2.1. Basics
As indicated in the introduction, several techniques exist for pursuing spin wave calculations. For the ferromagnetic case, which is the topic of this work, they all produce identical, and in fact exact results for the zero-temperature excitations. That is, the calculated energies and wavefunctions are the exact eigenvalues and eigenstates of the Hamiltonian.
The Hamiltonian considered in this work consists of Heisenberg exchange and ligand-field terms, and is written as:
with the standard meaning of the symbols. The length of the
-th spin is denoted as
, and the number of spin centers in the cluster is
. As alluded to in the introduction, the single requirement for the presented methods to be valid is that the ground state contains the fully polarized ferromagnetic state:
or its opposite
. Here,
denotes the magnetic quantum number or eigenvalue of the
-component of the total spin operator
and
the magnetic quantum number of the
-th spin center. The total spin quantum number will be denoted as
. The quantum numbers
and
are restricted by the maximum values
, which are achieved in the fully polarized state.
Any spin system for which the states are ground state will in the context of this work be called ferromagnetic. Accordingly, the coupling constants, , need not to be all ferromagnetic (), some can be antiferromagnetic (), but their overall effect must be such that the states Equation (2) are ground state. This happens in a relatively large class of molecules, which could be characterized as “ferromagnetically coupled with Ising anisotropy”.
The Hamiltonian Equation (1) is not the most general Hamiltonian which permits a fully polarized ground state. Examples would be systems with additional higher-order ligand-field terms (see also
Section 3.3), Ising exchange interactions, applied magnetic fields, and others. However, it is general enough to present the FCSWT procedures comprehensively, and includes systems of interest, such as several single molecule magnets and lanthanide containing clusters. It should not be difficult to extend the results to other systems.
In this work, two SWT techniques are presented. The first is, so to say, the simplest possible, and relies on basic quantum mechanical concepts. We will find it most convenient for practical calculations. The second method uses the formalism of boson raising and lowering operators, and is the standard approach in physics. It will provide us with deeper insight into the excitations calculated with the first method. The methods will be introduced in two steps: first, the case of a dimer will be treated, and then the results will be generalized to arbitrary clusters.
2.2. Dimers
The Hamiltonian of a dimer with isotropic exchange coupling reads:
and its eigenvalues and states are well known,
and
. Let’s first consider a spin-1/2 dimer, for which
and
. Its singlet and triplet eigenstates are:
with a standard notation of the states. Obviously, for a ferromagnetic interaction (
), the fully polarized states
or
, respectively, are ground states. Let’s pick the state
. We can then further observe that the subspace for
consists of two states. Moreover, we can construct a basis for this subspace by applying the lowering spin operator of each spin site,
, to the fully polarized state:
This method for constructing a basis
for the subspace
works always, for arbitrary number
of spin centers and arbitrary spin lengths
, because: (i) there are as many spin centers and thus operators
as there are states in the subspace; (ii) the states generated by applying
to
are mutually orthogonal; and (iii) the states generated by applying
to
all belong to the magnetic quantum number
. We thus have created a complete basis, and the eigenenergies and wave functions belonging to
are obtained by diagonalizing the Hamiltonian matrix in this basis,
. Note that by construction the dimension of the matrix
is equal to the number
of spin centers in the system, or two in the present case. The calculation of the matrix elements is straightforward; a general formula will be given in the next chapter. In the present case, the eigenstates and energies can be obtained by inspection, yielding:
where
is introduced as an arbitrary label to index the states. It is easily confirmed that the states
and
are exactly the states
and
, respectively. Thus, the procedure of constructing a basis for
by applying the lowering spin operators
to the fully polarized state
, and diagonalizing the Hamiltonian matrix in this basis, has provided us with the exact eigenstates and energies in that subspace. The resulting set of eigenstates are called “spin waves” (the justification is given later).
Some notational comments. The state created by applying the lowering spin operators to the fully polarized state describes a spin deviation at the site . These states will be called “local spin deviations” and the associated wave functions will be written as or . The spin wave states will be written as or , where the values of () may be chosen as appropriate (if not specified otherwise the states are indexed by integers running from to ).
For the spin-1/2 dimer, one arrives at the picture shown in
Figure 1. The spin-1/2 dimer exhibits four eigenstates, one of them is the fully polarized ground state, and two of them are calculated by using FCSWT. For the spin-1/2 dimer this is sufficient to completely construct the energy spectrum, since the “missing” fourth state belongs to the
multiplet, and thus must be degenerate to the fully polarized state and can be constructed by repeated application of the total spin lowering operator
. In the general case, the information obtained from FCSWT will not be sufficient to arrive at a complete energy spectrum (not even close to that), as upcoming examples will demonstrate. However, FCSWT provides a complete description of the excitations relevant at zero temperature. For instance, the neutron scattering intensity is subject to the selection rules
and
[
3]. These imply that, at zero temperature, only transitions with energies matching the excitation energies of the spin waves are allowed. The argument applies analogously to other experimental techniques. This is an important finding, and the foundation for the usefulness of FCSWT in practical work.
Before proceeding to the case of a general spin cluster, it is interesting to discuss some extensions. It might be tempting to apply the approach also to the antiferromagnetic dimer (
). The antiparallel alignment of the spins in the ground state, as it is obtained in the classical model, might be translated into the quantum state
(or its opposite), which is known as the Néel state [
2]. However, here, application of the lowering spin operators
does not produce a viable basis, nor does the diagonalization of the associated Hamiltonian matrix yield reasonable states. This is obviously due to the fact that the Néel state is not an eigenstate of the Heisenberg Hamiltonian Equation (3). Only for systems which undergo an antiferromagnetic phase transition is the ground state correctly described by the Néel state [
1,
2], and this can happen only in extended systems [
2]. Bounded spin systems such as magnetic molecules do not exhibit magnetic phase transitions. Therefore, applying SWT to antiferromagnetic molecules is not justified in general, and needs a careful case-to-case analysis [
18,
21]. An analogous deficiency occurs in ferromagnetic clusters with easy-axis anisotropy, where a state similar to
would be ground state. These findings explain the limitation in this work to “ferromagnetic systems with Ising anisotropy”, as defined before.
Also dimers with spins larger than 1/2 are illustrative examples (we continue with discussing ferromagnetic systems). The application of the lowering spin operators
to the fully polarized ground state yields two spin waves, which shall again be denoted as
and
(their expansion in terms of
functions is of course different from Equation (6)). For the case
,
, which is discussed exemplarily, the energy spectrum is shown in
Figure 2, where also the situation of a dimer with additional Ising type anisotropy is presented.
In both the isotropic and the Ising dimers, the fully polarized states are ground states. As before, FCSWT provides the complete zero-temperature excitation spectrum, as observable in e.g., INS. For an isotropic spin cluster the spin wave with the lowest energy, or the state
, respectively, must correspond to the
component of the ferromagnetic ground-state spin multiplet. That is, the state
must be identical to the state
, and its excitation energy must be zero (this was also observed before for the spin-1/2 dimer). In the presence of an Ising anisotropy, the state
is raised in energy with respect to the ground state, giving rise to what is known as zero-field splitting (ZFS) [
5,
29]. One thus arrives at the conclusion that, at zero temperature, the excitation spectrum consists of
excitations due to exchange splitting, plus one ZFS excitation due to anisotropy splitting. One can also infer that the energy of the lowest spin wave must be equal to or higher than the energy of the fully polarized states, since otherwise the fully polarized states would not be ground state, and the starting assumption for the validity of the method be violated. These rules are often useful in practice.
At this point, a further comment is appropriate. The physics which underlies the state
appears to be different from that of the other states
, in as much as the former corresponds to a ZFS, and is a component of the ground-state spin multiplet in the isotropic cluster. In fact, the
state plays a distinctive role, which in bounded spin clusters is associated with a number of subtleties which need careful attention [
18,
21]. It would thus be appropriate to exclude the
state from the spin wave spectrum [
18,
21]. However, in the present work, such details are irrelevant and the
state will also be called a spin wave.
2.3. General Concept
The observations for the dimer are generic, and valid for general clusters, consisting of arbitrary number of spin centers with arbitrary spin lengths (the exchange and anisotropy parameters are not arbitrary, they must result in a fully polarized ground state). The method presented previously is straightforward to extend.
The fully polarized state
is picked again. The local spin deviations are then defined as:
where the index runs from
. In this basis, the matrix elements
of the Hamiltonian Equation (1) are calculated to:
where
is the energy of the fully polarized state. This constant is irrelevant in the analysis of experiments, and often will be dropped (set to zero). The formula has an intuitive structure: the first three terms come from the
and
operators in the Heisenberg exchange and magnetic anisotropy terms, and thus contribute to the diagonal elements in the Hamiltonian matrix. The last term however comes from the
operators in the Heisenberg exchange term and thus produce non-diagonal elements. It is due to the latter that the local spin deviations
are mixed to form new states, the spin waves
. Diagonalization of the Hamiltonian matrix yields the eigenvalues and eigenstates, or spin wave energies and spin wave states,
and
, respectively,
where
represents the Hamiltonian matrix
, and
the eigenvector corresponding to the eigenvalue
(the eigenvectors can always be chosen to be real). With knowing the spin wave energies and states, experimental observables such as the neutron scattering intensity can be calculated and compared to experiment. It should be noted that this is achieved by diagonalizing a matrix of dimension
, which is generally massively smaller than the dimension of the Hilbert space of the spin system (even when symmetries are exploited). FCSWT can thus be handled easily with e.g., computer algebra systems, which makes it a useful and simple tool in practice.
As an example, for an isotropic, symmetric dimer with , one obtains from Equation (9): , , and . Diagonalizing this 2 × 2 matrix is a standard text book example, yielding the eigenvalues , , and the (un-normalized) eigenstates and . The two spin wave states are thus and , which correspond exactly to the states given in Equation (7). Note that these results apply to any value of .
The above procedure, with the key result Equation (9), establishes a straightforward recipe for calculating the observable zero-temperature excitations. It is, however, somewhat of a deus ex machina. It, for example, does not suggest an approach for calculating the remaining energies and states in the energy spectrum, nor higher temperature results. It also does not reveal further insight into the nature of the excitations. Physicists have thus developed more powerful techniques, which at first may appear abstract but are actually well suited for intuition.
One widely used technique is to associate the spins with bosons, or the spin lowering and raising operators
,
with boson creation and annihilation operators
,
, respectively [
2,
11]. This renders the spin problem into one of boson particles, which have some favorable properties. It comes at the cost of some approximations (and some subtleties). Nevertheless, the advantages are so numerous that it has become the standard approach in SWT. Gladly, for the topic of this work, this technique is again exact [
11,
30].
The boson-operator technique essentially replaces each spin center with a harmonic quantum oscillator, and the Heisenberg exchange describes a coupling of these oscillators. The spin Hamiltonian Equation (1) thus becomes a Hamiltonian of N coupled quantum oscillators, which is familiar in many branches of science, such as in the treatment of the elastic modes in molecules (vibrations) and solids (phonons).
The idea is to start, for a single spin, from its fully polarized alignment,
, and to treat the deflection of the spin by one step,
, as if it was due to the creation of one boson particle. In operator language (up to second order, which is sufficient for the purposes of this work), this leads to:
where
creates and
annihilates a boson at site
, and are defined as usually by the matrix elements
and
. Here,
denotes an eigenstate of the number operator
, and the boson number
can assume the values
. By applying these operators one, so to say, climbs up and down the ladder of states
,
,
, as sketched in
Figure 3.
Obviously, the creation of one boson at site
exactly produces a spin deflection at this site, or the state
, which implies:
In this picture, the fully polarized state
corresponds to a state with zero bosons, or the boson vacuum written as
, since in this state all spins are maximally aligned and spin deflections strictly absent. Exploiting Equation (11), the spin Hamiltonian Equation (1) can be reformulated in terms of the boson operators, which yields the boson Hamiltonian:
The matrix of the boson Hamiltonian in the space of the states or the one-boson sector, respectively, is easily calculated, and exactly reproduces the matrix given in Equation (9). The two techniques are thus indeed equivalent (with respect to the zero-temperature excitations).
The diagonalization of the boson Hamiltonian
can be accomplished by using the eigenvectors of the Hamiltonian matrix
for introducing new boson operators
,
, which shall be called magnon operators, and the associated particles as magnons:
The magnon Hamiltonian then assumes the simple form:
where the energies
are the eigenvalues obtained before in Equation (10). The eigenstates in the one-magnon sector, which are the zero-temperature spin wave excitations, are generated by:
So far, the previous results, obtained with using the spin lowering operator technique, have just been rediscovered. However, the physical picture has changed considerably, namely to a picture in which the Heisenberg interactions dress bare particles, leading to new particles, the magnons. There are different magnons, distinguished by the index , and each is associated with an excitation energy for its creation. More importantly, the Hamiltonian describing the magnons is exactly that of independent harmonic quantum oscillators, where each oscillator vibrates with frequency . This suggests to identify the zero-temperature excitations with the normal modes of the “magnetic harmonic oscillators”, or spin waves, and the magnons with their quanta. This also establishes a connection to the normal modes in elastic media, which is exploited in the next chapter.
2.4. Analogy to Vibrations and Pictorial Representation
In this chapter, we will present a graphical representation of the spin wave excitations, in order to reach a deeper insight into their nature, structure and properties. We will repeatedly stress the analogy with the elastic motion in molecules or vibrations, respectively. The analogy originates from the fact, that both systems are described by a set of coupled harmonic oscillators, and that thus mathematical and pictorial tools widely familiar for the elastic case can largely be carried over to the spin case.
The pictorial representation is based on the solutions of the classical version of the spin Hamiltonian Equation (1), which are essentially identical to the solutions of the quantum version discussed before [
2]. This is a well-known fact [
2,
31] (also well known for the elastic case), and roots in the equivalence of the classical and quantum equations of motion. The treatment is briefly reviewed here, for establishing the required results in a proper context. Only the isotropic case,
, is considered in the following for simplicity; the final results hold equally for the general case.
The equations of motions corresponding to the Hamiltonian Equation (1) read:
where
is the spin vector of the
-th spin,
an effective magnetic field, and the remaining symbols have the standard meaning. The right-hand side of the equation emphasizes the physical interpretation: The motion of
is determined by the torque on this spin due to an effective magnetic field generated by the exchange interaction with the other spins in the cluster. The FCSWT results are obtained by linearizing the equations of motions i.e., by assuming:
with constant projections on the
axis,
(the
axis is chosen for the direction of the spins in the fully polarized state). The equations of motions then lead to the eigenvalue equation:
with the matrix:
For the
-th normal mode, with eigenvalue
and eigenvector
(which can always be chosen real), the solution for the motion of an individual spin becomes:
where
and
are an arbitrary overall amplitude and phase, which in the following are set to
and
, without loss of generality.
The classical motion is described by a precession of each spin around the
axis, with a spin deflection given by the coefficient
in the eigenvector. Importantly, all spins
precess in phase (note that for negative
the spin appears to be out of phase by
π). The collective precession of the spins in a normal mode is thus characterized by different deflection angles for each spin, but equal phases for all spins. This is in difference to waves, or the solutions for extended systems, where the collective precession is characterized by equal deflection angles but different phases for the spin centers. It is, however, in close analogy to the elastic case, and we indeed recover the analogy of a celebrated result for molecular vibrations [
32,
33]:
All spins precess with identical frequency and all spins involved in the precession pass through, e.g., the x axis simultaneously.
This is a key finding of this chapter. The Hamilton matrix
in Equation (9) and the matrix
in Equation (20) are not identical, but are similar (in the mathematical sense). They thus exhibit identical eigenvalues, and the classical and quantum eigenvectors are related through:
where a factor
has been included for convenience, and
is an obvious normalization constant. The suggested procedure for obtaining the classical eigenvectors is to diagonalize the quantum Hamiltonian matrix Equation (9), and then to apply Equation (22).
Based on the classical solution we suggest two pictorial representations. The first consists of drawing the magnetic core of the molecule, and attaching a deflected arrow and a cone to each spin center, there the length of the cone corresponds to the spin length
and its radius to
. For larger molecules, this representation can become confusing. The second representation consists of drawing a planar schematic of the magnetic core, and attaching a circle or disk to each spin center, with a radius given by
, and an arrow to indicate the location of the tip of the precessing spin vector. In both representations attention is given to the phases of the spin precession by plotting the arrows accordingly, i.e., all spins with positive value of
are plotted in one direction, and those with negative
are plotted with opposite deflection. For the example of a dimer, the two spin wave states are shown for both representations in
Figure 4. It is noted that the latter representation could also be drawn without the circles or disks, i.e., with only the arrows representing the amplitudes of the spin deflections
, which would produce pictures in close analogy to those commonly used for representing molecular vibrations [
32,
33]. In the following, arrow heads are omitted for clarity.
The chapter is concluded with emphasizing that also the mathematical tools familiar for calculating and analyzing molecular vibrations can be carried over, essentially one-to-one, to the case of the spin wave excitations. This is especially true for the symmetry concepts, which is outlined exemplarily for a hypothetical spin cluster of four spin-5/2 spins centered around a spin-12 spin, see
Figure 5a. The presentation follows Reference [
23]. This five-spin model will also be of use in the discussion of the Mn
10 cluster in the next chapter. The spin Hamiltonian is given by:
The energies and eigenstates can be trivially calculated using Kambe’s approach [
29,
34], the emphasis here is however on the methods discussed in this paper.
The FCSWT procedure starts from the fully polarized state
with
. The basis states for the subspace with
are generated by applying Equation (8). This yields five states
, where
. The Hamiltonian matrix
is calculated with Equation (9) as:
where
.
This matrix can be analytically diagonalized by using its spatial symmetry properties [
33]. The point group of the model is
and the five-dimensional Hilbert space decomposes into the irreps
. Instead of using the symmetry group
, it is easier to view the spin cluster as a planar four-pointed star (
Figure 5a) and exploit its
symmetry (the technique is similar to the descent-in-symmetry technique in ligand-field theory [
33]). The
symmetry element implies a quantum number
κ with values
,
, and symmetry adapted linear combinations (SALCs)
. From the five basis states two symmetry adapted states with
are obtained, namely
and
. Since the symmetry group is
, the two symmetry adapted states generated for
are energetically degenerate and can be linearly combined. We select
and
. Finally,
. The following SALCs result:
which also transform according to the irreps of the
group. In this new basis the Hamiltonian matrix factorizes to:
It thus remains to diagonalize a block, which yields eigenstates with energy and , and eigenfunctions and , which belong to the irrep of the group. In addition, the energy spectrum exhibits a three-fold degenerate level at an excitation energy of , with eigenfunctions , which belong to the irrep .
In passing, it is noted that the
block could also have been block-diagonalized by exploiting properties deduced in
Section 2.1. The Hamiltonian Equation (23) is isotropic, and thus one of its eigenstates has to correspond to the
component of the fully polarized state
. Therefore, this eigenstate has the same energy as the ground state, i.e.,
, and the corresponding wave function can be obtained by applying the total spin lowering operator
to
, which yields
. Thus, one of the five eigenstates is already known, and can be used to block-diagonalize the sector transforming as the irrep
.
The resulting spin wave states are sketched in
Figure 5b. From the pictorial representation it is immediately obvious that the spin waves at energies of
and
indeed belong to the trivial irrep
. Also, they form a symmetric and anti-symmetric pair of states with respect to the spin precession of the central spin and the outer spins. In these two states, the collective spin precession is concentrated on the central spin, with weaker precession of the outer spins. In contrast, in the three
states the central spin does not contribute to the precession at all, i.e., here the excitation is completely localized on the outer spin centers. These features are related to non-trivial, observable effects in e.g., the neutron scattering intensity, as will be discussed further below.