply a decaying weighting function to the FID so as to force the signal to go to zero at the end. It is easy to show that the separation of successive maxima in these wiggles is 1/tacq Hz. Clearly these oscillations are undesirable as they may obscure nearby weaker peaks. Highly truncated time domain signals are a feature of three. . Each has its merits and drawbacks. The popular methods are maximum entropy. Unfortunately. linear prediction and FDM. Assuming that it is not an option to increase the acquisition time. they all need to be applied with great care. The more severe the truncation.
(Assume that the spectrometer has a B0 ﬁeld strength of 9.4 T). in a spectrum with many lines the dispersion mode lineshape is very undesirable – why? E 4–2 Suppose that we record a spectrum with the simple pulseacquire sequence using a 90◦ pulse applied along the x axis. E 4–4 Why is it undesirable to continue to acquire the FID after the signal has decayed away? How can weighting functions be used to improve the SNR of a spectrum? In your answer described how the parameters of a suitable weighting function can be chosen to optimize the SNR.08 × 108 rad s−1 T−1 . What lineshape would you expect to see in the spectrum. covering some 700 ppm. We then change the phase of the pulse from x to y. give the reasons for your answer.9 Exercises 4–17 E 4–1 In a spectrum with just one line. The resulting FID is Fourier transformed and the spectrum is phased to give an absorption mode lineshape. the dispersion mode lineshape might be acceptable – in fact we can think of reasons why it might even be desirable (what might these be?). Estimate the size of the phase correction which will be needed at the edges of the spectrum. However.9 Exercises 4. This nucleus shows a wide range of shifts. E 4–7 In a proton NMR spectrum the peak from TMS was found to show “wig .4. How would the spectrum be affected by: (a) applying the pulse about −x. What sets the limit on the improvement that can be obtained in practice? Is zero ﬁlling likely to improve the situation? E 4–6 Explain why use of a sine bell weighting function shifted by 45◦ may enhance the resolution but use of a sine bell shifted by 90◦ does not. (b) changing the pulse ﬂip angle to 270◦ about x? E 4–3 The gyromagnetic ratio of phosphorus31 is 1. acquire an FID in the same way and phase the spectrum using the same phase correction as above. Suppose that the transmitter is placed in the middle of the shift range and that a 90◦ pulse of width 20 µs is used to excite the spectrum. Are there any disadvantages to the use of such weighting functions? E 4–5 Describe how weighting functions can be used to improve the resolution in a spectrum.
. How can truncation artefacts be suppressed? Mention any difﬁculties with your solution to the problem. However.4–18 Fourier transformation and data processing gles” characteristic of truncation of the FID. the other peaks in the spectrum showed no such artefacts. Explain.
once the current is set running in the coil it will persist for ever. thereby generating a magnetic ﬁeld without the need for further electrical power. Chapter 5 “How the spectrometer works” c James Keeler. this reduces the amount of (expensive) liquid helium which boils off due to heat ﬂowing in from the surroundings. typically) the resistance goes to zero – that is the wire is superconducting. Broken down to its simplest form.1 The magnet Modern NMR spectrometers use persistent superconducting magnets to generate the B0 ﬁeld. The wire is of a special construction such that at low temperatures (less than 6 K. the really important parts of the spectrometer are not that complex to understand in outline. However. The whole assembly is constructed in a vacuum ﬂask so as to further reduce the heat ﬂow. 5. To maintain the wire in its superconducting state the coil is immersed in a bath of liquid helium. Thus.5 How the spectrometer works NMR spectrometers have now become very complex instruments capable of performing an almost limitless number of sophisticated experiments. Surrounding this is usually a “heat shield” kept at 77 K by contact with a bath of liquid nitrogen. and it is certainly helpful when using the spectrometer to have some understanding of how it works. 2002 . thereby generating a magnetic ﬁeld. The cost of maintaining the magnetic ﬁeld is basically the cost of liquid helium (rather expensive) and liquid nitrogen (cheap). homogeneous and stable magnetic ﬁeld • A “probe” which enables the coils used to excite and detect the signal to be placed close to the sample • A highpower RF transmitter capable of delivering short pulses • A sensitive receiver to amplify the NMR signals • A digitizer to convert the NMR signals into a form which can be stored in computer memory • A “pulse programmer” to produce precisely times pulses and delays • A computer to control everything and to process the data We will consider each of these in turn. Basically such a magnet consists of a coil of wire through which a current passes. Superconducting magnets tend to be very stable and so are very useful for NMR. the spectrometer consists of the following components: • An intense.
On its own. Essentially how this works is that the magnetic ﬁelds produced by the shims are cancelling out the small residual inhomogeneities in the main magnetic ﬁeld. Shims The lines in NMR spectra are very narrow – linewidths of 1 Hz or less are not uncommon – so the magnetic ﬁeld has to be very homogeneous. the magnetic ﬁeld is in the direction of this tube. which corresponds to a magnetic ﬁeld of 11. 5.1) where γ is the gyromagnetic ratio (2. once set on installation it is usually only necessary on a day to day basis to alter a few of the shims which generate the simplest ﬁeld proﬁles. say by a factor of 10. which is the direction of B0. which generate magnetic ﬁelds varying the corresponding directions. The current through each of these coils is adjusted until the magnetic ﬁeld has the required homogeneity. However. something we can easily assess by recording the spectrum of a sample which has a sharp line. . There are more shims whose labels you will recognize as corresponding to the names of the hydrogen atomic orbitals. the magnetic ﬁeld proﬁles that the shims coils create are in fact the spherical harmonic functions. Consider a proton spectrum recorded at 500 MHz. Expressed as a fraction of the main magnetic ﬁeld this is about 2 × 10−10 . for example. The shim z 2 generates a ﬁeld that varies quadratically along the z direction. Usually this room temperature zone takes the form of a vertical tube passing through the magnet (called the bore tube of the magnet). there are usually shims labelled x. What we have to do is to surround the sample with a set of shim coils. y and z. So. We can see that we need to have an extremely homogeneous magnetic ﬁeld for work at this resolution. Recall that the Larmor frequency is given by ν0 = − 1 γ B0 2π (5. We need to limit the variation in the magnetic ﬁeld across the sample so that the corresponding variation in the Larmor frequency is much less that the width of the line. The reason for this is easily demonstrated by an example.4 × 10−9 T. no superconducting magnet can produce such a homogeneous ﬁeld.1/γ = 2.5–2 How the spectrometer works Of course. we do not want the sample to be at liquid helium temperatures.75 T. each of which produces a tiny magnetic ﬁeld with a particular spatial proﬁle. which are the angular parts of the atomic orbitals. Modern spectrometers might have up to 40 different shim coils. the maximum acceptable change in Larmor frequency is 0.67 × 108 rad s−1 T−1 for protons). so adjusting them is a very complex task. This is no coincidence. meaning that it must not vary very much over space.1 Hz and so using Eq.1 we can compute the change in the magnetic ﬁeld as 2π 0. The shims are labelled according to the ﬁeld proﬁles they generate. With this condition. so a room temperature region – accessible to the outside world – has to be engineered as part of the design of the magnet.
2 The probe 5–3 The probe is a cylindrical metal tube which is inserted into the bore of the magnet. The power transfer to the transmitter and receiver (tx.) is optimized by adjusting the capacitor marked “match”.2 The probe 5. The two adjustments tend to interact rather. . It is also relatively easy to phase shift the output from such a synthesizer. We expect the tuning of the probe to be particularly sensitive to changing solvent or to changing the concentration of ions in the solvent. The inductance of the coil and the capacitance of the capacitor are set such that the tuned circuit they form is resonant at the Larmor frequency. Spectroscopists talk about “tuning the probe” which means adjusting the capacitor until the tuned circuit is resonant at the Larmor frequency.2 and 3. To aid us. The coil forms part of a tuned circuit consisting of the coil and a capacitor. 5. Usually a frequency synthesizer is used as the RF source. for example if we are doing experiments involving selective excitation (section 3.1 Schematic of the key parts of the probe. and rx.1 shows a typical arrangement. Extraordinary effort has been put into the optimization of the design of this coil. The radiofrequency transmitter is the part of the spectrometer which generates the pulses. As we only need the RF to be applied for a short time. That the coil forms part of a tuned circuit is very important as it greatly increases the detectable current in the coil.11). The small coil used to both excite and detect the NMR signal (see sections 3. Fig. but of course the coil needs to be made in such a way that the sample tube can drop down from the top of the magnet into the coil. To optimize the sensitivity this coil needs to be as close as possible to the sample. Various other pieces of electronics are contained in the probe.3 The transmitter match to tx/rx tune Fig. Such a device has all the desirable properties outlined above and is also readily controlled by a computer interface. 5. The key part of the probe is the small coil used to excite and detect the magnetization. the output of the synthesizer has to be “gated” so as to create a pulse of RF energy. The reason why we need to be able to set the frequency is that we might want to move the transmitter to different parts of the spectrum. so tuning the probe can be a tricky business. Usually we also need to “match the probe” which involves further adjustments designed to maximize the power transfer between the probe and the transmitter and receiver. a tuned circuit is formed by a capacitor (marked “tune”). Note that the coil geometry as shown is not suitable for a superconducting magnet in which the main ﬁeld is parallel to the sample axis. 5. along with some arrangements for heating or cooling the sample. We start with an RF source which produces a stable frequency which can be set precisely. which is something we will need to do in order to create phase shifted pulses. Such a gate will be under computer control so that the length and timing of the pulse can be controlled.4) is held in the top of this assembly in such a way that the sample can come down from the top of the magnet and drop into the coil. the instrument manufacturers provide various indicators and displays so that the tuning and matching can be optimized.5. The coil is shown on the left (with the sample tube in grey).
e. RF ampliﬁers are readily available which will boost this small signal to a power of 100 W or more. but it also results in unpredictable and erratic B1 ﬁelds.0 dB. As its name implies. not the natural logarithm. An attenuator which achieves this effect would be called “a 3 dB attenuator”. Usually the manufacturer states the power level which is “safe” for a particular probe. Eventually.2. i. high voltages are generated across the tuning capacitor. Pin note that the logarithm is to the base 10. For example.0 dB. the voltage will reach a point where it is sufﬁcient to ionize the air. This is useful as we may wish to set the B1 ﬁeld strength to a particular level. The attenuation is normal expressed is decibels (abbreviated dB and pronounced “deebee”). a 12 dB attenuator will reduce the power by a factor of 16. So. If the input power is Pin and the output power is Pout the attenuation in dB is Pout 10 × log10 . Clearly. The usual way of achieving this control is add an attenuator between the RF source (the synthesizer) and the ampliﬁer. as we know from elementary electrical theory. if the output power is half the input power. for example for selective excitation. So. Pin = 2×Pout . the attenuator and the gate used to create the pulses all under computer control. thus generating a discharge or arc (like a lightening bolt).2 Typical arrangement of the RF transmitter.0 Pin 2 So. there is a limit to the amount of power which can be applied because of the high voltages which are generated in the probe. an attenuation. the power ratio in dB is Pout 1 10 × log10 = 10 × log10 = −3. a power reduction by a factor of 4 corresponds to −6. It turns out that the B1 ﬁeld strength is proportional to the square root of the power applied. the more power that is applied to the probe the more intense the B1 ﬁeld will become and so the shorter the 90◦ pulse length. Power levels and “dB” The spectrometer usually provides us with a way of altering the RF power level and hence the strength of the B1 ﬁeld. The factor of 10 is the “deci” part in the dB. the attenuator reduces the signal as it passes through. Not only does this probe arcing have the potential to destroy the coil and capacitor. 5. power is equal to (resistance × current2 ). the current is proportional to the square root of the power.e. The synthesizer which is the source of the RF. In fact. When the RF power is applied to the tuned circuit of which the coil is part. computer control synthesizer gate attenuator amplifier to probe Fig. However. 5. Likewise. The reason for this is that it is the current in the coil which is responsible for the B1 ﬁeld and. because of the logarithmic relationship we can see that each 3 dB of attenuation will halve the power. the complete arrangement is illustrated in Fig. halving the power corresponds to a change of −3.5–4 How the spectrometer works The RF source is usually at a low level (a few mW) and so needs to be boosted considerably before it will provide a useful B1 ﬁeld when applied to the probe. the minus indicating that there is a power reduction i. .
this is not the case. τnew For example.4. τinit and τnew : power ratio in dB = 20 log10 τinit . We would therefore need to increase the attenuator setting by 4. in our experiment we want the ﬁeld strength to be ω1 /2π . Recall from the above that a change in the power by a factor of 4 corresponds to 6 dB.3 The transmitter Therefore. in order to double the B1 ﬁeld we need to double the current and this means multiplying the power by a factor of four. all we need to do is to shift the phase of the RF by 90◦ . This expression can be used to ﬁnd the correct setting for the attenuator. The required attenuation would be: power ratio in dB = 20 log10 15 25 = −4.9 we saw that pulses could be applied along different axes in the rotating frame. new However. As the duration of a pulse of a given ﬂip angle is inversely proportional to ω1 /2π the relationship can be expressed in terms of the initial and new pulse widths.4dB. . Suppose that with a certain setting of the init attenuator we have determined the B1 ﬁeld strength to be ω1 /2π . 6 dB of attenuation causes the B1 ﬁeld to fall to half its original value. and so on. . where we have expressed the ﬁeld strength in frequency units as described in section 3. This is very helpful to us as we can determine the attenuation needed for different B1 ﬁeld strengths. suppose we have calibrated the pulse width for a 90◦ pulse to be 15 µs but what we actually wanted was a 90◦ pulse of 25 µs. Phase shifted pulses In section 3. At ﬁrst sight you might think that to achieve this we would need a coil oriented along the y axis to apply a pulse along y. Usually the attenuator is under computer control and its value can be set in dB.4 dB.5. 12 dB would cause it to fall to one quarter of its initial value. Thus. for example x and y. The ratio of the powers needed to achieve these two ﬁeld strengths is equal to the square of the ratio of the ﬁeld strengths: power ratio = Expressed in dB this is power ratio in dB = 10 log10 = 20 log10 new ω1 /2π init ω1 /2π new ω1 /2π init ω1 /2π 2 new ω1 /2π init ω1 /2π 2 5–5 . However.
3.5–6 How the spectrometer works In section 3. whereas if it is modulated by a sine wave the result is a pulse y. there is no need to install more than one coil in the probe.8 where it starts out along x. the x component now takes the form of a sine wave. The resulting sine modulated ﬁeld can still be thought of as arising from + − two counterrotating ﬁelds. 5.8 where B1 starts out on the x axis. The crucial thing is that as the ﬁeld is zero at time zero the two counterrotating vectors start out opposed. 1 So. this is shown in Fig. We see that by phase shifting the RF we can apply pulses about any axis. φ. B1 and B1 . whereas in the latter it starts are zero and increases. Note that the component B − starts out along the y axis at time zero in contrast to the case of Fig. This ﬁgure should be compared to Fig.4 we described the situation where RF power at frequency ωRF is applied to the coil. as required. S x . varies as a cosine wave. 3. If we phase shift the RF by 90◦ the ﬁeld will be modulated by a sine function rather than a cosine: 2B1 sin ωRF t. In this situation the B1 ﬁeld along the x axis can be written as 2B1 cos ωRF t. 3. When we move to a rotating frame at ωRF this component will be along the y axis (the position it has at time zero). In (a) the vector starts out along x . as it rotates in a positive sense the x component. The difference between the cosine and sine modulations is that in the former the ﬁeld starts at a maximum and then decays. 5. as is shown in the graph on the right. As they evolve from this position the component along x − grows. 5. How this comes about is illustrated in Fig. . 3. as is shown in the graph. of −90◦ compared to (a). 5.3 Illustration of how a 90◦ phase shift takes us from cosine to sine modulation. Recall from section 3.4 Illustration of the decomposition of a ﬁeld oscillating along the x axis according to a sine wave into two counter rotating components. y B1 B1 + B1 + B1 B1 + B1 (a) y x Sx (b) y x φ Sx Fig. if the B1 ﬁeld is modulated by a cosine wave the result is a pulse about x.4. x + B1 B1 y x 2B1 field along x time Fig. This − is in contrast to the case in Fig. Situation (b) is described as a phase shift.4 that it is the component B1 which interacts with the magnetization. In (b) the vector starts on −y .8.
called the preampliﬁer or preamp is usually placed as close to the probe as possible (you will often see it resting by the foot of the magnet). The numbers that the ADC outputs are therefore an approximation . 110 and 111. We need to ensure that the highpower pulse does not end up in the sensitive receiver. this is illustrated in Fig. 5. and the largest number that the ADC can output is set by the number of binary “bits” that the ADC uses. 001. but output of the 3 bit ADC only has 8 levels so what it has to do is simply output the level which is closest to the current input level. Some diplexers are passive in the sense that they require no external power to achieve the required switching. 101. 010. Other designs use fast electronic switches (rather like the gate in the transmitter) and these are under the command of the pulse programmer so that the receiver or transmitter are connected to the probe at the right times.5 Digitizing the signal The analogue to digital converter A device known as an analogue to digital converter or ADC is used to convert the NMR signal from a voltage to a binary number which can be stored in computer memory.4 The receiver 5. There are various different ways of constructing such a device. This means that at one moment hundreds of Watts of RF power are being applied and the next we are trying to detect a signal of a few µV. One additional problem which needs to be solved comes about because the coil in the probe is used for both exciting the spins and detecting the signal. The output from the ADC is just a number. For example with only three bits the output of the ADC could take just 8 values: the binary numbers 000. These ampliﬁers need to be designed so that they introduce a minimum of extra noise (they should be lownoise ampliﬁers).5. resulting in a representation of the FID as data points.5. 011. This is so that the weak signal is boosted before being sent down a cable to the spectrometer console. Such an ADC would be described as a “3 bit ADC”. which is 2 raised to the power of the number of bits). The waveform which the ADC is digitizing is varying continuously. When the pulse is off the receiver is connected to the probe and the transmitter is disconnected. 100. thereby destroying it! This separation of the receiver and transmitter is achieved by a gadget known as a diplexer. but at the simplest level it is just a fast acting switch.4 The receiver 5–7 The NMR signal emanating from the probe is very small (of the order of µV) but for modern electronics there is no problem in amplifying this signal to a level where it can be digitized. 5. When the pulse is on the high power RF is routed to the probe and the receiver is protected by disconnecting it or shorting it to ground. The smallest number is 0 and the largest number is decimal 7 (the total number of possibilities is 8. The ADC samples the signal at regular intervals. The ﬁrst of these ampliﬁers.
5. if the time interval is too long we will miss crucial features of the waveform and so the digitized points will be a poor representation of the signal. Sampling rates Given that the ADC is only going to sample the signal at regular intervals the question arises as to how frequently it is necessary to sample the FID i. In (a) the data points (shown by dots) are quite a good representation of the signal (shown by the continuous line).e. i. f max . Clearly. 64 levels). Usually we think of this relationship the other way round i. We will see shortly that we are able to distinguish positive and negative frequencies. This sampling interval is often called the dwell time. 5. The approximation can be improved by increasing the number of bits. if the spectrum contains a very strong peak the sidebands from it can swamp a nearby weak peak. so a dwell time of means that the range of frequencies from − f max to + f max are represented correctly. 2 f max (a) (b) Fig. so the smoothly varying waveform has to be approximated by data points at one of the 8 levels. This is illustrated in Fig.e. 5. reduces the digitization sidebands. indicated by ﬁlled circles. the improvement over (a) is evident.e. 5. Spectrum (b) is from an FID which has been digitized using an 8 bit ADC (256 levels). 2 f max is called the Nyquist frequency. It turns out that if the interval between the points is the highest frequency which can be represented correctly.e. ADC with between 16 and 32 bits are commonly in use in NMR spectrometers. the number of bits it uses. this gives more output levels. Improving the resolution of the ADC. is given by 1 .6 Spectrum (a) is from an FID which has been digitized using a 6 bit ADC (i. Fig. Usually these are not a problem as they are likely to be swamped by noise. A signal at f max will have two data points per cycle. the vertical scale has been expanded 10 fold so that the digitization sidebands are clearly visible.7 Illustration of the effect of sampling rate on the representation of the FID. 5. The data points. The move to higher numbers of bits is limited by technical considerations.6. However. as is shown in Fig. At present.5 Digitization of a waveform using an ADC with 8 levels (3 bits). if we wish to represent correctly frequencies up to f max the sampling interval is given by: f max = = 1 .7. The output of the ADC can only be one of the 8 levels.5–8 How the spectrometer works to the actual waveform. . are therefore an approximation to the true waveform. what should the time interval between the data points be. (a) (b) (a) (b) Fig. The main consequence of the approximation process which the ADC uses is the generation of a forest of small sidebands – called digitization sidebands – around the base of the peaks in the spectrum. In (b) the data points are too widely separated and so are a very poor representation of the signal.
002 MHz and then subtract this from the NMR frequencies we end up with a range −2000 to +2000 Hz. This high frequency signal is easily separated from the required low frequency signal by a passing the output of the mixer though a low pass ﬁlter. Mixing down to a lower frequency Luckily for us.6 Quadrature detection A signal at greater than f max will still appear in the spectrum but not at the correct frequency.9. In spectrum (a) the peak (shown in grey) is at a higher frequency than the maximum set by the Nyquist condition.8.004 MHz. The other input is the NMR signal the probe. The process is visualized in Fig. The subtraction process is carried out by a device called a mixer. the 10 ppm of a proton spectrum recorded at 400 MHz covers just 4000 Hz.6 Quadrature detection 5–9 (a) fmax 0 (b) +fmax fmax 0 +fmax Fig. This signal is often called the local oscillator. In the previous section we saw that the signal which is passed to the ADC contains positive and negative frequencies – essentially this corresponds to positive and negative offsets. Such a small range of frequencies is easily within the capability of suitable ADCs. ( f 1 + f 2 ) and the difference ( f 1 − f 2 ). For example. The other output of the mixer is the sum frequency. Typically. although the NMR frequency is quite high. the frequency we subtract is set somewhere in the middle of the spectrum. such a peak is said to be folded.5. some are passive and are mainly constructed from transformers and diodes.000 MHz to 400. 5.8 Illustration of the concept of folding. number of bits) we need for NMR. Usually one of the inputs is generated locally. The ﬁltered signal is then passed to the ADC. The procedure is to take the NMR signals and then subtract a frequency from them all such that we are left with a much lower frequency. suppose that the 10 ppm range of proton shifts runs from 400. The solution to this problem is to mix down the signal to a lower frequency. A typical NMR frequency is of the order of hundreds of MHz but there simply are no ADCs available which work fast enough to digitize such a waveform with the kind of accuracy (i. If we set the receiver reference frequency at 400. For example. The question therefore arises as to whether or . 5.e. This Nyquist condition quickly brings us to a problem. such a peak would appear in the position shown in (b). the range of frequencies that any typical spectrum covers is rather small. as is described in the next section. For example a peak at ( f max + F) will appear in the spectrum at (− f max + F) as is illustrated in Fig. 5. this frequency is called the receiver reference frequency or just the receiver frequency. There are various ways of actually making a mixer – some are “active” and contain the usual transistors and so on. In practice. The details need not concern us – all we need to know is that a mixer takes two signal inputs at frequencies f 1 and f 2 and produces an output signal which contains the signals at the sum of the two inputs. which will be at the order of twice the Larmor frequency. In the spectrometer the local oscillator is set to a required receiver frequency so that the difference frequency will be quite small. typically by a synthesizer. 5.
5–10
How the spectrometer works not we are able to discriminate between frequencies which only differ in their sign; the answer is yes, but only provided we use a method known as quadrature detection.
positive frequency
y x
Mx
time
My
negative frequency
y x
Mx
time
My
Fig. 5.10 Illustration of the x and y components from a vector precessing as a positive and negative offset. The only difference between the two is in the y component.
Figure 5.10 illustrates the problem. The upper part shows a vector precessing with a positive offset frequency (in the rotating frame); the x component is a cosine wave and the y component is a sine wave. The lower part shows the evolution of a vector with a negative frequency. The x component remains the same but the y component has changed sign when compared with evolution at a positive frequency. We thus see that to distinguish between positive and negative frequencies we need to know both the x and y components of the magnetization. It is tempting to think that it would be sufﬁcient just to know the y component as positive frequencies give a positive sine wave and negative ones give a negative sine wave. In the corresponding spectrum this would mean that peaks at positive offsets would be positive and those at negative offsets would be negative. If there were a few well separated peaks in the spectrum such a spectrum might be interpretable, but if there are many peaks in the spectrum
5.6 Quadrature detection the result would be an uninterpretable mess. How then can we detect both the x and y components of the magnetization? One idea is that we should have two coils, one along x and one along y: these would certainly detect the x and y components. In practice, it turns out to be very hard to achieve such an arrangement, partly because of the conﬁned space in the probe and partly because of the difﬁculties in making the two coils electrically isolated from one another. Luckily, though, it turns out to be easy to achieve the same effect by phase shifting the receiver reference frequency; we describe how this works in the next section. The mixing process In section 5.5 we described the use of a mixer to subtract a locally generated frequency from that of the NMR signal. The mixer achieves this by multiplying together the two input signals, and it is instructive to look at the consequence of this in more detail. Suppose that the signal coming from the probe can be written as A cos ω0 t where A gives the overall intensity and as usual ω0 is the Larmor frequency. Let us write the local oscillator signal as cos ωrx. t, where ωrx is the receiver frequency. Multiplying these two together gives: A cos ω0 t × cos ωrx t = 1 A [cos(ω0 + ωrx )t + cos(ω0 − ωrx )t] 2 where we have used the well known formula cos A cos B = 1 (cos(A + B) + 2 cos(A − B)). After the low pass ﬁlter only the term cos(ω0 − ωrx )t will survive. We recognize the difference frequency (ω0 − ωrx ) as the separation between the receiver frequency and the Larmor frequency. If the receiver frequency is made to be the same as the transmitter frequency (as is usually the case) this difference frequency is the offset we identiﬁed in section 3.4. In summary, the output of the mixer is at the offset frequency in the rotating frame and takes the form of a cosine modulation; it is thus equivalent to detecting the x component of the magnetization in the rotating frame. Now consider the case where the local oscillator signal is − sin ωrx t; as we saw in section 5.3 such a signal can be created by a phase shift of 90◦ . The output of the mixer is now given by A cos ω0 t × − sin ωrx t = 1 A [− sin(ω0 + ωrx )t + sin(ω0 − ωrx )t] 2 where we have used the well known formula cos A sin B = 1 (sin(A + B) − 2 sin(A − B)). We now see that the difference frequency term is sine modulated at the offset frequency; it is the equivalent of the y component in the rotating frame. Thus, simply by shifting the phase of the local oscillator signal we can detect both the x and y components in the rotating frame. The complete arrangement for quadrature detection is outlined in Fig. 5.11. The NMR signal from the probe is split into two and fed to two separate mixers. The local oscillator signals fed to the two mixers are phase shifted by 90◦ relative to one another; as a result, the outputs of the two mixers
5–11
f1
f2
f1+f2 f1f2
Fig. 5.9 A radiofrequency mixer takes inputs at two different frequencies and produces an output which contains the sum and difference of the two inputs.
5–12
mixer
How the spectrometer works
Mx real part of FID
ADC low pass filter from probe mixer ADC low pass filter 90˚ phase shift local oscillator
My imag. part of FID
Fig. 5.11 Schematic of the arrangement for quadrature detection; see text for details.
are the equivalents of the x and y components in the rotating frame. These two outputs are digitized separately and become the real and imaginary parts of a complex time domain signal (as described in section 4.1). Such a quadrature detection scheme is the norm for all modern NMR spectrometers. 5.7 Time and frequency We are now in a position to tidy up a number of points of terminology relating to the FID and spectrum. It is important to understand these in order to report correctly the parameters used to record a spectrum. The range of frequencies correctly represented in the spectrum (i.e. not folded) is called the spectral width, f SW . As quadrature detection is used, the 1 frequency scale on the spectrum runs from − 1 f SW to + 2 f SW i.e. positive and 2 negative frequencies are discriminated. For a given spectral width, the sampling interval or dwell time, , is computed from: 1 = f SW which means that the Nyquist frequency is 1 f SW – the edges of the spectrum. 2 If we record the FID for a time tacq , called the acquisition time, then the number of data points, N , is tacq N= . Each data point is complex, consisting of a real and an imaginary part. To be clear about the conditions under which an FID is recorded it is important to quote the acquisition time and the spectral width; quoting the number of data points on its own is meaningless.
5.8 The pulse programmer 5.8 The pulse programmer
5–13
The pulse programmer has become an immensely sophisticated piece of computer hardware, controlling as it does all of the functions of the spectrometer. As the pulse programmer needs to produce very precisely timed events, often in rapid succession, it is usual for it to run independently of the main computer. Typically, the pulse program is speciﬁed in the main computer and then, when the experiment is started, the instructions are loaded into the pulse programmer and then executed there. The acquisition of data is usually handled by the pulse programmer, again separately from the main computer.
5–14 5. (ω1 /2π ). Your intention is to use it to record phosphorus31 spectra at Larmor frequency of 180 MHz. and you know that your typical linewidths are likely to be of the order of 25 Hz. of 2 kHz? E 5–3 A spectrometer is equipped with a transmitter capable of generating a maximum of 100 W of RF power at the frequency of carbon13. would have to be introduced into the transmitter in order to give a ﬁeld strength. but that a peak is present whose offset from the receiver is +60 Hz.8 µs. Using this transmitter at full power the 90◦ pulse width is found to be 20µs. Is the magnet sufﬁciently homogeneous to be of use? E 5–2 A careful pulse calibration experiment determines that the 180◦ pulse is 24. Why is it generally desirable to improve the number of bits that the ADC uses? E 5–5 A spectrometer operates at 800 MHz for proton and it is desired to cover a shift range of 15 ppm. What power would be needed to reduce the 90◦ pulse width to 7. How much attenuation.9 Exercises How the spectrometer works E 5–1 You have been offered a superconducting magnet which claims to have a homogeneity of “1 part in 108 ”.5 µs? Would you have any reservations about using this amount of power? E 5–4 Explain what is meant by “a two bit ADC” and draw a diagram to illustrate the outcome of such a ADC being used to digitize a sine wave. Assuming that the receiver frequency is placed in the middle of this range. what spectral width would be needed and what would the sampling interval (dwell time) have to be? If we recorded a FID for 2 s with the spectral width set as you have determined. Where will the peak appear in the spectrum? Can you explain why? . how many data points will have been collected? E 5–6 Suppose that the spectral width is set to 100 Hz. in dB.
introduced in Chapter 3. In quantum mechanics. momentum and energy. time etc. One often used interpretation of such electronic wavefunctions is to say that the square of the wavefunction gives the probability of finding the electron at that point. For example. The tools we need are provided by quantum mechanics.1 A quick review of quantum mechanics In this section we will review a few key concepts before moving on to a description of the product operator formalism. Wavefunctions are simply mathematical functions of position. 6. so we really must have some way of describing the behaviour of such systems under multiplepulse experiments. each observable has an operator associated with it. Operators "operate on" functions to give new functions. However. The wavefunction describes the system of interest (such as a spin or an electron) completely.6 Product Operators† The vector model. where r is the distance from the nucleus and a is a constant. When it comes to twodimensional NMR many of the experiments are only of interest in coupled spin systems. operators represent "observable quantities" such as position. specifically in the form of density matrix theory which is the best way to formulate quantum mechanics for NMR. In quantum mechanics. there is a way of proceeding which we can use without a deep knowledge of quantum mechanics: this is the product operator formalism. is very useful for describing basic NMR experiments but unfortunately is not applicable to coupled spin systems. we do not want to get involved in a great deal of complex quantum mechanics! Luckily. in words this operator says "differentiate † Chapter 6 "Product Operators" © James Keeler. the 1s electron in a hydrogen atom is described by the function exp(–ar). hence their name operator × function = (new function) An example of an operator is (d dx ) . much in the same way as we did for the vector model in Chapter 3. One particularly appealing feature is the fact that the operators have a clear physical meaning and that the effects of pulses and delays can be thought of as geometrical rotations. The product operator formalism is a complete and rigorous quantum mechanical description of NMR experiments and is well suited to calculating the outcome of modern multiplepulse experiments. 1998 & 2002 6–1 . The simplest example of this that is frequently encountered is when considering the wavefunctions which describe electrons in atoms (atomic orbitals) or molecules (molecular orbitals). two mathematical objects – wavefunctions and operators – are of central importance. if the wavefunction is known it is possible to calculate all the properties of the system.
1. H. In NMR.2 Hamiltonians The Hamiltonian. The x. it turns out that this is a vector quantity which points in a direction perpendicular to the plane of the rotation.and zdirections. and it is transitions between these energy levels which are detected in spectroscopy.1. This operator is exceptionally important as its eigenvalues and eigenfunctions are the "energy levels" of the system. In this chapter the form of the relevant Hamiltonians will simply be stated rather than derived. represented by a vector which points perpendicular to the plane of rotation. 6.y6–2 . There is one Hamiltonian for the spin or spins in the presence of the applied magnetic field. I y and Iz . The precise mathematical form of the Hamiltonian is found by first writing down an expression for the energy of the system using classical mechanics and then "translating" this into quantum mechanical form according to a set of rules. y. the Hamiltonian is seen as having a more subtle effect than simply determining the energy levels. For a single spinhalf.2. In quantum mechanics. To understand the spectrum. angular momentum and magnetization. such an energy. y. Its effect on the function sin x is d (sin x ) = cos x dx the "new function" is cos x. In NMR the Hamiltonian changes depending on the experimental situation. so for example the operator x2 just means "multiply by x2 ". This comes about because the Hamiltonian also affects how the spin system evolves in time. therefore. it is necessary to have a knowledge of the energy levels and this in turn requires a knowledge of the Hamiltonian operator.and zcomponents of this vector can be specified. The three components of this spin angular momentum (along x. By altering the Hamiltonian the time evolution of the spins can be manipulated and it is precisely this that lies at the heart of multiplepulse NMR. and these are the angular momenta in the x.1 Operators In quantum mechanics operators represent observable quantities. is the special name given to the operator for the energy of the system.1 Spin operators A mass going round a circular path (an orbit) possesses angular momentum.2 Operators for one spin p A mass going round a circular path possesses angular moment. Nuclear spins also have angular momentum associated with them – called spin angular momentum. y and z) are represented by the operators I x . there are operators which represent these three components of the angular momentum. but this Hamiltonian changes when a radiofrequency pulse is applied. 6. the x.with respect to x". Operators can also be simple functions. 6. 6.
and zcomponents of the magnetization are represented by the spin angular momentum operators Ix . 6.2 Hamiltonians for pulses and delays In order to work out how the density operator varies with time we need to know the Hamiltonian (which is also an operator) which is acting during that time. x = ω 1 I x and for a pulse about the yaxis it is Hpulse. of duration tp . Iy and Iz respectively. is computed from that at time 0. At equilibrium the density operator is proportional to Iz (there is only zmagnetization present). in the quantum mechanical picture the Hamiltonian involves the zangular momentum operator. is Hfree = ΩIz In the vector model free precession involves a rotation at frequency Ω about the zaxis. The constant of proportionality is usually unimportant. Iz – there is a direct correspondence. Suppose that an xpulse. The Hamiltonian for a pulse about the xaxis.or yaxes. can be represented as a sum of different amounts of these three operators σ (t ) = a(t ) I x + b(t ) I y + c(t ) Iz The amounts of the three operators will vary with time during pulses and delays. σ(t). using the following relationship σ (t ) = exp( −iHt ) σ (0) exp(iHt ) where H is the relevant hamiltonian. in quantum mechanics the density operator. In this situation H = ω1 Ix and σ(0) = Iz. is applied to equilibrium magnetization.1. This expression of the density operator as a combination of the spin angular momentum operators is exactly analogous to specifying the three components of a magnetization vector. Hpulse. σ(0) . The free precession Hamiltonian (i. that for a delay).1. If H and σ are expressed in terms of the angular momentum operators if turns out that this equation can be solved easily with the aid of a few rules. is Hpulse. Thus at any time the state of the spin system. the equation to be solved is σ t p = exp −iω 1t p I x I z exp iω 1t p I x ( ) ( ) ( ) 6–3 Such equations involving angular momentum operators are common in .e.3 Equation of motion The density operator at time t. so it is usual to write σeq = Iz 6. Hfree . y = ω 1 I y Again there is a clear connection to the vector model where pulses result in rotations about the x. σ.
so diagram III is required. {new operator} and Ia are determined from the three possible angular momentum operators according to the following diagrams. The required identity is therefore 6–4 .1.1]. [6. They all have the same form exp( −iθI a ) {old operator}exp(iθI a ) ≡ cosθ {old operator} + sin θ {new operator} where {old operator}. there are a limited number of identities of the type of Eqn. The diagram shows that Iy (the "old operator") is rotated to Iz (the "new operator"). with a sinusoidal dependence on the flip angle. 6.1] σ t p = cosω 1t p I z − sinω 1t p I y The result is exactly as expected from the vector model: a pulse about the xaxis rotates zmagnetization towards the –yaxis. The diagram shows that –Iz (the "old operator") is rotated to –Ix (the "new operator"). The required identity is therefore exp( −iθI x ) I y exp(iθI x ) ≡ cosθ I y + sin θ I z Second example: find the result of exp −iθI y ( ) {– I } exp(iθI ) z y This is a rotation about y.quantum mechanics and the solution to them are already all know.1] This is interpreted as a rotation of Iz by an angle β about the xaxis. [6. that is exp( −iθI x ) I y exp(iθI x ) For rotations about x the middle diagram II is required.4 Standard rotations Given that there are only three operators. β. the label in the centre indicates which axis the rotation is about I x –y z –x y y II z x –z –y –x III z y –z x ( ) Angle of rotation = Ωt for offsets and ω1tp for pulses First example: find the result of rotating the operator Iy by θ about the xaxis. The identity required here to solve this equation is exp( −iβI x ) Iz exp(iβI x ) ≡ cos β Iz − sin β I y [6. By putting β= ω1 tp this identity can be used to solve Eqn.
1. note that a rotation of an operator about its own axis has no effect e. the arrow notation is often used. In this. ω1 tp .2] is easy to handle as it is unaffected by 6–5 . for which the Hamiltonian is ΩIz. the following ( ) ( ) σ t p = exp −iω 1t p I x σ (0) exp iω 1t p I x is written 1p σ (0) x → σ t p ( ) ( ) ( ) ωt I ( ) For the case where σ(0) = Iz 1p I z x → cos ω 1t p I z – sin ω 1t p I y ωt I 6. is π radians. overall the result is πI x − cos Ωτ I y → cos Ωτ I y The second term on the left of Eqn. the term Ht is written over an arrow which connects the old and new density operators. So. [6.exp −iθI y {− I z } exp iθI y ≡ cosθ {− I z } + sin θ {– I x } ≡ − cosθ I z − sin θ I x Finally. In the shorthand notation this is − I y ΩτI z → σ (b) To solve this diagram I above is needed with the angle = Ωτ. is π so the second term on the right is zero and the first term just changes sign (cos π = –1).5 Shorthand notation To save writing. 6. In the shorthand notation 1p − cos Ωτ I y + sin Ωτ I x x → σ (e) ωt I [6. for example. The pulse about x has the Hamiltonian ω1 Ix . the relevant diagram is thus II 1p − cos Ωτ I y x → − cos Ωτ cos ω 1t p I y − cos Ωτ sin ω 1t p I z ωt I However. the delay τ therefore corresponds to a rotation about the zaxis at frequency Ω. the flip angle of the pulse.2] Each term on the left is dealt with separately. ω1 tp . The first term is a rotation of y about x.6 Example calculation: spin echo 90°( x ) a delay τ b 180°( x ) e delay τ f acquire At a the density operator is –Iy . a rotation of Ix about x leaves Ix unaltered.g. the pulse therefore corresponds to a rotation about x for a time tp such that the angle. The transformation from a to b is free precession.1. the "new operator" is Ix τIz − I y Ω→ − cos Ωτ I y + sin Ωτ I x In words this says that the magnetization precesses from –y towards +x.
Iy and Iz. One further point is that as far as the offset is concerned the spin echo sequence of Eqn. such systems cannot be treated by the vector model. [6. Thus the overall result of the spin echo sequence can be summarised τ− I z 90°( x )–τ −180°( x )−→ I y In words. this is a very useful feature which is much used in multiplepulse NMR experiments. Again. 6. the effect of the 180° pulse is then πI x − cos Ωτ I y + sin Ωτ I x → cos Ωτ I y + sin Ωτ I x [6. For two spins. In general the sequence – τ – 180°(x) – τ – [6. However.4] refocuses any evolution due to offsets. 6. Ω. even though there is evolution during the delays. product operators provide a clean and simple description of the important phenomena of coherence transfer and multiple quantum coherence. [6. spin 1: I1 x I1 y I1z spin 2 : I2 x I2 y I2 z I1 z represents zmagnetization of spin 1. indicates which spin they refer to. In the language of product operators I1 x is said to represent inphase magnetization of spin 1. . The next stage is the evolution of the offset for time τ.a rotation about x.1.4] is just equivalent to 180°(x). three such operators are needed for each spin. 1 or 2.8.3 Operators for two spins The product operator approach comes into its own when coupled spin systems are considered. The offset is said to be refocused by the spin echo. Overall. the spectrum consists of two doublets and the operator I1 x can be further identified with the two lines of the spin1 doublet.1 Product operators for two spins For a single spin the three operators needed for a complete description are Ix .3] As was shown using the vector model. with offsets Ω1 and Ω2 (rad s–1) and mutual coupling J 12 (Hz). the ycomponent just changes sign. the outcome is independent of the offset. I1 x represents xmagnetization on spin 1. This is exactly the result we found in section 3. As spin 1 and 2 are coupled. the description inphase 6–6 J12 J12 Ω1 Ω2 The spectrum from two coupled spins. an additional subscript.3] is considered separately τIz cos Ωτ I y Ω→ cos Ωτ cos Ωτ I y − sin Ωτ cos Ωτ I x τIz sin Ωτ I x Ω→ cos Ωτ sin Ωτ I x + sin Ωτ sin Ωτ I y Collecting together the terms in Ix and Iy the final result is (cos Ωτ cos Ωτ + sin Ωτ sin Ωτ ) I y + (cos Ωτ sin Ωτ − sin Ωτ cos Ωτ ) I x The bracket multiplying Ix is zero and the bracket multiplying Iy is =1 because of the identity cos 2 θ + sin 2 θ = 1. each term on the right of Eqn. and I2 z likewise for spin 2. and the delay τ.
an absorption mode lineshape will be assigned to magnetization aligned along x and a dispersion mode lineshape to magnetization along y. consider I1 x evolving under the offset of spin 1 and spin 2. For example.or yoperators) terms and correspond to multiplequantum coherences. Antiphase terms are thus sensitive to the spin states of the coupled spins. The relevant Hamiltonian is Hfree = Ω1 I1z + Ω 2 I2 z 6–7 Ω1 α β spin state of spin 2 The two lines of the spin1 doublet can be associated with different spin states of spin 2. 2I1 zI2 x . and that in an antiphase multiplet these two lines have different signs.1.e. but this magnetization is aligned along y and so will give rise to a different lineshape. 2I1 y I2 z. 2I1zI2x Ω2 2I1zI2y Ω2 Note that the two lines of the spin1 multiplet are associated with different spin states of spin2. Arbitrarily. The "cartoon" forms of the lineshapes are shown in the lower part of the diagram. There are four remaining product operators which contain two transverse (i. The rotations have to be applied separately to each spin and it must be remembered that rotations of spin 1 do not affect spin 2. The absorption lineshape is a maximum on resonance. . 2I1 zI2 y (the factors of 2 are needed for normalization purposes).2 Evolution under offsets and pulses The operators for two spins evolve under offsets and pulses in the same way as do those for a single spin. x. The operator 2I1 x I2 z is described as magnetization on spin 1 which is antiphase with respect to the coupling to spin 2.means that the two lines of the spin 1 doublet have the same sign and lineshape. they are not observable multiple quantum : 2 I1 x I2 y 2 I1 y I2 x 2 I1 x I2 x 2 I1 y I2 y Finally there is the term 2I1 zI2 z which is also not observable and corresponds to a particular kind of nonequilibrium population distribution. and vice versa. absorption dispersion I1x Ω1 Ω1 I1y I2x Ω2 I2y Ω2 There are four additional operators which represent antiphase magnetization: 2I1 x I2 z. 6. respectively. 2I1xI2z Ω1 Ω1 2I1yI2z The absorption and dispersion lineshapes. Following on in the same way I2 x represents inphase magnetization on spin 2. I1 y and I2y also represent inphase magnetization on spins 1 and 2. whereas the dispersion goes through zero at this point.
1. a 90° rotation of I1 x about y gives – I1 z (remembering that ω1 t = π/2 for a 90° pulse) 1 1 2 I1 x I2 z 1 y → cos ω1t 2 I1 x I2 z − sin ω1t 2 I1z I2 z 2 y → ω tI ω tI 2 I1 x I2 z 1 y → −2 I1z I2 z 2 y → The second arrow only affects the spin 2 operators. therefore I1 x Ω1tI1 z +Ω2tI2 z → cos Ω1t I1 x + sin Ω1t I1 y A second example is the term 2I1 x I2 z evolving under a 90° pulse about the yaxis applied to both spins. Such a process is called coherence transfer and is exceptionally important in multiplepulse NMR. 6.where Ω1 and Ω2 are the offsets of spin 1 and spin 2 respectively. Evolution under coupling causes the interconversion of inphase and antiphase magnetization according to the following diagrams π 2I π 2I π 2I π 2I 6–8 . The Hamiltonian representing this coupling is itself a product of two operators: HJ = 2πJ12 I1z I2 z where J12 is the coupling in Hz. The relevant Hamiltonian is H = ω 1 I1 y + ω 1 I2 y The evolution can be separated into two successive rotations 1 1 1 2 I1 x I2 z y → 2 y → ω tI ω tI The first arrow affects only the spin1 operators.3 Evolution under coupling The new feature which arises when considering two spins is the effect of coupling between them. a 90° rotation of z about y takes it to x 2 I1 x I2 z 1 y → −2 I1z I2 z 2 y → −2 I1z I2 x The overall result is that antiphase magnetization of spin 1 has been transferred into antiphase magnetization of spin 2. Evolution under this Hamiltonian can be considered by applying the two terms sequentially (the order is immaterial) I1 x Hfreet → I1 x Ω1tI1 z +Ω2tI2 z → tI I1 x Ω1→ Ω2tI2 z → 1 z The first "arrow" is a rotation about z tI I1 x Ω1→ cos Ω1t I1 x + sin Ω1t I1 y Ω2tI2 z → 1 z The second arrow leaves the intermediate state unaltered as spin2 operators have not effect on spin1 operators. Overall.
4 Spin echoes It was shown in section 6.1 Spin echoes in homonuclear spin system In this kind of spin echo the 180° pulse affects both spins i. to spin 1 and spin 2) – τ – At the start of the sequence it will be assumed that only inphase xmagnetization on spin 1 is present: I1 x . A delay of 1/J12 causes inphase magnetization to change its sign: πJ12t I1 z I2 z t =1 2 J12 I1 x 2→ 2 I1 y I2 z πJ12t I1 z I2 z t =1 J I2 y 2 12 → − I2 y 6. half the angle for the other rotations. using diagram V: πJ12t I1 z I2 z 2 I1 x I2 z 2→ cos πJ12 t 2 I1 x I2 z + sin πJ12 t I1 y The diagrams apply equally well to spin2.6 that the spin echo applied to one spin refocuses the offset.2. In this section it will be shown that the evolution of the scalar coupling is not necessarily refocused. I–III.2.IV x V y –yz zz yz xz zz –xz –x angle = πJt –y For example.4. inphase magnetization along x becomes antiphase along y according to the diagram d πJ12t I1 z I2 z I1 x 2→ cos πJ12 t I1 x + sin πJ12 t 2 I1 y I2 z note that the angle is πJ12t i.e. it is a nonselective pulse: – τ – 180°(x.e. for example πJ12t I1 z I2 z −2 I1z I2 y 2→ – cos πJ12 t 2 I1z I2 y + sin πJ12 t I2 x Complete interconversion of inphase and antiphase magnetization requires a delay such that πJ12 t = π 2 i. and this can be calculated by applying the 6–9 . For the first delay τ only the effect of evolution under coupling need be considered therefore: πJ12τ I1 z I2 z I1 x 2→ cos πJ12τ I1 x + sin πJ12τ 2 I1 y I2 z The 180° pulse affects both spins. this conclusion is not altered by the presence of a coupling so the offset will be ignored in the present calculation.6 that the offset is refocused in a spin echo. a delay of 1/(2J12). 6. In fact the starting state is not important to the overall effect of the spin echo.e. so this choice is arbitrary. It was shown in section 6. Antiphase magnetization along x becomes inphase magnetization along y. This greatly simplifies things.
6. for the purposes of the calculation it is permissible to ignore the 180° pulse and simply allow the coupling to evolve for 2τ.2 Interconverting inphase and antiphase states So far.1. spin echoes have been demonstrated as being useful for generating antiphase terms. to spin 1 and spin 2) in which the offset is ignored and coupling is allowed to act for time 2τ. The final result can therefore just be written down: −180° ( x )−τ I1 x τ→ cos 2πJ12τ I1 x + sin 2πJ12τ 2 I1 y I2 z From this it is easy to see that complete conversion to antiphase magnetization requires 2πJ12τ = π/2 i. The 180° rotation about x for spin 1 has no effect on the operator I1 x and I2 z. 6–10 . for a 180° pulse. and it simply reverses the sign of the operator I1 y I1 x I2 x cos πJ12τI1 x + sin πJ12τ 2 I1 y I2 z π→ cos πJ12τI1 x − sin πJ12τ 2 I1 y I2 z π→ The 180° rotation about x for spin 2 has no effect on the operators I1 x and I1 y . but the final result is just the same – the coupling evolves for 2τ: −180° ( x )−τ I1 y τ→ − cos 2πJ12τ I1 y + sin 2πJ12τ 2 I1 x I2 z In fact. to spin 1 and spin 2) – τ – is equivalent to the sequence – 2τ – 180°(x.180° rotation to each in succession I1 x I2 x cos πJ12τ I1 x + sin πJ12τ 2 I1 y I2 z π→ π→ where it has already been written in that ω1 tp = π. τ = 1/(4 J12). Equally useful is the sequence – 1/(4J12) – 180°(x) – 1/(4J12) – which will convert pure antiphase magnetization. The final result is thus I1 x cos πJ12τI1 x + sin πJ12τ 2 I1 y I2 z π→ cos πJ12τI1 x − sin πJ12τ 2 I1 y I2 z I2 x π→ cos πJ12τI1 x + sin πJ12τ 2 I1 y I2 z Nothing has happened. I1 y . For example. The calculation is not quite as simple if the initial state is chosen as I1 y . but simply reverses the sign of the operator I2 z.e. independent of offsets. such as 2I1 x I2 z into inphase magnetization. the general result is that the sequence – τ – 180°(x. the 180° pulse has left the operators unaffected! So. the sequence 90°(x) – 1/(4J12) – 180°(x) – 1/(4J12) – generates pure antiphase magnetization.
6. The evolution of the coupling is easy to calculate: πJ12τ I1 z I2 z I1 x 2→ cos πJ12τ I1 x + sin πJ12τ 2 I1 y I2 z This time the 180°(x) pulse is applied to spin 2 I2 x cos πJ12τ I1 x + sin πJ12τ 2 I1 y I2 z π→ cos πJ12τ I1 x − sin πJ12τ 2 I1 y I2 z The results is exactly as for sequence b (Eqn. it is possible to choose to apply the 180° pulse to either or both spins. Sequence a has already been analysed: the result is that the offset is refocused but that the coupling evolves for time 2τ. so the final result is the same i.5] Three different spin echo sequences that can be applied to heteronuclear spin systems. such as 13C and 1 H. the outcome of the sequence depends on the pattern of 180° pulses.1.e.3 Spin echoes in heteronuclear spin systems If spin 1 and spin 2 are different nuclear species. and attention will therefore be restricted to the effect of the coupling I1 x → cos πJ12τ I1 x + sin πJ12τ 2 I1 y I2 z The 180°(x) pulse is only applied to spin 1 I1 x cos πJ12τ I1 x + sin πJ12τ 2 I1 y I2 z π→ cos πJ12τ I1 x − sin πJ12τ 2 I1 y I2 z 2 πJ12τ I1 z I2 z a spin 1 spin 2 b spin 1 spin 2 c spin 1 spin 2 τ τ τ τ τ τ [6. Sequence b still refocuses the offset of spin 1. In words. The two terms on the right each evolve under the coupling during the second delay: πJ12τ I1 z I2 z cos πJ12τ I1 x 2→ cos πJ12τ cos πJ12τ I1 x + sin πJ12τ cos πJ12τ 2 I1 y I2 z πJ12τ I1 z I2 z − sin πJ12τ 2 I1 y I2 z 2→ − cos πJ12τ sin πJ12τ 2 I1 y I2 z + sin πJ12τ sin πJ12τ I1 x Collecting the terms together and noting that cos 2 θ + sin 2 θ = 1 the final result is just I1 x . the effect of the coupling has been refocused.5]). Sequence c refocuses the coupling but leaves the evolution of the offset unaffected. the offset of spin 1 is not refocused. [6. The open rectangles represent 180° pulses. Sequence b It will be assumed that the offset is refocused. but it turns out that the coupling is also refocused. but continues to evolve for time 2τ. Summary In heteronuclear systems it is possible to choose whether or not to allow the offset and the coupling to evolve. 6–11 . the coupling is refocused. Sequence c As there is no 180° pulse applied to spin 1. this gives great freedom in generating and manipulating antiphase states which play a key role in multiple pulse NMR experiments.
coherence order is a signed quantity. Other product operators can also be classified according to coherence order e. from Eqn. So.7] from which it is seen that Ix and Iy are both mixtures of coherences with p = +1 and –1. The operator product 2I1 x I2 x can be expressed in terms of the raising and lowering operators in the following way (note that separate operators are used for each spin: I1± and I2± ) 2 I1 x I2 x = 2 × 1 ( I1+ + I1− ) × 1 ( I2+ + I2− ) 2 2 = 1 2 ( I1+ I2+ + I1− I2− ) + 1 ( I1+ I2− + I1− I2+ ) 2 [6.8]. 2I1 zI2 y has p = 0 for spin 1 and p = ±1 for spin 2.8] The first term on the right of Eqn. p = ±1. respectively. or more generally coherences with order. In heteronuclear systems it is sometimes useful to classify operators according to their coherence orders with respect to each spin.1 Coherence order In NMR the directly observable quantity is the transverse magnetization. for example. 6. The pure double quantum part of 2I1 x I2 x is. The value of p can be found simply by noting the number of raising and lowering operators in the product.5. [6.8] has p = (+1+1) = 2 and the second term has p = (–1–1) = –2.9] can be reexpressed in terms of the Cartesian operators: 6–12 .g. Only single quantum coherences are observable. [6.5 Multiple quantum terms 6. which in product operators is represented by terms such as I1 x and 2I1 zI2 y . Such terms are examples of single quantum coherences. [6. I+ has coherence order +1 and I– has coherence order –1.6. These are defined as follows I+ = I x + iI y I− = I x − iI y [6. The third and fourth terms both have p = (+1–1) = 0 and are zero quantum coherences. [6.2 Raising and lowering operators The classification of operators according to coherence order is best carried out be reexpressing the Cartesian operators Ix and Iy in terms of the raising and lowering operators.9] The raising and lowering operators on the right of Eqn. 2I1 zI2 z has p = 0 and 2I1 x I2 y has both p = 0 (zeroquantum coherence) and ±2 (double quantum coherence). double quantum part[2 I1 x I2 x ] = 1 2 ( I1+ I2+ + I1− I2− ) [6. Using the definitions of Eqn.6] where i is the square root of –1.5. both are double quantum coherences. I+ and I– .6] Ix and Iy can be expressed in terms of the raising and lowering operators Ix = 1 2 ( I+ + I− ) Iy = 1 2i ( I+ – I− ) [6.
As in the case of two spins. No really new features arise. but antiphase with respect to the coupling to spin 3. the presence of more than one transverse operator in the product represents multiple quantum coherence. by a similar 2 method the pure zero quantum part can be shown to be 1 2 I1 x I2 x − 2 I1 y I2 y . The description will assume that spin 1 is coupled to spins 2 and 3 with coupling constants J12 and J13.6 Three spins The product operator formalism can be extended to three or more spins.6. The evolution under coupling follows the same rules as for a twospin system. The product 4I1 x I2 x I3 z is the same mixture. amongst other things. Products such as 4I1 x I2 x I3 x contain.2 Evolution Evolution under offsets and pulses is simply a matter of applying sequentially the relevant rotations for each spin. but some of the key ideas will be highlighted in this section. evolution of I1 x under the influence of the coupling to spin 3 generates 2I1 y I3 z πJ13tI1 z I1 x 2I3 z → cos πJ13t I1 x + sin πJ13t 2 I1 y I3 z Further evolution of the term 2I1 y I3 z under the influence of the coupling to spin 2 generates a double antiphase term 6–13 .6. triplequantum coherences.9 ( ( ) ) 6.1 2 ( I1+ I2+ + I1− I2− ) = 1 [( I1x + iI1y )( I2 x + iI2 y ) + ( I1x − iI1y )( I2 x − iI2 y )] 2 = 1 2 [2 I 1x 2 x I + 2 I1 y I2 y ] So. 6. For example. affects only the spin 1 operator I1 x . 6. remembering that rotations of spin 1 do not affect operators of spins 2 and 3. I1x 2I1xI2z 2I1xI3z 4I1xI2zI3z Representations of types of operators.and zeroquantum coherence between spins 1 and 2. For example. The third arrow also has no effect as there are no spin 3 operators present. in the diagrams it will be assumed that J12 > J13.1 Types of operators I1 x represents inphase magnetization on spin 1. 2 Some further useful relationships are given in section 6. 2I1 x I2 x is a mixture of double. 4I1 x I2 zI3 z represents magnetization which is doubly antiphase with respect to the couplings to both spins 2 and 3. For example. The second arrow has no effect as the spin 2 operator I2 z and this is unaffected by a zrotation. The spin states of the coupled spins are also indicated. the pure double quantum part of 2I1 x I2 x is 1 2 I1 x I2 x + 2 I1 y I2 y . 2I1 x I2 z represents magnetization antiphase with respect to the coupling to spin 2 and 2I1 x I3 z represents magnetization antiphase with respect to the coupling to spin 3. the term 2I1 x I2 z evolves under the offset in the following way: tI tI 2 I1 x I2 z Ω1→ Ω2tI2 z → Ω3→ cos Ω1t 2 I1 x I2 z + sin Ω1t 2 I1 y I2 z 1 z 3 z J12 J13 Ω1 α α α β β β spin 2 α β spin 3 The doublet of doublets from spin 1 coupled to two other spins. representing evolution under the offset of spin 1. different The first arrow.
NMR Spectrosc. The connection with the evolution of I1 y under a coupling can be made more explicit by writing 2I3 z as a "constant" γ πJ12tI1 z γ I1 y 2I2 z → cos πJ12 t γ I1 y − sin πJ13t 2γ I1 x I2 z which compares directly to πJ12 tI1 z I1 y 2I2 z → cos πJ12 t I1 y − sin πJ13t 2 I1 x I2 z 6. but significant extra complexity is introduced. In heteronuclear experiments a notation is sometimes used where the letter represents the nucleus. So. 2. it is important to try to simplify the calculation as much as possible.7 Alternative notation In this chapter different spins have been designated with a subscript 1. (Prog. for example. These can be very labour saving. The main difficulty with the product operator method is that the more pulses and delays that are introduced the greater becomes the number of operators and the more complex the trigonometrical expressions multiplying them. although it is often convenient (and tidy) always to write them in the same sequence. As will be seen in chapter 7. It can be extended to higher spins. For example.. carbon13 atoms the letter C and nitrogen15 atoms the letter N.. operators referring to protons are given the letter H. commonly I and S are used for two of the symbols 2 I1 x I2 z ≡ 2 I x Sz Note that the order in which the operators are written is not important. 16.8 Conclusion The product operator method as described here only applies to spinhalf nuclei. 4Cx HzNz denotes magnetization on carbon13 which is antiphase with respect to coupling to both proton and nitrogen15. for example by recognizing when offsets or couplings are refocused by spin echoes. 6–14 . 3 . A number of computer programs are available for machine computation using product operators within programs such as Mathematica or Maple. details can be found in the article by Sørensen et al. 6. carbonyl carbons are sometimes denoted C'. If pulses are either 90° or 180° then there is some simplification as such pulses do not increase the number of terms. Another common notation is to distinguish the spins by using a different letter to represent their operators.πJ12tI1 z 2 I1 y I3 z 2I2 z → cos πJ12 t 2 I1 y I3 z − sin πJ13t 4 I1 x I2 z I3 z In this evolution the spin 3 operator is unaffected as the coupling does not involve this spin. 163 (1983)).
and zeroquantum coherence in which spins i and j are active it is convenient to define the following set of operators which represent pure multiple quantum states of given order. passive spins contribute only zoperators.2 1 2 1 2 (2 I (2 I (2 I (2 I ix jx I − 2 Iiy I jy ≡ I + 2 Iiy I jx ix jy ) (I I ) ≡ (I I 1 2 1 2i 1 2 1 2i i+ j + i+ j + + Ii − I j − − Ii − I j − + Ii − I j + ) ) zero quantum. Active spins contribute transverse operators.and zeroquantum operators evolve under passive couplings in a way which is entirely analogous to the evolution of Ix and Iy . The operators can be expressed in terms of the Cartesian or raising and lowering operators. Double.9 Multiple quantum coherence 6.6. For double. Iy and I+. the resulting multiple quantum terms can be described as being antiphase with respect to the effective couplings: 6–15 .and zeroquantum operators evolve under offsets in a way which is entirely analogous to the evolution of Ix and Iy under free precession except that the frequencies of evolution are (Ωi + Ωj) and (Ωi – Ωj) respectively: ( ) ( ) DQ ( ) → cos(Ω + Ω )t DQ ( ) − sin(Ω + Ω )t DQ ( ) ZQ ( ) → cos(Ω − Ω )t ZQ ( ) + sin(Ω − Ω )t ZQ ( ) ZQ ( ) → cos(Ω − Ω )t ZQ ( ) − sin(Ω − Ω )t ZQ ( ) Ω tI + Ω tI ij y i iz j jz DQ (xij ) → cos Ω i + Ω j t DQ (xij ) + sin Ω i + Ω j t DQ (yij ) Ωi tIiz + Ω j tI jz i j ij y i j ij x ij x Ωi tIiz + Ω j tI jz Ωi tIiz + Ω j tI jz i j ij x i j ij y ij y i j ij y i j ij x Evolution under couplings Multiple quantum coherence between spins i and j does not evolve under the influence of the coupling between the two active spins. In a sense the spins contributing transverse operators are "involved" in the coherence. Iz. i and j. to the product. double quantum.9. p = 0 1 2 1 2 ix jx I + 2 Iiy I jy ≡ iy jx I − 2 Iix I jy ) (I I ) ≡ (I I i+ j − i+ j − − Ii − I j + ) ) Evolution of multiple quantum terms Evolution under offsets The double.9.1 Multiplequantum terms In the product operator representation of multiple quantum coherences it is usual to distinguish between active and passive spins. such as Ix . p = ±2 DQ (xij ) ≡ DQ (yij ) ≡ ZQ (xij ) ≡ ZQ (yij ) ≡ 6. while those contributing zoperators are simply spectators.
antiphase with respect to spin 3.eff t DQ (yij ) − sin πJDQ.eff t 2 Ikz DQ (yij ) DQ (yij ) → cos πJDQ.eff t 2 Ikz DQ (xij ) ZQ (xij ) → cos πJ ZQ.eff t ZQ (xij ) + sin πJ ZQ.eff is the sum of the couplings between spin i and all other spins minus the sum of the couplings between spin j and all other spins. 6–16 . 1987).3 of Ernst. J ZQ. t ZQ (yij ) − sin πJ ZQ. Bodenhausen and Wokaun Principles of NMR in One and Two Dimensions (Oxford University Press.eff t 2 Ikz ZQ (yij ) ZQ (yij ) → cos πJ ZQ.eff t 2 I3 z ZQ (y12 ) − sin πJZQ. evolves according to 2 I3 z ZQ (y12 ) → cos πJZQ.eff t ZQ (x12 ) where JZQ.eff t 2 Ikz ZQ (xij ) ZQ. DQ (xij ) → cos πJDQ.eff = J13 − J23 Further details of multiplequantum evolution can be found in section 5.eff t DQ (xij ) + cos πJDQ.eff JDQ.eff is the sum of the couplings between spin i and all other spins plus the sum of the couplings between spin j and all other spins. For example in a threespin system the zeroquantum coherence between spins 1 and 2.
one crosspeak multiplet centred at F1 = δA .1 Introduction The basic ideas of twodimensional NMR will be introduced by reference to the appearance of a COSY spectrum. which are coupled together. The figure shows a schematic COSY spectrum of a hypothetical molecule containing just two protons. in twodimensional spectroscopy intensity is plotted as a function of two frequencies. in the same way as a topographical map. usually called F1 and F2. F 2 = δX and a second crosspeak multiplet centred at F1 = δX. In twodimensional spectra the idea of a multiplet has to be expanded somewhat so that in such spectra a multiplet consists of an array of individual peaks often giving the impression of a square or rectangular outline. F2 = δX indicates that the two protons at shifts δA and δX have a scalar coupling † δX δA δA δX Schematic COSY spectrum for two coupled spins. A and X Chapter 7 "TwoDimensional NMR" © James Keeler 1998 and 2002 7–1 . These twodimensional multiplets come in two distinct types: diagonalpeak multiplets which are centred around the same F 1 and F 2 frequency coordinates and crosspeak multiplets which are centred around different F1 and F 2 coordinates. It is immediately clear that this COSY spectrum has some symmetry about the diagonal F1 = F2 which has been indicated with a dashed line. The appearance in a COSY spectrum of a crosspeak multiplet F1 = δA. the F 1 coordinates of the peaks in the twodimensional spectrum also correspond to those found in the normal onedimensional spectrum and to emphasize this point the onedimensional spectrum has been plotted alongside the F1 axis. In a onedimensional spectrum scalar couplings give rise to multiplets in the spectrum. There are various ways of representing such a spectrum on paper.7 Twodimensional NMR† 7. Conventional NMR spectra (onedimensional spectra) are plots of intensity vs. Several such arrays of peaks can be seen in the schematic COSY spectrum shown above. The onedimensional spectrum is plotted alongside the F2 axis. A and X. and consists of the familiar pair of doublets centred on the chemical shifts of A and X. The position of each peak is specified by two frequency coordinates corresponding to F1 and F2. δA and δX respectively. later in this chapter the product operator formalism will be used to predict the form of the spectrum. In the COSY spectrum. and this relation is often emphasized by plotting the onedimensional spectrum alongside the F2 axis. but the one most usually used is to make a contour plot in which the intensity of the peaks is represented by contour lines drawn at suitable intervals. Twodimensional NMR spectra are always arranged so that the F2 coordinates of the peaks correspond to those found in the normal onedimensional spectrum. frequency. Thus in the schematic COSY spectrum there are two diagonalpeak multiplets centred at F1 = F2 = δA and F1 = F2 = δX . F2 = δA.
called the preparation time. It is important to realize that the signal is not recorded during the time t1. After the mixing period the signal is recorded as a function of the second time variable.1 General Scheme for twoDimensional NMR In onedimensional pulsed Fourier transform NMR the signal is recorded as a function of one time variable and then Fourier transformed to give a spectrum which is a function of one frequency variable. The twodimensional signal is recorded in the following way. First. which consists of a further pulse or pulses. From a single COSY spectrum it is possible to trace out the whole coupling network in the molecule 7.1. 4∆1 and so on until sufficient data is recorded. the pulse sequence is executed and the resulting free induction decay recorded. The whole process is repeated again for t1 = 3∆1.between them. 7. the sample is excited by one or more pulses. This statement is all that is required for the analysis of a COSY spectrum. the sampling interval in t1. Thus recording a twodimensional data set involves repeating a pulse sequence for increasing values of t1 and recording a free induction decay as a function of t2 for each value of t1. t1 and t2. t1. called the mixing time. t2.2 Interpretation of peaks in a twodimensional spectrum Within the general framework outlined in the previous section it is now possible to interpret the appearance of a peak in a twodimensional spectrum at particular frequency coordinates. Again the spins are allowed to equilibrate. the sequence is repeated and a free induction decay is recorded and stored separately from the first.1. t1 is set to zero. The resulting magnetization is allowed to evolve for the first time period. but only during the time t2 at the end of the sequence. Then another period follows. and the resulting data Fourier transformed twice to yield a spectrum which is a function of two frequency variables. t1 is set to 2 ∆1. typically 50 to 500 increments of t1. This sequence of events is called a pulse sequence and the exact nature of the preparation and mixing periods determines the information found in the spectrum. and it is this simplicity which is the key to the great utility of such spectra. 7–2 . Then the nuclear spins are allowed to return to equilibrium. The data is recorded at regularly spaced intervals in both t1 and t2. The general scheme for twodimensional spectroscopy is preparation evolution t1 mixing detection t2 In the first period. t1 is then set to ∆1. In twodimensional NMR the signal is recorded as a function of two time variables. the pulse sequence repeated and a free induction decay recorded and stored.
F2 = 80 Hz (spectrum a above) The interpretation of this peak is that a signal was present during t1 which evolved with a frequency of 20 Hz. 1 and 2. The interpretation of this is that some signal was present during t1 which evolved at 20 Hz and that during the mixing period part of it was transferred into another signal which evolved at 80 Hz during t2. The initial state. consider the spectrum shown in c. Likewise. In the absence of a mixing time. To obtain an interesting and useful spectrum it is essential to arrange for some process during the mixing time to transfer signals from one spin to another. Finally. before the first pulse. if there is a peak at F1 = 20 Hz. F2 = 20 Hz (spectrum b) the interpretation is that there was a signal evolving at 20 Hz during t1 which was unaffected by the mixing period and continued to evolve at 20 Hz during t2. It is clear from the discussion in this section that the mixing time plays a crucial role in forming the twodimensional spectrum. Here there are two peaks. During the mixing time this same signal was transferred in some way to another signal which evolved at 80 Hz during t2. one at F1 = 20 Hz. The pulse sequence for EXSY is shown opposite.2 EXSY and NOESY spectra in detail In this section the way in which the EXSY (EXchange SpectroscopY) sequence works will be examined. 7. F2 = δB indicates that the spin resonating at δA is chemically exchanging with the spin resonating at δB. This kind of interpretation is a very useful way of thinking about the origin of peaks in a twodimensional spectrum.0 F2 80 Suppose that in some unspecified twodimensional spectrum a peak appears at F1 = 20 Hz. is equilibrium 7–3 t1 τmix t2 The pulse sequence for EXSY (and NOESY). On the basis of the previous discussion of COSY spectra.a b c F1 20 0. . the pulse sequence is shown opposite. but without any coupling between them. This experiment gives a spectrum in which a crosspeak at frequency coordinates F1 = δA . the part that changes frequency during the mixing time is recognized as leading to a crosspeak and the part that does not change frequency leads to a diagonalpeak. The processes by which these signals are transferred will be discussed in the following sections. The other part remained unaffected and continued to evolve at 20 Hz. The effect of the sequence will be analysed for the case of two spins. the frequencies that evolve during t1 and t2 would be the same and only diagonalpeaks would appear in the spectrum. F2 = 80 Hz and one at F1 = 20 Hz. F2 = 20 Hz. All pulses have 90° flip angles.
magnetization, represented as I 1z + I2z; however, for simplicity only magnetization from the first spin will be considered in the calculation. The first 90° pulse (of phase x) rotates the magnetization onto –y
2 2 I1z πI1 x → πI2 x → − I1 y
(the second arrow has no effect as it involves operators of spin 2). Next follows evolution for time t1 − I1 y Ω1t1→ Ω2t1 I2 z → − cos Ω1t1 I1 y + sin Ω1t1 I1 x I1 z again, the second arrow has no effect. The second 90° pulse turns the first term onto the zaxis and leaves the second term unaffected
2 2 − cos Ω1t1 I1 y πI1 x → πI2 x → − cos Ω1t1 I1z 2 2 sin Ω1t1 I1 x πI1 x → πI2 x → sin Ω1t1 I1 x
Only the I1z term leads to crosspeaks by chemical exchange, so the other term will be ignored (in an experiment this is achieved by appropriate coherence pathway selection). The effect of the first part of the sequence is to generate, at the start of the mixing time, τmix, some zmagnetization on spin 1 whose size depends, via the cosine term, on t1 and the frequency, Ω1, with which the spin 1 evolves during t1. The magnetization is said to be frequency labelled. During the mixing time, τmix, spin 1 may undergo chemical exchange with spin 2. If it does this, it carries with it the frequency label that it acquired during t1. The extent to which this transfer takes place depends on the details of the chemical kinetics; it will be assumed simply that during τmix a fraction f of the spins of type 1 chemically exchange with spins of type 2. The effect of the mixing process can then be written − cos Ω1t1 I1z mixing → −(1 − f ) cos Ω1t1 I1z − f cos Ω1t1 I2 z The final 90° pulse rotates this zmagnetization back onto the yaxis
2 2 −(1 − f ) cos Ω1t1 I1z πI1 x → πI2 x →(1 − f ) cos Ω1t1 I1 y 2 2 − f cos Ω1t1 I2 z πI1 x → πI2 x → f cos Ω1t1 I2 y
Although the magnetization started on spin 1, at the end of the sequence there is magnetization present on spin 2 – a process called magnetization transfer. The analysis of the experiment is completed by allowing the I1y and I2y operators to evolve for time t2. (1 − f ) cos Ω1t1 I1y Ω t I → Ω t I → (1 − f ) cos Ω1t2 cos Ω1t1 I1y − (1 − f ) sin Ω1t2 cos Ω1t1 I1x
1 2 1z 2 2 2z
f cos Ω1t1 I2 y Ω1t 2 I1 z → Ω 2 t 2 I 2 z → f cos Ω 2 t2 cos Ω1t1 I2 y − f sin Ω 2 t2 cos Ω1t1 I2 x If it is assumed that the ymagnetization is detected during t2 (this is an arbitrary choice, but a convenient one), the time domain signal has two terms: 7–4
(1 − f ) cos Ω1t2 cos Ω1t1 +
f cos Ω 2 t2 cos Ω1t1
The crucial thing is that the amplitude of the signal recorded during t2 is modulated by the evolution during t1. This can be seen more clearly by imagining the Fourier transform, with respect to t2, of the above function. The cos Ω1t2 and cos Ω 2 t2 terms transform to give absorption mode signals centred ( at Ω1 and Ω 2 respectively in the F2 dimension; these are denoted A1( 2 ) and A22 ) (the subscript indicates which spin, and the superscript which dimension). The time domain function becomes
(1 − f ) A1( 2 ) cos Ω1t1 +
( fA22 ) cos Ω1t1
If a series of spectra recorded as t1 progressively increases are inspected it would be found that the cos Ω1t2 term causes a change in size of the peaks at Ω1 and Ω2 – this is the modulation referred to above. Fourier transformation with respect to t1 gives peaks with an absorption lineshape, but this time in the F1 dimension; an absorption mode signal at Ω1 in F1 is denoted A1(1) . The time domain signal becomes, after Fourier transformation in each dimension
(1 − f ) A1( 2 ) A1(1) +
( fA22 ) A1(1)
Thus, the final twodimensional spectrum is predicted to have two peaks. One is at (F1, F2 ) = (Ω1, Ω1) – this is a diagonal peak and arises from those spins of type 1 which did not undergo chemical exchange during τmix. The second is at (F1, F2 ) = (Ω1, Ω ) – this is a cross peak which indicates that part of the 2 magnetization from spin 1 was transferred to spin 2 during the mixing time. It is this peak that contains the useful information. If the calculation were repeated starting with magnetization on spin 2 it would be found that there are similar peaks at (Ω2, Ω2) and (Ω2, Ω1). The NOESY (Nuclear Overhauser Effect SpectrocopY) spectrum is recorded using the same basic sequence. The only difference is that during the mixing time the crossrelaxation is responsible for the exchange of magnetization between different spins. Thus, a crosspeak indicates that two spins are experiencing mutual crossrelaxation and hence are close in space. Having completed the analysis it can now be seen how the EXCSY/NOESY sequence is put together. First, the 90° – t1 – 90° sequence is used to generate frequency labelled zmagnetization. Then, during τmix, this magnetization is allowed to migrate to other spins, carrying its label with it. Finally, the last pulse renders the zmagnetization observable. 7.3 More about twodimensional transforms From the above analysis it was seen that the signal observed during t2 has an amplitude proportional to cos(Ω1t1); the amplitude of the signal observed during t2 depends on the evolution during t1. For the first increment of t1 (t1 = 0), the signal will be a maximum, the second increment will have size proportional to cos(Ω1∆1), the third proportional to cos(Ω12∆1), the fourth to cos(Ω13∆1) and so 7–5
time
Fourier transform
Ω
frequency
The Fourier transform of a decaying cosine function cosΩt e x p ( –t/T2) is an absorption mode Lorentzian centred at frequency Ω; the real part of the spectrum has been plotted.
on. This modulation of the amplitude of the observed signal by the t1 evolution is illustrated in the figure below. In the figure the first column shows a series of free induction decays that would be recorded for increasing values of t1 and the second column shows the Fourier transforms of these signals. The final step in constructing the twodimensional spectrum is to Fourier transform the data along the t1 dimension. This process is also illustrated in the figure. Each of the spectra shown in the second column are represented as a series of data points, where each point corresponds to a different F2 frequency. The data point corresponding to a particular F2 frequency is selected from the spectra for t1 = 0, t1 = ∆ 1, t1 = 2 ∆1 and so on for all the t1 values. Such a process results in a function, called an interferogram, which has t1 as the running variable.
Illustration of how the modulation of a free induction decay by evolution during t1 gives rise to a peak in the
7–6
twodimensional spectrum. In the left most column is shown a series of free induction decays that would be recorded for successive values of t1; t1 increases down the page. Note how the amplitude of these free induction decays varies with t1, something that becomes even plainer when the time domain signals are Fourier transformed, as shown in the second column. In practice, each of these F2 spectra in column two consist of a series of data points. The data point at the same frequency in each of these spectra is extracted and assembled into an interferogram, in which the horizontal axis is the time t1. Several such interferograms, labelled a to g, are shown in the third column. Note that as there were eight F2 spectra in column two corresponding to different t1 values there are eight points in each interferogram. The F2 frequencies at which the interferograms are taken are indicated on the lower spectrum of the second column. Finally, a second Fourier transformation of these interferograms gives a series of F1 spectra shown in the right hand column. Note that in this column F2 increases down the page, whereas in the first column t1 increase down the page. The final result is a twodimensional spectrum containing a single peak.
Several interferograms, labelled a to g, computed for different F2 frequencies are shown in the third column of the figure. The particular F2 frequency that each interferogram corresponds to is indicated in the bottom spectrum of the second column. The amplitude of the signal in each interferogram is different, but in this case the modulation frequency is the same. The final stage in the processing is to Fourier transform these interferograms to give the series of spectra which are shown in the right most column of the figure. These spectra have F1 running horizontally and F2 running down the page. The modulation of the time domain signal has been transformed into a single twodimensional peak. Note that the peak appears on several traces corresponding to different F2 frequencies because of the width of the line in F2. The time domain data in the t1 dimension can be manipulated by multiplying by weighting functions or zero filling, just as with conventional free induction decays. 7.4 Twodimensional experiments using coherence transfer through Jcoupling Perhaps the most important set of twodimensional experiments are those which transfer magnetization from one spin to another via the scalar coupling between them. As was seen in section 6.3.3, this kind of transfer can be brought about by the action of a pulse on an antiphase state. In outline the basic process is I1 x coupling → 2 I1 y I2 z 90°( x ) to both spins → 2 I1z I2 y spin 1 spin 2
7.4.1 COSY The pulse sequence for this experiment is shown opposite. It will be assumed in the analysis that all of the pulses are applied about the xaxis and for simplicity the calculation will start with equilibrium magnetization only on spin 1. The effect of the first pulse is to generate ymagnetization, as has been worked out previously many times
2 2 I1z πI1 x → πI2 x → − I1 y
t1
t2
Pulse sequence for the twodimensional COSY experiment
This state then evolves for time t1, first under the influence of the offset of spin 1 (that of spin 2 has no effect on spin 1 operators): − I1 y Ω1t1→ − cos Ω1t1 I1 y + sin Ω1t1 I1 x I1 z 7–7
In addition. the lineshape will be dispersive. 2 I1 y I2 z . The squares are supposed to indicate the twodimensional double dispersion lineshape illustrated below Two peaks in F1 are expected at Ω1 ± πJ12 . Term {4} is the really interesting one. Consider again term {3}: it will give rise to an inphase multiplet in F2. remembering that the effect of the pulse on both spins needs to be computed. cos A sin B = 1 {sin( B + A) + sin( B − A)} to give 2 cos πJ12 t1 sin Ω1t1 = 1 2 {sin(Ω t 1 1 + πJ12 t1 ) + sin(Ω1t1 − πJ12 t )} J12 J12 F1 F2 Schematic view of the diagonal peak from a COSY spectrum. as shown opposite. 2 I1z I2 y . how cross. Some more consideration should be give to the form of the cross.Ω1) is predicted. this is an example of coherence transfer. Terms {1} and {2} are unobservable. and as it is along the xaxis. It has been shown. is transferred to antiphase magnetization on spin 2.and diagonalpeaks arise in a COSY spectrum. so the appearance of the twodimensional multiplet can best be found by "multiplying together" the multiplets in the two dimensions. so a diagonal peak centred at (Ω1. and have the double dispersion lineshape illustrated below 7–8 .Both terms on the right then evolve under the coupling πJ12t1I1 z I2 z − cos Ω1t1 I1 y 2→ − cos πJ12 t1 cos Ω1t1 I1 y + sin πJ12 t1 cos Ω1t1 2 I1 x I2 z πJ12t1I1 z I2 z sin Ω1t1 I1 x 2→ cos πJ12 t1 sin Ω1t1 I1 x + sin πJ12 t1 sin Ω1t1 2 I1 y I2 z That completes the evolution under t1. but it is modulated in t1 with the offset of spin 1. therefore. Note that both components in the spin 1 multiplet observed in F2 are modulated in this way.and diagonal peaks. all four components of the diagonalpeak multiplet have the same sign. Now all that remains is to consider the effect of the final pulse. In addition. Taking the terms one by one: 2I 2I − cos πJ12 t1 cos Ω1t1 I1 y π1 x → π2 x → − cos πJ12 t1 cos Ω1t1 I1z 2I 2I sin πJ12 t1 cos Ω1t1 2 I1 x I2 z π1 x → π2 x → − sin πJ12 t1 cos Ω1t1 2 I1 x I2 y {1} {2} {3} {4} time Fourier transform 2I 2I cos πJ12 t1 sin Ω1t1 I1 x π1 x → π2 x → cos πJ12 t1 sin Ω1t1 I1 x 2I 2I sin πJ12 t1 sin Ω1t1 2 I1 y I2 z π1 x → π2 x → − sin πJ12 t1 sin Ω1t1 2 I1z I2 y Ω frequency The Fourier transform of a decaying sine function sinΩt exp(–t/T2) is a dispersion mode Lorentzian centred at frequency Ω.Ω2). It shows that antiphase magnetization on spin 1. Term {3} corresponds to inphase magnetization of spin 1. these are just the two lines of the spin 1 doublet. The t1 modulation of this term depends on the offset of spin 1. thus it gives rise to a crosspeak centred at (Ω1. The form of the modulation in t1 can be expanded. aligned along the xaxis. Term {4} appears as observable magnetization on spin 2. since these are sine modulated they will have the dispersion lineshape. using the formula.
In F2 we know that this term gives rise to an antiphase absorption multiplet on spin 2. Term {4} can be treated in the same way. negative contours are indicated by dashed lines. the positive and negative peaks in the crosspeak multiplet begin to overlap and cancel one another out. at Ω1 ± πJ12 . where all the sign information is suppressed deliberately. This characteristic pattern of positive and negative peaks that constitutes the crosspeak is know as an antiphase square array. The result is shown opposite. more details of this are given in section 7. COSY spectra are sometimes plotted in the absolute value mode. In addition. especially for routine applications. and ultimately the crosspeak disappears into the noise in the spectrum. As the coupling constant becomes comparable with the linewidth.The double dispersion lineshape seen in pseudo 3D and as a contour plot. just as was done for the diagonalpeak multiplet. filled circles represent positive intensity. these are just the two lines of the spin 1 doublet. are expected.2). The double absorption lineshape seen in pseudo 3D and as a contour plot.6. Using the relationship sin B sin A = 1 {− cos( B + A) + cos( B − A)} the modulation in t1 can be expanded 2 sin πJ12 t1 sin Ω1t = 1 2 {− cos(Ω t 1 1 + πJ12 t1 ) + cos(Ω1t1 − πJ12 t )} Two peaks in F1. Although such a display is convenient. open represent negative intensity. it is generally much more desirable to retain the sign information. Spectra displayed in this way are said to be phase sensitive. The smallest coupling which gives rise to a crosspeak is thus set 7–9 . This leads to an overall reduction in the intensity of the crosspeak multiplet. The form of the crosspeak multiplet can be predicted by "multiplying together" the F1 and F2 multiplets. The circles are supposed to indicate the twodimensional double absorption lineshape illustrated below. Note that the two peaks have opposite signs – that is they are antiphase in F1. J12 J12 F1 F2 Schematic view of the crosspeak multiplet from a COSY spectrum. since these are cosine modulated we expect the absorption lineshape (see section 7.
and diagonalpeak multiplets. as all the observed signals are filtered through doublequantum coherence.5 2 I1 x I2 y = 2 × 1 ( I1+ + I1− ) × 21i ( I2+ − I2− ) 2 = 1 2i t1 t2 The pulse sequence for DQF COSY. The final pulse is needed to convert the double quantum coherence back into observable magnetization. this can be demonstrated explicitly by expanding this term in the raising and lowering operators. it is seen that after the second 90° pulse it is term {2} that contains doublequantum coherence. The pulse sequence is shown opposite. The result is a broad intense diagonal which can obscure nearby crosspeaks. the dispersive tails of these peaks spread far into the spectrum. The second term is antiphase magnetization of spin 2. This effect is particularly troublesome when the coupling is comparable with the linewidth as in such cases. However. this term can be reexpressed in Cartesian operators: 1 2i ( I1+ I2+ − I1− I2− ) = 21i [( I1x + iI1y )( I1x + iI1y ) + ( I2 x − iI2 y )( I2 x − iI2 y )] = 1 2 [2 I 1x 2 y I + 2 I1 y I2 x ] The effect of the last 90°(x) pulse on the double quantum part of term {2} is thus 2 2 − 1 sin πJ12 t1 cos Ω1t1 2 I1 x I2 y + 2 I1 y I2 x πI1 x → πI2 x → 2 ( − 1 sin πJ12 t1 cos Ω1t1 (2 I1 x I2 z + 2 I1z I2 x ) 2 ) The first term on the right is antiphase magnetization of spin 1 aligned along the xaxis. ( I1+ I2+ − I1− I2− ) + 21i (− I1+ I2− + I1− I2+ ) This term contains both double. "doublequantum filtered". this will give rise to a 7–10 .4. Up to the second pulse the sequence is the same as COSY. as was described above. this gives rise to a diagonalpeak multiplet. the delay between the last two pulses is usually just a few microseconds.and zeroquantum coherence. In the analysis of the COSY experiment. cancellation of antiphase components in the crosspeak multiplet reduces the overall intensity of these multiplets. This doublequantum derived signal is selected by the use of coherence pathway selection using phase cycling or field gradient pulses. again aligned along x. The pure doublequantum part is the term in the first bracket on the right. This difficulty is neatly sidestepped by a modification called double quantum filtered COSY (DQF COSY). 7. The components of a diagonal peak multiplet are all inphase and so tend to reinforce one another. it is arranged that only doublequantum coherence present during the (very short) delay between the second and third pulses is ultimately allowed to contribute to the spectrum. In addition. as was done in section 6.by the linewidth and the signaltonoise ratio of the spectrum. Hence the name.2 Doublequantum filtered COSY (DQF COSY) The conventional COSY experiment suffers from a disadvantage which arises from the different phase properties of the cross.
a given magnetization induces a larger voltage in the coil the higher the NMR frequency becomes. As these peaks all have the same lineshape the overall phase of the spectrum can be adjusted so that they are all in absorption. by the Boltzmann distribution. However.crosspeak multiplet. carbon13) which is coupled to the first nucleus. thus avoiding the strong inphase diagonal peaks found in the simple experiment. simply recording a carbon13 proton correlation experiment will give the assignment of all the protonated carbons.4. 7–11 .g. proton) and the coordinate in the other dimension is the chemical shift of another nucleus (e. these unwanted signals can be suppressed using coherence selection in a way which will be described below. which can be shown. By utilizing this large difference experiments can be devised which give maps of carbon13 shifts vs the shifts of directly attached protons. The onebond coupling between a carbon13 and the proton directly attached to it is relatively constant (around 150 Hz). if the proton spectrum has already been assigned. a greater population difference. and much larger than any of the longrange carbon13 proton couplings. Such spectra are very useful as aids to assignment.3 Heteronuclear correlation experiments One particularly useful experiment is to record a twodimensional spectrum in which the coordinate of a peak in one dimension is the chemical shift of one type of nucleus (e. Such spectra are often called shift correlation maps or shift correlation spectra.and crosspeak multiplets are in antiphase in both dimensions. it is very advantageous from the sensitivity point of view to record protons. Both of these terms have the same modulation in t1. In contrast to the case of a simple COSY experiment both the diagonal. For two reasons. to lead to an antiphase multiplet in F1. by a similar analysis to that used above. First. Carbon13 has a natural abundance of only 1%. the proton magnetization is larger than that of carbon13 because there is a larger separation between the spin energy levels giving. thus 99% of the molecules in the sample do not have any carbon13 in them and so will not give signals that can be used to correlate carbon13 and proton. Trying to record a carbon13 proton shift correlation spectrum by proton observation has one serious difficulty. 7. Only one kind of nuclear species can be observed at a time. see section 7. but the signals from these resonances will be swamped by the much stronger signals from noncarbon13 containing molecules. The 1% of molecules with carbon13 will give a perfectly satisfactory spectrum. for example. The DQF COSY experiment is the method of choice for tracing out coupling networks in a molecule.6 for further details. so there is a choice as to whether to observe carbon13 or proton when recording a shift correlation spectrum.g. Second.
1
H
∆
t1
∆
t2
13
C
The pulse sequence for HMQC. Filled rectangles represent 90° pulses and open rectangles represent 180° pulses. The delay ∆ is set to 1/(2J12 ).
7.4.3.1 Heteronuclear multiplequantum correlation (HMQC) The pulse sequence for this popular experiment is given opposite. The sequence will be analysed for a coupled carbon13 proton pair, where spin 1 will be the carbon13 and spin 2 the proton. The analysis will start with equilibrium magnetization on spin 1, I1z. The whole analysis can be greatly simplified by noting that the 180° pulse is exactly midway between the first 90° pulse and the start of data acquisition. As has been shown in section 6.4, such a sequence forms a spin echo and so the evolution of the offset of spin 1 over the entire period (t1 + 2 ∆) is refocused. Thus the evolution of the offset of spin 1 can simply be ignored for the purposes of the calculation. At the end of the delay ∆ the state of the system is simply due to evolution of the term –I1y under the influence of the scalar coupling: − cos πJ12 ∆ I1 y + sin πJ12 ∆ 2 I1 x I2 z It will be assumed that ∆ = 1/(2J12), so only the antiphase term is present. The second 90° pulse is applied to carbon13 (spin 2) only
2 2 I1 x I2 z πI2 x → −2 I1 x I2 y
This pulse generates a mixture of heteronuclear double and zeroquantum coherence, which then evolves during t1. In principle this term evolves under the influence of the offsets of spins 1 and 2 and the coupling between them. However, it has already been noted that the offset of spin 1 is refocused by the centrally placed 180° pulse, so it is not necessary to consider evolution due to this term. In addition, it can be shown that multiplequantum coherence involving spins i and j does not evolve under the influence of the coupling, Jij, between these two spins. As a result of these two simplifications, the only evolution that needs to be considered is that due to the offset of spin 2 (the carbon13). −2 I1 x I2 y Ω2t1I2 z → − cos Ω 2 t1 2 I1 x I2 y + sin Ω 2 t1 2 I1 x I2 x The second 90° pulse to spin 2 (carbon13) regenerates the first term on the right into spin 1 (proton) observable magnetization; the other remains unobservable
2 − cos Ω 2 t1 2 I1 x I2 y πI2 x → − cos Ω 2 t1 2 I1 x I2 z
This term then evolves under the coupling, again it is assumed that ∆ = 1/(2J12)
πJ12 ∆I1 z I2 z , ∆ =1 ( 2 J12 ) − cos Ω 2 t1 2 I1 x I2 z 2→ − cos Ω 2 t1 I1 y
7–12
This is a very nice result; in F2 there will be an inphase doublet centred at the offset of spin 1 (proton) and these two peaks will have an F1 coordinate simply determined by the offset of spin 2 (carbon13); the peaks will be in absorption. A schematic spectrum is shown opposite. The problem of how to suppress the very strong signals from protons not coupled to any carbon13 nuclei now has to be addressed. From the point of view of these protons the carbon13 pulses might as well not even be there, and the pulse sequence looks like a simple spin echo. This insensitivity to the carbon13 pulses is the key to suppressing the unwanted signals. Suppose that the phase of the first carbon13 90° pulse is altered from x to –x. Working through the above calculation it is found that the wanted signal from the protons coupled to carbon13 changes sign i.e. the observed spectrum will be inverted. In contrast the signal from a proton not coupled to carbon13 will be unaffected by this change. Thus, for each t1 increment the free induction decay is recorded twice: once with the first carbon13 90° pulse set to phase x and once with it set to phase –x. The two free induction decays are then subtracted in the computer memory thus cancelling the unwanted signals. This is an example of a very simple phase cycle. In the case of carbon13 and proton the one bond coupling is so much larger than any of the long range couplings that a choice of ∆ = 1/(2Jone bond) does not give any correlations other than those through the onebond coupling. There is simply insufficient time for the longrange couplings to become antiphase. However, if ∆ is set to a much longer value (30 to 60 ms), longrange correlations will be seen. Such spectra are very useful in assigning the resonances due to quaternary carbon13 atoms. The experiment is often called HMBC (heteronuclear multiplebond correlation). Now that the analysis has been completed it can be seen what the function of various elements in the pulse sequence is. The first pulse and delay generate magnetization on proton which is antiphase with respect to the coupling to carbon13. The carbon13 90° pulse turns this into multiple quantum coherence. This forms a filter through which magnetization not bound to carbon13 cannot pass and it is the basis of discrimination between signals from protons bound and not bound to carbon13. The second carbon13 pulse returns the multiple quantum coherence to observable antiphase magnetization on proton. Finally, the second delay ∆ turns the antiphase state into an inphase state. The centrally placed proton 180° pulse refocuses the proton shift evolution for both the delays ∆ and t1. 7.4.3.2 Heteronuclear singlequantum correlation (HSQC) This pulse sequence results in a spectrum identical to that found for HMQC. Despite the pulse sequence being a little more complex than that for HMQC, HSQC has certain advantages for recording the spectra of large molecules, such a proteins. The HSQC pulse sequence is often embedded in much more complex sequences which are used to record two and threedimensional 7–13
J12
F1
Ω2 Ω1
F2
Schematic HMQC spectrum for two coupled spins.
spectra of carbon13 and nitrogen15 labelled proteins.
y
∆
1
∆
2
∆
2
∆
2
H
2
t2
13
C A
t1 B C
The pulse sequence for HSQC. Filled rectangles represent 90° pulses and open rectangles represent 180° pulses. The delay ∆ is set to 1/(2J12 ); all pulses have phase x unless otherwise indicated.
If this sequence were to be analysed by considering each delay and pulse in turn the resulting calculation would be far too complex to be useful. A more intelligent approach is needed where simplifications are used, for example by recognizing the presence of spin echoes who refocus offsets or couplings. Also, it is often the case that attention can be focused a particular terms, as these are the ones which will ultimately lead to observable signals. This kind of "intelligent" analysis will be illustrated here. Periods A and C are spin echoes in which 180° pulses are applied to both spins; it therefore follows that the offsets of spins 1 and 2 will be refocused, but the coupling between them will evolve throughout the entire period. As the total delay in the spin echo is 1/(2J12) the result will be the complete conversion of inphase into antiphase magnetization. Period B is a spin echo in which a 180° pulse is applied only to spin 1. Thus, the offset of spin 1 is refocused, as is the coupling between spins 1 and 2; only the offset of spin 2 affects the evolution. With these simplifications the analysis is easy. The first pulse generates –I1y ; during period A this then becomes –2I1xI2z. The 90°(y) pulse to spin 1 turns this to 2I1zI2z and the 90°(x) pulse to spin 2 turns it to –2I1zI2y. The evolution during period B is simply under the offset of spin 2 −2 I1z I2 y Ω2t1 I2 z → − cos Ω 2 t1 2 I1z I2 y + sin Ω 2 t1 2 I1z I2 x The next two 90° pulses transfer the first term to spin 1; the second term is rotated into multiple quantum and is not observed
( 1x 2 x ) − cos Ω 2 t1 2 I1z I2 y + sin Ω 2 t1 2 I1z I2 x →
π 2 I +I
− cos Ω 2 t1 2 I1 y I2 z − sin Ω 2 t1 2 I1 y I2 x The first term on the right evolves during period C into inphase magnetization (the evolution of offsets is refocused). So the final observable term is cos Ω 2 t1 I1 x . The resulting spectrum is therefore an inphase doublet in F 2, centred at the offset of spin 1, and these peaks will both have the same frequency in F1, namely the offset of spin 2. The spectrum looks just like the HMQC spectrum. 7.5 Advanced topic: Multiplequantum spectroscopy A key feature of twodimensional NMR experiments is that no direct 7–14
observations are made during t1, it is thus possible to detect, indirectly, the evolution of unobservable coherences. An example of the use of this feature is in the indirect detection of multiplequantum spectra. A typical pulse sequence for such an experiment is shown opposite For a twospin system the optimum value for ∆ is 1/(2J12). The sequence can be dissected as follows. The initial 90° – ∆/2 – 180° – ∆/2 – sequence is a spin echo which, at time ∆ , refocuses any evolution of offsets but allows the coupling to evolve and generate antiphase magnetization. This antiphase magnetization is turned into multiplequantum coherence by the second 90° pulse. After evolving for time t 1 the multiple quantum is returned into observable (antiphase) magnetization by the final 90° pulse. Thus the first three pulses form the preparation period and the last pulse is the mixing period. 7.5.1 Doublequantum spectrum for a threespin system The sequence will be analysed for a system of three spins. A complete analysis would be rather lengthy, so attention will be focused on certain terms as above, as many simplifying assumptions as possible will be made about the sequence. The starting point will be equilibrium magnetization on spin 1, I1z; after the spin echo the magnetization has evolved due to the coupling between spin 1 and spin 2, and the coupling between spin 1 and spin 3 (the 180° pulse causes an overall sign change (see section 6.4.1) but this has no real effect here so it will be ignored)
πJ12 ∆I1 z I 2 z – I1 y 2→ − cos πJ12 ∆ I1 y + sin πJ12 ∆ 2 I1 x I2 z πJ13 ∆I1 z I 3 z 2→ − cos πJ13 ∆ cos πJ12 ∆ I1 y + sin πJ13 ∆ cos πJ12 ∆ 2 I1 x I3 z
∆
2
∆
2
t1
t2
Pulse sequence for multiplequantum spectroscopy.
[3.1]
+ cos πJ13 ∆ sin πJ12 ∆ 2 I1 x I2 z + sin πJ13 ∆ sin πJ12 ∆ 4 I1 y I2 z I3 z Of these four terms, all but the first are turned into multiplequantum by the second 90° pulse. For example, the second term becomes a mixture of double and zero quantum between spins 1 and 3
( 1x 2 x 3x ) sin πJ13 ∆ cos πJ12 ∆ 2 I1 x I3 z → − sin πJ13 ∆ cos πJ12 ∆ 2 I1 x I3 y
π 2 I +I +I
It will be assumed that appropriate coherence pathway selection has been used so that ultimately only the doublequantum part contributes to the spectrum. This part is
[− sin πJ13∆ cos πJ12 ∆ ]{ 1 (2 I1x I3 y + 2 I1y I3 x )} ≡ B13DQ(y13) 2
The term in square brackets just gives the overall intensity, but does not affect the frequencies of the peaks in the twodimensional spectrum as it does not depend on t1 or t2; this intensity term is denoted B13 for brevity. The operators in the curly brackets represent a pure double quantum state which can be denoted DQ (y13) ; the superscript (13) indicates that the double quantum is between spins 1 and 3 (see section 6.9). 7–15
Similar multiplet structures are seen for the doublequantum between spins 1 & 2 and spins 2 & 3. In F2 one twodimensional multiplet is centred at the offset of spins 1. namely an in–phase doublet. DQ (y13) and DQ (x13) do not evolve under the coupling between spins 1 and 3. Finally. such a doublequantum term evolves under the offset according to I3 z B13DQ (y13) Ω1t1 I1 z + Ω 2 t1 I 2 z + Ω 3 t1→ B13cos(Ω1 + Ω3 )t1 DQ (y13) − B13 sin(Ω1 + Ω3 )t1 DQ (x13) where DQ (x13) ≡ 1 2 (2 I 1x 3 x I − 2 I1 y I3 y .9. B13. split by (J 12 + J 23) and centred at (Ω 1 + Ω3). but they do evolve under the sum of the couplings between these two and all other spins. This evolution is analogous to that of a ) single spin where y rotates towards –x. Schematic twodimensional double quantum spectrum showing the multiplets arising from evolution of doublequantum coherence between spins 1 and 3. depends on the delay ∆ and all the couplings in the system. in this case this is simply (J12+J23). If has been assumed that J12 > J 13 > J 23. and one at the offset of spin 3. 7–16 . this is analogous to a normal multiplet.As is shown in section 6. only DQ (y13) is rendered observable by the final pulse Ω 1+ Ω 3 F1 − B13 sin(Ω1 + Ω3 )t1 cosπ ( J12 + J23 )t1 DQ (x13) ( 1x 2 x 3x cos(Ω1 + Ω3 )t1 cosπ ( J12 + J23 )t1 B13DQ (y13) )→ π 2 I +I +I Ω1 F2 Ω3 cos(Ω1 + Ω3 )t1 cosπ ( J12 + J23 )t1 B13 [2 I1 x I3 z + 2 I1z I3 x ] The calculation predicts that two twodimensional multiplets appear in the spectrum. As is also shown in section 6. Both have the same structure in F1. Of all the terms present at the end of t1. Ω 1. such terms are directly analogous to singlequantum antiphase magnetization. the overall amplitude. Ω3. The schematic spectrum is shown opposite. both multiplets are antiphase with respect to the coupling J13. Taking each term in turn πJ12 t1 I1 z I 2 z + 2 πJ 23 t1 I 2 z I B13cos(Ω1 + Ω3 )t1 DQ (y13) 23 z → B13cos(Ω1 + Ω3 )t1 cosπ ( J12 + J23 )t1 DQ (y13) − B13cos(Ω1 + Ω3 )t1 sinπ ( J12 + J23 )t1 2 I2 z DQ (x13) πJ12 t1 I1 z I 2 z + 2 πJ 23 t1 I 2 z I − B13 sin(Ω1 + Ω3 )t1 DQ (x13) 23 z → − B13 sin(Ω1 + Ω3 )t1 sin π ( J12 + J23 )t1 2 I2 z DQ (y13) Terms such as 2 I2 z DQ (y13) and 2 I2 z DQ (x13) can be thought of as doublequantum coherence which has become "antiphase" with respect to the coupling to spin 2.9.
The fact that the two twodimensional multiplets share a common F1 frequency and that this frequency is the sum of the two F2 frequencies constitute a double check as to whether or not the peaks indicate that the spins are coupled. during the two delays ∆ magnetization is lost by relaxation. 7. An unfortunate choice for ∆ will result in low intensity.5. but the doublequantum spectrum may show a peak with F1 coordinate (ΩA + ΩB) and F2 coordinate ΩA (or ΩB) even when spins A and B are not coupled. Double quantum spectra give very similar information to that obtained from COSY i. although doublequantum spectroscopy is used in some special circumstances.2 Interpretation of doublequantum spectra The doublequantum spectrum shows the relationship between the frequencies of the lines in the double quantum spectrum and those in the (conventional) singlequantum spectrum.1. in a COSY the crosspeaks from such a pair of spins could be obscured by the diagonal.5.7. (2) The value of the delay ∆ in the doublequantum experiment affects the amount of multiplequantum generated and hence the intensity in the spectrum. resulting in reduced peak intensities in the doublequantum spectrum. ΩB) the implication is that the two spins A and B are coupled. so that correlations between spins with similar offsets are relatively easy to locate. whereas in a doublequantum spectrum the multiplet is only antiphase in F2.and doublequantum frequencies for coupled spins. On the whole. (3) There are no diagonalpeak multiplets in a doublequantum spectrum. if J 13 = 0 the term B 13. All of the couplings present in the spin system affect the intensity and as couplings cover a wide range. In contrast. Such remote peaks.g. The form of these remote peaks in considered in the next section. If two twodimensional multiplets appear at (F1. goes to zero). as they are called. appear when spins A and B are both coupled to a third spin. F2 ) = ( ΩA + ΩB.3 Remote peaks in doublequantum spectra The origin of remote peaks can be illustrated by returning to the calculation of section 7. This may lead to stronger peaks in the doublequantum spectrum due to less cancellation.e. No such problems occur with COSY. ΩA) and (ΩA + ΩB. no single optimum value for ∆ can be given. Each method has particular advantages and disadvantages: (1) In COSY the crosspeak multiplet is antiphase in both dimensions.5. However. (4) In more complex spin systems the interpretation of a COSY remains unambiguous. and focusing on the doubly antiphase term which is present at 7–17 F2 F1 Ω A+ Ω B ΩA ΩB Schematic spectrum showing the relationship between the single. and it is then possible that correlations will be missed. but these require additional experiments. in the previous section. There are various tests that can differentiate these remote from the more useful direct peaks. COSY is regarded as a more reliable and simple experiment. . the identification of coupled spins. as it is only if there is a coupling present that doublequantum coherence between the two spins can be generated (e.
1 cos(Ω 2 + Ω3 )t1 cosπ ( J12 + J13 )t1 4 I1 y I2 z I3 z B23.1 cos(Ω 2 + Ω3 )t1 sinπ ( J12 + J13 )t1 (2 I2 x I3 z + 2 I2 z I3 x ) The first term results in a multiplet appearing at Ω1 in F2 and at (Ω2 + Ω3) in F1. The multiplet is doubly antiphase (with respect to the couplings to spins 2 and 3) in F2. as its frequency coordinates do not conform to the simple pattern described in section 7. all that is required is that both spins 2 and 3 have a coupling to the third spin. at the end of t1 these two terms are ( B23.2. The inphase multiplet is shown at the top. During t1 this term evolves under the influence of the offsets and the couplings.1]) sin πJ13 ∆ sin πJ12 ∆ 4 I1 y I2 z I3 z The 90° pulse rotates this into multiplequantum ( 1x 2 x 3x ) sin πJ13 ∆ sin πJ12 ∆ 4 I1 y I2 z I3 z → sin πJ13 ∆ sin πJ12 ∆ 4 I1z I2 y I3 y π 2 I +I +I The pure doublequantum part of this term is − 1 sin πJ13 ∆ sin πJ12 ∆ 4 I1z I2 x I3 x − 4 I1z I2 y I3 y ≡ B23. This multiplet is a remote peak. and if this coupling is zero there will be cancellation within the multiplet and no signals will be observed. Ω3 J23 decreasing J23 = 0 2I2zI3x Illustration of how the intensity of an antiphase multiplet decreases as the coupling which it is in antiphase with respect to decreases. in F1 it is inphase with respect to the sum of the couplings J12 and J13.6 Advanced topic: Lineshapes and frequency discrimination This is a somewhat involved topic which will only be possible to cover in outline here.the end of the spin echo (the fourth term in Eqn. The key thing is that no coupling between spins 2 and 3 is required for the generation of this term – the intensity just depends on J 12 and J 13. Only two terms ultimately lead to observable signals. 7–18 . spin 1. antiphase with respect to spin 1. The second and third terms are antiphase with respect to the coupling between spins 2 and 3. This is despite the fact that multiplequantum coherence between these two spins has been generated.1 2 I1z DQ(x 23) 2 In words.5.1 cos(Ω 2 + Ω3 )t1 cosπ ( J12 + J13 )t1 2 I1z DQx23) ( B23. and below are three versions of the antiphase multiplet for successively decreasing values of J23 . It is distinguished from direct peaks not only by its frequency coordinates.1 cos(Ω 2 + Ω3 )t1 sinπ ( J12 + J13 )t1 DQy23) ( ) and after the final 90° pulse the observable parts are J13 J23 B23. 7. [3. what has been generated in doublequantum between spins 2 and 3. but also by having a different lineshape in F2 to direct peaks and by being doubly antiphase in that dimension.
these do not correspond to the x– and ycomponents of the ideal timedomain function.7. The Fourier transform of S(t) carries forward the phase term S(ω ) = γ exp(iφ instr ){ A(ω ) + iD(ω )} 7–19 Re[ S(t )] = γ (cos φ instr cos Ωt − sin φ instr sin Ωt ) exp( −t T2 ) . φinstr S(t ) = γ exp(iφ instr ) exp(iΩt ) exp( −t T2 ) The real and imaginary parts of S(t) are Im[ S(t )] = γ (cos φ instr sin Ωt + sin φ instr cos Ωt ) exp( −t T2 ) Clearly. Usually. these two components are combined in the computer to give a complex timedomain signal.1.1 Onedimensional spectra Suppose that a 90°(y) pulse is applied to equilibrium magnetization resulting in the generation of pure xmagnetization which then precesses in the transverse plane with frequency Ω.1 Phase Due to instrumental factors it is almost never the case that the real and imaginary parts of S(t) correspond exactly to the x.6. If it is assumed (arbitrarily) that these components decay exponentially with time constant T2 the resulting signals. this is expressed by multiplying the ideal function by an instrumental phase factor. S(t) S(t ) = Sx (t ) + iSy (t ) = γ (cos Ωt + i sin Ωt ) exp( −t T2 ) = γ exp(iΩt ) exp( −t T2 ) The Fourier transform of S(t) is also a complex function. S(ω): S(ω ) = FT [ S(t )] = γ { A(ω ) + iD(ω )} [7.2] All modern spectrometers use a method know as quadrature d e t e c t i o n . These lineshapes are illustrated opposite.6.and ycomponents of the magnetization. which in effect means that both the x. ω Ω where A(ω) and D(ω) are the absorption and dispersion Lorentzian lineshapes: A(ω ) = 1 (ω − Ω )2 T22 + 1 D(ω ) = ω (ω − Ω )T2 (ω − Ω )2 T22 + 1 Absorption (above) and dispersion (below) Lorentzian lineshapes. NMR spectrometers are set up to detect the x. centred at frequency Ω. For NMR it is usual to display the spectrum with the absorption mode lineshape and in this case this corresponds to displaying the real part of S(ω). Sx (t ) and Sy (t ) .and ycomponents of the magnetization are detected simultaneously. 7. from the two channels of the detector can be written Sx (t ) = γ cos Ωt exp( −t T2 ) Sy (t ) = γ sin Ωt exp( −t T2 ) where γ is a factor which gives the absolute intensity of the signal.and ycomponents of this magnetization. Mathematically.
7. displaying the real part of S(ω ) will not give the required absorption mode spectrum. S(ω) is multiplied. assuming that the instrumental phase is zero. rather. the spectrum will show lines which have a mixture of absorption and dispersion lineshapes.1. NMR processing software also allows for an additional phase correction which depends on frequency. φ corr = – φinstr) the phase terms disappear and the real part of the spectrum will have the required absorption lineshape. The Fourier transform of the original signal is 7–20 ( ) . In this case the phase correction would be π /2. the "experimental" phase. the value of the phase correction is set "by eye" until the spectrum "looks phased". amongst other things. is – π /2 (recall that exp(iφ ) = cos φ + i sin φ .The real and imaginary parts of S(ω ) are no longer the absorption and dispersion signals: Im[ S(ω )] = γ (cos φ instr D(ω ) + sin φ instr A(ω )) Thus. In practice. the output of the two channels of the detector are Sx (t ) = γ sin Ωt exp( −t T2 ) Sy (t ) = −γ cos Ωt exp( −t T2 ) The complex timedomain signal can then be written S(t ) = Sx (t ) + iSy (t ) = γ (sin Ωt − i cos Ωt ) exp( −t T2 ) Re[ S(ω )] = γ (cos φ instr A(ω ) − sin φ instr D(ω )) γ ( −i )(cos Ωt + i sin Ωt ) exp( −t T2 ) = γ ( −i ) exp(iΩt ) exp( −t T2 ) = γ exp iφ exp exp(iΩt ) exp( −t T2 ) Where φ e x p .6. It is clear from the form of S(t) that this phase introduced by altering the experiment (in this case.2 Phase is arbitrary Suppose that the phase of the 90° pulse is changed from y to x. by altering the phase of the pulse) takes exactly the same form as the instrumental phase error. such a correction is needed to compensate for. imperfections in radiofrequency pulses. by a phase correction factor. T h e magnetization now starts along –y and precesses towards x. Restoring the pure absorption lineshape is simple. therefore.e. so that exp(–i π/2) = –i). It can. be corrected by applying a phase correction so as to return the real part of the spectrum to the absorption mode lineshape. in the computer. φcorr: S(ω ) exp(iφ corr ) = γ exp(iφ corr ) exp(iφ instr ){ A(ω ) + iD(ω )} = γ exp(i(φ corr + φ instr )){ A(ω ) + iD(ω )} By choosing φcorr such that (φ corr + φinst) = 0 (i.
7.2] in section 7.1). for that matter). 7. and the imaginary part shows the absorption lineshape.3 Relative phase is important The conclusion from the previous two sections is that the lineshape seen in the spectrum is under the control of the spectroscopist. Either can be phased to absorption.6. [7. if an experiment leads to magnetization from different processes or spins appearing along different axes. but if one is in absorption. for example.1. The time domain signal will then just have a real part (compare Eqn. The 90° phase shift simply swaps over the real and imaginary parts.6.6.Re[ S(ω )] = γ D(ω ) S(ω ) = γ ( −i ){ A(ω ) + iD(ω )} Im[ S(ω )] = −γ A(ω ) Thus the real part shows the dispersion mode lineshape.axis (or anywhere in between.4.4 Frequency discrimination Suppose that a particular spectrometer is only capable of recording one. component of the precessing magnetization. However.1) Using the identity cos θ = (exp(iθ ) + exp(−iθ )) this can be written S(t ) = 1 γ [exp(iΩt ) + exp( – iΩt )] exp( −t T2 ) 2 = 1 γ exp(iΩt ) exp( −t T2 ) + 1 γ exp( – iΩt ) exp( −t T2 ) 2 2 1 2 S(t ) = γ cos Ωt exp( −t T2 ) 7–21 . whether the pulse sequence results in magnetization appearing along the x. one will be in dispersion. It does not matter. say the x. there is no single phase correction which will put the whole spectrum in the absorption mode.1. This is the case in the COSY spectrum (section 7. whereas those leading to crosspeaks appear along y. It is always possible to phase correct the spectrum afterwards to achieve the desired lineshape.or y. The terms leading to diagonalpeaks appear along the xaxis. the two signals are fundamentally 90° out of phase with one another.
regardless of whether its real offset is positive or negative. The lack of frequency discrimination is associated with the signal being modulated by a cosine wave. each peak appears at both positive and negative frequency.a ω+ b 0 – The Fourier transform of the first term gives. [3. but the two peaks have opposite signs. If the sign of Ω changes the two peaks swap over. which has the property that cos(Ω t) = cos(–Ωt). Suppose now that only the ycomponent of the precessing magnetization could be detected. This spectrum is said to lack frequency discrimination. in the sense that it does not matter if the magnetization went round at +Ω or –Ω.2] in section 7. which has the property that sin(–Ω t) = – sin(Ω t). (exp(iθ ) − exp(−iθ )) this can be written S(t ) = 1 γ [exp(iΩt ) − exp( – iΩt )] exp( −t T2 ) 2 = 1 γ exp(iΩt ) exp( −t T2 ) − 1 γ exp( – iΩt ) exp( −t T2 ) 2 2 1 2i S(t ) = iγ sin Ωt exp( −t T2 ) and so Re[ S(ω )] = 1 γA+ − 1 γA– 2 2 This spectrum again shows two peaks. this is associated with the signal being modulated by a sine wave. The time domain signal will then be (compare Eqn. the transform of the second term gives the same but at ω = – Ω. the spectrum still shows peaks at both + Ω and –Ω. D+ and D– represent dispersion mode peaks at +Ω and –Ω. at ±Ω.and ycomponents are measured where one peak appears at either positive or negative ω depending on the sign of Ω. In onedimensional spectroscopy it is virtually always possible to arrange for the signal to have this desirable complex phase modulation. respectively. but in practice in a spectrum with many lines with a range of positive and negative offsets the resulting set of possibly cancelling peaks would be impossible to sort out satisfactorily. as opposed to a complex exponential. but in the case of twodimensional spectra it is almost always the case that the signal modulation in the t1 dimension is of the form cos(Ωt1) and so such spectra are not naturally frequency discriminated in the F1 dimension. likewise. but with the added complication that one peak is inverted.6. Spectrum c results from a sine modulated data set. like b each peak appears twice. but there are still two peaks. Re[ S(ω )] = 1 γA+ + 1 γA– 2 2 where A + represents an absorption mode Lorentzian line at ω = +Ω and A– represents the same at ω = –Ω. exp(iΩt) which is sensitive to the sign of Ω . In a sense the spectrum is frequency discriminated. This is in contrast to the case where both the x. in the real part. Spectra b and c lack frequency discrimination and are quite uninterpretable as a result. an absorption mode peak at ω = +Ω. 7–22 . as positive and negative frequencies can be distinguished. Spectrum b results from a cosine modulated timedomain data set.1) Using the identity sin θ = ω+ c 0 – ω+ 0 – Spectrum a has peaks at positive and negative frequencies and is frequency discriminated.
the details of which are given in a later chapter.e. in separate experiments. In COSY and DQF COSY a similar change can be brought about by altering the phase of the first 90° pulse. i. indeed all of the experiments analysed in this chapter have produced either sine or cosine modulated data. For example. at time t2 = 0.2. in which the evolution in t1 is encoded as a phase. It will turn out that the key to obtaining frequency discrimination is the ability to record. twodimensional experiments produce amplitude modulation.e. The second type is amplitude modulation. both sine and cosine modulated data sets. 7. This can be achieved by simply altering the phase of the pulses in the sequence.2 . and T2 (1) and T2 ( 2 ) are the decay time constants in t1 and t2 respectively. mathematically as sine or cosine S(t )c = γ cos(Ω1t1 ) exp −t1 T2 (1) exp(iΩ 2 t2 ) exp −t2 T2 ( 2 ) S(t ) s 1 2 1 1 1 2 2 2 2 2 ( = γ sin(Ω t ) exp( −t ) ( T ( ) ) exp(iΩ t ) exp( −t ) T ( )) Generally.2 Twodimensional spectra 7.6. can be written (1 − f ) cos Ω1t1 I1 y + f cos Ω1t1 I2 y If. consider the EXSY sequence analysed in section 7. The observable signal. the resulting confusion in the spectrum is not acceptable and steps have to be taken to introduce frequency discrimination. As explained above for onedimensional spectra.7.6.1 Phase and amplitude modulation There are two basic types of timedomain signal that are found in twodimensional experiments. In fact there is a general procedure for effecting this change. in which the evolution in t1 is encoded as an amplitude. t2 ) phase = γ exp(iΩ1t1 ) exp −t1 T2 (1) exp(iΩ 2 t2 ) exp −t2 T2 ( 2 ) ( ) ( ) where Ω1 and Ω2 are the modulation frequencies in t1 and t2 respectively. The first is phase modulation. however. Therefore most twodimensional spectra are fundamentally not frequency discriminated in the F1 dimension. mathematically as a complex exponential S(t1 .6. the first pulse in the sequence is changed in phase from x to y the corresponding signal will be −(1 − f ) sin Ω1t1 I1 y − f sin Ω1t1 I2 y i. the modulation has changed from the form of a cosine to sine. i.2 Twodimensional lineshapes The spectra resulting from twodimensional Fourier transformation of phase and amplitude modulated data sets can be determined by using the following Fourier pair FT exp(iΩt ) exp( −t T2 ) = { A(ω ) + iD(ω )} 7–23 [ ] .e.2.
illustrated below.ω2) = (+ Ω1. The real part of the resulting twodimensional spectrum is ( ( ( ( Re S(ω 1 .where A and D are the dispersion Lorentzian lineshapes described in section 7.+Ω2) with the phasetwist lineshape. ω 2 ) phase = γ exp(iΩ1t1 ) exp −t1 T2(1) A+2 ) + iD+2 ) ( )[ ] ( where A+2 ) indicates an absorption mode line in the F2 dimension at ω2 = + Ω2 ( and with linewidth set by T2 ( 2 ) . similarly D+1) is the corresponding dispersion line. The phasetwist lineshape is an inextricable mixture of absorption and dispersion.6.1). ω 2 ) phase = γ A+1) A+2 ) − D+1) D+2 ) [ ] ( ) This is a single line at (ω 1. Cosine amplitude modulation For the cosine modulated data set the transform with respect to t2 gives ( ( S(t1 .4. ω 2 ) phase = γ A+1) + iD+1) A+2 ) + iD+2 ) [ ][ ] ( where A+1) indicates an absorption mode line in the F1 dimension at ω1 = + Ω1 ( and with linewidth set by T2 (1) . No phase correction will restore it to pure absorption mode. The second transform with respect to t1 gives ( ( ( ( S(ω 1 .1 Phase modulation For the phase modulated data set the transform with respect to t2 gives ( ( S(t1 . and the dispersion component only dies off slowly. Pseudo 3D view and contour plot of the phasetwist lineshape. ω 2 )c = γ cos(Ω1t1 ) exp −t1 T2 (1) A+2 ) + iD+2 ) ( )[ ] The cosine is then rewritten in terms of complex exponentials to give 7–24 . similarly D+2 ) is the corresponding dispersion line. it is a superposition of the double absorption and double dispersion lineshape (illustrated in section 7. Generally the phase twist is not a very desirable lineshape as it has both positive and negative parts.
Sine amplitude modulation [ ]= 1 2 ( ( ( ( γA+1) A+2 ) + 1 γA−1) A+2 ) 2 For the sine modulated data set the transform with respect to t2 gives ( ( S(t1 . ω 2 )c = 1 γ 2 Re [{A Re ] ( ) (1) + ( ( ( ( + iD+1) } + { A−1) + iD−1) } A+2 ) ] The real part of the resulting twodimensional spectrum is Re S(ω1 . The real part of the resulting twodimensional spectrum is ( ( ( ( ( ( ( ( Re S(ω 1 .+Ω2) and the other is at (–Ω 1. located at (+ Ω1. i. ω 2 )c = 1 γ exp(iΩ1t1 ) + exp( −iΩ1t1 ) exp −t1 T2 (1) A+2 ) + iD+2 ) 2 [ ] ( )[ ] The second transform with respect to t1 gives S(ω 1 . ω 2 )c = Re S(t1 .+Ω2).( ( S(t1 . ω 2 )c = 1 γ 2 ( where A−1) indicates an absorption mode line in the F1 dimension at ω1 = –Ω1 ( and with linewidth set by T2(1) .+Ω2). but in contrast to the above both have the double absorption lineshape. ω 2 )c Re ( = γ cos(Ω1t1 ) exp −t1 T2 (1) A+2 ) [ ] ( ) Following through the same procedure as above: ( S(t1 . taking the real part of S(t1 . ω 2 )c S(t1 . ω 2 )c = 1 γ exp(iΩ1t1 ) + exp( −iΩ1t1 ) exp −t1 T2 (1) A+2 ) 2 Re [ S(ω1 . similarly D–1) is the corresponding dispersion line. one is located at (+Ω1. ω 2 )c This is two lines. There is still lack of frequency discrimination. A spectrum with a pure absorption mode lineshape can be obtained by discarding the imaginary part of the time domain data immediately after the transform with respect to t2. ω 2 )c = 1 γ A+1) A+2 ) − D+1) D+2 ) + 1 γ A−1) A+2 ) − D−1) D+2 ) 2 2 [{A (1) + ( ( ( + iD+1) + A−1) + iD−1) } { }][ A (2) + ( + iD+2 ) ] [ ] ( ) ( ) This is a two lines. ω 2 )s = γ sin(Ω1t1 ) exp −t1 T2 (1) A+2 ) + iD+2 ) ( )[ ] The cosine is then rewritten in terms of complex exponentials to give S(t1 . both with the phasetwist lineshape.e. ω 2 )s = 1 2i ( ( γ [exp(iΩ1t1 ) − exp( −iΩ1t1 )] exp( −t1 T2 (1) ) A+2 ) + iD+2 ) [ ] The second transform with respect to t1 gives S(ω 1 .+Ω2) and (–Ω 1. but the undesirable phasetwist lineshape has been avoided. ω 2 )s = 1 2i ( ( ( ( γ A+1) + iD+1) − A−1) + iD−1) [{ } { }][ A (2) + ( + iD+2 ) ] 7–25 The imaginary part of the resulting twodimensional spectrum is . As expected for a data set which is cosine modulated in t1 the spectrum is symmetrical about ω1 = 0.
each will be analysed here.+Ω2). processed as described in section 7.+Ω2). ω 2 )s Re [ ] ( = γ sin(Ω1t1 ) exp −t1 T2(1) A+2 ) ( ) Following through the same procedure as above: S(t1 . i. As expected for a data set which is sine modulated in t1 the spectrum is antisymmetric about ω1 = 0.1. ω 2 ) [ Re c ] − Im[S(ω . ω 2 )s [ Re ]= − 1 2 ( ( ( ( γA+1) A+2 ) + 1 γA−1) A+2 ) 2 The two lines now have the pure absorption lineshape. gives two positive absorption mode peaks at (+Ω1. 7.6.3 Frequency discrimination with retention of absorption lineshapes It is essential to be able to combine frequency discrimination in the F 1 dimension with retention of pure absorption lineshapes. Three different ways of achieving this are commonly used. a spectrum with a pure absorption mode lineshape can be obtained by discarding the imaginary part of the time domain data immediately after the transform with respect to t2.6. StatesHaberkornRuben method The essence of the StatesHaberkornRuben (SHR) method is the observation that the cosine modulated data set.2. Subtracting these spectra from one another gives the required absorption mode frequency discriminated spectrum (see the diagram below): Re S(ω 1 . ω 2 )s = − 1 γ A+1) A+2 ) − D+1) D+2 ) + 1 γ A−1) A+2 ) − D−1) D+2 ) 2 2 [ ] ( ) ( ) This is two lines.+Ω2) and (–Ω1. one is located at (+Ω1. both with the phasetwist lineshape but with opposite signs. ω 2 )s = Re S(t1 . taking the real part of S(t1 . ω 2 )s = Re 1 2i ( γ [exp(iΩ1t1 ) − exp( −iΩ1t1 )] exp −t1 T2(1) A+2 ) 1 2i S(ω1 . whereas the sine modulated data set processed in the same way gives a spectrum in which one peak is negative and one positive. ω 2 )s S(t1 . ω 2 )s = Re ( ( ( ( ( γ { A+1) + iD+1) } − { A−1) + iD−1) } A+2 ) [ ( ] ) The imaginary part of the resulting twodimensional spectrum is Im S(ω1 .e.+Ω2) and the other is at (–Ω1.ω ) ] 1 Re 2 s = [ 1 2 ( ( ( ( ( ( ( ( γA+1) A+2 ) + 1 γA−1) A+2 ) − − 1 γA+1) A+2 ) + 1 γA−1) A+2 ) 2 2 2 ] [ ] ( ( = γA+1) A+2 ) 7–26 . As before.( ( ( ( ( ( ( ( Im S(ω 1 .2.
ω 2 )SHR = γ A –Ω 1 +Ω 1 difference [ (1) + + iD (1) + ]A (2) + ( ( ( ( = γA+1) A+2 ) + iD+1) A+2 ) +Ω 1 Illustration of the way in which the SHR method achieves frequency discrimination by combining cosine and sine modulated spectra. where φ is the phase shift in radians.6. as was the case for the phase errors considered in section 7. MarionWüthrich or TPPI method The idea behind the TPPI (time proportional phase incrementation) or Marion–Wüthrich (MW) method is to arrange things so that all of the peaks have positive offsets. One simple way to make all offsets positive is to set the receiver carrier frequency deliberately at the edge of the spectrum.1 The change from cosine to sine modulation in the EXSY experiment can be though of as a phase shift of the signal in t1. such a phase shifted cosine wave is written as cos(Ω1t1 + φ ) .In practice it is usually more convenient to achieve this result in the following way. it is not really a very practical method as the resulting spectrum would be very inefficient in its use of data space and in addition offresonance effects associated with the pulses in the sequence will be accentuated. by shifting the phase of the first pulse by 90°. – sin(Ω1t1). could be turned into sine modulation. and so if t is in seconds Ω t is in radians. which is mathematically identical. The effect of such a phase shift can be represented mathematically in the following way. but it is made to appear that all the frequencies are positive by phase shifting systematically some of the pulses in the sequence in concert with the incrementation of t1. This expression can be expanded using the well known formula cos( A + B) = cos A cos B − sin A sin B to give 7–27 . Mathematically. Ωt can therefore be described as a phase which depends on time. ω 2 ) Re c ω1 cosine ω2 Ω2 + iS(t1 .2 it was shown that in the EXSY sequence the cosine modulation in t1. ω 2 )SHR = S(t1 . Then. frequency discrimination would not be required as there would be no ambiguity. The cosine and sine data sets are transformed with respect to t2 and the real parts of each are taken. Simple though this is. Then a new complex data set is formed using the cosine data for the real part and the sine data for the imaginary part: S(t1 . cos(Ω1t1). Recall that Ω is in units of radians s–1 .1. It is also possible to consider phases which do not depend on time. In section 7. ω 2 ) – –Ω 1 0 sine +Ω 1 + Re s = γ cos(Ω1t1 ) exp −t1 T2 = γ exp(iΩ1t1 ) exp −t1 ( ( ) A + iγ sin(Ω t ) exp(−t T ( ) ) A( ) (1) + (2) 1 1 1 T2 (1) )A + (2) 1 2 2 + Fourier transformation with respect to t1 gives a spectrum whose real part contains the required frequency discriminated absorption mode spectrum S(ω 1 . In the TPPI method the carrier can still be set in the middle of the spectrum.
After transformation the resulting spectrum displays correctly peaks with offsets in the range –(SW/2) to +(SW/2) where SW is the spectral width which is given by 1/∆ (this comes about from the Nyquist theorem of data sampling). Frequencies outside this range are not represented correctly. Suppose that the required frequency range in the F 1 dimension is from –(SW1 /2) to +(SW1 /2) (in COSY and EXSY this will be the same as the range in F2). will have to be halved i. The new timedomain function with the inclusion of this incrementing phase is thus cos(Ω1t1 + φ (t1 )) = cos(Ω1t1 + ω additional t1 ) = cos(Ω1 + ω additional )t1 In words.e. In onedimensional pulseFourier transform NMR the free induction signal is sampled at regular intervals ∆ . ∆1 = 1/(2SW1 ) in order that the range of frequencies present are represented properly. and t1 will be written ωadditional φ (t1 ) = ω additional t1 Clearly the units of ωadditional are radians s–1. Then the peaks will be in the range 0 to (SW1 ). phase shifted by π /2 radians (90°) is the same thing as a sine wave. the units of ω additional . but t 1 can be written as n∆1 for the nth increment of t1. The normal spectrum is shown in a. 7–28 . If the spectral width is doubled all peaks are represented correctly – spectrum c. ∆1. shown dotted) will then fall outside the spectral window – spectrum b. the effect of incrementing the phase in concert with t1 is to add a frequency ω additional to all of the offsets in the spectrum. where the 2π is to convert from frequency (the units of SW 1 ) to rad s –1. a cosine wave. is π/2 radians the result is cos(Ω1t1 + π 2) = cos Ω1t cos π 2 − sin Ω1t sin π 2 = − sin Ω1t In words. for the nth increment as a –SW1/2 b 0 0 +SW1/2 +SW1 c –SW1 0 +SW1 Illustration of the TPPI method. that is ωadditional is a frequency. what this means is that each time t1 is incremented the phase is also incremented in concert. it will be necessary to add (SW1 /2) to all the frequencies. the phase shift is made proportional to t1. As the maximum frequency is now (SW1 ) rather than (SW1 /2) the sampling interval.cos(Ω1t1 + φ ) = cos Ω1t cos φ − sin Ω1t sin φ If the phase shift. To make it appear that all the peaks have a positive offset. The phase increment is ω additional t1 . but some. The TPPI method utilizes this option of shifting all the frequencies in the following way. Suppose that instead of a fixed phase shift. The constant of proportion between the time dependent phase. Thus. in the EXSY experiment the effect of changing the phase of the first pulse by 90° can be described as a phase shift of the signal by 90°. Adding a frequency of S W/2 to all the peaks gives them all positive offsets. φ(t1). The required value for ω additional is 2π (SW1 /2) . with peaks in the range –SW/2 to +SW/2. φ. Putting all of this together ω additional t1 can be expressed.
. it is usual to attempt to use some method to eliminate the phasetwist lineshape. A data set from an experiment to which TPPI has been applied is simply amplitude modulated in t1 and so can be processed according to the method described for cosine modulated data so as to obtain absorption mode lineshapes. The first is called the Ptype or 7–29 TPPI phase incrementation applied to a COSY sequence. as it means that no special steps are required to obtain frequency discrimination..SW ω additional t1 = 2π 1 (n∆1 ) 2 SW 1 = 2π 1 n 2 2 SW1 π 2 The way in which the phase incrementation increases the frequency of the cosine wave is shown below: =n time The open circles lie on a cosine wave. The key to how this can be done lies in the fact that two kinds of phase modulated data sets can usually be recorded. cos(Ω × n∆ ). which are very undesirable. phase modulated data sets give rise to spectra with phasetwist lineshapes. 2 . The way in which this phase increment increases the frequency of the cosine wave is apparent. Echo antiecho method x t2 t1 = 0 t1 = ∆ y t2 –x t2 t1 = 2∆ –y t2 t1 = 3∆ x t1 t2 t1 = 4∆ Few twodimensional experiments naturally produce phase modulated data sets. while at the same time retaining frequency discrimination. So.. where ∆ is the sampling interval and n runs 0. here φ = π/8. but if gradient pulses are used for coherence pathway selection it is then quite often found that the data are phase modulated. However. As the spectrum is symmetrical about F1 = 0 it is usual to use a modified Fourier transform routine which saves effort and space by only calculating the positive frequency part of the spectrum.e. The way in which it can be decided which pulse to increment will be described in a later chapter. 1. The phase of the first pulse is incremented by 90° each time t1 is incremented. In words this means that each time t1 is incremented. the phase of the signal should also be incremented by 90°. for example by incrementing the phase of one of the pulses. The closed circles lie on a cosine wave in which an additional phase is incremented on each point i. In one way this is an advantage. the function is cos(Ω × n∆ + n φ).
meaning here that the sign of the frequencies in F1 and F2 are the same.2. The simplest way to proceed is to compute two new data sets which are the sum and difference of the P. The second data set is called the echo or Ntype S(t1 . t2 ) N = γ exp( – iΩ1t1 ) exp −t1 T2(1) exp(iΩ 2 t2 ) exp −t2 T2( 2 ) ( ) ( the "N" indicates negative. it is possible to recover the pure absorption spectrum by software manipulations of the spectrum. meaning here that the sign of the frequencies in F1 and F 2 are opposite.antiecho spectrum S(t1 . Usually. However. 7–30 . possibly in both dimensions.and Ntype data sets: 1 2 [ S(t . For example. the cosine modulated data set might be S(t )c = γ cos(Ω1t1 + φ1 ) exp −t1 T2(1) exp(iΩ 2 t2 + iφ 2 ) exp −t2 T2( 2 ) ( ) ( ) where φ1 and φ2 are the phase errors in F1 and F1. t2 ) P = γ exp(iΩ1t1 ) exp −t1 T2(1) exp(iΩ 2 t2 ) exp −t2 T2( 2 ) ( ) ( ) ) the "P" indicates positive.or Ntype spectrum. just as was described for the case of onedimensional spectra.4 Phase in twodimensional spectra In practice there will be instrumental and other phase shifts. t2 ) N = = γ cos(Ω1t1 ) exp( −t1 T2 (1) ) exp(iΩ 2 t2 ) exp( −t2 T2 ( 2 ) ) 1 2i γ [exp(iΩ1t1 ) + exp( −iΩ1t1 )] exp( −t1 T2 (1) ) exp(iΩ 2 t2 ) exp( −t2 T2 ( 2 ) ) ] [ S(t . t ) 1 1 2 2 P + S(t1 . 7. t ) 1 1 2i 2 P − S(t1 . Processing this data set in the manner described above will not give a pure absorption spectrum. t2 ) N = = γ sin(Ω1t1 ) exp( −t1 T2 (1) ) exp(iΩ 2 t2 ) exp( −t2 T2 ( 2 ) ) γ [exp(iΩ1t1 ) − exp( −iΩ1t1 )] exp( −t1 T2 (1) ) exp(iΩ 2 t2 ) exp( −t2 T2 ( 2 ) ) ] These two combinations are just the cosine and sine modulated data sets that are the inputs needed for the SHR method. respectively. NMR data processing software provides options for making such phase corrections to twodimensional data sets. As will be explained in a later chapter in gradient experiments it is easy to arrange to record either the P.6. which mean that the timedomain functions are not the idealised ones treated above. The pure absorption spectrum can therefore be calculated in the same way starting with these combinations.
In fact. where n α and n β are the populations of the two corresponding energy levels. In this chapter the language and concepts used to describe relaxation will be introduced and illustrated. At a practical level. Equilibrium is the state in which (a) the populations of the energy levels are those predicted by the Boltzmann distribution and (b) there is no transverse magnetization and. Magnetization parallel to the applied field is termed longitudinal. no coherences present in the system. This equilibrium magnetization arises from the unequal population of the two energy levels that correspond to the α and β spin states. As we are generally interested in the relative size of magnetizations and signals we may just as well write Mz = nα − nβ ( ) [2] † Chapter 8 "Relaxation" © James Keeler 1999 and 2002 8–1 . so a study of relaxation phenomena can lead to information on these properties. Having described the experiments which can be used to probe relaxation we will then go on to see what the source of relaxation is and how it depends on molecular parameters and molecular motion. the constant of proportion just determines the absolute size of the signal we will observe. The rate of relaxation is influenced by the physical properties of the molecule and the sample. Mz. the zmagnetization. Another example is the use of data on relaxation rates to probe the internal motions of macromolecules. the rate of relaxation determines how fast an experiment can be repeated. In Chapter 3 we saw that when an NMR sample is placed in a static magnetic field and allowed to come to equilibrium it is found that a net magnetization of the sample along the direction of the applied field (traditionally the zaxis) is developed. Ultimately. is proportional to the population difference Mz ∝ nα − nβ β α z x y ( ) A population difference gives rise to net magnetization along the zaxis.1 What is relaxation? Relaxation is the process by which the spins return to equilibrium.8 Relaxation† Relaxation is the process by which the spins in the sample come to equilibrium with the surroundings. 8. Perhaps the most often used and important of these phenomena in the nuclear Overhauser effect (NOE) which can be used to probe internuclear distances in a molecule. To begin with it will simply be taken for granted that there are processes which give rise to relaxation and we will not concern ourselves with the source of relaxation. so it is important to understand how relaxation rates can be measured and the factors that influence their values. more generally.
Doing the same with state β gives us the final. dt Similarly we can write for the population of state β : = −Wnβ + Wnα . which simply is not correct. the key thing to realize is that whereas a transition from α to β causes a loss of population of level α . We know that at equilibrium the population of state α exceeds that of state β . something must happen which causes a spin to move from the α state to the β state or vice versa. and is a first order process with rate constant W. dt These equations are almost correct. It is this process which results in relaxation. where nα is the population of state α at equilibrium. equations: dnα 0 0 = W nβ − nβ − W (nα − nα ) dt 8–2 dnβ A transition from state α to s t a t e β decreases the population of state α . correct. but a transition from state β to state α increases the population of state α. In the same way. then for the system to reach equilibrium the population of one level must increase and that of the other must decrease.β β W α α W 8. a transition from β to α causes the population of state α to increase. The simplest assumption that we can make about the rate of transitions from α to β is that it is proportional to the population of state α . With these assumptions the rate of loss of population from state α is Wnα. n α . This will lead to a description of the dynamics of the zmagnetization. This implies that there must be transitions between the two levels i. Suppose that the populations of the α and β states at time t are n α and n β. So we can write rate of change of population of state α = – Wnα + Wnβ The first term is negative as it represents a loss of population of state α and in contrast the second term is negative as it represents a gain in the population of state α .2 Rate equations and rate constants The populations of energy levels are in many ways analogous to concentrations in chemical kinetics. the rate of loss of population of state β is Wnβ. ( ) . However. This defect is easily remedied by replacing the population n α with the deviation of the 0 0 population from its equilibrium value (nα − nα ) . ( ) dnβ dt 0 0 = W (nα − nα ) − W nβ − nβ . and many of the same techniques that are used to describe the rates of chemical reactions can also be used to describe the dynamics of populations. this immediately implies that at equilibrium nα = nβ. The rate of change of the population can be written using the language of calculus as dnα/dt so we have dnα = −Wnα + Wnβ . but we need to make one modification. respectively. If these are not the equilibrium values.e. At equilibrium the populations are not changing so dnα/dt = 0.
the magnetization is Mz(0). On the other hand.1 Consequences of the rate equation The discussion in the previous section led to a (differential) equation describing the motion of the zmagnetization dMz (t ) = − Rz ( Mz (t ) − Mz0 ) [7] dt where the rate constant.2. In summary. the equilibrium zmagnetization. Mz(t). [2]. at time zero. Mz0 . The change will also be such as to return Mz to its equilibrium value. and this rate will be proportional to the deviation of Mz from Mz0 . This equation can easily be integrated: ∫ ( M (t ) − M ) = ∫ − R dt ln( M (t ) − M ) = − R t + const. Hence. [7] predicts that over time Mz will return to Mz0 . we can use these two equations to work out how the zmagnetization varies with time: dMz d nα − nβ = dt dt dn dn = α − β dt dt ( ) ) ( ) 0 0 0 0 = W nβ − nβ − W (nα − nα ) − W (nα − nα ) + W nβ − nβ 0 0 = −2W nα − nβ + 2W nα − nβ ( 0 0 where Mz0 = nα − nβ . ( = −2W ( Mz − Mz0 ) ( ) ( ) ) 8. the righthand side of Eqn. [7] is zero and hence so is the rate of change of M z: nothing happens. the constant of integration can be determined as ln( Mz (0) − Mz0 ) . to remind us that it may change. Using Eqn.You will recognise here that this kind of approach is exactly the same as that used to analyse the kinetics of a reversible chemical reaction. If M z = Mz0 . that is the system is at equilibrium. Mz0 . What this equations says is that the rate of change of Mz is proportional to the deviation of Mz from its equilibrium value. The rate at which this happens will depend on Rz. this is exactly what we expect. z 0 z z z 0 z z dM z ( t ) If. with some rearrangement: 8–3 . Eqn. if Mz deviates from Mz0 there will be a rate of change of Mz. Rz = 2W and Mz has been written as a function of time.
Finally. inversion. If the time when S(t) is zero is tnull it follows immediately from Eqn.e. So. i. which is T1 ≈ 1. The starting condition for Mz is Mz (0) = − Mz0 . S(t ) − S 0 ln [7C] 0 = − Rz t −2 S where S0 is the signal intensity from equilibrium magnetization. for example. decide on the time to leave between transients (typically three to five times T1 ). This method is rather crude. Equation [7C] implies that a plot of ln ( S(t ) − S 0 ) −2 S 0 against t should be a straight line of slope –Rz.Mz 0 Mz M (t ) − Mz0 ln z = − Rz t 0 Mz ( 0 ) − Mz or Mz (t ) = Mz (0) − Mz0 exp( − Rz t ) + Mz0 0 z [7A] Mz(0) t [ ] [7B] In words.3 A quick estimate for Rz (or T1) Often we want to obtain a quick estimate for the relaxation rate constant (or. but it good enough for estimating T1 . and this is often called T1 .e. the longitudinal or spinlattice relaxation time. is the basis of a method of determining the relaxation rate constant.10. Armed with this estimate we can then. 8. [7C] that: 1 ln = − Rz tnull 2 or Rz = ln 2 tnull or T1 = tnull ln 2 [ ] Probably the most useful relationship is the last. We can now analyse this experiment fully. One way to do this is to do an inversion recovery experiment but rather than varying t systematically we look for the value of t which results in no signal i. [7A] to be: M (t ) − Mz0 ln z = − Rz t 0 −2 Mz Recall that the amplitude of the signal we record in this experiment is proportional to the zmagnetization. a 180° pulse is applied. thereby inverting the magnetization.2 The inversion recovery experiment We described this experiment in section 3. 8.4 tnull. the relaxation time). then. 8–4 . Then a delay t is left for the magnetization to relax. we would find this from a simple 90° – acquire experiment. if this signal is S(t) it follows that t ln [ S(t)–S0 –2S0 ] Plot used to extract a value of Rz from the data from an inversion recovery experiment. With this condition the predicted time evolution can be found from Eqn. This. First.2. a 90° pulse is applied so that the size of the zmagnetization can be measured. a null. equivalently.2. The time constant of the exponential is 1/Rz. this says that the zmagnetization returns from Mz(0) to M following an exponential law.
Finally. It turns out that in such a system it is possible to have relaxation induced transitions between all possible pairs of energy levels. 8–5 . the levels are labelled with the spin of I first and the spin of S second. The rate constant for the transition between the αα and ββ states is denoted W2 .4 Writing relaxation in terms of operators As we saw in Chapter 6. and likewise for the spin S transitions. the rate constant for the transition between the αβ and βα states is denoted W0 . and then transcribing these into magnetizations. The dashed arrows indicate allowed transitions of the I spin. 8. why this is so will be seen in detail below. in quantum mechanics zmagnetization is represented by the operator I z. the spins are not coupled. Consider a sample consisting of molecules which contain two spins. and the solid arrows indicate allowed transitions of the S spin.3 Solomon equations The idea of writing differential equations for the populations. which can be labelled according to the spin states of the two spins. which is not what is meant.8. [7] in terms of operators rather then magnetizations. The rate constants for the two allowed I spin transitions will be denoted WI(1) and WI( 2 ) . As was seen in section 2. This will be illustrated in this section. to give: dIz (t ) [8] = − Rz ( Iz (t ) − Iz0 ) dt where Iz(t) represents the zmagnetization at time t and Iz0 represents the equilibrium zmagnetization. the "0" indicating that it is a zero quantum transition.2.4. the "2" indicating that it is a double quantum transition. It is therefore common to write Eqn. even those transitions which are forbidden in normal spectroscopy. especially in more complex system. 4 ββ> 4 WI(2) ββ> W2 W0 WS(2) βα> 3 2 αβ> βα> 3 2 αβ> WS(1) WI(1) 1 αα> 1 αα> (a) (b) Diagram (a) shows the energy levels of a two spin system. the two spins have between them four energy levels. As it stands this last equation seems to imply that the operators change with time. Diagram (b) shows the relaxation induced transitions which are possible amongst the same set of levels. We will use this notation from now on. What are changing are the populations of the energy levels and these in turn lead to changes in the zmagnetization represented by the operator. is a particularly convenient way of describing relaxation. I and S.
rate equations can be written for the flow of population from any of the levels. [10] shows that the same operator also represents the difference in population differences across the two Sspin transitions. likewise.2. The corresponding equations for the other populations are dn2 = −WS(1)n2 − WI( 2 )n2 − W0 n2 + WS(1)n1 + WI( 2 )n4 + W0 n3 dt dn3 = −WI(1)n3 − WS( 2 )n3 − W0 n3 + WI(1)n1 + WS( 2 )n4 + W0 n2 dt dn4 = −WS( 2 )n4 − WI( 2 )n4 − W2 n4 + WS( 2 )n3 + WI( 2 )n2 + W2 n1 dt All of this can be expressed in a more compact way if we introduce the I and S spin zmagnetizations. the magnetization has been represented as the corresponding operator. 1–3 and 2–4 Iz = n1 − n3 + n2 − n4 [9] As discussed above. (ni − ni0 ) . [9] reveals that 2IzSz represents the difference in population differences across the two Ispin transitions.2 the populations ought to be written as deviations from their equilibrium values. Iz. rather. to do this results in unnecessary complexity. [9] and then substituting for the derivatives of the populations gives 8–6 . the calculation will be carried forward as written and then at the last stage the populations will be replaced by their deviations from equilibrium. which is represented by the operator 2IzSz 2 Iz Sz = n1 − n3 − n2 + n4 [11] Comparing this with Eq. comparison with Eq. For example. However. The I spin magnetization is equal to the population difference across the two I spin transitions. Taking the derivative of Eq. Likewise for the Sspin magnetization Sz = n1 − n2 + n3 − n4 [10] A third combination of populations will be needed.Just in the same was as was done in Section 8. As was discussed in section 8. for level 1 dn1 = −WS(1)n1 − WI(1)n1 − W2 n1 + WS(1)n2 + WI(1)n3 + W2 n4 dt The negative terms are rates which lead to a loss of population of level 1 and the positive terms are ones that lead to a gain in its population.
the following differential equation is found for Iz dIz = − WI(1) + WI( 2 ) + W2 + W0 Iz dt − (W2 − W0 )Sz − WI − WI (1) ( ) ( (2) )2 I S ( [14] z z Similar algebra gives the following differential equations for the other operators dSz = −(W2 − W0 ) Iz − WS(1) + WS( 2 ) + W2 + W0 Sz − WS(1) − WS( 2 ) 2 Iz Sz dt d2 Iz Sz = − WI(1) − WI( 2 ) Iz − WS(1) − WS( 2 ) Sz dt ( ) ) ( ) ( ) − (W ( ) + W ( ) + W ( ) + W ( ) )2 I S 1 2 1 2 I I S S z z As expected. For example. After some tedious algebra. which is essentially the identity or unit operator E = n1 + n2 + n3 + n4 [13] and then realizing that the populations. As a result. from its definition. Eq. These three differential equations are known as the Solomon equations. Iz0 and Sz0 . Iz. Sz. S z etc. [9] – [13]. [14] becomes 8–7 . does not change with time. ni.dIz dn1 dn3 dn2 dn4 = − + − dt dt dt dt dt = −WS(1)n1 − WI(1)n1 − W2 n1 + WS(1)n2 + WI(1)n3 + W2 n4 + WI(1)n3 + WS( 2 )n3 + W0 n3 − WI(1)n1 − WS( 2 )n4 − W0 n2 −WS n2 − WI n2 − W0 n2 + WS n1 + WI n4 + W0 n3 + WS( 2 )n4 + WI( 2 )n4 + W2 n4 − WS( 2 )n3 − WI( 2 )n2 − W2 n1 This unpromising looking equation can be expressed in terms of Iz. Eqs. to be zero. the total population. the equilibrium value of 2IzSz is easily shown. and 2Iz Sz: (1) (2) (1) (2) [12] ( E + Iz + Sz + 2 Iz Sz ) n2 = 1 ( E + Iz − Sz − 2 Iz Sz ) 4 n3 = 1 ( E − Iz + Sz − 2 Iz Sz ) 4 n4 = 1 ( E − Iz − Sz + 2 Iz Sz ) 4 n1 = 1 4 where these relationships can easily be verified by substituting back in the definitions of the operators in terms of populations. can be written in terms of E. It must be remembered that the populations used to derive these equations are really the deviation of the populations from their equilibrium values. represented by E. by first introducing one more operator E. the I and S spin magnetizations should properly be their deviations from their equilibrium values.
given in the previous equation as a sum of W values. Finally. [14] the term (W ( ) + W ( ) + W + W )( I 1 2 I I 2 0 z − Iz0 ) describes the relaxation of I spin magnetization on its own. [14] describes the transfer of IzSz into I spin magnetization. this is sometimes called the self relaxation. It is clear that in the absence of the relaxation pathways between the αα and ββ states (W2 ). The self relaxation rate constant. Recall that WI(1) and WI( 2 ) are the relaxation induced rate constants for the two allowed transitions of the I spin (1–3 and 2–4). that the rate of change of Iz depends not only on Iz − Iz0 . there will be no cross relaxation. but also on Sz − Sz0 and 2IzSz. The rate constants for this transfer will be written ∆I = WI(1) − WI( 2 ) ( ) ∆S = WS(1) − WS( 2 ) ( ) According to the final Solomon equation. and is sometimes given the symbol σIS.d( Iz − Iz0 ) dt = − WI(1) + WI( 2 ) + W2 + W0 ( Iz − Iz0 ) − (W2 − W0 )( Sz − Sz0 ) − WI(1) − WI( 2 ) 2 Iz Sz ( ) ( ) 8. the term (W ( ) − W ( ) )2 I S 1 2 I I z z in Eq. Thus. (W2 –W0 ) is called the crossrelaxation rate constant. [14]. a further discussion of this is beyond the scope of these lectures. Even if W2 and W 0 are absent. This term is described as giving rise to transfer from S to I as it says that the rate of change of the I spin magnetization is proportional to the deviation of the S spin magnetization from its equilibrium value. is sometimes given the symbol RI or ρ I. is called cross relaxation.1 Interpreting the Solomon equations What the Solomon equations predict is. by which the magnetizations of the two different spins are connected. or between the αβ and βα states (W0 ). self relaxation still occurs. Only if these two rate constants are different will there be transfer from 2IzSz into I spin magnetization. This phenomena. the operator 2IzSz shows self relaxation with a rate constant 8–8 . The rate at which S magnetization is transferred to I magnetization is given by the term (W2 − W0 )(Sz − Sz0 ) in Eq. In Eq. This situation arises when there is crosscorrelation between different relaxation mechanisms. In other words the way in which the magnetization on the I spin varies with time depends on what is happening to the S spin – the two magnetizations are connected. if the S spin is not at equilibrium the I spin magnetization is perturbed. for example.3.
it is not proper to talk of a "T1 " relaxation time constant. it is assumed in this section that ∆I = ∆S = 0) d( Iz − Iz0 ) dt What this says is that if the S spin magnetization deviates from equilibrium there will be a change in the I spin magnetization at a rate proportional to (a) the crossrelaxation rate. This rate combined constant will be denoted RIS. The basis of the nuclear Overhauser effect can readily be seen from the Solomon equation (for simplicity. It is important to realize that in such a system Iz and S z do not relax with a simple exponentials. Before embarking on this discussion it is important to realize that although the states represented by operators such as Iz and S z cannot be observed directly. For such a twospin system. after Fourier transformation. the Solomon equations can be written d( Iz − Iz0 ) d( Sz − S dt 0 z = − RI ( Iz − I 0 z ) − σ (S IS z − S ) − ∆I 2 Iz Sz 0 z Iz ∆I Sz ∆S 0 0 [15] IS z − I z ) − RS ( Sz − Sz ) − ∆S 2 I z Sz dt d2 Iz Sz = − ∆I ( Iz − Iz0 ) − ∆S ( Sz − Sz0 ) − RIS 2 Iz Sz dt The pathways between the different magnetization are visualized in the diagram opposite. Using these combined rate constants.4 Nuclear Overhauser effect The Solomon equations are an excellent way of understanding and analysing experiments used to measure the nuclear Overhauser effect. In effect. 8. by computing the value of the coefficient a.RIS = WI(1) + WI( 2 ) + WS(1) + WS( 2 ) ( ) σIS Note that the W2 and W0 pathways do not contribute to this. a spectrum with a peak of size –a at frequency ΩI. They only do this if the differential equation is of the form ) = −σ ( I 2IzSz dIz = − RI ( Iz − Iz0 ) dt which is plainly not the case here. they can be made observable by the application of a radiofrequency pulse. the appearance of the subsequently observed spectrum is predicted. σIS and (b) the extent of the deviation of the S spin 8–9 = − RI ( Iz − Iz0 ) − σ IS ( Sz − Sz0 ) . ideally a 90° pulse ) x aIz (→ − aI y π 2 I The subsequent recording of the free induction signal due to the evolution of the operator Iy will give. Note that as dIz0 dt = 0 (the equilibrium magnetization is a constant). the derivatives on the lefthand side of these equations can equally well be written dIz dt and dSz dt . therefore.
To work out what will happen during the mixing time the differential equations dIz (t ) = − RI ( Iz (t ) − Iz0 ) − σ IS ( Sz (t ) − Sz0 ) dt dSz (t ) = −σ IS ( Iz (t ) − Iz0 ) − RS ( Sz (t ) − Sz0 ) dt need to be solved (integrated) with this initial condition. 8. there will be no such effect unless σIS is nonzero. Sequence (b) is used to record the reference spectrum in which the intensities are unperturbed. therefore. It will be seen later on that such pathways are only present when there is dipolar relaxation between the two spins and that the resulting crossrelaxation rate constants have a strong dependence on the distance between the two spins. One simple way to do this is to use the initial rate approximation.4. it can be assumed that the initial conditions set out in Eq. 8–10 . The observation of a nuclear Overhauser effect is therefore diagnostic of dipolar relaxation and hence the proximity of pairs of spins. Sequence (a) involves selective inversion of the S spin – shown here using a shaped pulse. a nonselective 90° pulse is applied and the spectrum recorded. The effect is of enormous value. and it is this change in intensity of the I spin when the S spin is perturbed which is termed the nuclear Overhauser effect. called the mixing time. This change in the I spin magnetization will manifest itself as a change in the intensity in the corresponding spectrum. in structure determination by NMR. The S spin is then not at equilibrium so magnetization is transferred to the I spin by crossrelaxation.1 Transient experiments A simple experiment which reveals the NOE is to invert just the S spin by applying a selective 180° pulse to its resonance. After a suitable period. which requires the presence of the W2 and W0 relaxation pathways.from equilibrium. so. After the selective pulse the situation is Iz (0) = Iz0 Sz (0) = − Sz0 [16] where Iz has been written as Iz(t) to emphasize that it depends on time and likewise for S. This involves assuming that the mixing time is sufficiently short that. on the righthand side of the equations. for the first equation dIz (t ) = − RI ( Iz0 − Iz0 ) − σ IS ( − Sz0 − Sz0 ) dt init = 2σ IS Sz0 This is now easy to integrate as the righthand side has no dependence on Iz(t) 180° (a) S 90° τm 90° (b) Pulse sequence for recording transient NOE enhancements. Plainly. [16] apply. τm.
ref Iz0 so in this case η is Visualization of how an NOE difference spectrum is recorded. This latter term results in a change in the intensity of the I spin signal.τm 0 ∫ dIz (t ) = ∫ 2σ IS Sz dt 0 0 τm Iz (τ m ) − Iz (0) = 2σ ISτ m Sz0 Iz (τ m ) = 2σ ISτ m Sz0 + Iz0 This says that for zero mixing time the I magnetization is equal to its equilibrium value. The difference spectrum. The enhancement is assumed to be positive. A single experiment for one value of the mixing time will reveal the presence of NOE enhancements. [16] and. defined as (perturbed spectrum – unperturbed spectrum) corresponds to the difference Iz (τ m ) − Iz . assuming for simplicity that Iz0 = Sz0 the following solutions are found 8–11 .1. σIS. but that as the mixing time increases the I magnetization has an additional contribution which is proportional to the mixing time and the crossrelaxation rate. but they can be integrated using standard methods (symbolic mathematical programmes such as Mathematica are particularly useful for this). is defined as S I enhanced reference difference η= intensity in enhanced spectrum . this is a method used for measuring the crossrelaxation rate constant. This initial rate approximation is valid provided that σ ISτ m << 1 and RSτ m << 1 the first condition means that there is little transfer of magnetization from S to I. 8. Using the initial conditions given in Eq.intensity in reference spectrum intensity in reference spectrum Iz (τ m ) − Iz .1 Advanced topic: longer mixing times At longer mixing times the differential equations are a little more difficult to solve. The normal procedure for visualizing these enhancements is to record a reference spectrum in which the intensities are unperturbed. η. ref = 2σ ISτ m Sz0 + Iz0 − Iz0 = 2σ ISτ m Sz0 The NOE enhancement factor. ref 2σ ISτ m Sz0 = η(τ m ) = Iz .g. and if I and S are of the same nuclear species ( e. and this change is called an NOE enhancement. their equilibrium magnetizations are equal so that η(τ m ) = 2σ ISτ m Hence a plot of η against mixing time will give a straight line of slope σIS. In terms of zmagnetizations this means that Iz .ref = Iz0 . and the second means that the S spin remains very close to complete inversion. both proton).4.
0 0. in fact it is the sum of two exponentials but their time constants are not sufficiently different for this to be discerned. (Iz – I z0 )/Iz0 .0 0. and so the two rate constants λ 1 and λ 2 differ by a quantity of the order of σIS. The figure below shows the typical behaviour predicted by these equations (the parameters are RI = RS = 5σIS) 1.5 1. on an expanded scale. The NOE enhancement is more easily visualized by plotting the difference magnetization. Differentiation of the expression for Iz as a function of τ m shows that the maximum enhancement is reached at time τ m. As expected for these two coupled differential equations.0 The S spin magnetization returns to its equilibrium value with what appears to be an exponential curve. If RI and RS are not too dissimilar. R is of the order of σIS. the plot now shows the positive NOE enhancement reaching a maximum of about 15%.Iz (τ m ) 2σ IS = exp( − λ2τ m ) − exp( − λ1τ m ) + 1 Iz0 R [ ] Sz (τ m ) RI − RS = exp( − λ1τ m ) − exp( − λ2τ m ) Iz0 R [ ] + 1 − exp( − λ1τ m ) + exp( − λ2τ m ) where 2 2 R = RI2 − 2 RI RS + RS + 4σ IS λ1 = 1 2 [ RI + RS + R] λ2 = 1 2 [ RI + RS − R] These definitions ensure that λ1 > λ 2 . The I spin magnetization grows towards a maximum and then drops off back towards the equilibrium value.5 0 10 × (Iz–Iz0)/Iz0 (Iz/Iz0) (Sz/Iz0) time 0.5 1. integration gives a time dependence which is the sum of two exponentials with different time constants.max = 1 λ ln 1 λ1 − λ2 λ2 and that the maximum enhancement is 8–12 .
have the advantage of reducing the size of the I spin signal which has to be removed in the difference experiment.3 Steady state experiments The steadystate NOE experiment involves irradiating the S spin with a radiofrequency field which is sufficiently weak that the I spin is not affected. the NOE enhancement is revealed by subtracting a reference spectrum which has equilibrium intensities.e. [15] can be written d( Iz − Iz0 ) dt therefore Iz.4.4.2 The DPFGSE NOE experiment From the point of view of the relaxation behaviour the DPFGSE experiment is essentially identical to the transient NOE experiment. Further discussion of this experiment is deferred to Chapter 9.SS − Iz0 ) − σ IS (0 − Sz0 ) = 0 Pulse sequence for recording steady state NOE enhancements. meaning Sz = 0. The irradiation is applied for long enough that the S spin is saturated. however. It does. This does not influence the build up of the NOE enhancement on I.Iz (τ m. (dIz dt ) = 0 .SS − Iz . 8. which means that none of the magnetizations are changing. ref σ IS Sz0 ηSS = = Iz .SS = SS 90° S (a) 90° (b) = − RI ( Iz. Under these conditions the first of Eqs. Sequence (a) involves selective irradiation of the S spin leading to saturation. The NOE enhancement. will be Iz. and that the steady state has been reached.max ) − Iz0 Iz0 2σ IS = R − λ1 −λ2 λ1 R λ1 R − λ2 λ2 8. as defined above. i. The only difference is that the I spin starts out saturated rather than at equilibrium. ref RI Iz0 8–13 . Sequence (b) is used to record the reference spectrum in which the intensities are unperturbed. σ IS 0 Sz + Iz0 RI As in the transient experiment.
the size of the NOE enhancement will be reduced. Starting with equilibrium magnetization on the I and S spins.4. although the size of the enhancement does depend on the crossrelaxation rate constant the size of the enhancement cannot be interpreted in terms of this rate constant alone. It is important to realise that the value of the steadystate NOE enhancement depends on the ratio of crossrelaxation rate constant to the self relaxation rate constant for the spin which is receiving the enhancement. for example as a result of interaction with many other spins.HA HB HC HD Irradiation of proton B gives a much larger enhancement on proton A than on C despite the fact that the distances to the two spins are equal. this zmagnetization on spin S is rendered 8–14 . due to the interaction with proton D. the magnetization has an amplitude label which depends on the evolution during tl. So. 8. The smaller enhancement on C is due to the fact that it is relaxing more quickly than A. the relaxation rate of the S spin does not enter into the relationship simply because this spin is held saturated during the experiment. the initial rate approximation will be used: dSz (τ m ) = −σ IS ( − cos Ω I t1 Iz0 − Iz0 ) − RS ( − cos Ω S t1Sz0 − Sz0 ) dt init = σ IS (cos Ω I t1 + 1) Iz0 + RS (cos Ω S t1 + 1)Sz0 Integrating gives τm t1 τmix t2 Pulse sequence for NOESY. Y X In contrast to the transient experiment. The key difference is that instead of the magnetization of the S spin being inverted at the start of the mixing time. the zmagnetizations present at the start of the mixing time are (other magnetization will be rejected by appropriate phase cycling) Sz (0) = − cos Ω S t1Sz0 The equation of motion for Sz is dSz (t ) = −σ IS ( Iz (t ) − Iz0 ) − RS ( Sz (t ) − Sz0 ) dt As before. This point is illustrated by the example in the margin. If this spin is relaxing quickly. the steady state enhancement only depends on the relaxation of the receiving spin (here I).4 Advanced topic: NOESY The dynamics of the NOE in NOESY are very similar to those for the transient NOE experiment. Iz (0) = − cos Ω I t1 Iz0 ∫ dS (t ) = ∫ [σ (cos Ω t z IS 0 0 τm I 1 + 1) Iz0 + RS (cos Ω S t1 + 1)Sz0 dt ] Sz (τ m ) − Sz (0) = σ ISτ m (cos Ω I t1 + 1) Iz0 + RSτ m (cos Ω S t1 + 1)Sz0 = σ ISτ m Iz0 + RSτ m Sz0 + cos Ω I t1[σ ISτ m ]I 0 z Sz (τ m ) = σ ISτ m (cos Ω I t1 + 1) Iz0 + RSτ m (cos Ω S t1 + 1)Sz0 − cos Ω S t1Sz0 {a} {b} {c} + cos Ω S t1[ RSτ m − 1]Sz0 After the end of the mixing time.
The intensity of the peak grows back towards zero linearly with the mixing time and at a rate depending on RS. it is seen that the axial peak is zero for zero mixing time and then grows linearly depending on RS and σIS. This peak arises from zmagnetization which has recovered during the mixing time. {ΩS. ΩI F1 0 ΩS ΩS ΩI {c} {a} {b} F2 8–15 . ΩI}. The intensity of the cross peak grows linearly with the mixing time and also depends on σIS. ΩI} are found. Term {c} evolves at ΩS during t1 and Ω S during t2.5 Sign of the NOE enhancement We see that the time dependence and size of the NOE enhancement depends on the relative sizes of the crossrelaxation rate constant σIS and the self relaxation rate constants R I and RS. This leads to a negative NOE enhancement. in t2 it evolves at Ω S . {b} and {c} all represent different peaks in the NOESY spectrum. decaying during that time at a rate determined by RS. It will be seen later that whereas R I and R S are positive. and hence a positive NOE enhancement. In this initial rate limit. If the calculation is repeated using the differential equation for Iz a complimentary set of peaks at {0. ΩS} and as Rsτm << 1 in the initial rate. ΩS}. this peak is negative. a situation typical for large molecules in viscous solvents. It turns out that these selfrates are always positive. but the crossrelaxation rate constant can be positive or negative. it is therefore a cross peak at {ΩI. σIS can be either positive or negative. The reason for this is that σIS = (W2 – W0 ) and it is quite possible for W0 to be greater or less than W2 . the magnetization is on spin S. This situation is typical for small molecules is nonviscous solvents. Under some conditions W0 and W2 can become equal and then the NOE enhancement goes to zero.F2} = {0. whereas if σIS is negative all the peaks will have the same sign. Term {a} has no evolution as a function of t1 and so will appear at F1 = 0. The three terms {a}. Term {b} evolves at ΩI during t1 and ΩS during t2. 8. A negative crossrelaxation rate constant means that if spin S deviates from equilibrium crossrelaxation will decrease the magnetization on spin I. A positive crossrelaxation rate constant means that if spin S deviates from equilibrium crossrelaxation will increase the magnetization on spin I.observable by the final 90° pulse. This is therefore an axial peak at {F1. it is therefore a diagonal peak at {ΩS. This leads to an increase in the signal from I. ΩI} and {ΩI.4. If σIS is positive. this is analogous to the transient NOE experiment. and so will precess at ΩS during t2. the diagonal and cross peaks will be of opposite sign. ΩS }. This peak arises from S spin magnetization which remains on S during the mixing time.
When an RF pulse is applied all of the spins experience the same oscillating field. In fact.8. Aij ( ) This factor depends on the quantum mechanical details of the interaction. We already know from Chapter 3 that transitions are caused by transverse magnetic fields (i. An RF pulse gives rise to just such a field. if the spins need to increase their energy to go to equilibrium. Likewise. Thus we will see that NMR relaxation rate constants are particularly sensitive to molecular motion. these fields vary randomly in direction and amplitude. However. They are made time varying by the random motions (rotations. we are in a position to discuss the kinds of factors which influence these rate constants.e. If the spins need to lose energy to return to equilibrium they give this up to the motion of the molecules. so what we seek is the origin of these transitions. This is the origin of the original name for longitudinal relaxation: spinlattice relaxation. in particular) which result from the thermal agitation of the molecules and the collisions between them. Recall from section 8. the amounts of energy given up by the spins are tiny compared to the kinetic energies that molecules have.5 Origins of relaxation We now turn to the question as to what causes relaxation.1 that relaxation involves transitions between energy levels. often due to interactions of spins with one another or with their environment in some way. Of course. In addition. However. The lattice does not refer to a solid. meaning that they only affect a few spins and not the whole sample. Let us consider the rate constant Wij for transitions between levels i and j.1 Factors influencing the relaxation rate constant The detailed theory of the calculation of relaxation rate constants is beyond the scope of this course. in the xyplane) which are oscillating close to the Larmor frequency. this energy comes from the motion of the molecules. this turns out to depend on three factors: Wij = Aij × Y × J ω ij We will consider each in turn. so they are hardly affected. Relaxation is essentially the process by which energy is allowed to flow between the spins and molecular motion. The kind of transitions which lead to relaxation are different in that the transverse fields are local. it is precisely their random nature which drives the sample to equilibrium. The spin factor. 8. but to the motion of the molecules with which energy can be exchanged. for example if the population of the β state has to be increased.5. For 8–16 . there is an important distinction between the kind of transitions caused by RF pulses and those which lead to relaxation. The fields which are responsible for relaxation are generated within the sample.
On account of the large gyromagnetic ratio of the electron (when compared to the nucleus) such paramagnetic species are often a significant source of relaxation.2). this orientation will vary randomly with time (it is said to be a random function of time). At any instant. However. Y This is just a measure of how large the interaction causing the relaxation is. 8. all spins in the system) 8–17 Paramagnetic species have unpaired electrons. Fi(t). as we saw with RF pulses. ωij. Correlation functions To make the discussion concrete. In a two spin system the transition between the αα and ββ cannot be brought about by a simple oscillating field in the transverse plane. this random field will not be the same as that experienced by the first spin.5. There is no way that a detailed knowledge of each of these random fields can be obtained. has a large effect on relaxation rate constants. The average field experienced by the spins is found by taking the ensemble average – that is adding up the fields for all members of the ensemble (i. suppose that a spin in a sample experiences a magnetic field due to a dissolved paramagnetic species. These generate magnetic fields which can interact with nearby nuclei. Its size depends on the detailed origin of the random fields and often it is related to molecular geometry. and as both are subject to random thermal motion. and so the magnetic field will be a random function of time. This also experiences a random magnetic field. Let the field experienced by this first spin be F1(t). so it is well worthwhile spending some time in understanding the form that this function takes. The spectral density. J(ω ). Recall that molecular motion is the effect which makes the random fields vary with time. the field will only have an effect on the spins if it is oscillating at the correct frequency. Now consider a second spin in the sample. not all oscillating fields can cause transitions between all levels.e. We can think of Aij as representing a kind of selection rule for the process – like a selection rule it may be zero for some transitions.6. each spin experiences a different random field. . in fact it needs a more complex interaction that is only present in the dipolar mechanism (section 8. For a macroscopic sample. due to the interaction with the paramagnetic species. but in some cases it is possible to characterise the overall behaviour of the system quite simply. The size of the magnetic field will depend on the relative orientation of the spin and the paramagnetic species. to cause the transitions. J(ω ij ) This is a measure of the amount of molecular motion which is at the correct frequency. The size factor.2 Spectral densities and correlation functions The value of the spectral density. F2(t).example. The spectral density is a measure of how much of the motion is present at the correct frequency.
and F 1(t+τ ) is the field experienced at a time τ later. This is illustrated in the figure below. (a) F(t) (b) F(t)F(t+1) (c) F(t)F(t+15) 5 1. τ ) = F1 (t ) F1* (t + τ ) + F2 (t ) F2* (t + τ ) + F3 (t ) F3* (t + τ ) + K = F (t ) F * (t + τ ) c a b d Visualization of the different timescales for random motion. all possible pairs are plotted. (b) is a plot of the value of F multiplied by its value one data point later – i. (b) is a snap shot a very short time after (a). more spins have moved.τ). the ensemble average is taken – the result will be for them to reinforce one another and hence give a finite value for G(t. is thus a function which characterises the memory that the system has of a particular arrangement of spins in the sample. this is a property of stationary random functions. An important property of random functions is the correlation function. (a) is the starting position: the black dots are spins and the open circle represents a paramagnetic species. hardly any of the spins have moved. As a result. The correlation function. As τ gets longer. randomly distributed about zero. this ensemble average turns out to be independent of the time. there are about 100 separate time points. In the limit. goes to zero.2 1. it will therefore be written G(τ). and G(τ ) falls. the product F1 (t ) F1* (t + τ ) need not necessarily be positive. defined as G(t. the product F1 (t ) F1* (t + τ ) will be positive. the star indicates the complex conjugate. plot (b). (c) is the same as (b) but for a time interval of 15 data points. so this ensemble average will be zero. G(τ ). the correlation function is independent of the time t. For 8–18 . If the time τ is short the spins will not have moved very much and so F1(t+τ) will be very little different from F 1(t). The ensemble average of all these Fi (t ) Fi * (t + τ ) is thus less than it was when τ was shorter.τ). This is illustrated in plot (c) above.2 5 (a) is a plot of the random function F(t) against time. G(t. once τ becomes sufficiently long.τ). (c) is a snapshot at a longer time. one data point to the right. the initial arrangement becomes more and more disturbed. so when the Fi (t ) Fi * (t + τ ) are added together for a particular time. the Fi (t ) Fi * (t + τ ) are randomly distributed and their ensemble average. but part of the original pattern is still discernible. G(t. As time proceeds. – that is. For times τ which are much less than the time it takes for the system to rearrange itself G(τ) will be close to its maximum value. t. (d) is after a long time. For stationary random functions. F1(t) is the field experienced by spin 1 at time t.2 5 1.F(t ) = F1 (t ) + F2 (t ) + F3 (t ) + K For random thermal motion. which allows for the possibility that F(t) may be complex. G(t. The two arrows indicate the spacing over which the correlation is calculated.e. The same is true for all of the other members of then ensemble.τ) thus has its maximum value at τ = 0 and then decays to zero at long times. Typically. the spin will have had more chance of moving and so F1(t+τ) will differ more and more from F1(t). the Fi(t) are signed quantities. all the spins have moved and the original pattern is lost.
For a simple exponential correlation function. the corresponding spectral density is a Lorentzian exp( − τ τ c ) Fourier Transform→ 2τ c 2 1 + ω 2τ c J(ω) 0 0 1 2 ωτc 3 4 5 Plot of the spectral density as a function of the dimensionless variable ωτ c.sufficiently long times. note how it drops off significantly once the product ωτc begins to exceed ~1. τ c is called the correlation time. The simplest form for G(τ) is G(τ ) = G(0) exp( − τ τ c ) [18] 0 0 0 G(τ ) G(0) the variable τ appears as the modulus. at the frequency indicated by the dashed line the contribution at correlation time b is greater than that for either correlation times a or c. such as oscillating back and forth about a mean position. 8–19 J(ω) a b c 0 ω . However. when the time is of the order of τc significant rearrangements have taken place and the correlation function has fallen to about half its initial value. The amount at frequencies higher that 1/τc tails off quite rapidly as the frequency increases. For times much longer than τc the spins have moved to completely new positions and the correlation function has fallen close to zero. frequencies from zero up to about 1/τc are present. the spins are subject to random motions with a range of different periods. [18]. Spectral densities τc τ The correlation function is a function of time. given in Eq. However. The plot opposite compares the spectral densities for three different correlation times. the area under the plot remains the same. just like a free induction decay. it can be Fourier transformed to give a function of frequency. Note that as the correlation time decreases the spectral density moves out to higher frequencies. the spectral density gives a measure of the amount of motion present at different frequencies. For times much less than the correlation time the spins have not moved much and the correlation function is close to its original value. for random motion characterised by a correlation time τc. resulting in the same value of G(τ) for positive and negative values of τ. as the name implies. the spectral density would show a peak at that frequency. b an intermediate value and c the shortest. So. The curve is a lorentzian This function is plotted in the margin. Generally. so the contribution at lower frequencies is decreased. The resulting frequency domain function is called the spectral density. curve a is the longest. so the spectral density shows a range of frequencies rather than having peaks at discreet frequencies. This means that the correlation is the same with time τ before and time τ after the present time. In particular. G(τ) tends to zero. The spectral density is usually denoted J(ω) G(τ ) Fourier Transform→ J (ω ) If the spins were executing a well ordered motion.
ω0. varies with the correlation time. Motion which gives rise to correlation times which are much shorter than 1/ω0 is described as being in the fast motion (or extreme narrowing) limit.4 ns. This is hardly surprising as to cause relaxation – that is to cause transitions – the field needs to have components oscillating at the Larmor frequency. the maximum contribution at frequency ω is found when τc is 1/ω. the only relevant spectral density is that at the Larmor frequency. mathematically this limit is ω0τ c >> 1. We see that. falls. so to have the most rapid relaxation the correlation time should be 1/ω0. but here we have plotted the relaxation time constant. and the corresponding relaxation time (T1 = 1/W) plotted as a function of the correlation time. As the correlation time increases the amount of spectral density at the Larmor frequency increases and so the relaxation rate constant increases. In terms of the relaxation time. After this point. and hence the rate constant. which limit we are in depends on the Larmor frequency. Small molecules have correlation times 8–20 .5.3 The "T1 minimum" In the case of relaxation of a single spin by a random field (such as that generated by a paramagnetic species). the spectral density is a maximum when τc is 1/ω0. there is a maximum in the rate constant when τ c = 1/ω0. τ c . for a given Larmor frequency. which in turn depends on the nucleus and the magnetic field. W. reaching a maximum when τc = 1/ω0. We have just seen that for a given frequency. like Goldilocks and the Three Bears. efficient relaxation requires a correlation time which is neither too fast nor too slow. this has important consequences which are described in the next section. T1. but not that much as the energy of the motion is spread over a very wide frequency range. Plot (b) shows the same effect. 8. (a) W T1 (b) 1/ω0 τc 1/ω0 τc Plot (a) shows how the relaxation rate constant. Motion which gives rise to correlation times which are much slower than 1/ ω0 is described as being in the slow motion (or spin diffusion) limit. T1 . fast motion means ω0τ c << 1. W. the spectra density at the Larmor frequency.For this spectral density function. This is illustrated in the plots below which show the relaxation rate constant. For a Larmor frequency of 400 MHz we would expect the fastest relaxation when the correlation time is 0. Put mathematically. this shows a minimum. Clearly. At very short correlation times (τc << 1/ω0) there is some spectral density at the Larmor frequency. there is a minimum in T 1 which corresponds to the maximum in W. ω 0.
The relaxation induced by the dipolar coupling is proportional to the square of the coupling. As a molecule tumbles in solution the direction of this vector changes and so the magnetic field changes. and these clearly fall in the slow motion limit. leading to intra. the overall interaction is a property of the pair of nuclei. Two spins are thus required for this interaction.2 The dipolar mechanism Each spin has associated with it a magnetic moment. Changes in the distance between the nuclei also result in a change in the magnetic field. are generally at far too high frequencies to give significant spectral density at the Larmor frequency.6. nuclei in the same molecule can approach much more closely than those in different molecules so that intramolecular relaxation is dominant. Generally. it is generally changes in orientation which are responsible for relaxation. measured relative to that of the applied magnetic field. their roles are reversible.6. The large magnetic moment of the electron means that paramagnetic species in solution are particularly effective at promoting relaxation. The size of the interaction depends on the inverse cube of the distance between the two nuclei and the direction of the vector joining the two nuclei.and intermolecular relaxation. such as proteins. 8. which do give such changes. in the sense that the second spin creates a field which is experienced by the first. however. However. The pair of interacting nuclei can be in the same or different molecules.significantly shorter than this (say tens of ps). can easily have correlation times of the order of a few ns. Thus it goes as 8–21 .6 Relaxation mechanisms So far. Each such source is referred to as a relaxation mechanism. So. but of these only a few are really important for spin half nuclei. therefore. 8. both very small and very large molecules tend to relax more slowly than mediumsized molecules. the source of the magnetic fields which give rise to relaxation and the origin of their time dependence have not been considered. Large molecules. Somewhat strangely. Such species include dissolved oxygen and certain transition metal compounds. As a result. and this is turn gives rise to a magnetic field which can interact with other spins. so such molecules are clearly in the fast motion limit. 8. one to "create" the field and one to "experience" it. molecular vibrations. There are quite a range of different mechanisms that can act.1 Paramagnetic species We have already mentioned this source of varying fields several times. However.
In liquids. Essentially. It is important to realize that in dipolar relaxation the effect is not primarily to distribute the energy from one of the spins to the other. This would not.2 γ 12γ 2 1 6 r12 where γ1 and γ2 are the gyromagnetic ratios of the two nuclei involved and r12 is the distance between them. on its own. the magnetic field experienced by a nucleus is different to that applied to the sample. bring the spins to equilibrium. As the size of the dipolar interaction depends on the product of the gyromagnetic ratios of the two nuclei involved. the lattice is the molecular motion.6. Relation to the NOE The dipolar mechanism is the only common relaxation mechanism which can cause transitions in which more than one spin flips. but not by very much. 8. In contrast. Thus.g. not surprisingly it follows that W0 > W2 and a negative NOE enhancement is seen. A detailed calculation shows that W2 > W 0 and so we expect to see positive NOE enhancements (section 8. every thing else being equal. with reference to section 8. the dipoledipole interaction turns molecular motion into an oscillating magnetic field which can cause transitions of the spins. W 0 corresponds to transitions at ( ω0. In the case where the two spins are the same (e. In this case. all that is observable is the average chemical shift.3 The chemical shift anisotropy mechanism The chemical shift arises because. and the resulting relaxation rate constants depends on the square of this. In the fast motion limit (ω0τc << 1) J(2ω0) is somewhat less than J(0).3. due to the effect of the electrons in a molecule. a protonproton pair will relax 16 times faster than a carbon13 proton pair. two protons) the two relevant spectral densities are J(2ω0) and J(0).4. pairs of nuclei with high gyromagnetic ratios are most efficient at promoting relaxation. At a more detailed level. in the slow motion limit (ω0τc >> 1) J(2ω0) is all but zero and so J(0) >> J(2ω0).S) and so it is the spectral density at this sum frequency which is relevant. Specifically.I + ω0. which results from the molecule rapidly experiencing all possible orientations by rapid molecular tumbling.5).S ) and so for these it is the spectral density at this difference frequency which is relevant. In contrast. the dipolar interaction provides a path by which energy can be transferred between the lattice and the spins. For example.I – ω 0. The rate constant W2 corresponds to transitions which are at the sum of the Larmor frequencies of the two spins. Rather. the magnetic field experienced by the nucleus 8–22 . the dipolar mechanism gives rise to transitions between the αα and ββ states (W2 ) and between the αβ and βα states (W0 ). ( ω0.
This is called chemical shift anisotropy (CSA). we have only discussed the fist of these two. which is a mathematical object represented by a three by three matrix.7 Transverse relaxation Right at the start of this section we mentioned that relaxation involved two processes: the populations returning to equilibrium and the transverse magnetization decaying to zero. it is not only the magnitude of the field which is altered but also its direction. σ xx σ = σ yx σ zx σ xy σ xz σ yy σ yz σ zy σ zz The element σxz gives the size of the extra field in the xdirection which results from a field being applied in the zdirection. 8. CSA relaxation is expected to be significant for nuclei with large shift ranges observed at high fields. So far. A rough estimate of the size of this anisotropy is that it is equal to the typical shift range. The second. it is the transverse fields which will give rise to changes in population. As has already been discussed. is called transverse (or spinspin) relaxation. and so they have the necessary time variation to cause relaxation.depends on the orientation of the molecule relative to the applied magnetic field. Detailed calculations show that the relaxation induced by CSA goes as the square of the field strength and is also proportional to the shift anisotropy. in which the transverse magnetization decays. The size of the CSA is specified by a tensor. The changes are very small. These fields vary in size as the molecule reorients. 8–23 . It is usually insignificant for protons. One convenient way of imagining the effect of CSA is to say that due to it there are small additional fields created at the nucleus – in general in all three directions. In addition. likewise. These elements depend on the electronic properties of the molecule and the orientation of the molecule with respect to the magnetic field. but sufficient to be detectable in the spectrum and to give rise to relaxation. σyz gives the extra field in the ydirection and σ zz that in the zdirection. So.
It is immediately clear that if these phases are random the net transverse magnetization of the sample will be zero as all the individual contributions will cancel. and in general have a component along x. See text for details.(a) (b) (c) Depiction of how the individual contributions from different spins (shown on the left) add up to give the net transverse magnetization (on the right). or in other words the pulse generates a coherence. y and z. Each spin in the sample can be thought of as giving rise to a small contribution to the magnetization. An RF pulse applied to equilibrium magnetization generates transverse magnetization. the phases must not be random. these contributions are precessing at the Larmor frequency in the transverse plane. as 8–24 . This is the situation that pertains at equilibrium and is shown in (a) in the figure above. We can represent each of these contributions by a vector precessing in the transverse plane. In quantum mechanics this is described as a coherence. Transverse relaxation destroys this coherence by destroying the alignment of the individual contributions. For there to be net magnetization. rather there has to be a preference for one direction. this is shown in (b) in the figure above. The transverse contributions behave in a more complex way as. just like the net transverse magnetization. these contributions can be in any direction. The direction in which these vectors point can be specified by giving each a phase – arbitrarily the angle measured around from the x axis. The individual contributions along z add up to give the net zmagnetization of the sample.
We can see that this makes sense as this part of transverse relaxation requires no transitions. So. the transverse rate constant carries on increasing and shows no such maximum. 8–25 .2 we see that J(0) = 2τc.shown in (c) above. the part which has the same origin as longitudinal relaxation is called the nonsecular part. this is in contrast to longitudinal relaxation is most rapid for a particular value of the correlation time. we can see that these jumps could be brought about by local oscillating fields which have the same effect as pulses. at random. J(0). The second way of destroying the coherence is to make the vectors get out of step with one another as a result of them precessing at different Larmor frequencies. Again. rather.5. This is exactly what causes longitudinal relaxation. The plot below compares the behaviour of the longitudinal and transverse relaxation rate constants. Looking at the result from section 8. The longitudinal rate constant shows a maximum. Our picture indicates that there are two ways in which the coherence could be destroyed. all we need for it to do is to be different at different locations in the sample. W transverse longitudinal τc 1/ω0 Comparison of the longitudinal and transverse relaxation rate constants as a function of the correlation time for the fixed Larmor frequency. As the correlation time increases the longitudinal rate constant goes through a maximum. The first is to make the vectors jump to new positions. so too does the relaxation rate constant. anything that causes longitudinal relaxation will also cause transverse relaxation. and so as the correlation time gets longer and longer. but the transverse rate constant simply goes on increasing. in which we imagine the local fields causing the spins to flip. Detailed calculations show that in the fast motion limit the two relaxation rate constants are equal. This latter contribution is called the secular part of transverse relaxation. Thus large molecules in the slow motion limit are characterised by very rapid transverse relaxation. We can attribute this to the secular part of transverse relaxation which depends on J(0) and which simply goes on increasing as the correlation time increases. However. a local field plays the part we need but this time we do not need it to oscillate. Drawing on our analogy between these vectors and the behaviour of the bulk magnetization. just a field to cause a local variation in the magnetic field. It turns out that the secular part depends on the spectral density at zero frequency.
There are two general ways in which one pathway can be isolated from the many possible. Unlike phase cycling. As a result. We shall see that this dephasing and rephasing approach can be used to select particular coherences. The second method of selection is to use field gradient pulses. The effect of changing the phase of a pulse is easy to visualise in the usual rotating frame. For example. and the coherence restored. any coherences present dephase and are apparently lost. 9. we may wish to pass the spins through a state of multiple quantum coherence at a particular point. Both of these selection methods can be described in a unified framework which classifies the coherences present at any particular point according to a coherence order and then uses coherence transfer pathways to specify the desired outcome of the experiment. for example. this dephasing can be undone. In this method the phases of the pulses and the receiver are varied in a systematic way so that the signal from the desired pathways adds and signal from all other pathways cancels. a 90° pulse about the xaxis rotates † Chapter 9 "Coherence Selection: Phase Cycling and Gradient Pulses" © James Keeler 2001 and 2003. by application of a subsequent gradient.9 Coherence Selection: Phase Cycling and Gradient Pulses† 9. For example. The first is phase cycling. However. a pulse whose role is to generate doublequantum coherence from antiphase magnetization may also generate zeroquantum coherence or transfer the magnetization to another spin. or plan for the magnetization to be aligned along the zaxis during a mixing period. These are short periods during which the applied magnetic field is made inhomogeneous. However. the use of field gradient pulses does not require repetition of the experiment.2 Phase in NMR In NMR we have control over both the phase of the pulses and the receiver phase. So. something which is probably required in any case in order to achieve the required signaltonoise ratio. If no steps are taken to suppress these unwanted pathways the resulting spectrum will be hopelessly confused and uninterpretable. Each time a radiofrequency pulse is applied there is this possibility of branching into many different pathways. 9–1 .1 Introduction The pulse sequence used in an NMR experiment is carefully designed to produce a particular outcome. it is usually the case that the particular series of events we designed the pulse sequence to cause is only one out of many possibilities. Phase cycling requires that the experiment is repeated several times.
the frequencies of the signals which are digitized in the spectrometer are the offset frequencies at which the spins evolve in the rotating frame. In the spectrometer the incoming signal at ω0 is downshifted by subtracting the receiver reference which. called the offset Ω. which we will assume is oscillating at ωrf Sref = cos ω rf t The output of the mixer is the product SFIDSref SFID Sref = A cos ω rf t cos ω 0 t = 1 A cos(ω rf + ω 0 )t + cos(ω rf − ω 0 )t 2 [ ] The first term is an oscillation at a high frequency (of the order of twice the Larmor frequency as ω0 ≈ ωrf) and is easily removed by a lowpass filter. this effectively multiplies the incoming signal by a locally generated reference signal. whereas the same pulse applied about the yaxis rotates the magnetization onto the xaxis. In this frame. ωrf. Ultimately. 9. which we will write as SFID = cos ω 0 t where ω 0 is the Larmor frequency. called the receiver reference or carrier. The details of this downshifting process can be quite complex. but the overall result is simply that a fixed frequency. This is in line with the previous comment that this downshifting process is equivalent to detecting the precession in the rotating frame.1 Detector phase The quantity which is actually detected in an NMR experiment is the transverse magnetization. in effect. a spin whose Larmor frequency is ω0 precesses at ( ω0 – ωrf). Ω. will be equal to the frequency of the radiofrequency pulses. is subtracted from the frequency of the incoming NMR signal. this appears at the probe as an oscillating voltage. So. The rotating frame which we use to visualise the effect of pulses is set at the transmitter frequency. Frequently. Often this whole process is summarised by saying that the "signal is detected in the rotating frame". this receiver reference frequency is the same as the transmitter frequency used to generate the pulses applied to the observed nucleus. We shall assume that this is the case from now on.magnetization from z onto –y. so that the field due to the radiofrequency pulse is static.2. 9–2 . as we have already decided. Within the spectrometer this signal is shifted down to a much lower frequency in order that it can be digitized and then stored in computer memory. Sref. The second term is an oscillation at the offset frequency. The idea of the receiver phase is slightly more complex and will be explored in this section. The downshifting process in the spectrometer is achieved by an electronic device called a mixer. The NMR signal – that is the free induction decay – which emerges from the probe is a radiofrequency signal oscillating at close to the Larmor frequency (usually hundreds of MHz).
and the projection of the magnetization onto the reference axis is proportional to cos (Ωt – φ). something which is easily achieved in the spectrometer. After time t the magnetization has precessed through an angle θ = Ωt. How this is put to good effect is described in the next section. The projection of the magnetization onto the reference axis is proportional to cos Ωt.We can go further with this interpretation and say that the second term represents the component of the magnetization along a particular axis (called the reference axis) in the rotating frame. this is illustrated below θ t=0 time t At time zero the magnetization is assumed to be aligned along the reference axis. Suppose now that the phase of the reference signal is shifted by φ. The component along the new reference axis is proportional to cos(Ωt − φ ) . Now the apparent angle of precession is θ = Ω t – φ. as shown below θ–φ φ t=0 time t Shifting the phase of the receiver reference by φ is equivalent to detecting the component along an axis rotated by φ from its original position (the previous axis is shown in grey). Effectively. Such a component varies as cos Ωt. 9–3 . assuming that at time zero the magnetization is aligned along the chosen axis. this shifts the reference axis by φ.
The final step in this whole process is regard the outputs of the two mixers as being the real and imaginary parts of a complex number: cos Ωt + i sin Ωt = exp(iΩt ) The overall result is the generation of a complex signal whose phase varies according to the offset frequency Ω. exp(iφsig): exp(iΩt ) exp iφsig ( ) As we saw in the previous section. and indeed we saw 9–4 . would be confusing as each peak would appear at both +Ω and –Ω. Ω . obtained by Fourier transformation. which certainly is under our control.2. The resulting spectrum. the projection along one axis is proportional to cos(Ωt) and to sin( Ωt) along the other. This phase is a combination of factors that are not under our control (such as phase shifts produced in the amplifiers and filters through which the signal passes) and a phase due to the pulse sequence. As is illustrated opposite. 9. the projection along x would be the same. The way out of this problem is to detect the signal along two perpendicular axes. The overall result of this phase is simply to multiply the final complex signal by a phase factor. as we have seen in the previous section the effect of the downshifting scheme used is to generate a signal of the form cos Ωt.y θ x The x and y projections of the black vector are both positive. If the receiver reference frequency is the same as the transmitter frequency.2 Quadrature detection We normally want to place the transmitter frequency in the centre of the resonances of interest so as to minimise offresonance effects. Knowledge of both of these projections enables us to work out the sense of rotation of the vector i. and at the same frequency. The sin modulated component is detected by having a second mixer fed with a reference whose phase is shifted by π /2. φsig. but along y it would be minus that of the black vector. may be both positive and negative. However. If the vector had precessed in the opposite direction (shown shaded).3 Control of phase In the previous section we supposed that the signal coming from the probe was of the form cos ω 0 t but it is more realistic to write the signal as cos(ω0t + φsig) in recognition of the fact that in addition to a frequency the signal has a phase. 9.e. Since cos(θ) = cos(–θ ) such a signal does not discriminate between positive and negative offset frequencies.2. it immediately follows that the offset frequencies. Following the above discussion the output of the mixer is cos(Ωt − π 2) = cos Ωt cos π 2 + sin Ωt sin π 2 = sin Ωt The output of these two mixers can be regarded as being the components of the magnetization along perpendicular axes in the rotating frame. we can also introduce another phase shift by altering the phase of the reference signal fed to the mixer. the sign of the offset.
5 Relative phase and lineshape We have seen that we can alter the phase of the spectrum by altering the phase of the pulse or of the receiver. S(ω ) exp(iΘ ) = B[ A(ω ) + iD(ω )] exp(iΦ ) exp(iΘ ) = B[ A(ω ) + iD(ω )] exp(i[Φ + Θ ]) As this is a numerical operation which can be carried out in the computer we are free to choose Θ to be the required value (here –Φ) in order to remove the phase factor entirely and hence give an absorption mode spectrum in the real part. In the complex notation. If we want just the absorption lineshape we need to somehow set Φ to zero. the output of the two mixers becomes cos(Ωt − φ rx ) and sin(Ωt − φ rx ) . the overall signal thus acquires another phase factor exp(iΩt ) exp iφsig exp( − iφ rx ) Overall. 9. usually called the receiver phase. 9–5 . then. exp(t/T2 ) has been added to impose a decay on the signal.4 Lineshapes Let us suppose that the signal can be written S(t ) = B exp(iΩt ) exp(iΦ ) exp( −t T2 ) where Φ is the overall phase (= φsig − φ rx ) and B is the amplitude. 9.2.that the cosine and sine modulated signals are generated by using two mixers fed with reference signals which differ in phase by π/2. This is what we do when we "phase the spectrum". in this case. If the phase of each of these reference signals is advanced by φrx. which is easily done by multiplying S(ω) by a phase factor exp(iΘ). Fourier transformation of S(t) gives the spectrum S(ω): S(ω ) = B[ A(ω ) + iD(ω )] exp(iΦ ) [1] absorption ( )