Notes Magnetism
Notes Magnetism
Notes Magnetism
Lecture Notes
Dr. Jorge Quintanilla SEPnet and Hubbard Theory Consortium, University of Kent and STFC Rutherford Appleton Laboratory
We will have 28 lectures, roughly split as 14 on Magnetism and 14 on Superconductivity. I will make handouts available after each lecture. The handouts can be found at http://blogs.kent.ac.uk/strongcorrelations/teaching/superconductivity-and-magnetism/ Be sure to check that page regularly. Check also the reading list available online. familiar with by the end of this course: There are two recommended texts that you should be very
Stephen Blundell, Magnetism in Condensed Matter (OUP 2001). James F. Annett, Superconductivity, Superuids and Condensates (OUP 2004). An
References to these two books will be abbreviated SB and JFA, respectively, in these lecture notes. expanded reding list with additional references will be made available as the course progresses.
There will be 6 problem sheets, with a few practice problems each. The problem sheets will be handed out or made available online 2 weeks before they are due. Every problem sheet will have an assesed component. There will also be two class tests: one on Magnetism on week 14 (one week after the Winter break) and one on Superconductivity on week 19. Finally, in the third term we will have an exam. The assesment pattern can be found on the web. In addition to the lectures, we will have a number of workshops: two just before the rst class tests, two just before the second one, and a number of optional ones after the lectures and class tests are over. convenient for everyone nearer the date. My room is Ingram 230 however I do spend a lot of time doing research the ISIS Facility, STFC Rutherford Appleton Laboratory, Harwell Science Campus, Didcot, Oxfordshire, OX11 0QX. The best form of contact in any case is email: [email protected]. You may also want to try Skype: j.quintanilla. In principle these have been scheduled in the latter part of the second term but we will agree whatever is most
2 2.1
Magnetism and superconductivity are two of the most fascinating phenomena known to humankind.
Magnetism, as manifested in the ability of some materials to attract iron and each other and to sense the
direction of the magnetic North, has been known for millenia. The earliest known magnetic artifact dates from the 4th centruy BC. It is a magnetic compass from China:
(This was not used to navigate, but to nd the spiritually most desirable alignment for new houses!) Even the word magnetism itself is very old. It comes from the Greek
magnes lithos,
2
The magnetic compass uses ferromagnetism. Other magnetic phenomena include paramagnetism and antiferromagnetism. Some of them have been discovered recently - in fact new forms of magnetism are being discovered even nowadays. In contrast to magnetism,
superconductivity serendipitously in 1911 (this year is the 100th anniversary!) when he was studying the electrical properties of metals at very low temperatures. He found that the resistivity of Mercury (Hg) completely vanished when it was cooled below a critical value of about 4.2K. Here is an old photo of the gentleman alongside his historic plot of resistivity vs temperature:
2.2
Applications
applications.
Computer hard disks store bits
Magnetism is ubiquitous in our technology. Sometimes the applications are quite humble, e.g. we use it to keep doors closed. More important is its application in
magnetic storage.
in the form of microscopic magnetisation domains on a ferromagnetic surface. The information is then read using a device that converts the magnetic eld into an electric signal (a magnetic read head). The read heads improved considerably with the discovery of giant magnetoresistance (GMR). In GMR, an electron current is passed through a stack of ultra-thin ferromagnetic layers. The magnetic elds created by dierent layers point in dierent directions. This scatters the electrons as they move from one layer to the next, increasing resistance. Under the inuence of an externally-applied magnetic eld, the dierent layers align their contributions, leading to a large drop in resistance:
2 3
Douglas Harper, Magnet, in The Online Etymology Dictionary, http://www.etymonline.com/index.php?term=magnet Images from Rudolf de Bruyn Ouboter, Heike Kamerlingh Onnes's Discovery of Superconductivity, Scientic American The gure is from Giant Magnetoresistance, in Physics Central, American Physical Society,
(accessed 29 July 2011). (March 1997) and from Dirk van Delft and Peter Kes, The discovery of superconductivity, Physics Toda y (September 2010).
These drops in resistence can be used to read the bits recorded on the magnetic disc underneath a read head. Thanks to GMR read heads, the density of magnetic storage kept increasing exponentially with time into the 21st century. This is what allowed, among other tings, the miniaturisation of the iPods. Fert and Grunberg, the discoverers of GMR, shared the Nobel prize in 2007. The increase in magnetic information storage density with time is even faster than the famous Moore's law for transistors in microchips. It is called Kryder's law. It is the great technological triumph of magnetism. Here is a graph:
Superconductivity also has important applications, although it is not as ubiquitous as magnetism because, while many useful magnetic properties occur at room temperature, all known superconductors need to be kept at liquid nitrogen or lower temperatures in order to remain superocnducting. Nevertheless superconductivity is routinely used wherever it is important to generate large magnetic elds, such as in medical
room-temperature su-
What is special, from a more fundamental point of view, about magnetism and superconductivity? First of all, they are both properties of matter whose understanding necessitatates they are both
emergent phenomena.
is from Giant
quantum mechanics.
American
Secondly,
The
gure
Magnetoresistance:
Research,
in
Physics
Central,
Physical
Society,
The laws of physics that govern the behaviour of dierent Yet some materials are magnetic, others are
superconducting, and many others do none of this. What is more, a superconducting or magnetic material will lose its properties if we change its thermodynamic conditions - for example, by raising the temperature. It is clear that just knowing the laws of physics is not enough to understand phenomena such as magnetism and superconductivity. We need to understand the principles of self-organisation whereby the prize-winning American physicist, Phil W. Anderson, summed this magic up with the motto
1023
particles
that make up the sample cooperate to allow new behaviour to emerge from the physical laws. The Nobel
More is dierent
We will do Magnetism rst because it is conceptually simpler and the theoretical tools required are less advanced, so by looking at the subjects in that order the learning curve will be less steep. (On the other hand as you will see superconductivity is a simpler subject in the sense that it deals with a single phenomenon. In contrast, in the rst part of the course we will be looking at paramagnetism, ferromagnetism, antiferromagnetism...)
3.1
E = / B = 0 E =
(1) (2)
B t
0 0
(3)
B = 0 J +
These equations determine the electric eld current density
E t B
given the charge density
(4)
t.
0 and
and 0 are
the electric permitivity and magnetic permeability of free space, respectively. What the equations tell us, respectively, is the following:
Gauss' law for electric elds: electric elds ow out of electric charges. We prove this by integratin g both sides of the equation over a region of volume outside surface
V,
theorem to show that the integral on the LHS is the same as the ux of the vector eld
through the
of
V.
0.
Gauss' law for magnetic elds: there are no magnetic charges (also called magnetic monopoles). The proof is similar to that for the electric eld.
Faraday's law: time-varying magnetic induction leads to circulating electric elds. in the wire.)
(A direct conse-
quence of this is that a time-dependent magnetic eld going through a loop of wire will induce a current
To understand we calculate the ux of the quantities on the LHS and RHS through an open surface
S.
We then invoke Gauss' theorem according to which the intergal on the LHS is the same as a line
Ampere's law: currents induce circulating magnetic eld (assuming that the electric eld is static). Again we can use Stokes' theorem to see this.
We know that magnets have certain properties like their ability to attract each other, etc. What is the reason
generate a magnetic induction one would have to set up a current going around a wire, however we nd that in the presence of any magnetic material we can detect and additional magnetic induction which is an intrinsic property of the magnet. The study of magnetism is the study of how this magnetic induction comes about, and what form it takes.
3.2
How do we know that magnets have their own, intrinsic magnetic induction? We know it because we can measure this induction. The device used to do this is called a
magnetometer.
magnetometer, but the majority are based on Faraday's law. One way to attain this is to build an electric circuit with a closed loop in it. We then place the sample in the middle of the loop. Then we either withdraw the sample from the loop (extraction magnetometer) or move our magnetic sample repeatedly in and outirrespectful of the mag tirrespectful of the maghrough the loop (vibration magnetometer). All the time we measure the electric current through the loop with an Amp-meter. There will be current if the circulation of the electric eld is nite. But this is given by
E du = t C
B ds,
S
(5)
which is the integral form of Faraday's law. If the sample does not have a magnetic induction associated with it then the RHS is zero and we will see no current. If, on the other hand, there is a nite magnetic induction generated by the sample, then as we move the sample the ux of induction through the surface will change with time:
This will make the RHS of the above equation nite and induce a circulating electric eld. That will create a current around the loop. Thus the detection of the current shows that there is a nite magnetic induction around the sample. In this way we nd that some magnetic materials do indeed generate their own magnetic induction. This is not the only way of measuring the magnetic induction associated with magnetic materials - and in fact it wouldn't work for all magnetic materials. For an antiferromagnet, for example, this would not work, beacuse it is only capable of detecting a macroscopic magnetic inductance existing outside the sample. For antiferromagnets, there is an alternating arrangement of magnetic inductance that cancels outside the sample and can only be detected inside the sample and on a microscopic (10 more sophisticated methods that we will see later in the course.
Magnetisation
For static elds the electric eld and magnetic induction completely decouple and Maxwell's equations simplify considerably. We can then describe magnetism with only two equations:
B = 0 B = 0 J
(1) (2)
The second of these equations, Ampere's law, tells us that loops of magnetic induction form around currents:
B du = 0
C
J ds
S
(3)
In the absence of a net electron ow, this means that all magnetic inductance is created by closed loops of electric current. For example, a wire bent around in a loop will generate a magnetic eld:
That means that the intrinsic magnetic induction of a magnetic sample must be due to small loop-like currents inside the sample:
and
B = 0 (M + H)
The rst component, through a material:
(4)
This eld can spill out as it does when we use an electric circuit to generate an inductance:
M,
is called the
due to charge owing in microscopic closed loops inside the sample. We say that
arises from bound charges, which can only circulate around a microscopic region inside the
arises from free charges, which can ow along the whole sample. We
decompose the current similarly to how we have decomposed the magnetic induction:
J = Jbound + Jf ree . M
and
(5)
H du = C M du =
C
Jf ree ds S Jbound ds
S
(7) (6)
Now suppose we have set up an electric circuit to generate a given magnetic induction that we wish to apply to our magnetic sample. induction In the absence of the sample, there is no magnetization so we have a magnetic
Bapplied = 0 Happlied .
When the sample is introduced, sample has) and
(8)
H = M.
(9)
This means that in general when we apply a given magnetic induction to a sample we don't know what the value of the magnetic eld one can approximate
inside the sample is. However if the magnetization is not very large,
H,
(10)
H Happlied .
when this is not true can be found in SB Sec. 1.1.4.
We will always assume this to be the case except when we state otherwise. The details of what happnes
M = 0.
(11)
The actual form that the magnetization takes depends on the type of magnetism. The study of the dierent forms of magnetization and how they come about is the subject of study of magnetism.In the previous lecture we saw that magnetic induction
J.
one due to free charges that can roam around a sample and another one due to charges that are bound to move in very small microscopic loops inside a sample. The two types of current give rise to the magnetic eld
M,
respectively:
B = 0 (H + M) .
M.
In this
Magnetic moments
M (and therfore of H) has been rather loose - how do we distinguish between bound M
as the volume density of magnetic moments.
and free charges. Here we are going to give a somewhat more rigorous denition in terms of magnetic moments. We shall dene the magnetization
What is a magnetic moment? A magnetic moment is a very small object made up of an electric current going around in a small closed loop
I r
C: B (r)
in the space around it. For distances
The magnetic moment gives rise to a nite magnetic induction which are small), this is given by
than the linear dimensions of the current loop (this is what we mean by the current loop being
B=
where we have dened the vector quantity moment.
0 [3 (m ) m] r r 4r3
(12)
m which measures the strength and direction of the magnetic I and a vector giving the area and orientation of the minimal m=IS
(13) not on other features such as the shape of the
C:
B (r)
at a given
depends only on
m,
loop of current (e.g. an ellipse or a circle). This is because the moment is small. Now let us consider a sample that has a nite density of innitesimally-small magnetic moments, magnetization is the volume density of such moments, obtained by dividing element:
M=
The magnetic induction at a point
dm . d3 r
B (r) =
We mow substitute
dB =
(15)
Md3 r = dm
B (r) =
(16)
As we can see the relationship between the magnetic induction generated by the sample and its magnetization is not at all trivial. That is why it is convenient to focus on the magnetization rather than the magnetic induction when discussing magnets.
This is a well-known result from the theory of electromagnetism. It follows from Maxwell's equations, which we recalled in
the previous lecture. There are many textbooks that prove particular cases of this result. See, for example, Feynman Lectures on Physics, Vol. II, Sec. 14-5 (note that this text does not use S.I. units; to convert the equations to the S.I. we must substitute 1/4 0 c2 0 /4 ).
Let us now see how a magnetic moment is aected by an applied eld. First we will establish the relationship between magnetic moment and angular momentum. particle of charge and has radius Consider a
and mass
mq
orbiting the origin of coordinates. Let us assume that the orbit is circular
r.
The orbiting particle creates a current along the ciruclar orbit. This current is
I=
where
qv 2r
(17)
is the velocity of the particle. The magnetic moment due to this orbiting particle is this times the
r2 : m= qv qvr r2 = . 2r 2
(18)
L = r mq v L = rmq v.
Thus there is a relationship between magnetic moment and angular momentum:
(19)
m=
Since both
q L. 2mq
(20)
and
m = q L,
where the quantity
(21)
q =
is called the
q 2mq
(22)
gyromagnetic ratio
of our particle. Note that it depends on the charge and mass of the
Larmor precession
The relationship found above between a magnetic moment and angular momentum allows us to work out how a magnetic moment
B.
mechanics that the force on a charged particle in the presence of an electromagnetic eld is
F = q (E + v B) .
In the absence of an electric eld, we are left only with the second term: the
(23)
Lorentz force
(24)
F = q (v B) .
This will act on the charge or charges making up a magnetic moment, causing a torque that makes it re-orient itself. The total torque on the particles in is given by
G = r F. 1 = q r (v B) = q (r v) B 2 q = L B = q L B, 2mq
(25)
where in the second line we have assumed only the total torque averaged over an orbit need be considered and in the third line we have used that the angular momentum is the rate of change of the angular momentum:
L = r mq v .
d L = G dt d L = q L B. dt
(26) (27)
We now use the gyromagnetic ratio to turn this into an equation for the time-evolution of the magnetic moment:
d m = q m B. dt
(28)
The key thing to notice here is that the rate of change of the magnetic moment is perpendicular to the magnetic moment itself - therefore a magnetic induction can never lead to an increase or decrease of the size of the magnetic moments - they merely precess in periodic orbits. To see this explicitly let us solve the above equatio of motion. First, let us choose our system of coordinates to that
is parallel to the
axis:
B = (0, 0, B) .
The vector product then gives
(29)
mB=
i j k mx my mz 0 0 B
= (my B, mx B, 0)
(30)
d mx = q Bmy dt d my = q Bmx dt d mz = 0 dt
The third of these equations tells us that the component of change. The other two coupled equations are solved by
mx = m sin (L t + ) my = m cos (L t + ) ,
where
(34) (35)
and
Larmor frequency
(36)
L = q B.
(This can be proven by substitution and we leave it as an exercise.)
This tells us that in the presence of a magnetic induction a magnetic moment will presrve its tilt with respect to the magnetic induction and precess around it at a constant frequency which is proportional value of the induction and to the gyromagnetic ratio:
cannot be used to
The spins will start precessing around the applied eld, all of them at the same
frequency, and therefore the total amount of magnetization will change direction but not size (it is not just that each magnetic moment will remain the same size, but also that the way they add up will also remain the same).
an induction to almost any system the total size of the magnetization changes, either for or against the eld (paramagnetism and diamagnetism, respectively). Also, in many magnetic systems such as ferromagnets and antiferromagnets there is a spontaneous magnetization even in the absence of an applied eld which is due to each magnetic moment reacting to the induction created by all the other moments. But Larmor precession suggests that none of this can take place. Indeed the Lorentz force (24) is always perpendicular to a particle's displacement, therefore it cannot do any work on a system.
Something we have ignored until now is that angular momentum, and therefore also magnetic moments, are quantised. Consider the magnetic moment:
action associated with the motion in a circular orbit of a particle giving rise to a
S=
C
mq v dr
(1)
According to the Bohr-Somerfeld quantisation rules that action must be an integer number of times the quantum of action, which is Planck's constant:
S = h, = 0, 1, 2, . . .
But the action is nothing but
(2)
S = mq v2r = 2L,
where
(3)
L = rmq v
is the magnitude of the angular momentum, so this is just the quantisation of angular
momentum,
L= .
This, in turn, quantises magnetic moment,
(4)
: 7 = q L = q .
(5)
L.
See the previous Lecture 2, Eqs. 21,22.] Magnetic moment is thus quantised in units of
q =
For an electron we have
q . 2mq
(6)
mq = me , q = e
Bohr magneton:
(7)
B =
Just as is the quantum of angular momentum,
e . 2me
= B .
The negative sign of the Bohr magneton means that, for electrons, the magnetic moment points in the opposite direction to the angular momentum (this is due to the negative charge of electrons: when they turn around clockwise, the corresponding current turns anti-clockwise). As we will see the quantisation of magnetic moment plays an essential role in magnetism.
m).
Let us now be a bit more rigorous. In Quantum Mechanics the electron has two components to its angular momentum: orbital angular momentum and spin angular momentum. Both of them contribute to the electron's magnetic moment. Let us treat rst the
limit, to the classical angular momentum that we considered before. The corresponding quantum-mechanical operator is therefore obtained straight-forwardly from the corresponding classical expression:
L=rpL = rp = r = i i y
r
(9) (10)
z ,z x ,x y z y x z y x Lx , Ly , Lz
Lx , Ly , Lz .
(11)
From the last expression one can show that the three components commutation relations:
Lx , Ly = i Lz , Ly , Lz = i Lx , Lz , Lx = i Ly .
Thus dierent components of the angular momentum do not commute.
(12)
coordinates of the spin vector has a well-dened value, the other two are undened. These commutation relations are very unusual. Indeed they can be written as
LL=i L
which means the vector
(13)
is perpendicular to itself !
On the other hand the magnitude squared of the angular momentum, with each of the individual components:
L2 = L2 + L2 + L2 , x y z
does commute
L2 , Lx = L2 , Ly = L2 , Lz = 0.
Thus we can nd a set of states that are simultaneous eigenstates of set takes the form
(14)
L2
and
Li
i = z.
That (15)
{|l, ml }
2 and
l = 0, 1, 2, . . . , ml = l, l + 1, . . . , l 1, l
, respectively. In other words
l (l + 1)
ml
L2 |l, ml Lz |l, ml
8
= l (l + 1)
|l, ml ,
(16) (17)
= ml |l, ml .
Bra-ket notation:
.
to denote the eigenstates of the angular momentum operator we have used `ket' notation. This means
that instead of denoting a state by its wave function we just write its quantum nubers, e.g. `l, ml ', between `|' and ` ', like this:
|l, ml
What's wrong with just giving the wave function, In the momentum basis, for example, the same the ket notation
l,ml (r)? There is nothing wrong, of course, but it does mean that what r represents position, l,ml (r) is the wave function in the position basis. quantum-mehcanical state would have a dierent wave function, say l,ml (p).
It is useful to have a way of designating the state that does not require specifying which basis we are working on. This is what
|l, ml
aords us.
From the ket of the state we are interested in, we can easily construct wave functions in dierent bases. This requires using a `bra' to make a `bra-ket', like this: projections of the state at position
and
|l, ml
onto each of these two states, giving the amplitudes of probability that the partcile will be found
or with momentum
p,
respectively.
L=
and its projection along the
l (l + 1)
(18)
direction is
Lz = ml .
Note that although
and
Now we can use the gyromagnetic ratio of the electron, which we derived in the last lecture, to work out the contribution of an electron's quantised angular momentum to its magnetic moment,
L :
(20) (21)
L = e L = B L/ .
From this follows that the magnetic moment's size and projection along a given direction are quantised:
L =
Also, if in
l (l + 1)B
where
l = 0, 1, 2, . . . , ;
L,z = B ml
where
ml = l, l + 1, . . . , l 1, l.
L,z has a well-dened value then L,x and L,y are completely undetermined (likewise which L,x is well-dened L,y and L,z are undened, and so on). So we can see that from a
mechanical point of view the magnetic moment becomes quite a strange object compared to our previous, classical treatment.
spin'
L,
and denoted by
S.
Unlike
L,
expression [c.f. Eq. (9)]. Spin is an intrinsically quantum-mechanical quantity with no classical analogue. The components of the spin obey analogous commutation relations to those of orbital angular momentum [Eqs. (12-14)]. We can therefore nd simultaneous eigenstates of numbers are
S2
and
Sz .
ms
ml ).
s can take either integer or half-integer values, depending on wether the particle is a boson or a fermion. For electrons s = 1/2 (this makes electrons fermions).
Moreover we shall now see that the gyromagnetic ratio is dierent for spin than for orbital angular momentum.
The gyromagnetic ratio for spin angular momentum is dierent than for orbital angular momentum - it diers a factor
called the
g-factor:
S = gS e S, = B gS S/ . g = 2. g = 2.0023...
(24) (25)
For an electron, the g-factor can be calculated using relativistic Quantum Mechanics and turns out to be Quantum electrodynamics corrects this to but we will ignore that. Thus the magnetic moment of an electron's spin is given by
S = gS s (s + 1)B 3B ; S,z = gS B ms B .
with
with
s = 1/2
or
and
gS 2
(26) (27)
ms = 1/2
+ 1/2
(28) (29)
Note that the magnetic moment coming from the spin angular momentum of an electron is bound to the electron. Thus, in the language we used in Lecture 1, it behaves like a magnetic moment arising from bound charges even if the electron itself is unbound. important for Because of this the spin angular momentum is particuarly
itinerant magnetism.
Nuclear magnetism
Since the nucleus also contains orbiting charges, namely the protons, there is also a magnetic moment of the nucleus that results from the quantised angular momentum of the protons. As in the case of the electrons, the protons have spin
thus be described in a very similar way to electronic moments, with one important dierence: the quantum of magnetic moment for a proton is not the Bohr magneton but the nuclear magneton for the proton charge
N .
and the electron
+e
mass
me
mp : N = e . 2mp
(30)
Since
mp 103 me ,
we have
N 103 B ,
can often be neglected. We will do this from now on unless explicitly mentioned otherwise.
The magnetic moment of an atom or ion in a solid is made up of contributions from dierent electrons. Also, even the magnetic moment of a single electron is made up of orbital and spin magnetic moments. To nd the combined magnetic moment of the atom we will need to combine dierent types of angular momentum. Let us consider the sum of two angular momenta:
J = J1 + J2 .
For example, we could be dealing with the orbital and spin angular momenta of the electron:
(1)
J = L + S.
(2)
Alternatively, we could be talking about the angular momenta of two electrons on the same atom or the total orbital andgular momentum of all the electrons in one atom and the total spin angulat momentum of the same electrons. In any case since the two angular momentum operators act on dierent degrees of freedom, they commute. Therefore we can nd a complete basis set of simultaneous eigenstates of label
J2 , J1,z , J2 , 1 2
and
J2,z ,
|j1 , m1 , j2 , m2 .
These states are built by the direct product of the bases for the two individual angular momenta,
(4)
J2
j1 , j2
m1 = j1 , j1 + 1, . . . , j1 1, j1 m2 = j2 , j2 + 1, . . . , j2 1, j2 ,
giving One
(2j1 + 1) (2j2 + 1) states in total. can prove that J is also an angular momentum,
J J = i J.
Because of this, there must be another complete basis set formed by simultaneous eigenstates of Moreover
J2
and
Jz .
Jz
and
J2
commute with
J2 1
|j1 , j2 , j, m .
The two bases are related through
(8)
|j1 , j2 , m1 , m2 =
j,m
j1 , j2 , j, m|j1 , j2 , m1 , m2 |j1 , j2 , j, m .
The projections
j1 , j2 , j, m|j1 , j2 , m1 , m2
[for a derivation of their values see any good Quantum Mechanics textbook, e.g. Bransden & Joachain] but we will note that they are non-zero only for certain combinations of out, for given will give us allowed values of
j, m, m1
j1 , j2 , which values are allowed for the new quantun numbers we starts with the the m1 and m2 , which we know (we jkust gave them above), and each of these combinations a value of j and m. The combinations for which the corresponding coecient is zero are not j = |j1 j2 | , |j1 j2 | + 1, . . . , j1 + j2 1, j1 + j2 m = m1 + m2 = j, j + 1, . . . , j 1, j,
m2 . j and m:
and
(2j1 + 1) (2j2 + 1)
states.
The rst of these equations tell us that the two angular momenta can combine in oppposition or re-inforce each other, the most extreme cases corresponding to total angular momentum of length direction can take values
j = |j1 j2 |
and
j = j1 + j2 ,
J=
with
Jz = m,
j (j + 1) has been formed, its component along the quantisation m = j, j + 1, . . . , j as usual for an angular momentum.
Our interest in the combination of angular momenta is that the combined angular momentum will give rise to its own magnetic moment through a gyromagnetic ratio
:
(11)
= J.
The gyromagnetic ratio
and
2 :
1 = 1 J1 2 = 2 J2
To put
(12) (13)
in terms of
and
we write
= 1 + 2
J = 1 J 1 + 2 J 2 .
We now multiply by
J,
obtaining
J2 = 1 J J1 + 2 J J2 ,
and use
(15)
J2 = J J2 1 J2 = J J1 2
to re-arrange the expression as
2 2
1 2 2 2 J + J2 J1 = J2 + J2 2J J2 J J2 = 2 2 1 2 2 2 = J 2 + J 2 2J J 1 J J 1 = J + J1 J2 1 2
(16) (17)
J 2 = 1
1 2 2 2 1 2 2 2 J + J1 J2 + 2 J + J2 J1 . 2 2
(18)
Let us now make each term on the LHS and RHS of the above equation act on the same basis state
|j1 , j2 , j, m
J2 |j1 , j2 , j, m 1 1 2 1 2 2
Here
J2 + J2 J2 |j1 , j2 , j, m 1 2 J2 + J2 J2 |j1 , j2 , j, m 2 1
J=
j (j + 1) , J1 =
(22)
are the size of each of the three angular momenta, respectively. Putting all this back into (18) we nd
= 1
Evidently if
(23)
1 = 2
= 1 = 2
(even if the
quantum numbers of the two combined angular momenta were dierent). It is customary to express the gyromagnetic ratio in terms of the Bohr magneton by introducing the
g-factor
Land
(24)
gJ B / = gJ B J/ .
gJ = g1
(25)
For the total magnetic moment of an electron, resulting form its spin and orbital angular momenta, we have
gL = 1
and
gS = 2
gJ =
3 3/4 l (l + 1) + . 2 2j (j + 1)
(26)
More generally if we are adding a large spin, resulting from combining several spins, and a large orbital moment, with quantum numbers
stot gJ =
(27)
Fine structure
or
to describe a com-
pound angular momentum has been treated as immaterial. However, if what we are trying to describe are stationary states (i.e. eigenstates not just of the above operators, but also of the Hamiltonian describing the dynamics of our particles) then we must choose a basis that corresponds to a set of operators that commute with the Hamiltonian. Consider a Hamiltonian featuring a term of the form
V = 2 J1 J2 ,
i.e. a quadratic coupling betweem the two spins. This can be, for example:
(28)
1. the spin-orbit relativistic correction that couples the orbital and spin angular momenta of an electron, an atom or an ion; 2. the exchange interaction between two adjacent magnetic moments (we will see where that can come from later on).
Then or
|j1 , m1 , j2 , m2
commute with
J1,z
V = 2 J2 J2 J2 , 1 2 2
which commutes with
(29)
J2 , J2 , J2 1 2
Jz
(since
J2 , J2 , 1
and
J2 2
E ,
order, this corresponds to simply evaluating the expectation value of the perturbation, ground state, which can be taken to be a simultaneous eigenstate of
V , in the unperturbed
J,J1
and
J2 .
We get (30)
E V
We now think about what happens when a term of the form (28) is switched on. levels were specied by
j1
and
j2
coupling term the levels split into dierent ones corresponding to dierent values of degeneracy thus gets lowered to just
Example:
j , as given by (30).
2j + 1. p
orbital (l
= 1, s = 1/2)
has available
(2l + 1) (2s + 1) = 3 2 = 6
and
mj . j
goes from
from
to
j:
j = 3/2 , mj = 3/2 j = 3/2 , mj = 1/2 j = 3/2 , mj = 1/2 j = 3/2 , mj = 3/2 j = 1/2 , mj = 1/2 j = 1/2 , mj = 1/2
Now we turn on perturbatively a spin-orbit coupling term no longer quantum numbers, but
L S.
ml
and
ms
are
and
mj
E L S
Thus the 6-fold degenerate energy level splits, under the inuence of spin-orbit coupling, into a 4-fold degenrate level and a 2-fold degenerate one:
This degeneracy can be lifted further by application of a magnetic eld (we will see that later on).
Let us assume for a moment that electrons in atoms do not interact with each other, in other words, that all the electron sees is the potential created by the atomic nucleus. We will also, for the time being, ignore spin-orbit coupling. Then each electron in the atom goes on a particular energy shell with quantum numbers
n = 1, 2, 3, . . .
and
and
l = 0, 1, 2, . . . , n 1.
2 (2l + 1) j = |l s| , . . . , l + s
mj = j, . . . , j .
Once all the electrons have been placed, their angular momenta can be added up as
discussed in the previous lecture, giving rise to the magnetic moment of the atom. Now, if all the states in a given energy level are occupied, then the contribution of that energy level to the total angular momentum, and therefore the magnetic moment, is zero, because for every electron with
Jz = mj
Jz = mj ,
lled like this, then it is not very interesting from the point of view of magnetism.
For example,
n = 1, l = 0
(i.e. (i.e.
ml + ms = 1/2
which add up to zero.
and
1/2,
partially-lled shell
necessarily compensate and potentially the atom can have a magnetic moment that may contribute to The problem is that when the shell is not full there are potentially many dierent ways of distributing the available electrons among the available orbitals. Each way has the same energy but it leads to a dierent nal value for the magnetic moment. To sort this out we need to remove our initial assumptions: let electrons interact with each other so that dierent ways of distributing electrons among orbitals will lead to dierent energies. Let us also allow spin-orbit coupling to lift some of the degeneracy of the individual energy levels. The accurate calculation of the energy of a few-electron system forming an atom is actually a very dicult problem. Here we give a set of rules-of-thumb called ` how the electrons are distributed among the dierent orbitals. They allow us to determine the spin, orbital, and total angular momentum quantum numbers of an atom or ion resulting from all their individual electrons:
jtot ,
respectively. This in turn enables us to work out the g-factor and therefore the magnetic
Here are the rules for a shell with angular momentum quantum number be applied in the order in which they are given here:
l containing N
Zeroth rule: Pauli's exclusion principle.- there can be no two electrons with the same value of
ml
and
ms .
Hund's rst rule.- Make the total spin quantum number
stot
as big as possible.
By the rules for addition of angular momenta, which we saw above, this will force us to align the spins of all the electrons. That will in turn make them occupy dierent orbitals (by Pauli's exclusion principle). In this way their mutual Coulomb replusion will be minimized as we avoid payimng the pairing energy price of having two electrons in the same orbital.
Hund's second rule.- Make the total orbital angular momentum quantum number possible, while remaining consistent with rule number 1.
ltot
as large as
Again, this will lower Coulomb repulsion because of the form the angular momentum wave functions happen to have.
Hund's third rule.- Make the total total angular momentum quantum number if
N 2l + 1
half-lled). This minimizes the spin-orbit coupling term for more than half-lled shells.
2 2 LS = 22 jtot (jtot + 1) ltot (ltot + 1) s2 (stot + 1) , tot taking into account that the coupling constant tends to be > 0 for less than half-lled shells and < 0
3+
ion.
11 6s2 .
f -shell
has
2 (2l + 1) = 14 s
1/2
l = 3 and therefore enough room for our 10 electrons. These are the states
l
3
ml
3 2 1 0 -1 -2 -3
ms
1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2
Let us apply Hund's rst rule: distribute electrons among the spin states so as to generate the highest possible value of
stot . ms
1/2
l
3
s
1/2
ml
3 2 1 0 -1 -2 -3
3
more electrons, and those will lower our spin by
-1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2 We still have to put
3/2
where to put them we need, therefore, to consult Hund's second rule: in order to maximise put them on the available states with the highest value of
ltot ,
we
ml :
l
3
s
1/2
ml
3 2 1 0 -1 -2 -3
ms
1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2
We now have completely distributed all the electrons among the available states. The spin and orbital
stot = 7/23/2 = 2 and ltot = 3+2+1 = 6. and ltot combine to give the total angular momentum quantum number I need to apply the third rule: since the shell is more than half-lled, I choose the maximum possible
angular momenta quantum numbers have worked out to be Now to work out how
stot
value for
= gJ B
with the g-factor given by
jtot (jtot + 1)
(31)
gJ =
This yields
(32)
5 = B 4
The spin conguration of an atom or ion is denoted
A = 2stot + 1, B = jtot
and
C = ltot
(with
ltot = 0, 1, 2, 3, 4, 5, 6, . . .
For example,
denoted
S, P, D, F, G, H, I, . . .)
for Ho we have
2stot + 1 = 2 2 + 1 = 5, jtot = 8
and ltot
=6I
so we write
5I
8.
Crystal elds
Interactions between electrons in dierent magnetic ions or atoms are crucial to magnetism, as they are responsible for the collective behaviour that we anticipated in Lecture 1. They often involve virtual quatummechanical tunnelling of electrons from one lattice site to another (exchange interactions). This will be seen in future lectures. For the time being we are interested only in the eect electrostatic interactions with electrons in the immediate environment of a given atom or ion. These electrostatic interactions would not be present in an isolated atom or ion but manifest themselves when the atom/ion is placed inside a crystal where it is surrounded by other atoms or ions. They take the form of an electric eld that the electrons see: the `
crystal eld'.
4f
3d
shells (the rst row of transition metals) they do not work. In general, they work better the deeper inside the atom the unlled shell is located. In that case, the electrons in the unlled shell are quite isolated from the crystal eld. This is good because Hund's rules assume an atom or ion which is suspended in empty space, in a spherically-symmetric environament. In reality, the magnetic atom or ion is often encaged by a number of non-magnetic atoms or ions that surround it. This makes the atom or ion's environment non-spherically
symmetric due to the crystal eld. If the unlled shell is deep inside the atom, the eect of the crystal eld is notwhen the so great. The crystal eld prevents the electrons from orbiting freely around the nucleus, which may result in an ltot = state rather than the higher value suggested by Hund's second rule. This is called orbital ` '.
quenching
L = 0 J = S.
For example, an
In addition to orbital quenching, the crystal eld splits the energy levels as dierent congurations have dierent level of overlap with the adjacent atoms.
l = 2
is surrounded by negatively-charged ions placed in an octahedral conguration (with the magnetic ion in the centre of the octahedron, and a non-magnetic ion in each vertex) will see its degenerate level split into a called
2 3 = 6-fold
t2g
and a
eg .
Hund's rst rule still applies, but the electrons are distributed
among the 5
t2g
and
eg
orbitals depending on whether the pairing energy is smaller than or larger than the
t2g eg
splitting, respetively. [See Fig. 3.4 in the text by Blundell for a picture of this.]
the combined
behaviour of many quantum-mechanical entities (the other major example being superconductivity, which we will treat later on in the course) and we stated the problem as follows: there are materials that create a nite Then, in Lecture 2, we formulated the problem in terms of a vector quantity called the
magnetic induction in and around them, and the science of magnetism tries to explain its origin. magnetization, M, which is nite within a magnetic sample and zero without. We also introduced the notion of magnetic moments, , which are the microscopic entities that combine to give a macroscopic magnetization. In Lecture 3 we established the quantum nature of the magnetic moments of electrons and in Lecture 4 we saw how those combine to give the total magnetic moments of atoms and ions in solids. We are now in
a position where we understand what the building blocks of manetism are and we can start to tackle the problem of combining them to yield the macroscopic phenomena of interest. In these two lectures we will introduce the basic tools of to describe the simplest magnetic phenomenon:
statistical mechanics that we will be using to do that and we will apply them paramagnetism.
M=
d . d3 r
Rj
of all the
magnetic sites
j = 1, 2, . . . , N ,
and at each of those sites we knew the magntiude and direction of the corresponding
j j d3 r
(1)
M.
If all dipoles are identical, and they are distributed uniformly around the sample with
then
M = .
(2)
We have seen in previous lectures how a magnetic moment can arise for an electron or a whole atom. We have not said anything about what those magnetic moments are doing, though. They could, for example, be acting in unison, all pointing in the same direction; or they could be oriented randomly. In the latter case the above sum might give moment
d = 0
M = 0,
individual magnetic moments, their behaviours combine to re-inforce or counter-act each other for a given state of thermal equilibrium at a given temperature,
T.
For our purposes, the important statement that Statistical Mechanics makes is the following. Suppose that we know all the stationary states of the system,
|n = |0 , |1 , |2 , |3 , . . .
and their energies,
(3)
En = E0 , E1 , E2 , E3 , . . .
These means that
(4)
|n
and
En
H:
(5)
H|n = En |n , n = 0, 1, 2, . . .
Once we know the eigenstates, we can compute the expectation value of any operator, including the operator th giving the components of the j magnetic moment. Thus in the ground state it will be given by
0|j |0 .
More generally, in an arbitrary stationary state
(6)
|n
n|Rj |n .
possible state of the system.
In this way we could, in principle, compute the magnitude and direction of each magnetic moment in each
Now, in reality the system is not normally in a stationary state (either the ground state or any excited state). Instead it is in
thermal equilibrium
at some temperature
T,
|n
in this mixture is
pn =
where
eEn /kB T Z Z
kB
some
state is 1:
pn = 1.
We thus obtain the
(9)
n=0
Z=
n=0
The above formula giving
eEn /kB T .
(10)
pn
n
can be used to compute the average value of any given magnetic moment:
= =
n
pn n|j |n eEn /kB T n|j |n Z eE0 /kB T 0|j |0 + eE1 /kB T 1|j |1 + eE2 /kB T 2|j |2 +
(11)
(12)
1 Z
(13)
Having introduced the tools of Statistical Mechanics necessary to describe magnetic phenomena, we will now use them to tackle
paramagnetism.
not involve in any essential way interactions between magnetic moments (i.e. they are only collective in the sense that they result from the sum of many independent magnetic behaviours, not in the deeper sense that we will encounter later on). Our rst take on the problem of paramagnetism will in fact neglect interactions between magnetic moments completely and assume that we can Hamiltonian is
Hj
system can be specied by the individual states of the magnetic moments, considered in isolation, i.e. in the above formuale we can intepret energy.
|n
En
as its
Before we plunge into the many-body problem, we need to learn one last thing about magnetic moments: their interaction with an applied mangetic eld. In Lecture 2 we already saw that a magnetic moment will precess in a magnetic eld with the Larmor frequency
L = q B.
magnetic moment posesses by virtue of being in the presence of the magnetic eld it is precessing about. The torque that is responsible for Larmor precession is
G = q L B
(Lecture 2, Eq. 25) which gives
G = B.
Let us call
(15)
and
= 0,
we have
G = 0.
At nite angle
. If the magnetic induction and the magnetic moment are parallel, we have G = B sin . Therefore when the magnetic moment and adiabatically from = 0 to a nite angle the energy changes by an
E =
0
Now, using induction
G d =
0
B sin d = B (1 cos ) .
the energy change of a magnetic moment
(16)
when a magnetic
E = B .
Now from (Lecture 4, Eq. 24) we have
(17)
= gJ |B | J/ .
Moreover the component of the total angular momentum
(18)
Jz = m, m = J, J + 1, . . . , J 1, J.
All this leads to
z = gJ |B | m,
so we have (choosing the quantization direction to be parallel to the magnetic eld)
(20)
E = gJ |B | m B, where m = J, J + 1, . . . , J 1, J
(21)
Thus a magnetic eld splits degenerate angular momentum states, reecting that the energy of a magnetic moment depends on the relative orientation to the applied eld. Let us look at an example from Lecture 4: the stationary states of an electron in a
of spin-orbit coupling and without an applied magnetic eld, these states are six-fold degenrate. In Lecture 4 we saw that spin-orbit coupling partially lifts this degenracy by splitting levels diering in the value of the total angular momentum,
In a magnetic eld, the degenracy is much less: even without spin-orbit coupling, dierent values of non-degenerate, so the only degeneracy we have left is that between states with the same value of dierent values of
m are m but
j:
Since in reality the latter degeneracy was already lifted by spin-orbit coupling, then that cannot be neglected we have no degeneracy left at all:
Let us now consider a system made up of independent magnetic moments: a solid with magnetic ions, i.e. those with unlled shells that have a net magnetic moment (as discussed in previous lectures). Let us assume thatthere arer no interactions between the magnetic moments (we will see what is the eect of interactions in later lectures). Consider rst the situation with no applied magnetic eld. Then assuming all the magnetic moments are identical and have angular momentum quantum number magnetic moment with magnetic quantum number
mj = jtot , ..., jtot is degenerate. Thus each one will compensate with another one with the opposite value of that quantum number,mj (this is true whether or
not spin-orbit coupling is present). Evidently these two states have values of energy)
jtot .
and equal in magntidue, so the results is that (choosing the equal energy fo all the states as the origin of
jtot
=
m=jtot
1 gJ |B | m Z
jtot
(1)
1 gJ |B | Z
m
m=jtot
(2)
= 0
Now let us put in an applied magnetic induction:
(3)
jtot
=
m=jtot
1 gJ |BT| m B kB gJ |B | m e Z
jtot m=jtot
(4)
1 gJ |B | Z
megJ |B |Bm
(5)
[ 1/kB T ].
jtot
(6)
where we have introduced the very useful notation of Statistical Mechanics for the inverse temperature. Now
Z =
m=jtot
egJ |B |Bm
jtot
(7)
Z H
This allows us to write
= gJ |B |
m=jtot
megJ |B |Bm .
(8)
z =
id est
So all we have to compute is the partition function
1 Z Z B ln Z. B
Now from Eq. (1) we have
(9)
z = kB T Z.
(10)
jtot
Z =
m=jtot
rm
2jtot
with
r = egJ |B |B
(11)
= r
jtot
rm
m=0
(12)
rm =
m=0
Applying this we get
rM +1 1 r1
(13)
Z = r =
2j
2j +1 jtot r tot
r1 2jtot +1 r 1
tot
2 +1 2j
(14)
rjtot +1 rjtot
2j
(15) tot
+1
[
Now we put in
r r
tot
2
+1
tot 2
2j
+1
] =
r 2 r1/2 r1/2
(16)
r = egJ |B |B = ex
with
(17)
x
and we get
gJ |B | B kB T
(18)
Z=
Thus for
(19)
we obtain
= kB T
(20)
= kB T gJ |B | = gJ |B |
The magnetization
(21)
(22)
M = z
is thus
M = gJ |B |
(23)
Proof: Let M +1
Then
r (. . .) = r + r2 + . . . + rM + rM +1 = (. . .) + rM +1 1r (. . .) = (. . .) +
M = gJ |B |
The variable
was given by
gJ |B | B kB T
is the density of magnetic dipoles, jtot is the total angular momentum quantum number of each dipole, and gJ its gyromagnetic ratio. B is the externally-applied magnetic induction and T is the temperature. We
Here are now going to investigate what the above result means. Since
coth =
for large
e + e 1 e e
as
(1)
(2)
M Ms gJ |B | jtot
This is the
for
gJ |B | B
kB T.
(3)
saturation magnetization
with the eld so their component in the eld direction has the maximum value, maximum value corresponds to the maximum allowed value of the magnetic
, coth
1 1++1 = , 1 + (1 )
(4)
(5)
M 0
This happens when the magnetic energy
as
gJ |B | B/kB T 0. kB T ,
(6)
z B = gJ |B | jtot B
that tends to align the moments in the which tend to scramble the
eld direction is negligible compared to the energy of thermal uctuations, direction of all the magnetic moments so they average to zero.
The above results suggest expressing our results in terms of the new variable
gJ |B | jtot B kB T
(7)
which is the ratio between the magnetic energy of a fully-aligned moment and the typical energy of a thermal uctuation. Evidently
y = xjtot .
M = Ms Bj
tot
gJ |B | jtot B kB T
(8)
where
Bj
tot
(y)
is the
Brillouin function,
Bj
tot
(y) =
(9)
As we can see it is easier to overcome thermal uctuations and fully polarise a spin energy) and against it (high energy):
1/2
magnetic moment
than one with a large spin. This is because the low spin has to choose between pointing along the eld (low
There have to be very strong thermal ucatuations before the spin decides that it is worth pointint against the eld from time to time. In contrast, a large spin has many possible congurations that are only pointing slightly away from the applied eld:
j=2
Thermal uctuations can easily tilt the spin slightly in that way at little cost in terms of magnetic energy.
Magnetic susceptibility
To conclude our initial study of paramagnetism, we are going to characterise the magnetic behaviour of our paramagnet in terms of an important quantity called the
magnetic susceptibility.
This is dened by
M . H
(1)
Note that the derivative is with respect to the applied magnetic eld, rather than the magnetic induction:
B = 0 H .
0 =
M H
.
H0
coth =
to obtain
1 1 + + 2 2 + 1 + 1 2 ex + ex 2 + 2 1 2 = = + , 1 2 1 2 x ex e 2 3 1 + + 2 1 + 2
(2)
Bj
tot
(y) =
2jtot + 1 2jtot (2jtot + 1) y 1 2jtot y + + 2jtot (2jtot + 1) y 2jtot 3 2jtot y 2jtot 3 2jtot + 1 2jtot
2
(3)
y 3
1 2jtot
(4)
Thus
M Ms
and hence
for
(5)
0 = Ms
(6)
(7)
and therefore
(9)
gJ |B | /
is the gyromagnetic ratio for our magnetic moment (Lecture 4, Eq. 3), so the expression
between square brackets is nothing but the magnitude of the magnetic moment,
2 = gJ |B |
(10)
Hence
0 = 0
which is the celebrated
2 , 3kB T
(11)
Curie law:
i.e.
jtot = 1/2
Plotting
vs
Although at low temperatures there may be deviations due to interactions between mag-
netic moments (whcih we have neglected here), at suciently high temperatures, when than the characteristic energy of those interactions, the Curie behaviour is recovered.
kB T
becomes greater
Additionally, one could use a measurement of the saturation magnetization, Eq. (3), to determine elds and do not alter the value of the saturation magnetization. Thus from these two measurements we can deduce both the g-factor number
gJ jtot . Again, interaction eects may be important at moderate elds, but they become negligible at large enough gJ
and the angular momentum quantum
jtot
It is experiments of this type that enable us to check that Hund's rules do work, and also to establish where they don't. More generally, they allow us to conrm, even in systems with quite strong interactions, the
validity of our microscopic picture of individual, quantum-mechanical magnetic moments as the building blocks of magnetism. The quantum-mechanical nature of magnetic dipole moments has a spectacular manifestation in the above measurements. For a classical magnetic moment, we would expect saturation to occur when the full magnetic moment
z = z,max = (z
z,max / = jtot /
(12)
jtot (jtot + 1) jtot and the classical behaviour is recovered. But for magnetic For large jtot , we have moments with small jtot there can be a signicant dierence. In the extreme quantum-mechanical limit of
the smallest possible angular momentum,
jtot = 1/2,
we have
jtot /
i.e. the
maximum polarization a magnetic moment can achieve is little more than half the actual size of the spin! This can be appreciated in the gure on page 2.
Diamagnetism
So far we have always focused on systems where the atoms have unlled shells which can possess nite magnetic moments in zero eld. It turns out that even systems with lled shells have a magnetic response, as the atoms can acquire a magnetic moment when they polarise in an applied magnetic eld. However, unlike paramagnetism, diamagnetism is quite unrelated to the more intersting interaction-induced magnetic phenomena that we will study later on in the course. We will therefore not say more about diamagnetism for the time being. (We will return to it when we tackle superconductivity, an interaction-induced phase of metals one of whose dening characteristic is perfect diamagnetism. In that sense supeconductivity can be regarded as a very exotic type of itinerant magnetic state.)
In Lectures 1-4 we built the Lego bricks out of which magnets are made, namely the magnetic moments of individual electrons and atoms. In Lectures 5-7 we showed how the collective behaviour of many individual magnetic moments may lead to macroscopic magnetic behaviour - in particular, to paramagnetism, which is the ability of some materials to develop a magnetization in repsonse to an applied magnetic eld. responding to the applied eld and to thermal uctuations independently. magnetic phenomena that do not conform to this description. In To show this, we treated the many magnetic moments in a paramagnet as non-interacting entities, each one
showed for our model of indepedent magnetic dipoles, where each dipole points in a dierent direction so the magnetization averages to zero unless a eld is applied. Note that ferromagnetism was the rst magnetic state to be discovered and indeed that it was through ferromagnetism that magnetic forces in general came to be known thousands of years ago (remeber from Lecture 1 that
magnet
comes from
magnet lithos,
the
stone from Magnesia, a lodestone and therefore a ferromagnet). An inescapable consequence of this is that the knowledge we have built up since the start of this course is still a couple of millenia behind! Of course we do have now some understanding of where the magnetization of a ferromagnet comes from: it is an
These interactions
tend to make the moments line up. This means that it becomes energetically favourable for the magnetic dipole moments to collectively make up their minds on a single direction that they all point in, so they can all minimize that interaction energy. The above paragraph gives the correct qualitative picture, which we will make quantitative in the following lectures, of how ferromagnetism (and also other magnetized ground states) emerges. However, it still leaves us facing the problem of the
mechanism:
in the rst place? In the reminder of this lecture we will take our rst steps towards tackling this problem, by identifying where the energy dierence which lines up magnetic moments could come from.
Magnetostatics
The rst, obvious candidate are the interactions between magnetic moments and the eld created by other magnetic moments. This is, in fact, what makes two macroscopic magnets (e.g. two pieces of lodestone) attract each other. As we saw in Lecture 2 an individual magnetic dipole given by the formula
B=
of the magnetic induction due to the rst dipole. Substituting the above formula for already derived for the energy of a dipole in a generic magnetic induction,
E = B.
(Lecture 5, Eq. 17),
(13)
0 3 ( ) , r r (14) 4r3 where r is the vector that goes from dipole to dipole . If we assume that the magnitudes of the magnetic E=
dipole moments are
we have
, |B | =
and that the distance between the to dipoles is
(15)
r 5A,
we have, using
(16)
0 = 4107 NA2 , E 0 2 4107 NA2 B = 9.274 1024 Am2 4r3 4 (5 1010 m)3
2
1025 J.
(17)
kB T = E T =
(18)
This is clearly far too small to explain ferromagnetic transition temperatures of order
nered in many materials. For example, the Curie temperature of Fe is 1043K, while its magnetic moments (determined from high-temperature susceptibility data in the paramagnetic state above when we dealt with paramagnets) have a size of 2.2B , corresponding to a temperature scale of to interaction energies sought. (That said, in some materials the dipolar interactions are important. In spin ices, which are pyrocholorestructure magnets with Dy or Ho ions, the magnetic dipole per ion is larger meaning that dipolar interactions become important at
0.05K.
So whatever the interactions between magnetic moments leading to ferromagnetism in Fe, they have to lead
orders of magntidue greater! Clearly another source for these energies has to be
the interesting physics of these materials such as emergent magnetic monopoles happens.)
Electrostatics
1000K
in adjacent atoms. The energy of one electron in the eld of another electron a distance
apart is
E=
1 e2 . 4 0 r
(19)
Substituting the same distance as above, i.e. electric permitivity of free space,
r 5A, 1
e 1.6 1019 C,
and the
8.85 1012 Fm1 , we obtain (1019 C)2 = 1018 J 1010 Fm1 1010 m
(20)
E
aso the corresponding temperature is
1018 E 23 = 105 K. kB 10
(21)
Now we are talking! The Coulomb repulsion between electrons in dierent atoms is a much greater energy scale than any magnetic interactions, and it is certainly large enough to explain phase transitions with critical temperatures
1000K
or more.
The problem is, how does an electrostatic repulsion become an interaction between magnetic moments? The key is that the Coulomb interaction inuences the way the electrons move and this is correlated with the spin of the electrons through the Pauli exclusion principle. This is called the physics of what is going on. (Note: a similar problem presents itself for high-temperature superconductivity: the energy scale of the
exchange interaction.
In the
next lecture we will see how this works through a very simplied calculation which nevertheless captures the
phonon-mediated electron-electron interaction that is responsible for superconductivity in conventional superconductors is far too small to explain the superconducting critical temperatures
100K
or more of the
high-Tc 's. As in magnetism, the electrostatic interaction between electrons seems to be strong enoug. On the other hand for the high-Tc superconductors the mechanism whereby the Coulomb replusion between electrons is converted into a pairing interaction that can lead to superconductivity is still unclear. In contrast, for ferromagnets we do understand quite well how Coulomb repulsion can become an interaction tending to align magnetic moments. This is what we will do in the next lecture.)
Consider a magnetic moment of strength equal to one Bohr magneton. If the moment where due o to electric charges orbiting a nucleus at a distance of 2A, what would be the corresponding current?
If this was due to a single electron, how many r.p.m.'s (revolutions per minute) would the electron be making around the nucleus?
Problem 2:
Consider a magnetic moment, of strength equal to one Bohr magneton, sitting at the origin of
direction. Compute the direction and magnitude (in SI units) o o of the corresponding magnetic induction at distances of 5A and 10A along the z axis, the x axis and o the (1, 0, 1) diagonal (1A = 0.1nm). On the basis of these six values and of symmetry considerations,
l, s, j, mj
f-
shell (l
=3
orbitals). For each state, give the values of the electron's orbital, spin, and total angular
momenta as well as the projection of the latter along the quantisation axis. Sketch the partial lifting of degeneracy due to spin-orbit coupling.
Problem 4: (*)
s, l, j
of a Pm
3+
the standard `term symbol' describing this ion's mangetic state (describe in detail the application of each of the Hund rules). How much degeneracy does this state have? Derive the g-factor and use it to calculate the size of the corresponding magnetic moment. Sketch how the degeneracy is lifted under an applied eld of increasing strength. Assume a magnetic eld equivalent to
1T
3+
the results obtained above, the value of the energy split between the state that is most aligned with the applied eld and the next one down xxxxup in energy. Use this value to estimate the temperature at which the magnetization of the salt will be, to a good approximation, completely saturated by the applied o o o eld. If there are 2 magnetic moments per unit cell, and the unit cell dimensions are 5A 5A 5A, what is the staurated value of the magnetization? How does it compare to the strength of the applied eld? Finally, compute the magnetic susceptibility of the salt at
Problem X5: 6
1K.
jtot = 1 in an Derive expressions for the magnetization and the magnetic susceptibility at
Verify that the expressions you obtain are particular cases of the more general formulae
jtot
Calculate the magnetic moment of Mn3+ rst using Hund's rules and then assuming orbital
ltot = 0).
Experimentally for a particular salt containg this ion it is found that the susceptibility is
0 = 23.2 0 2 B kB T . Is the orbital angular momentum quenched in this system? Would the result be the same if the
[Solution]
Problem 2 solution:
[Solution]
Problem 3 solution:
[Solution]
Problem 4 solution:
The atom Pm has the electronic structure [Xe]4f 6s will leave us with [Xe]4f
[1 point].
The
f -shell
2 (2l + 1) = 14
and
l = 3 and therefore enough room for only 4 electrons (note I am using l, s to refer to numbers of individual electrons; I will use stot , ltot
jtot to denote the spin, orbital, and total angular momentum quantum numbers of the ion as a whole). These are the states that are available for each individual electron: l
3
s
1/2
ml
3 2 1 0 -1 -2 -3
ms
1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2
Let us apply Hund's rst rule: distribute electrons among the spin states so as to generate the highest possible value of stot . This is achieved by placing all 4 electrons in the same spin state, ms = 1/2, with their spins pointing up (we can place all electrons in the same spin state because the shell is less than half-lled). This still leaves some freedom as to where the electrons go, so we need to invoke Hund's second rule: in order to maximise ltot , we put ll rst the
ms = 1/2
ml :
l
3
s
1/2
ml
3 2 1 0 -1 -2 -3
ms
1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2 1/2 -1/2
We now have completely distributed all the four electrons among the available states. The spin and orbital angular momenta quantum numbers have worked out to be
stot = 4 1/2 = 2
and
[4 points] points]
and ltot combine to give the total angular momentum quantum number I need to apply Hund's third rule: since the shell is less than half-lled, I choose the minimum possible Now to work out how value for
points]
A = 2stot + 1, B = jtot and C = ltot (with ltot = 0, 1, 2, 3, 4, 5, 6, . . . 3+ we have 2s denoted S, P, D, F, G, H, I, . . .) For Pm tot + 1 = 2 2 + 1 = 5, jtot = 4 and ltot = 6 I ,
term symbol
= 5 I4 .[2
points] points]
Since
jtot = 4,
we have degeneracy
= 2jtot + 1 = 2 4 + 1 = 9 , [10
distinct states
mj = 4, 3, 2, 1, 0, 1, 2, 3, 4.
The g-factor is given by
gJ
= =
=
The magnetic moment is
where
= gJ |B | /
2 =
jtot (jtot + 1)
= gJ |B |
This yields
3 |B | 5
points]
E = B = z B,
where we have chosen the
z -axis
z = Jz = (gJ |B | / ) mj = gJ |B | mj
with tot
tot
mj
tot
the 9-fold degeneracy into 9 distinct energies that change linearly with the magnetic eld with dierent slopes:
gJ |B | B , y E ):
[10 points]
3
mj
tot
= 4, 3, 2, . . . , 3, 4
Emost
The next one up in energy has
aligned
= 4gJ |B | B = 3g |B | B
Enext
The dierence is therefore
one up
Emost
aligned
Enext
one up
[10 points]
At
1T.
will the moments be completely aligned? That will happen when the thermal
kB T ,
kB T T =
Emost Emost
aligned aligned
Enext Enext
one up one up
/kB
points]
Below this temperature, to a good approximation, all magnetic moments are aligned with the eld. The density of magnetic moments is
points]
Ms
= z,max = 4gJ |B | 3 = 1.6 1028 m3 4 9.27 1024 JT1 5 = 3.84 1028 |B | = 3.56 105 JT1 m3 . [7
points]
B = 0 (M + H) .
Outside the sample
B = Bext , H = Hext
M =0
so
Bext = 0 Hext .
Thus the externally-applied eld is
Hext =
1 1 B = 1T 0 ext 4 107 Hm1 7.96 105 JT1 m3 2.24 Ms Ms 0.46 Hext . [10
points]
0 = 0
At
2 . 3kB T
T = 1K
this gives
4 10
Hm
1.6 10
28
0.3 . [10
points]
Problem 5 solution:
[Solution]
Problem 6 solution:
[Solution]
paramagnetism
align their magnetic moments with an applied magnetic eld, leading to an additional magnetic induction. Paramagnetism is a collective phenomenon, since many magnetic moments have to act in unison, and we thus had to describe it using the tools of Statistical Mechanics (i.e. the partition function). On the other hand, our discussion of paramagentism treated each magnetic moment in the material as an independent entity. Such theory falls short when we try to use it to describe magnetic materials that present a spontaneous magnetization even in the absence of an applied eld, such as is not what happens in ferromagnets or antiferromagnets. At the end of the last lecture we put forward the idea that in order for the magnetic moments in a material to spontaneously self-organise without an external stimulus it is necessary to take into account purely
ferromagnets or antiferromagnets.
Indeed
our theory of paramagnets predicted (correctly) zero magnetization in the absence of an applied eld, which
interactions
between dierent magnetic moments. We discussed the possible origin of such interactions. We discarded a
magnetostatic origin on the grounds that the energy scales that come out of such assumption are far
1K, 1000K
or more. We then pointed out that
too small - the eects should not be noticable above temperatures temperatures can be
electrostatic
electrons are much stronger and indeed perfectly capable, at least as far as energy scales are concerned, to lead to magnetism. However, we did not discuss the
mechanism, i.e.
give rise to an interaction tending to align magnetic moments with respect to each other in a particular way? That is the question that we address in what follows.
Consider, for simplicity, an ion that has an unlled shell with a single electron. Moreover let us assume that the shell has only
orbitals so the magnetic moment of the ion comes exclusively from the electron's spin.
Now let us bring near it other identical ions, to form a crystal. Our unpaired electron will be repelled by the electrons in the other ions and will therefore tend to localise more tightly on its own ion. In this way, the electron minimizes the
Coulomb interaction energy between our electron and all the other electrons:
U= 1 4 0
j
e2 |r rj | j th
electron, respectively.
(1)
where
and
rj
around its ion. Indeed, since the Coulomb interaction energy does not depend on where the spins of our electrons are pointing, the magnetic moments of such system would be completley disordered in the absence of an applied magnetic eld:
kinetic energy,
T = p2 , 2me
where
p = i
2 r.
(2)
To minimize this energy the electron would want to be in a plane wave state,
r|p eip.r/ .
(3)
However the probability density of a plane wave extends uniformly across the lattice:
2 1.5 wave function 1 0.5 0 -0.5 -1 -20 -15 -10 -5 0 position (a.u.) 5 10 15 20 Real part Imaginary part |...|2
Evidently this is very bad from the point of view of the Coulomb interaction as the electron overlaps with all the other electrons in the crystal! The optimal situation is intermediate between the electron being completely localised and being completely spread out. Indeed
values taken by an electron's position along a given coordinate axis, same axis,
x,
xp
If the electron is highly localised,
(4)
x is small, implying large p which means that the electron's wave packet
is made up of many plane waves, some of them with quite high momentum, impying a higher kinetic energy. Thus the electron will be neither completely localised on its ion, nor extended throughout the whole lattice, but will be centred on its ion but spread onto a few of the neighbouring ions:
The system remains a Mott insulator, but there is some additional interaction between dierent ions. (The magnetism of metals will be discussed later on in the course.) Now, here comes the crucial part: when the electron is sitting tightly on its ion its spin can point either up or down indepedently of what the other electrons in the other ions are doing; however, hopping onto another ion is more or less likely depending on what the direction of our electron's spin is relative to that of the surrounding electrons. For instance:
The electron may hop onto an orbital that already contains one electron, so by
principle (Hund's rule #0) its spin must be the opposite to that of that electron. (by Hund's rule #1)
Pauli's exclusion
The electron may hop onto a shell where there are free orbitals and all electrons have their spins aligned, in which case direction. it will prefer to have its own spin aligned with the preferred
Either way the ability of the electron to hop onto an adjacent ion, and therefore to lower its kinetic energy by spreading out its wave function, depends on the direction of its spin. The electron will therefore achieve lower energies when its spin direction maximises this ability. In this way the direction of its spin has become
correlated
to
exchange interaction. The two cases mentioned above correspond anti-ferromagnetic exchange (which tends to make adjacent spins point in opposite directions) and ferro-magnetic exchange (tending to make all spins parallel to each other).
Let us now make this a bit more quantitative using a simple calculation. We will model the behaviour of our electron using a very small Hilbert space consisting of three states:
|L , |L , |R
(5)
These states correspond to the electron being, respectively, in its original ion (which we have artbitrarily labelled
for Left) with its spin pointing up, in the same ion with its spin pointing down, and in the
adjacent ion with its spin also pointing down. The electron cannot hop onto
|R
adjacent ion already contains an electron with its spin pointing up - so by Pauli's exclusion principle our
Let us now construct a matrix describing the Hamiltonian of our electron for this simple model. Since there are no applied elds, if we ignore the inuence of the neighbouring ion the energies that the electron has when it points up or down are equivalent:
L |H|L = L |H|L .
has energy = 0:
(6)
We can choose this to be the origin of energy i.e. the electron staying on its site with the spin up or down
L |H|L = L |H|L = 0.
On the other hand if the electron is on the other ion, a certain amount site:
(7)
R,
due to the stronger repulsion with the other electron when they are both on the same
R |H|R = U.
This extra energy cost is called the Hubbard the concept in his eponymous model of 1963. Finally, the Hamiltonian must be able to make our electron hop spontaneously from
(8)
U
to
R,
or
vice versa :
(9)
L |H|R = R |H|L = t.
Thus this simple model contains the three essential ingredients of the exchange interaction:
3. A local constraint on the electron's spin when it is on the neighbouring ion (in this case, coming from the Pauli exclusion principle).
of one electron while considering the spin and position coordinates of the other electron xed - evidently if our electron can hop from
R,
to
L!
result we will obtain is quantitatively correct, which is why we are using it. In the above basis, and using the above results, the Hamiltonian matrix is
0 0 0 H = 0 0 t . 0 t U
The diagonal matrix elements represent the energies associated with being in a particular state:
(10)
The o-diagonal matrix elements represent the quantum-mechanical probability amplitude to spontaneously go from one quantum-mechanical state to another one.
Note we have used that the Hamiltonian matrix has to be Hermitian and therefore, given that all its elements happen to be real, symmetric. We have also assumed that the Hamiltonian does not feature any terms capable of ipping the elctron's spin. To nd the energies
of stationary states
Schrdinger equation:
H| = E| .
We do that in the next lecture.
(17)
|H E1| = 0,
where
1=
1 1 1
(2)
E t t U E = 0
(3)
E 2 (U E) + t2 E = 0 t U U 2 2t +1 , 2t E = 0, or U t U 2 2t + +1 2t
We see that the energy depends on
(4)
(5)
1 U , 2t
a dimensionless constant measuring the relative strength of the hopping the repulsion
(6)
U.
Our description of our electron hopping only to the adjacent atom is valid if this quantity is large, i.e. if
hemming in the electron is much larger than the hopping amplitude (otherwise, we have a
metal, not an insulator; the magnetism of metals will be discussed later on in the course). Expanding the expressions between curly braces to linear order in
we obtain
E=
1 2
+ 1 t = t + O 2 U
t2 U
t3 U2
0, or (similarly) U +
+O
t3 U2
1 1 + 1 = + + o 2 . 2 2
Let us now see what is the structure of the ground state
(7)
L | 0 0 0 L | t2 0 0 t L | = L | . U 0 t U R | R |
Note that we have expressed the state states
(8)
|L , |L , |R .
L | , L | , R
( L | , L | , R | ) = 0, 1, U 2t
+1 U 2t 0, 1, 2
(9)
= =
0, 1, 0, 1, t U
1 1 +1 2
(10)
(11)
(note that we have not normalised this state). Thus the ground state does correspond to basically the state with a bit of
|L
|R
E = 0,
is localised on the
( L | , L | , R | ) = (1, 0, 0) .
Our main result is thus that the ground state involves the electron being localised on down, and a little bit of spillover onto
L, with its spin pointing R, and the corresponding ground state energy is lowered by an amount J t2 /U
(13)
L,
excited state. This explains why the particle will choose to have its spin pointing down.
Superexchange
we have described above. Superexchange occurs when the magnetic ions are separated by an intervening, non-magnetic ion. Suppose for example, following Blundell, that we have a transition metal oxide featuring two transition metal ions (M) with, for simplicity, one unpaired electron each occupying identical orbital:
orbitals.
Imagine that the two ions are separated by a non-magnetic oxygen ion featuring two electrons in a single
orbitals cannot hop directly from M to M because they are too far away. Neither
orbital can lower their kinetic energies by hopping onto the two
M ions. That makes them couple antiferromagnetically to the two electrons in the two M ions. Since the
orbital must also have their spins pointing in opposite directions (by Hund #0), an
Evidently the situation in a real magnet is a lot more complicated than what we have described above. However, for magnetic insulators it is often possible to write a Hamiltonian that depends only on the angular momentum operators on each ion:
H (J1 , J2 , J3 , . . .) .
(14)
This is based on the fact that the unpaired electrons are mostly localised on each magnetic ion (see above). There are some things that can be said quite generally about the specic form of such Hamiltonian. Let us start by carrying out a Taylor expansion in terms of all those variables:
10
H (J1 , J2 , J3 , . . .) = constant +
i
Hi J i +
i,j
(1)
Ji Hi,j Jj + . . .
(2)
(17)
The rst term on the RHS can be ignored by choosing the origin of energy appropriately
H=
i
Hi J i +
i,j
(1)
Ji Hi,j Jj + . . .
(2)
(18)
In the absence of an externally-applied magnetic eld the system must be under the ipping of all the angular momenta:
symmetric (i.e.
invariant)
Jj Jj ,
meaning that the linear term, and inded all odd-order terms, must vanish:
(19)
... = 0 H=
i,j
Ji Hi,j Jj + O J4
(2)
Another symmetry that we can use for many crystals is the invariance of the system under rotations of all the spins. Then the tensor
(2)
(2) Hi,j
identical
(21)
H=
i,j
Hi,j Ji Jj + O J4
Moreover, we can assume, for an insulator,that the size of the magnetic moment on each site is xed (so the magnetic interactions only change the moments' directions, not their size). This will be given by its angular momentum quantum number, value of
jtot we have Ji Ji = = of the quadratic term thus take the form Hi,i J2 = i
i,i (2)
J2
2j
jtot . Assuming for simplicity that all sites have the same tot (jtot + 1) [see Lecture IV, page 2]. The diagonal elements Hi,i
i,i (2) 2
jtot (jtot + 1)
for
i = j.
(22)
This does not depend on the direction the spins are pointing in and therefore it is another constant that can be eliminated from the Hamiltonian by a further oset of the origin of energy
H=
i=j
10
Here we have used a shorthand notation whereby
Hi,j Ji Jj + O J4 .
(1)
(2)
(23)
Hi
is a vector,
Hi,j
(2)
Hi
i
(1)
Si
=
i =x,y,z
(1)
Si Hi,j Sj
i,j
(2)
= ...
i,j ,=x,y,z
(16)
To conclude we shall assume that the higher-order terms can be neglected to obtain the (very famous!)
Heisenberg model
H=
i=j
where we have introduced the so-called
Ji,j Ji Jj ,
(24)
exchange constants11
Ji,j 2 Hi,j .
(2)
(25) for ferromagnetic coupling (favouring
Note that we have introduced a sign convention whereby parallel spins) and
J >0
J <0
Ji
Ji Ji /
so that the exchange constant
Ji,j
L=0J=S
Ji,j
is negligible for all but nearest-neighbour sites, the Heisenberg model reads
H = J
i,j
where
Si Sj ,
(26)
i, j
indicates that
i, j
expect that for nearest-neighbour exchange coupling that we discussed above. More generally, the exchange coupling
J t2 /U
and of
individual electrons in a more complicated way. In the next lecture we will start to tackle the problem of solving Hamiltonians of the type given above using
11
and
j , Ji,j ,
Ji
and
Jj .
theory.
1
between magnetic moments. We will use a very successful, though approximate approach called
mean-eld
Our starting point is the Heisenberg Hamiltonian which we derived in the last lecture:
H=
j i=j
Ji,j Ji Jj .
(Lecture X, Eq. 24) Here we have introduced the dimensionless angular momentum
Jj Jj /
so the exchange constants magnetic moment that is at site
Ji,j have dimensions of energy. Ji,j gives i and another one that is at site j :
Ji,j > 0.
If
Ji
and
Jj
are parallel (or, at any rate, the component of one along the axis
Ji Jj
and
contribute to lowering of the energy (note the negative sign in the front - this is a convention). In
contrast, if
Ji
and
Jj
have an anti-parallel component then the corresponding term in the Hamiltonian will
increase the energy and the pair will make a positive contribution to the energy. Thus two magnetic moments have a to each other. Conversely, if
Ji,j < 0
ith
and
j th
sites have an
anti-ferromagnetic interaction
i.e. one that tends to make the spins point in opposite directions. The above Hamiltonian describes magnetic moments interacting with each other but not subject to any external inuences. As in the case of paramagnets (Lectures V-VIII), we can also apply externally a magnetic induction:
H=
j i=j
Ji,j Ji Jj
j
j B.
(2)
The applied magnetic induction couples to the magnetic moment of each individual site
j , j .
= gJ |B | Jj / = gJ |B | Jj , Ji,j 2 gJ 2 i j B
so
H=
j i=j
j B
j
(5)
Unfortunately, the above problem is incredibly dicult - in fact, exact solutions exist only for one-dimensional systems (chains of spins) and some particularly simple examples in two dimensions (the two-dimensional Ising model, whose solution is due to Lars Onsager). In three dimensions, for accurate knowledge of the behaviour of such systems formed by many interacting magnetic moments we rely on computer simulations (e.g Monte Carlo) and, of course, on experiments. However there is an approximation which not only works quite well in many instances but moreover aords a mathematically simple and physically insightful description of this and many other many-body problems: the
We start by noticing that the above Hamiltonian can be re-written in the form
H=
i=j
Ji,j i + B j . 2 gJ 2 B
(6)
We now note that this looks the same as the Hamiltonian of a paramagnet made up of non-interacting th magnetic moments if we say that the magnetic moment at the j site experiences an applied eld of the form
i=j
2 Ji,j /gJ 2 i + B. B
In eect, putting in
Bef f,j =
i=j
we have
Ji,j +B 2 g J 2 i B
(7)
H=
j
Bef f,j j .
(8)
which looks, formally at least, exactly like the Hamiltonian of a paramagnet. Of course
is itself a uctuating magnetic moment - in other words, unlike the case of a true non-interacting
in the form
H = H1 + H2 + . . . + HN ,
(Lecture V, Eq. 14) where each
Hj
depends on the
j th
molecular eld
i=j
i,j i
depends also on all the other magnetic moments replace the molecular eld with a moments that interact with
mean eld
Bef f,j
i .
i=j
Ji,j i + B . 2 gJ 2 B
(9)
This captures the notion that the environment of one magnetic moment is just like the environment of any other: once the moments settle on a particular form of collective behaviour, they will all follow it. This is a physically-reasonable approximation that works very well in many cases. The mean eld approximation, developed initially by Curie and Weiss in the context of magnetism, is the bedrock on which much of our knowledge of emergence and many-body physics rests. Further applications of mean eld theory include the Bardeen-Cooper-Schrieer theory of superconductivity (which was one of the great triumphs of Twientiethcentury physics and we will see later on).
We can now write the Hamiltonian in the separable form recalled above with the Hamiltonian for each individual site
given by
Hj = Bef f,j j ,
or
(10)
Hj = Be,j z,j
where
(11)
z,j
is the comopnent of
Bef f,j .
M = Ms Bj
(Lect. VII, Eq. 8) Using
tot
gJ |B | jtot B kB T
M = j,z
and
Ms gJ |B | jtot
(Lect. VII, Eq. 3) and substituting the eective eld
Be,j
instead of
we obtain
j,z = z Bj
sat
sat Be,j z
tot
kB T
(12)
where
sat z
moment along the eld direction; note that it depends on the angular momentum quantum number and therefore, in general, on the site index
jtot
j,
we will assume below that all ions are equivalent i.e. jtot is the same everywhere and, as in the above two sat on jtot explicitly). equations, we will not show the dependence of z The above result (12) is just a recycling of our earlier formulae for a paramagnet but with the applied magnetic induction
Be,j depends on the value of the magnetic moment itself. Thus equation (12) is to be solved self-consistently with (9), which also relates
our earlier theory of a non-interacting paramangnet is that the eective mean
z,j
2
to
Be,j .
This is very complicated in general so we will assume now a few particularly simple cases.
Let us say a little bit about the validity of the mean eld approximation. temperature
F = H T S.
(14)
The rst term is the energy of the system. The second term is the temperature times the entropy, with a negative sign. Thus the thermal equilibrium state of any system will strike a balance between minimizing the energy
S.
Consider the mean eld theory of a ferromagnetic system. The magnetic moments have collectively settled on a single direction which they are pointing along. If they were all perfectly `locked' into that position, there would be no uctuations, so the entropy would be zero (only one microstate compatible with the observed macrostate). In mean eld theory uctuations are allowed, but only at the level of a single magnetic moment, i.e. each magnetic moment uctuates, as in a paramagnet, but it does so in the presence of an eective mean eld that does not. Mean eld theory starts to break down when there are collective uctuations that involve more than one individual magnetic moment ipping or re-orienting in sync. Such
collective excitations
cost some energy, so at very low temperatures (where the rst term in the above equation is dominant) they do not take place. However at any nite temperature they can potentially proliferate, as they lead to an increase in entropy. To ascertain whether collective excitations will proliferate, leading to a failure of the mean eld theory, imagine a system with ferromagnetic interactions in its thermal equilibrium state. hypothesize a collective uctuation, in the form of a small region, of size moments have spontaneously ipped to point in a dierent way: For simplicity let us say that interactions take place only between nearest neighbours, with exchange nocstant
J.
Let us now
l,
Locally, both inside and outside the region where the uctuation has happened, each magnetic moment sees other magnetic moments pointing the same way. However, on the surface separating the uctuation from the bulk, there are magnetic moments that are pointing in opposite directions and which therefore raise the total energy of the system. The increase in total energy is
J l2
where
l2
the surface separating the bulk from the uctuation. Therefore as the uctuation grows its energy increases. For low enough temperatures, this will tend to make the uctuations small and short-lived, validating the mean-eld approximation. Now consider the case of a one-dimensional system, i.e. a chain of magnetic dipole moments where each
magnetic moment interacts with the one in front and with the one behind it. Let us ip all the spins in a region of the chain to create a uctuation, as we did above:
As we can see, there are a couple of bonds along which two adjacent spins are pointing in opposite directions. This costs an energy
J (2J ,
Unlike the 3D case, where the energy increased as the uctuation grew, in the 1D case there are still only two unstatised bonds, so the energy cost of the uctuation is still is independent of its size, l. Therefore, for any given
J.
T > 0,
the entropy, and once they form their size will be un-bounded. More generally, the energy cost associated with the boundary of a uctuation of linear size
grows like
Jld1 ,
where
d,
only single-moment uctuations like the ones considered in mean eld theory are important. Indeed mean eld theories become exact in the academic, but conceptually important limit For small
(it is conceptually
important because it gives us an anchoring point where we can prove that mean eld theory must work).
d, the larger uctuations become very cheap as a means to increase entropy so they proliferate and
mean eld theory becomes inadequate. The main consequence of this is that the mean eld approximation often works in 3D, though not always, never works in 1D, and rarely works in 2D.
Ferromagnetism
Let us assume that the interactions between magnetic moments are all ferromagnetic:
Ji,j > 0,
(1)
tending to align the magnetic moments in the same direction. It is therefore reasonable to suppose that all magnetic moments
i = z uz
for all
i.
(2)
This will of course be the direction of the applied magnetic eld. (Lect. XI, Eqs.9,12) become
= sat B z
sat Be z kB T
(3) (4)
Be = j z + B
where
B (y)
is the Brillouin function (Lecture VII, Eq. 9; note that we have ommitted the dependence on
jtot
jtot is
1
2 g J 2 B i=j
Ji,j j
(5)
is a dimensinless coupling constant quantifying the strength of interaciton of the spin at spins in the sample. In what follows we will assume that the exchange constant
Ji,j
and
j,
(6)
Ji,j = J (Ri Rj ) ,
and therefore
will be site-independent:
j = .
The above equations can be recast as
(7)
= B
sat Be z kB T
2
(8)
(9)
sat z kB T
Dening the dimensionless parameter
z sat z
(10)
y
we obtain
(11)
(12)
z 1 = sat z
2y sat z
kB T
B sat z
=T yB .
vs the parameter
(13)
To solve this system of equations we plot the quantity controlling the problem are
z /sat z
above two formulae. Where the curves cross we have found a solution. The three dimensionless parameters
jtot T B kB T
2 sat z
(14)
(15)
(16)
The rst of these three parameters is the angular momentum quantum number of the magnetic moments; the second compares the temperature to the energy of interaction between magnetic moments; and the third compares the applied magnetic induction to the molecular eld. paramagnetism and whose shape is given by the vertical axis, at of The two cruves whose intersection gives our mean eld solutions are the Brillouin function, which we have already encountered in the theory of
y = 0,
is
B .
jtot ,
The second curve is a straight line. In what follows we will talk about
small temperatures, large temperatures, small elds and large elds meaning small and large values
and
B,
and
Example:
Magnetization of ferromagnet whose magnetic moments have angular momentum quanrtum num-
ber
B =
giving three solutions, two of which point against the direction of the eld and which are therefore unphysical. The solution where
points in the eld direction is the physical one. Once we have located
that solution approximately we can zoom in graphically on the solution or determine it numerically by, for example, the bisection method. A few bisections give, for the above parameters, y 4.76(6), for sat which we nd z 0.992 z (i.e. for this value of the eld the magnetization is almost completely saturated).
1.1
Magnetisation
This is to be expected: as in the case of a non-interacting paramagnet, when we apply a eld we immmediately polarize the magnetic moments. If the interactions between magnetic moments are also tending to make them point in a given direction, the eect is only re-inforced. In fact we can nd up to three solutions, as in the previous example:
Evidently the solution with the highest magnetization in the direction of the eld will have the lowest energy, as we already mentioned in the example.
Now look at what happens when we decrease the eld. In the case of the paramagnet, remember that for zero eld the magnetic moment (and therefore the magnetization) was zero. Here it depends:
For large
the straight line is very steep and as the magnetic eld is turned o the magnetization we
The three straight lines in the above plot correspond to three dierent values of the eld, decreases and for zero eld it is equal to zero. Thus for large
B = 2/5, 1/5, 0
(lower, middle and upper lines, respectively). As we can see as we lower the eld the magnetization
For low
then even at
B=0
In this plot we have decreased eld from unphysical, as discussed above). The
B = 0.6 (lower straight line) through B = 0.4 and 0.2 to B = 0. We go from having one solution at positive y to having three solutions (the two new ones being
B=0
curve in the above example deserves additional comment. Here there are still three solutions of
z = 0
B 0
limit of one
of the two unphysical solutions going against the applied eld direction). On the other hand the positive and negative solutions are equivalent in the absence of an applied magnetic eld. of an applied eld the system will choose one of the two solutions at random: As soon as we add an innitesimal eld, the one in the direction of the eld becomes slightly preferable. However int he absence
breaking.
spontaneous symmetry
Within this theory, the explanation is that the molecular eld remains nite even in zero applied eld. This is the way the mean-eld theory captures the tendency of magnetic moments to their mutual interactions. Another way to look at this is as a function of temperature, for xed magnetic eld. At zero eld, since the temperature controls the slope of the straight line, we see that there is a slope is such that there will be a nite magnetization:
self-organize as a rsult of
T.
y=0
T,
i.e.
(17)
The latter can be easily obtained from the low-y approximation to the Brillouin function that we obtained
B (y)
Equating these two slopes at
B(y) y y=0
(18)
(19)
T = Tc
we obtain
T = kB TC =
(20)
(21)
i.e. where
(22)
=
(Lect. VIII, Eq. 10)
2 = gJ |B |
jtot (jtot + 1)
in the absence of an applied eld. It tells us that ferromagnetism will develop at a higher temperature if
either the interactions between magnetic moments are strong (large are large (large
).
and
TC
jtot
and on
Ji,j :
(23)
1 kB TC = jtot (jtot + 1) 3
A particularly simple case is obtained when
Ji,j .
i=j
Ji,j =
Then
J 0
if
i, j
otherwise.
(24)
kB TC =
where
(25)
M = .
(Lect. V, Eq. 2). As the temperature is lowered, we go from zero magnetization at transition takes place continuously, i.e. to
(26)
T > TC
to non-zero at
TC ,
T < TC . The + T = TC
This is called a
TC
it becomes
metry breaking.
and
M :
sym-
the symmetry has broken, leading to a more ordered state (lower entropy). Such concepts were invented to deal with superconductivity and magnetism and then found much wider applicaiton, e.g. in particle physics (the Higgs mechanism). Under a magnetic eld, the temperature dependence of magnetization is dierent. It looks as follows (red curve):
We see that
T,
on
Ms
just as in the zero-eld case. This can be obtained using the same geometric
1.2
As in the case of the paramagnet, let us look at the low-eld behaviour of the magnetization to derive the zero-eld magnetic susceptiblity,
0 =
Remember that for a paramagnet
M H
= 0
H0
M B
= 0
B0
z B
.
B0
0
(Lecture VIII, Eq. 11)
para
0 2 1 . 3kB T
(27)
We wish to see how this expression is modied when interactions between magnetic moments are taken into account. We will focus on the case where
T > TC , i.e. we are in the disordered (paramagnetic) state. Now y Eq. 11 we see that at low eld and in the disordered state, y= sat Be sat ( z + B) z = z kB T kB T T > TC the only net magnetization TC we can do a low-y expansion
we use
(28)
is small
B0
because both
and
z B
are small (B is small because we are calculating the susceptibility in the is that induced by the small (18) in (12): and above
limit and
(29)
= sat z
(30)
= = =
sat z
(31)
(32)
(33)
= gJ B J/
with
J2 = 1
2j
tot
(jtot + 1) . = =
From this
2 3kB T z
2 B 3kB T 1 2 3 B kB T 1 2 3
(34)
(35)
1 kB TC = 3 2 ,
therefore
z =
whence
1 2 3
kB T kB TC
(36)
ferro = 0
0 2 1 3kB T TC
(for T > TC )
(37)
which, interestingly, is the same as the zero-eld magnetic susceptibility of a paramagnet but with the temperature dependence replaced with
1/T
1/ (T TC ):
The divergence of the susceptibility is one of the main features of a continuous phase transition (also called, as we mentioned above, a second-order phase transition or an instability). As we lower the temperature, and we approach the critical temperature at which the system will develop its own, spontaneous magnetization, inducing a magnetization by application of an external eld becomes easier and easier. above Eventually, right
All instabilities are accompanied by diverging susceptibilities, though it is not always the same susceptibility. In a ferromagnet it is the magentic susceptiblity, as we have just shown, but in a superconductor, for example, the susceptiblity that diverges at the superconducting critical temperature is the pairing susceptibility, which measures how susceptible electrons are to pair up in the presence of a pairing eld created by another superconductor.
H=
j i=j
Ji,j 2 gJ 2 i j B
j B,
j
Ji,j < 0.
(1)
As we will see this can lead to an instability into a form of order where the magnetic moments point alternately up and down in such way that the total magnetization averages to zero (unlike the case of a ferromagnet). However, let us rst discuss the magnetic susceptiblity of a system with antiferromagnetic interactions in the disordered state, i.e. above the critical temperature.
The discussion of the zero-eld magnetic susceptibility of an antiferromagnet above its critical temperature is entirely analogous to that of a ferromagnet. We will nevertheless go through the entire argument again here, just keeping it a bit more general. Recall that within the mean-eld approximation, the thermal average of the th moment at the j magnetic ion is given by
j,z = sat Bj z
sat Be,j z kB T
tot
j th
ion is given by
Bef f,j
i=j
Ji,j i + B 2 gJ 2 B
z -axis
Now let us assume that we start from a situation where the applied magnetic indcution is are at high enough temperature that the system is disordered (paramagnetic) so second of the above two equations gives
Bef f,j = 0
to a nite
j Bef f,j = 0 (we can check that this is self-consistent: back j,z = 0, as we assumed). B = 0.
This will make
Bef f,j
j,z .
The direction of the induced magnetic moment and the associated mean eld will be the
same everywhere, so the situation is the same as the paramagnetic state of a ferromagnet (i.e. a ferromagnet above its Curie temperature). The same arguments we used in the last lecture apply so we get
z =
1 2 3 B, kB T 1 2 3
1
2 gJ 2 B i=j
Ji,j .
(Lecture XII, Eqs. 5,7) In Lecture XII we then went on to identify ture, and deduced
1 2 3 with
kB TC ,
ferro = 0
1 0 2 , 3kB T TC
(Lect. XII, Eq. 37) which generalised the expression valid for a paramagnet ( para
0 2 1 . 3kB T
(Lecture VIII, Eq. 11) On the other hand, for negative exchange constants we may have expression by dening the quantity
<0
1 kB 2 , 3
i.e.
(2)
1 kB = jtot (jtot + 1) 3
Ji,j .
i=j
(3)
This new temperature scale allows us to write the zero-eld magnetic susceptibility in the following, very general way:
0 =
0 2 1 , 3kB T
(paramagnetic
state)
(4)
which is valid, within mean eld theory, whatever the sign of the interactions, as long as we are at high enough temperatures to be within the paramagnetic state. For a ferromanget ( Finally, for a
> 0), coincides with the Curie temperature, TC , and thus (4) coincides with (Lect. XII, = 0) we have = 0 and thus (4) reproduces (Lect. VIII, Eq. 11), also above. system with antiferromagnetic interactions ( < 0) is negative and it does not, in general, 0
as a function of
T /
for
>0
(right),
=0
(centre) and
<0
(left):
Taking the
= 0
magnetic moments (Lecture VIII), we see that ferromagnetic interactions tend to make the system
ceptible
more sus-
have the opposite eect: they react against the tendency of the magnetic eld to align all the moments in the same direction and thus decrease the magnetic susceptibility, compared to the non-interacting case. In the ferromagnetic case, the susceptiblity eventually diverges when
T = = TC
the ferromagnetic instability that we discussed in Lecture XII. In the antiferromagnetic case, the magnetic susceptibility never diverges. However there can be, as in the case of a ferromagnet, an instability towards an ordered state - it is just not an ordered state characterised by a nite magnetization. The description of such anti-ferromagnetically ordered state will be the subject of the main lecture.
state. We saw that unlike a ferromagnet, where the susceptiblity is enhanced by interactions between magnetic moments, compared to the non-interacting case, in an antiferromagnet the zero-eld susceptibility is suppressed compared to the non-interacting case. Now we will work out the mean eld theory of the antiferromagnetic instability that leads, in this case, to an ordered state below a critical temperature called the Neel temperature,
TN
Unlike the ferromagnetic state, which always takes the same form (all magnetic moments aligned in the same way), ferromagnetism can take many forms depending on the exact type of anti-ferromagnetic interactions in the system. We will assume here the simplest possible type of interaction, where each magnetic moment interacts antiferromagnetically with its immediate neighbours, and that's it:
Ji,j =
J < 0 0
if
and
otherwise.
(1)
We will also assume, unlike the case of ferromagnetism, that there is no externally-applied eld (this introduces considerable complications in the case of the anti-ferromagnetic state - see Blundell). Let us now think of what kind of magnetic order we should nd at low temperatures in this case. To illustrate it, consider the simple case of a square lattice:
We are saying that the magnetic moment on a certain site this square lattice example,
z = 4):
is always negative, it will tend to align the four neighbouring magnetic j th one:
This is the reason why the susceptibility is suppressed (see the last lecture): since each magnetic moment wants to point against the direction chosen by the surrounding ones, there is a penalty, coming from interactions, for aligning all the magnetic moments in the same direction. What kind of magnetic order can such interaction lead to? It is easy to guess if we forget now about the magnetic moment that we coloured in red and think of the four moments coloured in blue. Each of them has four neighbours (including the red magnetic moment) and each of those neighbours will tend to be antialigned to the corresponding blue spin - i.e. they will tend to adopt the same orientation as the original red spin:
Clearly as this propagates through the lattice we end up with two square sublattices, one made up of red magnetic moments and another one composed of blue ones:
We will call the two sublattices the A sublattice and the B sublattice. This suggests that we look for solutions of the form
z,j =
if if
j j
(2)
Bef f,j
i=j
Ji,j i + B 2 gJ 2 B
j,z = z Bj
sat
sat Be,j z
tot
kB T
(Lect. XI, Eq. 12) Substituting the above type of solution, with the interaction given by Eq. (1), above, and applied magnetic induction
B = 0,
12
=
i,j
J 2 g J 2 B
z,B z,A
and
J
2 gJ 2 B i,j
z,A z,A
(3)
(4)
. 2 g J 2 B i,j
(5)
z,A z,B
z,A z,A
= sat Bj z
tot
sat z kB T
(6)
= sat Bj z
tot
(7)
= sat Bj z
12
We introduce the notation
tot
(8)
i,j
...
and
ith
and
j th
i.e.
sat z Bef f kB T
(9)
sat Bef f z kB T
z,A = Ty sat z
and
with
T =
kB sat z
2
(10)
(11)
These self-consistency equations are entirely analogous to the ones we obtained for the ferromagnetic state in Lecture XII, with the magnetic moment in the A sublattice playing the role of the uniform magnetic moment. We solve the equation graphically in the same way as before
and nd the same temperature-dependence of the magnetic moment then for the
uniform
in the A sublattice
magnetic moment:
The critical temperature at which the magnetic moment becomes nite, the Curie temperature [this is left as an exercise] and is given by
TN ,
1 2 kB TN = 2 = Jzjtot (jtot + 1) , 3 3
(12)
2 = gJ |B |
is the number of
jtot (jtot + 1) is the r.m.s. expectation value of the magnetic moment nearest nieghbours (z = 4 is the square lattice example).
Note, however, that unlike ferromagnetism the magnetization in the antiferromagnetic state is
M =
(13)
Mstaggered
(14)
antiferromagnetic instability. It is the susceptibility to develop such staggered magnetization that diverges as the system undergoes the antiferromagnetic instability at
TN
Broken symmetry
The dierent nature of ferromagnetism and antiferromagnetism can be better appreciated in terms of the concept of
broken symmetry.13
H=
j i=j
Ji,j 2 gJ 2 i j B
j B.
j
(Lect. XI, Eq. 5) In the absence of an externally-applied magnetic induction (B=0) the above Hamiltonian becomes
H=
j i=j
Ji,j , 2 gJ 2 i j B
(15)
which is symmetric under the reversal of all the magnetic moments in the system. That is to say that under the operation
j j
for all
(16)
the Hamiltonian stays exactly the same. This is because the dot product
i j
does not change when I reverse both
(17)
and
j .
Such symmetry under the inversion of all the magnetic moments can also de regarded as a symmetry under time reversal, since in order to ip a magnetic moment all we need to do is reverse the currents giving rise to it, which could be achieved by turning time backwards so as to reverse the trajectories of all the charges:
14
Time reversal
13 14
The wonder and deep implications of broken symmetry are beautifully brought forth in Nobel laureate Phil W. Anderson's
177,
393-396 (1972).
Of course, spin does not emerge from circulating currents in any trivial way, but nevertheless it behaves in the same way
In the paramagnetic state, the time-reversal operation (inverting all the magnetic moments) would leave all observables unchanged. In particular the magnetization, which is zero in the paramagnetic state (remember that there is no applied eld in this discussion), would remain zero after time-reversal. In contrast, in the ferromagnetic state, where the magnetic moments have spontaneously chosen to point in one particular direction, time reversal would have the eect of inverting the magnetization, so it would have a macroscopically-observable eect. We thus say that
how does the system decide which value of The key to this conundrum is the
the magnetisation to adopt, given that the Hamiltonian (which describes fully the dynamics of the system) cannot distinguish between values related by time-reversal symmetry? divergence of the susceptibility to magnetise as the Curie temperature is approached from above. Because of this, any small perturbation, e.g. a tiny applied magnetic eld (e.g. the Earth's magnetic eld), will make a macroscopic region of the magnet point in one direction rather than another. In antiferromagnets, time-reversal symmetry is also broken, since ipping all the magnetic moments in the sample would have the eect of swapping the A and B sublattices, changing the sign of the staggered magnetization. However, in addition, another symmetry is broken, namely the symmetry under lattice translations. This says that if all the sites on the lattice are displaced by exactly one lattice constant in the same direction, so that each occupies the site of one of its nearest neighbours, the Hamiltonian, again, stays the same (this is obviously only valid for ininite systems with no boundaries - but for all practical purposes within this discussion any macroscopic chunk of matterial can be ocnsidered innite). This symmetry holds true in the paramagnetic state, and also for a ferromagnet (since all sites develop the same magnetization and therefore are still equivalent). However, when we carry out the operation in the antiferromagnetic state we nd that it has the eect of interchanging the A and B sublattices, which is the same as ipping all the magnetic moments, again changing the sign of the order parameter (that is, the staggered magnetization):
Thus the antiferromagnetic state breaks time-reversal symmetry and, in addition, also the symmetry under lattice translations.
be detected in this way. In this lecture we will discuss two of the main techniques used to detect the microscopic arrangement of magnetic moments in a material:
Neutrons have no charge, therefore they are very penetrating and can be used to probe the bulk of a magnetic material without being absorbed or reected near the surface. At the same time, they do have a spin (with angular momentum quantum number
s = 1/2)
moments inside the sample. This is what makes them valuable in the study of magnetism. They are not just useful for magnetism. In addition, they also bounce o the nuclei in the sample due to the strong-force interaction with the other nuclei in them, so they can also be used to study the crystal structure. Neutrons can be produced in two ways: through
nuclear ssion (in a sision reactor) and through spallation (using a proton accelerator). In nuclear ssion a heavy nucleus is hit by by a neutron from an earlier ssion event and splits into two
lighter nuclei, ejecting some excess neutrons in the process. Some of these neutrons go on to split other nuclei while the rest escape the reactor core and can be used to generate energy (by heating some water that evaporates and moves a turbine) or, even better (!), to do neutron scattering! In a
spallation neutron source, a proton beam from a synchrotron or cyclotron is red at a target made of moderator:
a chunk of material that is kept at some
some heavy element (e.g. Tungsten). The nuclei get hot and evaporate some of their neutrons. In either case, the neutrons have to go through a
suitable temperature where the neutrons can thermalise. When they leave the moderator, the neutrons have a known, thermal (or quasi-thermal) energy distribution, corresponding to the moderator's temperature, They have become thermal neutrons. The typical enegies of thermal neutrons are given by
T.
kB T
where
p2 , 2mn
(1)
and
mn
are the linear momentum and mass of the neutron, respectively. The momentum is related
p=
Combining these two equations yields the
h .
(2)
(4)
1 nm T /K
(5)
T 100K
1 nm/10 = 1
Angstrom, which
of the order of the atomic spacings of solids. At these wavelengths interference eects result from features that vary within the sample on the atomic scale. This is what makes neutron scattering useful for the study of magnetism. The choice of moderator will depend on what temperatures (i.e. what wavelengths) we wish to achieve,
e.g. to study biological systems or long-wavelength helical magnetism (see Lecture XVI) we may wish for somewhat longer wavelengths than to study, say, Neel order. Some common substances to use are water, solid hydrogen and solid methane. Once the neutrons have been thermalised, they are guided towards our sample of magnetic material. While going through the sample, the neutrons will get deected a certain angle
Each neutron is detected after being scattered and its exact energy is recorded. Recording the energy is of course the same as recording the wavelength, since, using again two of the relations employed above,
E = =
p2 2mn h2 . 2mn 2
(6)
(7)
1. In a
pulsed neutron source, the neutrons are produced in bunches, each lasting a very short time (of
t
it takes it to hit the detector (
the order of a nanosecond). Since we know when each pulse of neutrons hits the sample, we can work out the energy of a neutron from the additional time its momentum is
time-of-ight).
If the distance between the detector and the sample is l, then the neutron's velocity is
p = mn v = mn (l/t) .
continuous source, there is a constant ow of neutrons so the time-of-ight technique cannot be chopper between the source and the sample.
The chopper is a rapidly-rotating
used. This is the case of most reactor-based neutron sources. In that case there are two potions (sic!): (a) We can place a
cylinder with two apertures which eectively chops the neutron beam into pulses. 3. Alternatively, and more commonly, we can use a
neutrons of dierent energies in dierent directions (using Bragg diraction, which we explain below). By orienting the monochromator crystal in dierent directions we can select neutrons of dierent wavelengths to reach our sample:
Scattered neutrons
chopper
Sample
Detectors
ed t er n s at o S c e ut r n le mp Sa
cete rs D o t
neutron shield
While reactors generate much higher uxes of neutrons than spallation sources, in a pulsed source data from all neutrons can be collected, while in a continuous one we have to throw away most of the neutrons. The result is that in the end both technologies have their advantages and disadvantages. The scattering angle
depends on the crystalline and magnetic structure of the system. It is very useful to
crystallographic planes.
many parallel, equi-spaced planes containing an identical array of atoms. For a given crystal lattice, there are many such sets of crystallographics planes, each of them with a given periodicity (the spacing between the
planes) and direction (their orientation). For example, here we draw two of the many sets of crystallographic planes that make up a square lattice:
When neutrons scatter o a crystal, trajectories corresponding to dierent scattering angles interfere. This will lead to more scatering intensity for some values of
scatering if the angle is such that trajectories correpsonding to reections on parallel crystallographics planes interact construcitvely. Fro this to happen two conditions are necessary: 1. The angle crystal. 2. The lengths of the optical paths corresponding to reection on subsequent planes must dier by an integer number of wavelengths, so as to ensure constructive interference. The latter condition leads to the
where
n = 0, 1, 2, 3, . . .
/2
/2
elastic scattering,
energy with the system. Most neutron scattering events are like this, so it is in general a good approximation
actually exchanged some energy are deliberately selected. We will discuss this later.) Under this assumption the magnitudes of the neutron momentum before and after it went through the sample are identical. terms of the neutron's wave vector
k,
p = k,
this translates to
(9)
|k| = k .
For elastic scattering, the scattering vector
(10)
Q = k k,
relative direction of
(11)
that is, the change in the neutron's wave vector on going through the sample, depends exclusively on the
and
k,
(12)
k' 2 k Q
sponding set of crystallographic planes. Regarding its magnitude, it follows from the Bragg diraction law (8). First we notice that the wavelength of the neutrons via de de Broglie relation, Eq. (2):
is related to
k=
thus
h 2 k= , 4 sin .
2d 4 Q = n
i.e.
(13)
Q=
(14)
Q=n
2 . d
(15)
related to the planes' spacing by the above formula form the crystal's A neutron diraction experiment will show bright
reciprocal lattice.
lattice vectors. When the system orders magnetically, the scattering of neutrons will become stronger. The crucial reason why neutron scattering is so useful in the study of magnetism is that anti-ferromagnetic can alter the reciprocal lattice. Take the square lattice we looked at before. Assume we cool down the crystal and it goes through a Neel instability, so it develops anti-ferromagnetic order:
Since the neutrons couple to the magnetic moments, they are sensitive to the broken translational symmetry. This has been broken by the AF order, with the result that, for example, the set of crystallographic planes that we colored above in green has its periodicity halved (since spin-up and spin-down planes are no longer equivalent):
QT >TN = n
Giving values to
2 2 1 QT <TN = n = QT >TN . d 2d 2
(16)
n = 0, 1, 2, . . .
we see that all the reciprocal lattice vectors that we had originally are still
there, but now there are smaller ones that weren't present in the paramagnetic crystal: At
T > TN
At
T < TN
As the temperature is lowered further, the magnetic peaks become more intense as the magnetic moments become stronger:
This is how antiferromagnetism was discovered and the main way to study it nowadays.
Muons are sub-atomic particles. Like the electron, they are leptons, and they carry the same charge charge as the electron,
(negative muon,
) or the same charge of a positron, +e (positive muon, + ). Also like have s = 1/2. Thus lke the neutron the muon can interact with magnetic m 200 me , and a nite lifetime, 2.2 s.
moments - though the way we exploit that is quite dierent, as we shall see. Unlike the electron, however, the muon has a farily large mass,
Muons consitute some of the most common forms of cosmic ray. We are regularly bombarded with muons. They are generated when high-energy particles hit nuclei in the ionisphere. Interestingly, at the velocities with which they reach the planet's surface they wouldn't be able to make the journey form the ionosphere in the brief time spanned by their lifetimes, were it not for time dilation (resulting from their relativistic speeds) which means that from their pespective time passes mucho more slowly than from the point of view of our stationary frame of reference. The muons we use in condensed matter research are not of the cosmic variety. They are generated here,
on Earth, using proton beams. The muons are created by hitting a nucleus (any nucleus will do, but C or Be are commonly used) with a proton beam. When the proton hits the nucelus, it produces a pion which quickly disintegrates, emitting a muon and the corresponding neutrino:
Here there's an extra bit of particle physics that helps us a lot: it turns out that the pion's disintegration involves the electro-weak interaciton, which violates parity conservation. As a result, the muon is produced with its spin pointing anti-parallel to its momentum (the neutrino is also produced with its spin anti-parallel to its momentum; this way both spins add up to the pion's produce has a well-dened spin polarisaiton. Usually for condensed matter experiments positive muons are employed. The are implanted in a sample and then we wait for them to spontaneously disintegrate. Typically the muons will come into the sample in short pulses (~100ns or so, which is much shorter than the muon lifetime, see above). When they go into the sample, they will very quickly come to rest in side it. When the muons disintegrate, they emit a positron and two neutrinos. Again the electro-weak interaction comes to help us here, because the positron is emitted with higher probability in the direction in which the muon's spin is pointing at the time of disintegration:
s = 0).
It is the positrons that we detect, but because of this coupling between their direction of emission and the muon's spin direction, we gain information about where the muon's spin was pointing inside the sample. Because the spins of all the mouns were pointing in the same direction to begin with, if there are absolutely no magnetic elds (either internal to the sample or externally-applied) then the moments would just continue to point in that direction indenitely and all the positrons will be detected in the same direction, too. However, if there is an applied magnetic eld or, more interestingly, the sample has an intrinsic, internal magnetic induction (e.g. an eective exchange eld then the muon'sm magnetic moment will
Bef f
This precssion will happen at the Larmor frequency of the muon. The derivation is entirely as what we did for electrons early in this course, so we quote the result here:
Larmor = g Blocal
The gyromagnetic ratio and g-factor of the muon are
(17)
= g
giving
|e| 2m 2M Blocal
T
(18) (19)
(20)
This precession can be seen as a rotation in the direction in which we detect positrons for about 10s after the muon pulse has been implanted in the sample (after that time, most of the muons in the pulse have died out, so we have to wait until the next pulse comes along before we can collect more data). Very often, what is measured is the asymmetry between the positron detections in two detectors pointing in dierent directions. This asymmetry oscillates as the mouns precess around the internal elds in the sample. The following two animations illustrate much better than any words one could write how a muon rotation experiment works:
15
http://neutron.magnet.fsu.edu/images/uSR.gif http://neutron.magnet.fsu.edu/images/uSR2.gif Like neutron scattering, muon spin rotation alolows us to detect the magnetic moments that turn up below
TN
in anti-ferromagnets.
The key is that because the muon sits at a particular site within the crystal Muon spin rotation therefore allows us to determine the
lattice, it can see the magnetic moment of the nearest magnetic ion, without averaging the magnetization from dierent ions on dierent magnetic sites. temperature-dependence of the magnetic moment of each particular sub-lattice:
15
These
videos
come
from
Muon
spin
relaxation,
Quantum
Materials
Group,
Florida
State
University,
10
Muon spin rotation and, specially, neutron scattering underpin all modern research in magnetism and the technologies that are based on it (such as magnetic data storage). Although individual experiments are carried out by small groups of researchers, the neutron and muon sources themselves are huge enterprises (each of them has many dierent instruments on which dierent teams can be working with dierent samples at the same time). The following video describing the ISIS facility at the Science and Technology Facilities Council's Rutherford Appleton Laboratory, in Oxfordshire, gives an idea of the scale:
16
http://www.isis.stfc.ac.uk/science/a-video-tour-of-the-isis-facility4469.html
16
11
Competing orders
H=
j
Bef f,j j .
Lect. XI, Eq. 8) In the mean eld approximation the eective eld
Bef f,j
is given by
Bef f,j
i=j
Ji,j i + B . 2 gJ 2 B
Lect. XI, Eq. 9) This equation relates the eective eld to the expectation value of the magnetic moment, which can, in turn, be computed from the above Hamiltonian, giving
j,z = sat Bj z
sat Be,j z kB T
tot
Lect. XI, Eq. 12) Note that this gives the component of
j th
site,
Bef f,j .
T = 0) Bj
we have
sat Be,j z
tot
kB T
for
Be,j
0,
(1)
so the magnetic moments saturate i.e. they are perfectly aligned (as far as is quantum-mechanically allowed
17 )
with the direction of the local eective eld. Yet this equation does not give the direction of the local eld
Bef f,j
itself. In fact, there can in general be many solutions to the self-consistency equations corresponding
to dierent directions of the eective elds (e.g. for the bi-partite anti-ferromagnet of Lecture XIV, a ferromagnetic arrangement with
z,A = z,B
eective eld to point always in the same direction; but that solution is not physical).
17
Even the an angular momentum reaches its maximum alignment with a given direction, there is always uncertainty left in
the value of the comopnents perpendicualr to that direction - see Lectures III and VIII.
To determine which one of several low-temperature solutions to the self-consistency equations is the correct one we need to compute the
ground-state energy.
=
j
(3)
i=j
Ji,j i + B j 2 gJ 2 B j
j
(4)
j i=j
Ji,j i j B 2 g J 2 B
(5)
Given several competing solutions to the self-consistency equations, the above formula tells us which one has the lowest energy. The RHS of the above equation has two terms: an of magnetic moments; and a
interaction term featuring the dot products of pairs Zeeman term giving the energy due to the externally-applied magnetic eld.
The interaction term will depend on the relative orientations of the magnetic moments and therefore will allow us to decide between co-linear and non co-linera magnetism.
Ferrimagnetism
While our discussion of ferromagnetism (Lectures XI-XII) was fairly general, as was that of the correlated paramagnetic state of ferromagnets and anti-ferromagnets above their critical temperatures (Lecture XIII), our treatment of the Neel instability and of anti-ferromagnetic order below the corresponding critical temperature,
TN
(Lecture XIV) relied on particularly simplied assumptions about the crystal lattice and about In particular we assumed a
bipartite lattice
i.e.
divided into two equivalent lattices made up of totally identical sites. Let us discuss the state that emerges when this assumption is relaxed. We consider a magnet with nearest-neighbour antiferromagnetic interactions but where, unlike the case that we already discussed, rather than a bi-partite lattice with all identical sites we have a bi-partite lattice where the sites in the A sublattice are of one type and those in the B sublattice are of another - for example, the magnetic moment residing at an A site may have one angular momentum quantum number that at a B site may have a dierent size given by a dierent angular momentum quantum
We will still expect the magnetic moments to form a bipartite anti-ferromagnetic arrangement, but this time the spins that are pointing one way will have one value of their magnetic moment, while those pointing the other way will haev a dierent value:
Indeed, even from the more fundamental point of view of symmetry-breaking, this state is more akin to a ferromagnet than to an anti-ferromagnet, because it only breaks time-reversal symmetry, and not translational symmetry. To emphasize the fact that the moments are anti-ferromagnetically aligned, however, we call such states
ferromagnetic interactions. Why does this state not break translational symmetry? Even at high temepratures in the absence of a
magnetic eld, where the system is paramagnetic and the magnetic moments will all average to zero, the sites are inequivalent, so the spatial periodicity in the situation with no magnetic order is the same as in the magnetically-ordered state - to reproduce the whole crystal, we have to make copies of the region within the dotted lines:
This is the same unit we would need to repeat in order to reproduce the magnetic structure drawn above.
Our discussions of paramagnetism (Lectures VI-VIII), ferromagnetism (Lecture XII), anti-ferromagnetism (Lectures XIII-XIV), and ferrimagnetism (above) all relied on the assumption that the magnetic moments on distinct magnetic ions are
co-linear, i.e.
with respect to other magnetic moments. Co-linearity is, indeed, a natural assumption as, on the one hand, the applied magnetic eld will tend to make moments parallel to each other and, on the other hand, moment-moment interactions of the form
Ji,j i j
will tend to make the moments either parallel (Ji,j
(6)
> 0)
or anti-parallel (Ji,j
< 0).
situations where such co-linearity does not necessarily follow. We will give two examples below.
3.1
Spin-op phase
Firstly let us consider an anti-ferromagnet in the presence of an externally-applied magnetic induction. Then the two tendencies we mentioned above (for each magnetic moment to align parallel to the eld and for magnetic moments to align anti-parallel to each other) compete. This may lead the system to choose a phase where the orientation is intermediate between anti-ferromagnetic (each magnetic moment anti-aligned to its neighbours) and paramagnetic (aligned with the externally-applied eld). To achieve this, the system can exploit the Heisenberg Hamiltonian's symmetry under global rotations of all the spins (Lecture X). This allows it, for small applied elds, to orient the antiferromagnetically-aligned
moments in a direction perpendicular to the applied eld, so as to maintain the anti-ferromagnetic order while avoiding having any spins pointing against the eld, and allows, as the applied eld grows, for the magnetic moments to tilt increasingly in the direction of the eld until eventually, for high elds, a ferromagnetic arrangement in the eld direction develops:
The component of magnetization parallel to the eld develops gradually and is present, albeit small, as soon as an external eld is applied (however tiny). Such phase is termed a
anisotropy
spin-op
magnetic
(i.e. some intrinsic tendency, due to spin-orbit coupling, for each magnetic moment to point The anti-ferromagnetic moments can no When the eld
along a given easy axis, independently of moment-moment interactions). In that case the symmetry under global rotations of the Heisenberg Hamiltonian no longer holds. longer rotate freely, and therefore, for weak elds, they stay in their un-altered anti-ferromagnetic conguration, even if it means some magnetic moments are pointing against the applied eld.
transition.
is strong enough, they suddenly op onto a more ferromagnetic conguration. This is called a
spin-op
The spin-op phase then evolves continuously, as the eld is increased further, towards the
fully-ferromagnetic conguration:
If the anisotropy is very large, so that magnetic moments can really only point either parallel or anti-parallel to the eld, there is no spin-op but there is instead a ferromagnetic to the ferr-magnetic conguration:
3.2
Helical anti-ferromagnetism
Another feature of many real anti-ferromagnets that our discussion of the anti-ferromagnetic state has ignored are next-nearest-neighbours may induce
interactions beyond nearest-neighbours. In particular, anti-ferromagnetic interactions between helical anti-ferromagnetic order. This often happens, for example,
in a ferromagnet with layered crystal structure when in addition to the ferromagnetic interactions between the magnetic moments within a layer and between immediately adjacent layers there is an anti-ferromagnetic interaction between moments in next-nearest-neighbour layers. Each layer will order ferromagnetically and be mostly aligned with the two nearest-neighbour layers, but with a slight tilt so as to lower the antiferromagnetic interaction energy with the next nearest-neighbour ones. As we go from layer to layer, tha small additional angle builds up so that after some distance
2/q
q q
is called the pitch of the helix. may not be related to the lattice constant in any obvious way i.e. the helical order may be An example of this behaviour is Dy.
incommensu-
An important class of systems with anti-ferromagnetic interactions are those where such interactions are
frustrated.
moments that minimises the energy. Instead, there is a manifold of degenerate ground states. For example, consider magnetic moments on a triangular lattice. If there is an anti-ferromagnetic interaction between every pair of moments, then once two of the moments are aligned anti-parallel with one another the third one cannot be anti-parallel to both of its nearest-neighbours:
18
Indeed, whether the third moment points parallel to one of the other two or anti-parallel makes no dierence to the total energy, leading to equally likely at
T = 0.
macroscopic ground-state degeneracy: a large number of states are nite entropy in the ground state. (In general, what
will happen then is that other terms in the Hamiltonian, such as longer-ranged components of the interaction, will break the degeneracy, but sometimes we have to go to even lower temperatures to see that.) Frustrated magnets are a very active area of current research.
18
Image from Ludovic D. C . Jaubert, Topological Constraints and Defects in Spin Ice (PhD thesis, University of Lyon,
2009).
composed of Ho
3+
19
All other atoms and ions in the material are non-magnetic. The Curie temperature of this material is
TC = 500K.
Assuming that the magnetism is dominated by exchange interactions between nearest-neighbour magnetic ions, what is the value of the corresponding exchange constant? Compute the zero-eld magnetic susceptibility at cool down to
T = 1000K.
600K?
Use the mean-eld self-consistency equations to derive the size of the magnetic moments and the corresponding local eective magnetic induction
Be (in Tesla) at the ion site at T = 700K, 300K, and 100K. Estimate the precession frequency of muons implanted in the sample at those temperatures (assuming
that the muon sees the same eective eld as the magnetic ions).
Hint:
Problem 2:
You will need the parameters for a Ho3+ ion, which we worked out in Lect. IV.
20
A magnetic material is made up of two types of magnetic ions arranged in a NaCl structure:
As shown in the gure, this crystal structure has two types of site. Let us call them A sites and B sites. Suppose that A sites have magnetic moments of strength half that strength,
/2.
Let us also assume that the magnetic interactions take place between nearest
neighbours only and are antiferromagnetic. Sketch the ground-state magnetic structure and name what type of magnetism it corresponds to. What symmetries are broken by that state? Derive (explaining how you arrive at the result) the zero-eld magnetic susceptibility in the paramagnetic state (i.e. at temperatures above the critical temperature). Compare the formula you have obtained for the suscetibility with the result valid for ferromagnetism, anti-ferromagnetism and paramagnetism (Lecture XIII, Eq. 4) and comment on the similarities and/or dierences.
19 20
Image from Wikipedia, Cubic crystal system (retrieved 21 November 2011). Image from P. G. Nelson, Introduction to Inorganic Chemistry, http://www.hull.ac.uk/chemistry/intro_inorganic/ (re-
Problem 3:
=J
if
and
otherwise). Now add some large anisotropy i.e. assume that each site has
an easy axis such that its magnetic moment can only point in one of the two directions along that axis (up or down). This is called the Ising model. Describe the two ground states of the Ising model for applied elds. For the case when the exchange constant is set to a xed value
J >0
and
J <0
applied eld that will make the system prefer the other ground state. At this value of the eld there is a so-called spin-ip transition. Describe what that consists of.
Problem 4:
Consider a magnetic material made up of identical magnetic ions, of magnetic dipole moment
strength
21
This is best viewed as a stack of two-dimensional layers, each layer having a square-lattice geometry. All other atoms and ions in the material are non-magnetic. Let us say that the magnetism of our between moments between moments lying on top of material is dominated by exchange interactions between nearest-neighbours (i.e. within the same layer) and between next-nearest-neighbours (i.e. each other on dierent layers), with exchange constants Assuming that
and
, respectively.
J >0
and
J < 0,
be reected in a neutron diraction experiment? Derive, in analogy with the thoery of bipartite anti-ferromagnetism of Lecture XIV, the self-consistency equations describing this state at nite temperature. Dene an appropriate order parameter. Derive a formula for the critical temperature. How does it depend on
and
21
Image from Wikipedia, Tetragonal crystal system, http://en.wikipedia.org/wiki/Tetragonal (retrieved 21 November 2011).
J (Ri Rj ) i j .
Here
i Ji / = Si + Li /
is the total angular momentum of the positions of the two ions,
ith
Ji
in Lecture
X). We have also showed that the exchange coupling between ions
and
Ri
and
Rj ,
respectively:
Ji,j = J (Ri Rj ) .
We will assume all sites to have the same angular momentum quantum number, The ferromagnetic ground state
(2)
jtot .
|0
will have all the magnetic moments pointing along the same direction.
(3)
i,z |0 = jtot |0
i.
(4)
We are going to check that this state is an eigenstate of the Hamiltonian. This is not trivial - we hypothesized the above ground state as part of a mean eld theory but there was no guarantee that it was the exact ground state of the Hamiltonian. We will see that it is. It is useful to introduce the raising and lowering operators
i, i,x ii,y .
One can use the commutation relations for angular momentum to show that
(5)
22
(site i) i, | . . . mj . . . = C jtot , mj
tot where
tot
(site i) | . . . mj 1... ,
tot
(6)
C jtot , mj
tot
is a normalisation constant:
C jtot , mj
In other words,
tot
jtot
mj
tot
jtot + 1 mj mj
tot
(7)
i,
by one or, equivalently, they tilts tot axis so as to make its component along that axis change by :
B.H. Bransden and C.H. Joachain,
22
We can nd the proof in ny good introductory Quantum Mechanics textbook, e.g.
z-axis
If the spin at
z -axis
as possible, then
i,+
(site i) i, | . . . jtot . . . = 0 .
In terms of these operators, the Hamiltonian is
(8)
H =
i=j
Ji,j (i,x j,x + i,y j,y + i,z j,z ) Ji,j (i, j,+ + i,z j,z ) .
i=j
(9)
(10)
=
i=j
2 Ji,j 0 + jtot |0
(11)
= E0 |0 .
i.e.
(12)
|0
E0 .
In other words
|0
is a stationary
H|0 = E0 |0 ,
with energy
(13)
E0 =
i=j
This proves that
2 Ji,j jtot .
(14)
|0
is an eigenstate of
H.
here, but basically because this state is getting as low an energy as you can get from every bond, any chnage we made ot that state would actually increase the energy). So indeed the kind of ground state given by the
T 0
(For antiferromagnetism, it is a dierent story: the ground state of the Neel mean eld theory turns out not to be an eigenstate of the Hamiltonian, and it is therefore only an approximation. The true ground state is a lot more complicated, riddled with quantum ucutations.) Let us now study the low-energy
excited states.
lth
l, |0
(15)
(16) (17)
Note that we have introduced the shorthand notation angular momentum of the
|l
|0
lth
|l
is
not an eigenstate of H.
H|l =
i=j
with
(l)
i, j,+ |l
= j,l 2jtot |i
and
i,z j,z |l =
Thus
if if
l = i, j l = i or l = j.
(23)
H|l
=
i:i=l
Ji,l 2jtot |i
i,j:i=j,i=l,j=l
2 Ji,j jtot |l
j:l=j
(24)
=
i:i=l
Ji,l 2jtot |i
i,j:i=j
2 Ji,j jtot |l
(25)
+
j:l=j
= =
i
(In the last line we used that
(26)
(27)
i = l |i |l = 0.)
state are not simply re-orientations of individual quantum-mechanical spins. This already tells us that the mean eld theory will be more limited at nite temperature since it assumed a description in terms of a paramagnet in an eective eld. In a paramgnet, the excitations at nite temperature are changes in the orientation of individual spins. In a true ferromagnet, the excitations are more complicated. Let us now show that the following linear superposition of states with turned spins is an eigenstate:
1 |k N
This is a mentum,
eikRl |l .
l
(28)
plane wave
p = k,
k (
wavelength
= 2/k )
i.e.
and which is therefore completely delocalised throught the material. This is consistent
px z
axis.
the object which delocalises throught the sample here is not a particle, but the site at which the magnetic moment is slightly rotated away from the
|k
is
an eigenstate of
H.
|k
: (29)
H|k
1 N 1 N
eikRl H|l
l
eikRl
l i
(30)
1 = 2jtot N 1 +2jtot N 1 + N e
l
(31)
eikRl
l ikRl i
J (Ri Rl ) |l
(32)
E0 |l .
(33)
1 2jtot N
eikRl
l i
J (Ri Rl ) |i
(34)
Ri Rl R, Ri R
eik(R R) J (R) |R
R,R
(35)
eikR |R
R
(36)
= 2jtot
R
for the seoncd term we have
(37)
1 2jtot N
eikRl
l i
J (Ri Rl ) |l
(38)
Ri Rl R, Rl R
eikR J (R) |R
R,R
(39)
1 J (R) N J (R) |k ;
eikR |R
R
(40)
= 2jtot
R
and for the third term we have
(41)
1 N
eikRl E0 |l = E0 |k .
l
(42)
H|k
= 2jtot
R
J (R) |k + E0 |k |k
(43)
=
This shows that
E0 + 2jtot
R
J (R) 1 eikR
(44)
|k
H|k = Ek |k ,
with energy
(45)
Ek = E0 + 2jtot
R
J (R) 1 eikR
(46)
The states |k are spin waves and correspond to the lowest-energy excited states of the system. Ek is the spin-wave dispersion relation giving the energy Ek of the spin wave as a function of its wave vector, k. A spin wave is a delocalised spin-1 excitation, since it corresponds to lowering the angular momentum of the system by (this is what
i,
s = 1,
bosons
s = 1/2,
magnons.
(The independent spin-wave approximation is all right at low temperatures, when there are only a few of them spread throughout the system, but it is not so ne at higher temepratures when they start to collide with each other.) Let us now consider the above expression and expand for low wave vectors:
1 eikR = 1 1 + ik R = ik R +
This leads to
1 (k R)2 + . . . 2
(47)
1 (k R)2 + . . . 2
(48)
Ek
= E0 + 2jtot
R
J (R) ik R +
1 (k R)2 + . . . 2
(49)
E0 + jtot
R
J (R) (k R)2 + . . . , k:
(50)
which has a quadratic dependence on the wave vector of the spin wave,
(The plot shows the excitation energy with respect to the ground state, This is a very generic and very important result. gap to spin-wave excitations at low energies. wavelengths,
Ek E0 .)
in the limit of long
k 0,
(i.e. rotating the magnetic moments away from the preferred direction) uniformly throughout the sample.
Since the total size of the angular momentum, do this, this corresponds to a is
uniform rotation
J2 =
2 =
of all the magnetic moments in the sample. But as we As a result, one can uniformly roate all the magnetic
know the Heisenberg model is symmetric under such uniform rotations, i.e. the Hamiltonian of the system
invariant
k becomes nite, then the rotation is not the same everywhere, = 2/k ...
/2
...and that twist, of course, does cost energy. Quite generally, the ungapped, low-energy excitations that appear in a system whose ground state breaks a continuous symmetry of the Hamiltonian are called very general idea that goes well beyond magnetism. Magnons are very important in ferromagnets. For a start, because they are ungapped, they appear as soon as the temperature is non-zero. Since each magnon lowers the component of the magnetization along the direction of the ferromagnetic order, the magnons lead to a lowering of the magnetization, below the value predicted by mean-eld theory, at any non-zero temperature. This is called
This is a
the dependence of the suppression of magnetization on temperature (for small temperatures). The inuence of these Goldstone bosons goes even further for systems that are eectively two- or onedimensional (i.e. made up of un-coupled planes or chains of magnetic moments). density of states available for magnons at low energies in one or two dimensions, their proliferation at nite temperature is so great that they completely suppress the magnetisation, killing the ferromagnetism. This is the
Mermin-Wagner-Berezinskii theorem.
very general. In its more general formulation it says that you can have no spontenous symmetry breaking at nite temperature in systems with a continuous symmetry (i.e. with Glodston bosons) in less than three dimensions. In addition to their inuence on thermodynamic properties, magnons can actually be observed directly using
inelastic neutron scattering (INS). INS works by selecting the energy of the neutrons before and after
amount of energy
they scatter from the sample, which allows us to count the number of neutrons that have exchanged a given
and momentum
elastically (this is the main assumption of neutron diraction experiments like the ones we discussed before). Some of the inelastically-scattered neutrons will have ipped their spin, transmiting a quantum of angular momentum, , to the sample. This creates one magnon. By looking at many neutrons, and recording, for those whose spin has ipped on interacting with the sample, their individual values of at what combinations of those quantities scattering is maximum. Taking the spin-wave dispersion relation.
and
k,
we can see
Ek = ,
superconductivity.
23
23
Images from Rudolf de Bruyn Ouboter, Heike Kamerlingh Onnes's Discovery of Superconductivity, Scientic American
(March 1997) and from Dirk van Delft and Peter Kes, The discovery of superconductivity, Physics Toda y (September 2010).
Ek
jtot = 1/2.
Assume a 1D chain geometry:
Compute the quadratic approximation to the dispersion relation valid at low values of the wavevector
k.
Derive the mass of the magnons,
mmag ,
Ek = constant +
Problem 2: (*)
2 k2
2mmag
(1)
Ek
jtot = 1/2.
Assume a one-dimensional square lattice geometry:
[DRAWING]
The ring is kept in the superconducting state at all times. Initially there is zero current going through the solenoid. When we pass a current, is there any magnetic ux going through the ring? What is happening inside the supercondcutor to produce that result? Now imagine that the ring were much broader than the solenoid:
[DRAWING] Will any net magnetic ux go through the ring when we
start passing current theorugh the solenoid? Will the magnetic induction on the plane of the ring be the same everywhere? Sketch the magnetic eld lines around the ring.
http://www.scienticamerican.com/article.cfm?id=kryders-law (request from library and use pictures) Since then many magnetic and superconducting states with ever more exotic electromagnetic properties have been discovered, and the study of superconductivity and magnetism remains one of the most challenging areas of research. Although some materials are capable of displaying magnetic or superconducting propeties while others are not, it is important to emphasize that neither magnetism nor superconductivity are intrinsic properties of a given substance. Instead, they occur only when the temperature of the sample is lowered below its
temperature, Tc .
at
critical
For example, in Kammerlingh Onnes' experiment the sample of Hg went from having a
T 4.1K
(see Fig.
??).
R 0.1,
at a temperature of
T 4.3K
to vanishing resistance,
R < 105 ,
Tc 4.2K.
temperatures. For example, iron is ferromagnetic only below its Curie temperature,
Tc = 770K.
Since this
is well above ordinary room temperatures, we are used to thinking of Fe as an intrinsically ferromagnetic material. However, if we heat Fe above its critical temperature it will suddenly loose its ferromagnetism, even though other properties such as crystal structure or electrical resistivity will be changing only smoothly with temperature within this range. How can a material undergo, without any change in composition or in the interactions between its constituents, radical changes in its properties such as those that take place when it enters a magnetic or superconducting state? It all boils down to the balance between energy and disorder. At a given temperature, any system in equilibrium must minimize its Helmholtz
free energy,
F = U TS
Here
(1)
is to increase
low becomes more important. Thus magnetism and superconductivity are more ordered states than their
high-temperature non-magnetic and non-superconducting counterparts. The ordered states develop in order to lower the internal energy of the system. Just as the solid state is a state of matter, distinct from, for example, the liquid state, superconductivity and magnetism are also
states of matter.
or magnetic state as soon as it is placed in the right thermodynamic conditions, independently of the path through parameter space that was taken to place it there (as long as the sample is always in thermal equilibrium
24 ).
When a more ordered state of matter is reached upon cooling there is often an associated
broken symmetry.
To understand this let us consider the example of crystallisation (the change from liquid to solid - Fig. 1). The equations describing the microscopic motion of water molecules have a number of symetries. For example, they are homogeneous (the same weverywhere) and isotropic (the same in all directions). Yet when water crystallises, forming ice, the molecules arrange themselves in an inhomogeneous, anisotropic state. This has macroscopic manifestations such as the beautiful shape of a snowake (Fig. 1). The snowake is symmetric under rotations by
/3
and its multiples around its central axis but not, for example, under
all
/2
rotations.
arrangement of the molecules on the microscopic scale. Since the collective state of the macroscopically-large number of molecules forming the snowake has lower symmetry than the microscopic equations of motion that describe their motion we say that the symmetry is 'broken'.
24 25
What this retriction amounts to, in practice, is that any change in the sample's conditions must have taken place very The wonder and deep implications of broken symmetry were discussed by Nobel laureate Phil W. Anderson in his classic
177,
393-396 (1972).
Fig. 1:
states that a crystal may enter upon cooling further, below symmetries are we will get to in due course.
Tc ,
In principle, we can obtain the free energy of any system, and therefore all its thermodynamic properties, by computing its
partition function,
Z=
n=1,2,3,...
where the
kEn T
B
(2)
kB = 1.3807 1023 JK1 (Boltzmann's constant). This requires knowing the possible microstates of system, labelled by n = 1, 2, 3, . . . and their energies En . In other words we need to solve the mechanics
of the system under consideration. For any crystal, the mechanics are given by the following
Hamiltonian:
(3)
H = ...
The above Hamiltonian is the theory of everything as far as condensed matter physics is concerned.
[Two approaches: brute force (but no one knows how - you have to lower your demands on what you want to get and then you obtain DFT, which requires uncontrolled approximations) and models (the route we will follow here, which is used for the interesting cases when DFT does not work and also is more useful in a course because it leads to understanding).]
26
In a
metal, the crystal lattice is made up of positively-charged ions, with the remaining charge belonging to
27
Superconductivity takes place when
free electrons that are responsible for the conduction of electricity. resistance. In
those free electrons re-arrange themselves into a collective state where such conduction occurs without any
insulators, on the other hand, every electron is bound to a specic atom, ion or molecule.
28
Conduction
26 27
For conciseness, we have ommitted relativistic eects and assumed that there are no applied elds. However these are easy The positive ions are xed in deep potential wells at their lattice positions. The exception to this rule are super-ionic
to put back in without altering the throust of our discussion. conductors (not to be confused with superconductors) where both the free electrons and the ions are mobile. However, this is rare, and throughout this course we will assume it doesn't happen.
28
Tc
It is clear from the above that both magnetism and superconductivity are instances of
electronic order
Tc .
This is
i.e. it is the electrons inside the material which go from a disordered to and ordered state at nuclei that go into a more ordered state.
quite dierent from crystallisation, the example of broken symmetry that we discussed above, where it is the
The crucial consequence of the electronic nature of magnetism and superconductivity is that it makes
quan1nm.
is conned to a space the size of an atom or molecule. The uncertainty in its position is therefore
Now consider the case of the free electrons in a metal. These electrons have very low uncertainty in their
x L
(where
T , the typical thermal energy of one kB T . At high temperatures, the unbound electrons will increase their energy by 2 increasing their kinetic energy, p /2me kB T. From this we deduce a typical momentum for one of these electrons: p 2me kB T . Now, in quantum mechanics a particle is also a wave and its wavelength is given
This gives a very low uncertainty of the momentum. At temperature by the famous de Broglie expression
=
This leads to
h . p
(4)
h 2me kB T
(5)
29
29
implies some form of order as discussed above. However in general the crystal structure does not change at
Tc .
Reading list
The main texts for this course:
On superconductivity:
P G de Gennes; Superconductivity of Metals and Alloys (Westview Press, March 31, 1999)
This is a reprint of the classic text on superconductivity - still the most lucid exposition half a century after it was rst published.
R M White, Quantum Theory of Magnetism; (Springer; softcover reprint of hardcover 3rd ed. 2007 edition - November 23, 2010)
This is a more advanced text on magnetism, emphasizing theoretical methods and also with a very good section on neutron scattering.
Aschroft and Mermin; Solid State Physics (Thomson Press India, paperback edition, 2003) Charles Kittel; Introduction to Solid State Physics
These give a good grounding on solid state concepts useful to follow the course.
Michael Tinkham; Introduction to Superocnductivity (2nd Edition, 1996) S. Elliot: The Physics and Chemistry of Solids (1998) D.C. Mattis The theory of magnetism made simple (2004) Tilley and Tilley; Superuidity and Superconductivity, (1990)
Index
broken symmetry, 1
critical temperature, 1
free energy, 1 Hamiltonian, 2
E E
= / =
B = 0 B t
0 0
B = 0 J +
E t
R:
(6)
p2 1 ZR e2 HR = + . 2me 4 0 | R| r
The rst term is the kinetic energy and the second term is the Coulomb potential due to the attraction between the electron and the ion. The solutions to the corresponding time-independent Schrdinger equation give the atomic orbitals and energy levels:
R R HR |m = Em |m .
Note that the energy levels
(7)
Em
R.
Now consider the Hamiltonian of the same electron but in the presence of a whole Bravais lattice of ions. Let
{R}
be the set of all Bravais lattice vectors (the positions of the ions). The Hamiltonian is
H =
p2 + 2me p2 + 2me
Ze2 1 4 0 | R| r VR
(8)
= HR +
R =R
where in the second line we have introduced the notation
VR .
(9)
VR =
1 Ze2 4 0 |R | r R.
(10)
for the Coulomb interaction between the electron and the nuceus located at position approximate solution to the corresponding time-independent Schdinger equation,
We wish to nd an
H| = E| .
The Schrdinger equation is
(11)
HR +
R =R
and we will treat the term
VR | = E|
(12)
R =R VR as a perturbation. |
, in terms of the atomic orbitals
R |m
m,R
30
We follow quite closely the treatment in Ashcroft and Mermin, Solid State Physics.
| =
m,R
This expression assumes that
R R |m m | . R m,R |m R m | = 1. R m |
(13)
31 the coecients of the expansion. We now require that the state of the electron obey Bloch's theorem:
R |m
m,R
is a complete set:
The brackets
are
r + R| = eik.R r|
We can attain this condition by choosing
R.
(14)
R 0 m | = eik.R m | ,
Under this assumption the expansion of the state of the electron becomes
(15)
=
m,R
(16)
=
where we have introduced the shorthand notation
0 m m |
for the amplitudes to be determined.
32
0 n |.
The LHS takes the form
We now substitute (16) into (12) and project both sides of the equation on
HR +
R =R
VR
m,R R eik.R m |m =
m {Em
m
while the RHS becomes
0 R eik.R n |m R
+
R
0 eik.R n | R =0
R VR |m }
m {E
m
Thus what we have
0 R eik.R n |m }. R
m {(Em E)
m
31 32
0 R eik.R n |m R
+
R
0 eik.R n | R =0
R VR |m } = 0,
For a proof see the derivation of Eq. (8.6) in Ashcroft and Mermin.
Projecting (16) on
r+R
we obtain
r + R|
= =
r + R|
m,R
0 R eik.R m | |m
e
m,R
ik.R
0 R m | r + R|m
=
m,R
0 R eik.R m | r|m R
R R R
=
m,R
eik.(R eik.R r|
m,R
+R)
0 R m | r|m
= =
We have used the fact that the origin is the same as
0 R r|m = r + R|m the amplitude at r + R
eik.R
0 R m | |m
eik.R r| , Q.E.D.
i.e. the amplitude at
of the
mth
R.
0 = 1 ,
. . . namely
A. = 0.
Here
An,m
R
0 R eik.R (Em E) n |m
+
R
Take the simplest case of a 1D system when when
0 eik.R n | R =0
R VR |m .
R R n |m
R =0 VR
R |m
= 0
except
n=m
R = a, when it equals tm . Then E0 E 2t0 cos (ka) 0 0 0 E1 E 2t1 cos (ka) 0 A= 0 0 E2 E 2t2 cos (ka)
and . . . . . . . . . .. .
so we have
A=
chracteriand eigenvalue problem whose solutions are given by the charcteristic ploynomial equation
(17)
E E0 + tm a2 k 2
leading to
2 k2
2mef f
(18)
mef f =
To have this equal to an electron mass we would need
2tm a2
2
t m a2 = 2me x2
2 2
2me
is
i.e.
2 2 2me a
2 + v0 1 0 1 2 + v1 1 0 1 2 + v2
. . .
0 1 2 + v3 1
.. .
0 1 2 + v4
.. .
0
. . .
1 0
. . .
0 .. . 1
.. . .. .
0 1 2 3 4
. . .
=E
0 1 2 3 4
. . .
where
vj = V (ja) /
2me a2
Thus solving this problem reduces to nding the eigenvalues of the above
2 + v0 1 0
. . .
1 2 + v1 1 0
. . .
0 1 2 + v2 1 0
. . .
0 1 2 + v3 1
.. .
0 1 2 + v4
.. .
0 1
.. .
.. .. . .
= 0.
Additional material
Broken symmetry
Fitzgerald: The rich are dierent from us. Hemingway: Yes, they have more money.
More is dierent,
Quoted in P. W. Anderson,
Interestingly, the increased order in the superconducting and magnetic states manifests itself in the form of
broken symmetries.
Consider a system of
H=
i=1,2,...,N
Here
p2 i + 2mi
V (|ri rj |) .
i,j=1,2,...,N
(19)
mi
ith
particle,
pi
is its momentum,
describing the interaction between the particles at positions only depends on distance
potential potential
|ri rj |
between the two particles. Similarly, ... This Hamiltonian has a number
ri ri + R
for all
i,
(20)
the Hamiltonian stays the same. Similarly, if we rotate the system by a given angle the Hamiltonian also does not vary. ... that is very well illustrated by comparing the structure of a water droplet to that of a snowake (Fig. 1). Both are made up of identical consituents, namely large quantities of H2 O molecules. Yet the snowake is made up of crystalline water, which has much lower symmetry than liquid water. This is quite clearly visible in the shape of the snowake, which is symmetric under rotations by exactly 60 , or one of its multiples. Liquid water, on the other hand, is symmetric under all rotations (i.e. it stays the same after we rotate it by
an arbitrary angle). So the snowake is much less symmetric than the droplet - we say that in the snowake rotational symmetry is broken. Indeed liquid water has all the symmetries of free space, while crystalline water (ice) lacks many translational and rotational symmetries. What is remarkable about broken symmetry is that the state of the system has lower symmetry than the Hamiltonian that describes it.
33
As we will see magnetism and superconductivity provide particularly rich examples of symmetry-breaking. Can one predict a broken-symmetry state from a microscopic model of a given system? some given temperature we can compute the In principle, the
methods of statistical mechanics allow us to compute any observable for a system in thermal equilibrium at
T.
This presumes that we can solve the mechanics of the system, i.e. that we can
partition function
Z=
n=1,2,3,...
Here
(21)
is given by
pn =
microstates:
1 kEn e BT . Z On
This in turn allows us to compute the thermal average of any microscopic observable
O On =
n
Alternatively, we can use
pn On .
F = kB T ln Z
and employ thermodynamic relations to obtain other observables by dierentiating the free energy. For example, the internal energy is the thermal everage of the energy
(24)
En
n:
(25)
U= E =
n
This could be used to compute use the thermodynamic relation
pn En . F
and then
U = kB T 2
This can be expressed more compactly using
F kB T
(26)
U=
(F )
(27)
1 . kB T
(28)
Exercise:
Prove that (27) is equivalent to (26) using the chain rule. Then prove that this thermodynamic
Hints: it is much better to put all expressions in terms of rather than T rst. Then substitute (21) into (24) and that into (27). Substitute (22) into (25) and compare the results.
33
The deep implications of broken symmetry were discussed by Nobel laureate Phil W. Anderson in his classic article More is dierent, Science
177,
393-396 (1972).
What makes magnetism and superconductivity special as broken-symmetry states is that to compute the spectrum
{En }n=1,2,...
it is indispensable to employ
quantum mechanics.
In magnetism, time-reversal
symmetyr is borken, while in superconductivity is the symmetry with respect to an overall change of the wave function. Classical physics gives no mechanism for the breaking of either of these symmetries (in the case of superconductivbity, the pahse of the wave function is not even a classical concept). The problem is
H = ...
but of course we cannot solve this problem so we work with models.
Problem:
Using dimensional analysis, show what the superocnducting and magnetic Tc can depend on.
Now let us put more esh into the above discussion by considering the particular case of a ferromagnetic transition. To keep things simple, let us assume that the system is an insulator so we do not have the added complications of electrical conductivity and possible superocnducting behaviour. [...] This is made up of
1023
molecules of water that move about in a complicated fashion following (to Nevertheless on the macroscopic leand Quantum
Mechanics is the deepestngth scales corresponding to the above picture the liquid making up the droplet seems stationary, uniform and isotropic. This is not suprising i.e. it has the same symmetries quite uniform represents the water droplet level Insideaccording to the droplet, we nd a liquid that is homogeneous and isotropic: although the chaotic trajectories of the water molecules inside it ... are entered What happened at mixture of
Tc
1023
34 and conduction
electrons moving around them), but of the collective state of those constitutents. ions, the electrons : both above and below
Tc ,
1023
Mercury ions, vibrating around their equilibrium positions on the crystal lattice , with the same conduction electrons moving through it. The interactions between the electrons, between the electrons and the ions, and between the ions were also exactly the same above and below the critical temperature. Indeed the whole physics of most superocnductors can be described by the relatively simple Hamiltonian
H =
Pi2 + 2Mi i 1 + 4 0
p2 j 2me Zi Zi e2 + Ri Ri
35
(29)
i,i
j,j
e2 rj rj
i,j
Zi e 2 Ri rj
p,
(30)
=
Here
h . p
At a given temperature
T , the kinetic
(31)
34 35
j, j
designate electrons,
Tm = 234.32K [http://en.wikipedia.org/wiki/Mercury_%28element%29 (accessed 19 August 2011)]. i, i designate ions, e and me are the charge and mass of an electron, respectively,
and mass of an ion, and the position and momenta of electrons and ions are given by the operators
where the
me = 9.1095 1031 Kg
is the rest mass of an electron (for free electrons we would have written ;
sign take into account that the electron may be bound inside an atom or interacting with other Combining the last two expressions we obtain the
electrons). electron,
of the
At room temperature,
T =
h . 2me kB T
Given that
(32)
T 300K,
we have
Let us rst rst consider an isolated atom. As we know from undergraduate quantum mechanics, the atom has a postiviely-charged nucleus surrounded by a cloud of electrons that occupy a series of discrete energy levels. We will ignore for the time being the interactions between the electrons. Then the Hamiltonian of each electron can be written individually. Assuming that we can take the nucleus to be xed (using the fact that its mass is much larger than that of the electron) the Hamiltonian in question is
(33)
me
is its mass,
Ze
The rst term on the RHS is the kinetic energy and the second term is
the potential energy due to the interaction with the nucleus. interaction features the permitivity of free space this is given by Schrdinger's equation
1/4
, and describe exclusively the interaction between each electron and the nucleus. At the non-relativistic level
p2 1 2mr