Spinor Bose-Einstein gases

Spinor Bose-Einstein gases

G. Edward Marti and Dan M. Stamper-Kurn

In a spinor Bose-Einstein gas, the non-zero hyperfine spin of the gas becomes an accessible degree of freedom. At low temperature, such a gas shows both magnetic and superfluid order, and undergoes both density and spin dynamics. These lecture notes present a general overview of the properties of spinor Bose-Einstein gases. The notes are divided in five sections. In the first, we summarize basic properties of multi-component quantum fluids, focusing on the specific case of spinor Bose-Einstein gases and the role of rotational symmetry in defining their properties. Second, we consider the magnetic state of a spinor Bose-Einstein gas, highlighting effects of thermodynamics and Bose-Einstein statistics and also of spin-dependent interactions between atoms. In the third section, we discuss methods for measuring the properties of magnetically ordered quantum gases and present newly developed schemes for spin-dependent imaging. We then discuss the dynamics of spin mixing in which the spin composition of the gas evolves through the spin-dependent interactions within the gas. We discuss spin mixing first from a microscopic perspective, and then advance to discussing collective and beyond-mean-field dynamics. The fifth section reviews recent studies of the magnetic excitations of quantum-degenerate spinor Bose gases. We conclude with some perspectives on future directions for research.

Back in 1998, an “Enrico Fermi” summer school on Bose-Einstein Condensation in Atomic Gases was organized and held at Varenna. One of the topics touched lightly upon was the novelty of optically trapped Bose-Einstein condensates, which allowed one to study ultracold atoms occupying several spin states, including those that were not amenable to magnetic trapping. And one of the applications of that new capability was the creation of spinor Bose-Einstein condensates. Results from the first experiment on spinor Bose-Einstein gases [1] were presented by Wolfgang Ketterle in one short segment of a several lecture series [2]. In the intervening 16 years, the study of spinor Bose-Einstein gases has broadened to the point that the topic of just a few overhead transparencies grew into the focus of a three-lecture series given at the 2014 “Enrico Fermi” summer school on Quantum Matter at Ultralow Temperatures by one of the authors of the present manuscript.

The present document and the lectures they summarize are meant to share with newcomers to the cold-atom community some of the central concepts that spinor Bose-Einstein gases bring to the fore, and some of the experimental techniques by which these concepts have been, and continue to be, studied. A lack of depth in parts of our presentation does not reflect a shortage of interesting ideas and work by the community exploring these topics, but rather reflects our attempt to present material in an accessible, pedagogical style. We hope those interested will turn their attention to several review articles [3, 4, 5] and book chapters [6, 7, 8] that have been written about spinor Bose-Einstein gases. We have chosen to make two of the sections in these notes – Sec. LABEL:sec:imaging on the topic of imaging spinor gases, and Sec. LABEL:sec:excitations magnetic excitations – more technical and detailed than the others, for the reason that the material in those sections has not been covered deeply in previous reviews.


Before 1995, the list of quantum fluids that were available to study experimentally was short. Superfluid He is the oldest item on that list, first created when Kamerlingh Onnes liquified helium in 1908. However, at the time Onnes did not recognize that he had produced a new state of matter. It was not until 1937 that it was “discovered” that, at low temperatures, the He liquid was indeed a superfluid [9, 10]. This fluid is an example of a “scalar superfluid” because its order parameter is a single complex number: the ground state is described by a single complex number everywhere, while its dynamics are described by a complex number assigned to each point in space (a scalar field). The simplicity of this order parameter reflects the fact that He is a spinless atom: its electrons are paired into a spin singlet state, and so are the neutrons and protons within its nucleus.

Electronic superconductors are the second oldest on the list. These were discovered by Onnes. The simplest and best understood superconductors are described by the theory of Bardeen, Cooper and Schrieffer as having an s-wave order parameter. This order parameter can be understood as describing the center-of-mass wavefunction of pairs of electrons, the Cooper pairs, that have formed a spinless composite boson. So, again, the quantum fluid is described by a scalar complex field, though one that it is quite different from liquid He owing to the long-range Coulomb interaction among Cooper pairs. There are examples of superconductors in which the crystal structure environment of the electrons favors their adopting an electron-pair state of different symmetry. In p-wave superconductors, Cooper pairs are formed between electrons of like spin [11]. In this case, the electron pair carries both orbital and spin angular momentum. In d-wave superconductors, the Cooper pairs are spinless but have a non-zero orbital angular momentum. However, the possible states of both the p-wave and d-wave superconductors are highly constrained by the broken rotational symmetry of the crystal lattice. Thus very few – usually just one – of the conceivable angular momentum states of the Cooper pairs are accessible, and the order parameter is greatly simplified.

The last quantum fluid to be discovered before 1995 is superfluid He. The description of this superfluid is far from simple. Like the electrons in a superconductor, He atoms are fermions. At temperatures down to a few mK and pressures below 30 atmospheres, He forms a Fermi liquid. Like electrons in a metal, the distribution of helium atoms is characterized by a Fermi surface. Low-energy excitations involve the correlated motion of helium atoms near the Fermi surface and their neighbors. Below 3 mK, these quasiparticles undergo Cooper pairing. It is found that the short-range repulsion among helium atoms causes p-wave pairing to be favored, similar to the case of p-wave superconductors. However, unlike the electronic superconductor, here the helium fluid is free of the anisotropy of an underlying crystal. The physics that ensues is fascinating [12, 13, 14].

The realization of Bose-Einstein condensation in atomic gases has produced a panoply of new quantum fluids. Quantum fluids have already been formed of more than a dozen atomic species, in one-, two-, or many-electron atoms, some in their electronic ground state and others in metastable states. Molecular quantum gases have been produced by associating ultracold atoms into high lying molecular states, and efforts are underway to repeat the feat with molecules in their internal ground state as well. Some of these are Bose fluids while other are Cooper-paired Fermi gases; the distinction between these two fluids is blurred by experiments span the crossover from a Cooper-paired Fermi gas to a Bose fluid of bound molecules. The strength of interactions among atoms, and thereby of correlations among them, is broadly tunable [15, 16]. So, rather than studying only specific instances of broad theories of possible quantum fluids, we can now study many of them experimentally. The quantum fluids landscape is newly beckoning, inviting our exploration.

One of the appealing areas for investigation is the behavior that comes about when several quantum fluids are mixed together. Such a system will be represented by a multicomponent order parameter, for example, with one complex field for each component of the quantum fluids. In choosing which mixtures to study, it is clear that some are going to be more interesting than others. For example, if we immerse a superconductor in a cup of superfluid He, we have indeed mixed quantum fluids, but we haven’t produced something that seems very interesting; there is no motion or mixing between the fluids, and the only exchange is the (boring) thermodynamic exchange of energy. The same perhaps can be said of mixtures of He, He, and He that were considered previously [17, 18].

To refine our sense of when something interesting will arise, we might look for three things (these are personal preferences, not strict rules):

  1. Cases where the components of the quantum fluids can be transformed into one another, where we’re not considering apples and oranges (or He and He liquids), but rather situations where the experimentalist can drive transitions that convert one component of the quantum fluids into the other.

  2. Cases where the system is not “trivially” ordered, but rather where order is established from a set of degenerate or at least nearly degenerate possibilities. The degeneracy of the ground state can be brought about either by fine tuning, or else by symmetries, continuous or discrete, possessed by the system Hamiltonian. For a bulk fluid (outside of a crystal lattice), the symmetries most likely at play are spatial and/or spin-space rotations.

  3. Cases where the system can undergo dynamics that explore the different choices of order parameter implied above.

Spinor Bose-Einstein gases, which are ones comprised of atoms (and potentially, some day, molecules too) whose internal state is free to lie anywhere within a single manifold of angular momentum states, satisfy all three of these criteria. (1) Converting one component of the gas into another is accomplished, for example, by driving magnetic dipole transitions between Zeeman levels. (2) In the absence of conditions that explicitly break rotational symmetry (such as anisotropic trapping containers or applied fields), we are certain that any many-body state with non-zero angular momentum is perforce degenerate. A ground state with angular momentum zero would not necessarily be degenerate, but we assert that such a state (discussed in Sec. Spinor Bose-Einstein gases) would already be interesting enough to satisfy the “non-trivial” requirement. (3) As we will see, interactions quite generically allow for “spin-mixing” dynamics where the composition of the gas among accessible Zeeman states can vary.

We would be remiss if we did not mention another class of quantum fluids that is appearing on the landscape, namely an assortment of Bose-Einstein condensates of quasi-particle excitations in non-equilibrium settings. This class includes Bose-Einstein condensation of excitons, exciton-polaritons, photons, and magnons (probably others as well). These quasi-particle condensates occur in systems that are pumped with a population of excitations that is larger than would exist at thermal equilibrium. Under favorable conditions, this excess number of quasi-particles decays over a time that is sufficiently long that the quasi-particles can undergo enough collisions amongst themselves or with the medium in which they propagate in order to reach a near-equilibrium distribution in position and momentum space. There are fundamental questions about whether these driven and decaying quasi-particle gases can form quantum fluids that are the same as those achieved in thermal equilibrium systems in which the particle number is truly conserved. For lack of space, we do not elaborate on the properties of such quasi-particle quantum gases here and simply refer the interested reader to the literature for further reading [19].


Experiments on spinor Bose gases have taken place using either alkali atoms (notably sodium and rubidium) or high-spin atoms (notably the transition metal chromium). The selections are summarized in Table I. In all cases, the spin manifold that is occupied by the atoms of the spinor gas is one of the hyperfine spin manifolds of the electronic ground state.

Stable Unstable
conserved not conserved
Li, (f) Cr, (not f) Li,
Na, (af) Dy, (unknown) Na,
K, (f) Er, (unknown) K
Rb, (f) Rb
Rb, (af or cyc) Cs
Rb pseudo-spin: Tm, (unknown)
, Ho, (unknown)
Table I.: Experimental candidates for the study of ultracold spinor Bose gases. Species are divided according to whether they are stable at zero magnetic field (information on thulium and holmium is lacking), and whether the dipolar relaxation rate is small enough to allow the longitudinal magnetization () to be conserved in an experiment. The nature of the spin-dependent contact interactions is indicated in parentheses (f: ferromagnetic, af: antiferromagnetic, cyc:cyclic or tetrahedral). Stable pseudo-spin-1/2 gases of Rb are indicated, with states labeled with quantum numbers having the same low-field magnetic moment. Table taken from Ref. [5]

To identify atomic properties that come up later in this document, let us remind the reader briefly of the structure of an alkali atom. The alkali atom is a one-electron atom in the sense that, in the electronic ground state and accessible excited states, all but one of the electrons in the atom remain in closed atomic shells while only the state of the last electron varies. Focusing on the electronic ground state, that one electron is unpaired in an s-orbital, so that the atom has no internal orbital angular momentum. The atom does have two sources of spin angular momentum in this electronic ground state: the 1/2 spin of the unpaired electron, and the spin of the nucleus denoted by the angular momentum quantum number . These two spins are energetically coupled by the hyperfine interaction.

At zero magnetic field, the atomic energy eigenstates are also eigenstates of the total angular momentum of the atom, denoted by the vector operator and known as the hyperfine spin111for notational convenience, we will make the spin vector and spin quadrupole operators dimensionless, and include factors of explicitly where needed. The angular momentum quantum number for the hyperfine spin, , takes one of two values: . We observe that for bosonic alkali atoms is a half-integer number222We note that the quantum statistics of a neutral atom, i.e. whether it is a boson or fermion, depend on whether the number of neutrons in the nucleus is even (boson) or odd (fermion), because the sum of the number of electrons and protons is always even. The nucleus of a bosonic alkali atom will contain an even number of neutrons and an odd number of protons; hence the nuclear spin will be non-zero and half-integer. As a counter-example, for alkali-earth atoms, the bosonic isotopes have an even number of nucleons. Nuclear stability most often then implies that the nuclear spin is , so that the bosonic atom will have zero net spin in its ground state and will not be a candidate for spinor physics.. These two hyperfine spin states are separated by an energy on the order of GHz. Each of these energy states is -fold degenerate.

Thus, an alkali atom supports, in principle, two different species of spinor Bose-Einstein gases: an spinor gas, and an spinor gas. In reality, one finds often that the gas of atoms in the higher-energy hyperfine spin manifold is unstable against collisions that relax the hyperfine energy, limiting the amount of time one can allow such a gas to evolve; Rb is an exception in that both its hyperfine manifolds are long lived. The remaining choices are listed in Table I.

The identification of an atomic state with a hyperfine spin quantum number is strictly valid only at zero magnetic field. A magnetic field breaks rotation symmetry, and introduces coupling and energy repulsion between the two states of the same magnetic quantum number that characterizes the spin projection along the field axis. Strong mixing between states is obtained when the magnetic field is strong enough so that the magnetic energy, roughly , is greater than the zero-field hyperfine energy splitting. With MHz/G being the Bohr magneton ( is the electron charge, is the electron mass) and using a typical hyperfine energy, this strong-field regime applies for G. So, if we restrict the magnetic field to be less than 100 G or so in strength, we can still take the atomic states as comprising spin manifolds of well-defined spin.


More recent additions to the family of atoms that can be laser-cooled and brought to quantum degeneracy are elements in the transition metal and lanthanide series. These are many-electron atoms, with several electrons lying in the valence shell, and for which the optical spectrum is rich with different series of states corresponding to different arrangements of these many electrons. It came as a surprise that laser cooling and trapping techniques work so well for these complex atoms and their complex spectra [20, 21].

The valence shell for these atoms has a high angular momentum. When these shells are only partly filled, the electrons arrange themselves to maximize the total electron spin that can thus become large. The partly filled shells can also have a non-zero orbital angular momentum . Both and contribute substantially to the magnetic moment of the atom. Add in the possible contribution of nuclear spin, and one has the potential for creating spinor Bose-Einstein gases of very high spin .

The ground state angular momentum and the magnetic moment of these atoms can be determined by applying Hund’s rules for atomic structure, an exercise that is described in standard atomic and condensed-matter physics textbooks. For bosonic chromium [22], with the configuration , one finds the quantum numbers and magnetic moment . For bosonic dysprosium [23], with the configuration , one finds the quantum numbers and magnetic moment . For bosonic erbium [24], with two extra electrons in the orbital, we find and . The point here is not only that the spin quantum number of gases formed of these atoms is large, which, for example, leads to an interesting variety of magnetically ordered states that might be realized in such gases [25, 26, 27, 28, 29, for example], but also that the effects of dipolar interactions, scaling in strength as , become increasingly dominant.


One consequence of the large magnetic moment is that the high-spin gases undergo rapid dipolar relaxation whereas the alkali gases do not. Let us distinguish here between two types of collisions: s-wave contact collisions and dipolar relaxation collisions (see Ref. [30] for a nice review of quantum scattering in the context of ultracold atomic gases). The s-wave contact interaction describes low-energy collisions where atoms interact with one another via a short-range molecular potential, where “short-range” means that the range of the potential is small compared with the deBroglie wavelength defined by the small relative velocity of the incident colliding atoms. In an s-wave contact collision, two atoms enter into and emerge from the collision with no relative, center-of-mass orbital angular momentum. As we discuss further below, this situation requires that the total spin angular momentum of the two colliding bodies be conserved in the collision. In contrast, in a dipolar relaxation collision, the colliding atoms emerge with a non-zero value of the center-of-mass angular momentum. This change in orbital angular momentum implies a change also in the total hyperfine spin of the colliding pair.

In alkali spinor gases, dipolar relaxation collisions occur (typically) much less frequently than s-wave contact interactions. Consider the evolution of such a gas in the presence of a small, uniform applied magnetic field. Single-particle dynamics in the presence of this field conserve the populations in each of the Zeeman sublevels of the spinor gas. The s-wave contact interaction between two atoms conserves their total hyperfine spin, but not necessarily the spin of each atom. Generically, then, one finds that such collisions redistribute atoms among the Zeeman sublevels of the gas. However, the conservation of the total hyperfine spin implies that the net hyperfine spin along the field axis remains constant.

Remarkably, the conservation of the longitudinal (along the field axis) total hyperfine spin implies that, to first order, the Zeeman energy shifts imposed by applied magnetic fields are irrelevant to the evolution of the spinor gas. If the gas were completely free to change its spin composition, then the linear Zeeman energy from magnetic fields of typical strength – say 10’s of mG – would completely overwhelm all other energy scales of the gas (the temperature, the chemical potential, the spin-dependent s-wave interaction strength, the magnetic dipole-dipole interaction, etc.). The ground state of such a spinor gas would trivially be one in which the gas is fully magnetized along the magnetic field direction. However, absent dipolar relaxation collisions, this Zeeman-energy equilibration does not occur. Rather, the gas evolves as if these Zeeman energy shifts did not exist at all, i.e. as if it were in a zero-field environment (except for residual effects of the magnetic dipole interaction). The s-wave contact interactions then provide a means for the gas to evolve to an equilibrium constrained to a constant longitudinal magnetization, the value of which is established by the experimentalist by the application of radio or microwave frequency fields that can drive atoms between Zeeman states.

In contrast, in the high-spin spinor gases, dipolar relaxation collisions occur more rapidly, and the constraint of constant longitudinal magnetization is lifted. In an applied magnetic field of moderate strength, the high-spin spinor gas will indeed evolve toward the trivial magnetically ordered state where all atomic spins are aligned with the field. The resulting quantum fluid only occupies a single magnetic state (), and so the physics is described by only a single complex number, just like a scalar fluid. If one wants to access non-trivial magnetic order in such gases, one must reduce the applied magnetic field to very small values. For example, one may want to make the applied field smaller than the field generated by the magnetization of the gas in order to realize the chiral spin textures proposed in Refs. [31, 32]. We can estimate this field as where is the atomic density (say ) and is the atomic magnetic moment (say dysprosium’s ). We arrive at a field of . So experiments on these high-spin spinor gases require that the magnetic field be controlled with great care.

This distinction between “small” and “large” magnetic moments can be made quantitative by comparing two energies: the van der Waals energy, and the magnetic dipole energy. The van der Waals interaction leads to a potential of the form where is the distance between two atoms and quantifies the dc electric polarizability of the atoms. We can turn this potential into an energy scale by calculating the distance , known as the van der Waals length, where the confinement energy equals the magnitude of the van der Waals potential. We obtain , where is the atomic mass. We can now assess the magnetic dipole energy as the energy of two atomic dipoles separated by a distance ; this energy depends on the orientation of the two dipoles relative to the separation between them, but let’s just define the energy as being where is the vacuum permeability and is the atomic magnetic moment. So now the importance of the dipolar interactions in the collisions between atoms is quantified by the ratio


A similar ratio is obtained by taking the ratio of the mean-field contact interaction energy and the mean-field magnetic dipole energy of a spherical ball of atoms with uniform density . The mean-field contact interaction energy is given as , where is the s-wave scattering length, while the magnetic dipole energy is . So we obtain the ratio


Recall that the van der Waals potential plays a dominant role in low-energy atomic collisions, so that is indeed on the order of except near a collisional resonance [33]. Thus, our two expressions for are the same apart for a numerical factor of order unity.

On the right hand side of the expression above, we quantify the ratio with the magnetic moment being measured in units of the Bohr magneton, being measured in amu, and the being measured in units of the Bohr radius . With this expression we can compare the relevance of dipolar interactions in the collisions of various gases. For example, for Rb, with and , we have , confirming the fact that s-wave interactions dominate the collisional physics of alkali gases. In contrast, for Dy, with and [34], we have so that dipolar interactions are highly significant.


Spinor Bose-Einstein gases are not just multicomponent quantum fluids, but specifically ones in which the components of the fluid are simply different orientations of the same atomic spin. The spin operator is a vector operator, one that transforms as a vector under physical rotations. The implications of this statement for the properties of spinor gases are profound. Just as we find the notion of continuous symmetry groups powerful in explaining the nature of quantum field theories or condensed-matter systems, so too do we find the notion of rotational symmetry a powerful one in simplifying and constraining the physics of spinor gases.

Consider the constraints that rotational symmetry places on the possible interactions among atoms in the spinor gas. In the absence of applied fields, and in a spherically symmetric trapping volume, the interaction between two atoms must conserve their total angular momentum, the quantum number for which we will denote by the symbol . This total angular momentum includes both the internal angular momenta of the two atoms, and also the orbital angular momentum of their relative motion.

Let us assume that the collision between two particles occurs only in the s-wave channel, where the relative center-of-mass orbital angular momentum of the two atoms, with quantum number , is zero. This assumption is justified if we consider low-energy collisions between two particles with a short-range two-body potential. As discussed below, this assumption is questionable in the case of dipolar interactions.

Under this assumption, the total angular momentum of the incoming state is just the sum of the two atomic hyperfine spins, with total spin quantum number . As we are considering the collision between two identical bosons, exchange symmetry implies that the total angular momentum quantum number of the colliding pair must be an even integer; spin combinations with an odd-integer value of the total pair angular momentum simply do not collide in the s-wave regime. So the total atomic spin (in this case also the total atomic angular momentum ) can take different values: . The states of the colliding incident particles can thus be written in the following basis:


where is the projection of the total (dimensionless) angular momentum onto a selected axis.

The conservation of angular momentum under collisions now implies that the collisional interaction between two particles can be written in the form


where projects the two-atom state onto the subspace with total angular momentum , and is a scalar two-body operator.

The effects of interactions are now fairly well characterized. Consider collisions where the outgoing atom pair also has zero center-of-mass orbital angular momentum, and hence can be written in the same basis as above (denoted by primed symbols). Since is a scalar operator, and recalling the Wigner-Eckart theorem, non-zero matrix elements for the two-body interaction are only obtained when the outgoing total spin angular momentum is the same as the incident total spin angular momentum:


The simple dependence of the matrix element on and (diagonal, with only one overall constant ) again comes from the fact that is a scalar operator. The constants are determined from different scattering lengths, . To capture this part of the interaction, we can adopt the pseudo-potential approach and replace the collisional interaction with the operator


where is the interatomic separation. The factor of is inserted to account for double-counting in a summation over atom pairs.

It remains to consider the effects of dipolar interactions that can couple the incident s-wave collision channel to an outgoing channel with non-zero center-of-mass orbital angular momentum with total angular momentum quantum number (coupling to is forbidden by parity). In addition to the treatment above, we now have three more outgoing basis sets to consider:


Matrix elements to each of these final basis states are again independent of . The dipolar interaction does not conserve the hyperfine spin; it leads to a “spin-orbit” interaction through which spin angular momentum can be converted into orbital angular momentum, a phenomenon known as the Einstein-de Haas effect [35, 36, 25, 37]. This coupling also leads to dipolar relaxation among Zeeman levels, allowing the Zeeman energy of a spinor gas be exchanged with kinetic energy.

It turns out to be advisable to treat the magnetic dipolar interactions as an explicit long-range interaction, rather than adopting a pseudo-potential approach and replacing the dipolar interaction with a zero-range contact interaction [38]. The pseudo-potential method is appropriate for describing the non-dipolar molecular potential between two atoms, which falls off as at long distance. This rapid fall-off ensures that only s-wave interactions will persist at low energies of the incident particles. In contrast, the slower fall-off of the dipole-dipole interaction allows many partial waves to remain relevant to low-energy collisions.


Having established some of the basic properties of spinor Bose-Einstein gases, we ask how they behave at low temperature. Our expectations are set by the example of scalar Bose-Einstein gases: At low temperature, these gases undergo Bose-Einstein condensation, a phase transition driven by Bose-Einstein statistics. This phase transition is observable by the growth to a non-zero value of a local order parameter, the complex field which is the superfluid order parameter. This non-zero order parameter represents a spontaneous breaking of a symmetry obeyed by the system’s Hamiltonian. In this particular case, the broken symmetry is the gauge symmetry for which the conserved quantity is the particle number.

Following this example, we expect that order will be established in low-temperature spinor Bose-Einstein gases. We expect that this order may come about through the effects of Bose-Einstein statistics, and that it may represent a spontaneous breaking of symmetries of the many-body Hamiltonian, in particular the symmetry under rotation discussed in the previous sections.

We analyze the subject of magnetic ordering in spinor gases in three stages. For simplicity, each will focus on the spin-1 spinor Bose-Einstein gas. First, we present the simple case of a non-interacting spinor gas that equilibrates in the presence of an applied magnetic field. This case illustrates the statistical origin of magnetic ordering in a degenerate spinor Bose gas, and also exemplifies some basic features we should expect of the thermodynamic properties of a spinor gas. However, we will show that this non-interacting model provides little guidance as to the exact nature of magnetic ordering, particularly for highly degenerate gases at very low temperature. So, second, we discuss the role of those rotationally symmetric interactions outlined in the previous section. We analyze what type of magnetic order is favored under different conditions, often through the spontaneous breaking of rotational symmetry, and summarize experiments that have revealed such order and such symmetry breaking. Finally, we find an exact solution to the spin-dependent many-body Hamiltonian – exact at least under the assumption of there being no spatial dynamics and no dipolar interactions. We find that the ground state of one type of spin-1 spinor Bose-Einstein gas may be a highly correlated, non-condensed, rotationally symmetric spin state, providing a theoretical counter-example to our prediction of local mean-field ordering and spontaneous rotational symmetry breaking.


We consider a simple model of a non-interacting spinor Bose-Einstein gas in a uniform box of volume , and exposed to a uniform magnetic field [39]. We assume the gas comes to thermal equilibrium, constrained only by the constant particle number . The magnetic field shifts the single-atom energy levels according to the Hamiltonian


where is the atomic magnetic moment (assumed positive), and where we include only the linear Zeeman energy. This Zeeman energy causes the chemical potentials μ for atoms in the different magnetic sublevels to differ, i.e.


Let us pause and ask whether the assumptions of this model are appropriate. For high-spin atoms, rapid dipolar relaxation collisions allow the gas atoms to redistribute themselves among the Zeeman sublevels, and thus the model should apply directly. Experiments on the thermodynamics of condensed chromium atoms in moderate-strength magnetic fields are described well by our model (though with higher spin) [40].

For the low-spin, alkali metal spinor gases, the model we have presented is still valid so long as we regard the applied magnetic field as being an effective, fictitious magnetic field, the value of which is determined by the magnetization of the gas. In these gases, collisions among atoms will redistribute the kinetic (and trap potential) energy among them, but will not change the magnetization. The linear Zeeman energy determined by the applied magnetic field is inaccessible to the gas, and is thus irrelevant to its dynamics. That is, without dipolar interactions, the gas is subject to collisions of the form


only so long as . These processes evolve independent of the linear Zeeman shift, and thus such shifts cannot enter into the dynamics of a spinor gas. However, spin-mixing collisions do generally occur, allowing the Zeeman-state populations to redistribute and come to chemical equilibrium. For example, a spin mixing collision within a spin-1 gas can be viewed as the following reaction:


Chemical equilibrium for this reaction (i.e. where the forward and backward reaction rates are equal) implies the chemical potentials of the Zeeman sublevels are related as follows:


Rearranging this equation, and considering all other possible spin-mixing collisions (in case of a higher spin), we see that the chemical-potential difference is independent of . In other words, spin-mixing collisions in the absence of dipolar relaxation allow an equilibration at constant magnetization that is equivalent to the equilibration in an applied magnetic field, where now this effective magnetic field is not the actual magnetic field present in the laboratory, but rather a Lagrange multiplier that serves to enforce the constraint of constant magnetization.

We continue to explore our simple model. The number of non-condensed atoms in the gas in each of the magnetic sublevels is given by the Bose-Einstein distribution,


where , are fugacities and is the thermal deBroglie wavelength. The Bose-Einstein distribution is defined for fugacities being in the range , i.e. for chemical potentials .

We consider three generic scenarios. First, at high temperature, all atoms in the gas can be accommodated within the thermal Bose-Einstein distribution of the Zeeman components, i.e.


In this non-degenerate gas, the chemical potentials of all Zeeman components are negative. In an applied magnetic field (assumed positive, without loss of generality), the populations in the three magnetic sublevels of a spin-1 gas will differ, with the largest population being in the state (Fig. 1a). The non-degenerate gas thus acquires a non-zero magnetization.

Figure 1.: Composition of a non-interacting spinor Bose gas at thermal equilibrium. An applied magnetic field of strength separates the chemical potentials for the three Zeeman sublevels, labeled by magnetic quantum number , by an amount . (a) Above the critical temperature for Bose-Einstein condensation, the chemical potentials are all below zero, and the gas is non-degenerate. The number of atoms in each sublevel, shown in the bar chart, is below the critical number for Bose-Einstein condensation (dashed line). This gas is paramagnetic, with a finite magnetic susceptibility that lends it a magnetization under an applied magnetic field. (b) Upon cooling below the critical temperature for Bose-Einstein condensation, the chemical potential of the lowest-energy sublevel, here with , reaches zero, and a Bose-Einstein condensate forms in this sublevel. The chemical potentials for the remaining sublevels are negative, and the populations in these sublevels remain non-degenerate. Here, the gas is ferromagnetic, acquiring a large magnetization even in an infinitesimal magnetic field, implying an infinite magnetic susceptibility.

At lower temperature, the gas undergoes Bose-Einstein condensation, marked by the fact that the chemical potential of at least one of the magnetic sublevels goes to zero. In an applied, positive magnetic field, the linear Zeeman shift implies that because the chemical potential of the non-interacting Bose gas cannot be positive. Therefore, the remaining chemical potentials, and , are both negative. The number of non-condensed atoms saturates at the value


with . The Bose-Einstein condensation transition occurs when . The transition temperature, defined implicitly by the above equation, has the following limiting values


with being the Riemann zeta function.

Bose-Einstein condensation occurs in just one of the magnetic sublevels (Fig. 1b). The Bose-Einstein condensate, containing atoms, is completely spin polarized along the direction of the applied field, while the non-degenerate fraction of the gas is only partly polarized.

A third scenario to consider is Bose-Einstein condensation at zero magnetic field. In this case, the chemical potentials of all Zeeman sublevels go to zero simultaneously. The condensate number is defined as before. However, the state of the Bose-Einstein condensate is not determined within this model, requiring instead that we begin accounting for the effects of spin-dependent interactions.

In Fig. 2b, we sketch the magnetization of the Bose gas vs. applied field at several representative temperature. At temperatures above , the magnetization varies smoothly, with finite slope at , indicating that the gas is paramagnetic. For temperatures in the range , the onset of Bose-Einstein condensation at non-zero magnetic field, occurring when the paramagnetic gas is sufficiently magnetized so that the density of one of the Zeeman sublevels rises above the critical density for condensation, causes a discontinuity only in the second derivative of the magnetization.

Figure 2.: Magnetization of a non-interacting spinor Bose gas at thermal equilibrium in an applied field. (a) A contour plot shows the magnetization as a function of applied field strength and temperature. The Zeeman energy and temperature are shown scaled by the energy and temperature defined by the critical temperature for Bose-Einstein condensation at zero magnetic field, , defined in the text. The solid black line shows the Bose-Einstein condensation transition. The maximum transition temperature is indicated by the dashed black line. The dashed yellow line demarcates a line of constant thermal energy, while the dotted white line demarcates a line of constant entropy, both calculated using expressions in Eqs. Spinor Bose-Einstein gases and Spinor Bose-Einstein gases. (b) The magnetization vs. magnetic field is plotted at several representative temperatures, indicated by the arrows in (a). The lines here are dashed in regimes where the gas is non-degenerate, and solid where the gas is degenerate. At temperatures , the sudden jump in magnetization at zero field indicates the gas is ferromagnetic.

At temperatures below , the magnetization is discontinuous at zero magnetic field. The magnetic susceptibility for this gas (slope of magnetization vs. field at zero field) is infinite. The spinor Bose-Einstein gas in this regime is thus ferromagnetic, even in the absence of any interaction among the atomic spins. Magnetic ordering in this case can be thought of as a parasitic phenomenon, taking advantage of the fact that Bose-Einstein condensation creates a fraction of the gas with no entropy.

Our model also provides an approximation for the thermodynamic properties of the spinor Bose-Einstein gas. We can use ideal-gas formulae to calculate the kinetic energy and entropy within each Zeeman sublevel of the gas:


Lines of constant energy and entropy determined by these formulae are shown on Fig. 2a. In the quantum-degenerate regime, we see that, at constant temperature, the energy and entropy of the gas decrease with the magnitude of the applied magnetic field. As discussed above, there is one Zeeman sublevel for which the chemical potential is zero, and thus the normal fraction of atoms within this sublevel is saturated at an energy and entropy that is determined solely by the temperature. Therefore, as the magnitude of the magnetic field is increased, the populations of the remaining magnetic sublevels decrease, decreasing the amount of thermal energy and entropy contained in these populations while increasing the number of condensed atoms.

This thermodynamic tendency is at the root of demagnetization cooling in a spinor Bose gas, which has been demonstrated both for alkali [41] and for high-spin spinor gases [42, 43]. In the work with alkali gases, a fully longitudinally magnetized spinor gas was Bose-Einstein condensed in an optical trap. Within the model discussed here, the full magnetization of the gas is equivalently described by the gas’ coming to free equilibrium (without the magnetization constraint for alkali spinor gases) at an effective magnetic field . The magnetization of the gas was then deliberately lowered, simply by rotating the atomic spin slightly from the longitudinal direction, using rf fields so that the spin rotation was spatially uniform. The gas was then allowed to come to thermal equilibrium under the constraint of constant magnetization. Owing to the rotational symmetry of the s-wave contact interactions, the spin rotation pulse did not change the internal energy of the gas. Of course, it did change the Zeeman energy of the gas greatly, but, as stated previously, the alkali gas is impervious to such energy changes. Thereafter, the gas was allowed to equilibrate at constant magnetization. The equilibration proceeds roughly at constant energy, assuming that the optically trapped gas is completely isolated from its surroundings. As such, the gas undergoes isoenergetic demagnetization, following the yellow dashed line in Fig. 2a to an equilibrium at a lower value of . As shown in the figure, following this line implies that demagnetization (of the coherent transverse magnetization produced by the spin-rotation pulse) leads to cooling. Such cooling was indeed observed and used to drive down both the temperature and, upon ejecting the spin-flipped atoms, the entropy per particle of the spinor gas.

In the experiments on high-spin gases, demagnetization cooling is amplified by the fact that the gas can access the Zeeman energy. A spin polarized gas was prepared with its magnetic moment oriented along an applied magnetic field. Upon reducing the strength of the magnetic field, atoms were found to populate the magnetic sublevels with higher Zeeman energy, reaching those levels by converting thermal kinetic energy into Zeeman energy in a dipolar relaxation collision. In the recent work of Ref. [43], performed with a Bose-Einstein condensed chromium gas, both these demagnetization cooling effects contributed, i.e. both the conversion of kinetic energy to Zeeman energy and also the cooling effect of allowing magnetic excitations into the previously longitudinally polarized gas led to a reduction in temperature and entropy.


The above model emphasized the role of Bose-Einstein statistics in causing a spinor Bose-Einstein gas to become magnetically ordered. Now we turn to the role of interactions in determining what type of magnetic order will emerge in conditions that are poorly described by our simple non-interacting-atom model, namely under the application of very weak magnetic fields, or, equivalently, in gases that are only weakly magnetized in the longitudinal direction.

We consider the effects of spin-dependent s-wave interactions, which we characterized in Eq. Spinor Bose-Einstein gases in terms of projection operators, , onto two-particle total spin states. It is helpful to rewrite these projection operators onto more familiar forms of spin-spin interactions. These interaction terms should retain the rotational symmetry of the interaction.

For the case of spin-1 gases, we find that two rotationally symmetric terms suffice to describe the s-wave interaction. Considering operators that act on the states of two particles, labelled by indices and , we note the following identities:


Here, the subscript “sym” reminds us that we should include only states that are symmetric under the exchange of the two particles. We can then replace the spin-dependent interaction of Eq. Spinor Bose-Einstein gases with the following expression:


where the coefficients and are defined as


The spin-dependent interactions are thus distilled to a form of isotropic Heisenberg interactions (). The sign of the interaction divides spin-1 spinor gases into two types. For , the spin-dependent interaction is ferromagnetic, favoring a state where atomic spins are all maximally magnetized, i.e. the maximum spin-projection eigenstate along some axis . For , the spin-dependent interaction is anti-ferromagnetic. We recall that in solid-state magnets, anti-ferromagnetic interactions between neighbors in a crystal lead to a variety of interesting magnetic states, such as Néel-ordered anti-ferromagnets, valence-bond solids, and spin liquids, the selection among which requires careful examination of the specific geometry and connectivity of the lattice. Similarly, in anti-ferromagnetic spin-1 spinor gases, identifying the specific nature of magnetic ordering requires some careful thinking.

It may be useful here to remind the reader that the spin state of a spin-1 particle is characterized, at the one-body level, by a 3 3 density matrix. Such a Hermitian matrix is defined by nine real quantities: three real diagonal matrix elements plus six real numbers that define the three independent off-diagonal complex matrix elements. Corresponding to each of these real numbers, there is a Hermitian matrix that defines an observable quantity. Thus, the space of one-body observables for a spin-1 particle is spanned by nine basis operators. We may select these operators to be the identity operator, the three vector spin operators (denoted by symbols ), and the five spin quadrupole operators (denoted by symbols ). In the basis of eigenstates, these operators can be written in the following matrix form:


All nine operators (except the identity operator) are traceless. Except for , the traceless operators also have zero determinant and the same eigenvalues: .

With these operators in mind, let us express the operator , written in Eq. Spinor Bose-Einstein gases, in terms of Bose field creation and annihilation operators which are written in the basis of the projection of the atomic spin along some axis (owing to the symmetry of the spin-dependent interaction, it does not matter which axis we choose). We obtain the following,


where, noting the delta-function spatial dependence, we have dropped the explicit dependence on position which is the same for all field operators in the integrand.


One common approach to identifying the state of the degenerate gas is to assume that a Bose-Einstein condensate forms in a coherent state, i.e. a many-body state that is a product state of identical single-particle states. Let us simplify even further by stipulating that the single-particle state is a product state of a spatial wavefunction (normalized to unity) and a spin wavefunction , and consider only the lowest-energy choice for the spin wavefunction. This is a variational, mean-field, and single-spatial-mode approximation.

We are left to minimize the following energy functional:


where is our variational wavefunction, includes the kinetic and potential energy terms. Let us assume that is spin independent. Next, we assume that the spatial wavefunction is fixed, and that only the spin wavefunction is to be varied. With these assumptions, the only term remaining to consider is the spin-dependent energy, which we conveniently normalize by the number of atoms in the condensate:


Here, is the particle number, is the average density, and .

This mean-field energy functional has two extremal values. One extremum is obtained for the state that maximizes the length of the average spin vector, i.e. the state with the maximum projection of its spin along some direction , giving . This state minimizes the energy for gases with ferromagnetic interactions, with . The state is maximally magnetized, with defining the direction of the magnetization (or the opposite of that direction, depending on the sign of the gyromagnetic ratio). In our single-mode approximation, the spinor part of the wavefunction is constant in space. So a gas in the state is “ferromagnetic” in the sense that, like in a solid-state ferromagnet, it has a macroscopic magnetization that has a constant direction across the entire volume of the material.

A second extremum in the mean-field energy function occurs for the state that minimizes the length of the average spin vector. This is the state , which is the eigenstate for the spin projection along the axis with eigenvalue zero. The average spin in directions orthogonal to is also zero, so one obtains . This state then minimizes the energy for gases with antiferromagnetic interactions, with . We can regard this state as being nematic, in that it is aligned with an axis (which we may call the director), but that it is oriented neither in the or directions. A condensate with a spatially uniform nematic spin wavefunction is “nematically ordered” in that the director is constant in space. The is also termed polar because of the fact that rotating the state by radians about an axis transverse to produces a final state that still is a zero-valued eigenstate of the operator ; however, the rotation results in the final state differing from the initial state by a minus sign. This property is exhibited also by the p electronic orbital of atomic and chemical physics, in which the electron density is symmetric under a rotation about, say, the axis, but where the phase of the electronic wavefunction is antisymmetric under such rotation.


In Sec. Spinor Bose-Einstein gases, we discussed the thermodynamics of a non-interacting spinor gas in an applied magnetic field, or, equivalently, under the constraint of constant longitudinal magnetization. Let us extend this consideration also to the interacting spinor gas in order to understand how the application of symmetry breaking fields tears at the rotationally symmetric interaction Hamiltonian and its solutions.

We consider two symmetry-breaking external influences. The first is a linear Zeeman shift, or equivalently, as discussed before, a constraint of constant longitudinal magnetization. Keeping with nomenclature in early papers on spinor Bose-Einstein condensates, we account for this energy through the term . In early experiments on spinor Bose-Einstein condensates, a magnetic field gradient was applied deliberately to the gas to provide clear experimental signatures of the spin-dependent interactions [1]. Such a gradient can be accommodated by making spatially varying; we revisit this point in Sec. Spinor Bose-Einstein gases.

Secondly, we include also a quadratic Zeeman energy. As mentioned earlier, the spinor gas can evolve by the spin-mixing collision. This process is insensitive to the linear Zeeman energy, but depends on the quadratic Zeeman shift, which shifts the average energy of the states (Zeeman energy of the right side of Eq. Spinor Bose-Einstein gases) by an energy with respect to that of the state (Zeeman energy of the left side of Eq. Spinor Bose-Einstein gases). The quadratic Zeeman shift comes about from the fact than an applied magnetic field mixes hyperfine levels, causing energy repulsion between states in, say, the upper and lower hyperfine states with the same value of in the electronic ground state of alkali gases. A quadratic Zeeman energy can also be imposed by driving the atoms with microwave-frequency magnetic fields that are near resonant with hyperfine transitions [44]. The latter approach allows one to reverse the sign of the quadratic Zeeman energy (microwave-dressed hyperfine levels are made to move closer, rather than farther, in energy). For simplicity, and to match with most experiments, we consider quadratic Zeeman energy shifts in the same basis as that determined by the linear Zeeman energy, i.e. through the term . As such, the Zeeman energies still preserve the symmetry of rotations about the longitudinal direction (the axis). More generally, the quadratic and linear Zeeman energies could, in principle, be applied along different axes, breaking rotational symmetry altogether.

We now obtain a spin-dependent energy functional of the form


Expressing the spin wavefunction in the eigenbasis, the energy functional has the form


This expression, while correct, hides from us the fact that some of the information in the spin wavefunction is irrelevant for calculating the energy. For one, the spinor is normalized to unity, so we need not keep track of all the populations . We’ll keep track just of two, choosing the quantities


Here, is the longitudinal magnetization of the gas, normalized to lie in the range , and is the fractional population in the state. Under a constraint of constant longitudinal magnetization, is constrained to vary in the range .

In addition, the wavefunction contains three complex phases, for example, the complex arguments of the three probability amplitudes . However, two linear combinations of these phases do not affect the energy. Clearly, the energy is unaffected by multiplying the wavefunction by a global phase. Thus, the combination is unimportant to us. Also, the mean-field energy functional is invariant under rotations of the spin about the axis. Such a rotation would change the value of the phase difference , but would not change the energy. The only linear combination of phases that is still potentially important is the combination


In our later discussion of spin-mixing dynamics, we will see that it is this phase combination that controls the direction in which populations will flow in a spin-mixing collision (Eq. Spinor Bose-Einstein gases).

We now rewrite the spin-dependent energy as


The task of minimizing this energy is now much clearer. The phase controls the degree of magnetization present when the spin wavefunction contains a superposition of states with . We choose for ferromagnetic interactions () to maximize that magnetization, and for antiferromagnetic interactions () in order to minimize that magnetization. All that is left is to minimize the energy with respect to the populations in the spin wavefunction, expressed through and .

Such minimization was performed initially in Ref. [1]. The results are presented in Fig. 3. Focusing first on the non-interacting case, the phase diagram is simple: For , the lowest energy state is either of the longitudinally magnetized states, , depending on the sign of . The situation here matches that of the non-interacting gas considered in Sec. Spinor Bose-Einstein gases. For , the quadratic Zeeman shift favors the state where .

Figure 3.: Mean-field ground states predicted for an spinor gas with spin-dependent s-wave interactions. The energy functional of Eqs. 31 and 35 is considered. Here (linear Zeeman energy, or Lagrange multiplier for magnetization constraint) and (quadratic Zeeman energy) are measured in units of the interaction energy . Diagrams are shown for (a) antiferromagnetic interactions, (b) neglecting the effects of interaction, and (c) ferromagnetic interactions. Gray regions in (a) and (c) indicate the ground state is a mixture of Zeeman sublevels. Solid lines indicate discontinuous phase changes, while dashed lines indicate continuous changes. Also shown are the mean-field energies of three extremal states – the longitudinally magnetized state, the longitudinal polar state (, for which the nematic director is the axis) and the transverse polar state (for example , for which the nematic director is the axis) – for either (d) antiferromagnetic or (e) ferromagnetic interactions, where we set . In the antiferromagnetic case, the polar state is always the lowest energy state, changing its alignment at . For the ferromagnetic case, the longitudinal polar state is the ground state for , while the longitudinally magnetized state is the ground state for . Between these regions, a state of non-zero transverse magnetization is favored (energy given by black line), with spin wavefunction given in Eq. 36.

The spin-dependent s-wave interaction modifies this phase diagram. We observe that, for example, the region in the - plane occupied by the ground state (this is the longitudinal polar state, with the nematic vector aligned with the longitudinal axis) is increased in the case of antiferromagnetic interactions, and decreased in the case of ferromagnetic interactions.

The diagram also shows evidence of the miscibility and immiscibility of different magnetic sublevels. The miscibility of two Bose-Einstein condensed components and of a gas (with equal masses, as here) is governed by the relation between the s-wave scattering lengths, , and where the indices indicate the two components colliding. If , then the components are miscible in that a superposition of the two components has less interaction energy than two phase-separated domains of the two components. The s-wave scattering lengths for different collisions among atoms are all determined by the values of and , as shown in Table II.

Table II.: Interaction strengths that characterize binary s-wave collisions among atoms in the manifold. The interaction strength is characterized both in terms of the scalar and Heisenberg-like interaction strengths, and respectively, that are defined in Eqs. Spinor Bose-Einstein gases, and also in terms of the scattering lengths that describe s-wave collisions between atoms with total angular momentum and .

We see that for antiferromagnetic interactions, the and components are immiscible. This fact is reflected in the ground-state phase diagram (Fig. 3a) by the sharp boundary between the regions in which either of these components is the ground state. In contrast, in the case of ferromagnetic interactions, the transition between these two regions is gradual. The gray region between them in Fig. 3c indicates that the mean-field ground state is a superposition of states including the and state; such a mixture allows the gas to be partly magnetized and thus take advantage of the spin-dependent interaction. One also sees that, for antiferromagnetic interactions, the and states are miscible. This miscibility is indicated in Fig. 3a by the gray region that lies between the and ground state regions at , in which a superposition of these two states reduces the mean-field energy. Indeed, for and , the minimum energy state for antiferromagnetic interactions is the transversely aligned polar state, i.e. the state for which the nematic director lies in the plane transverse to (in the - plane). Such a state is a superposition of the states with equal population in the two states.


Compelling experimental evidence for the realization of such mean-field ground states for spinor gases has been obtained. The ferromagnetic case is exemplified by the spinor gas of Rb. The first experiments on this system were performed by the groups of Chapman [45] and Sengstock [46]. They examined optically trapped, Bose-Einstein condensed gases that were prepared with zero longitudinal magnetization (i.e. with ), by preparing all atoms in the state. Then, the gases were allowed to evolve under an applied, uniform magnetic field, which produced an adjustable, positive value of .

At high , the state being lower than energy than the average energy of the states that would be produced by spin mixing, we expect the state to be the preferred ground state of the atoms. At low , one expects the gas to acquire magnetization. Consistent with the initial magnetization of the gas, one expects the gas to become transversely magnetized, described by the single-atom spin wavefunction


over the range . Here, the transverse magnetization has the amplitude , i.e. ranging from zero magnetization at the transition from the paramagnetic to the ferromagnetic phase () to full magnetization at where the spin-dependent interactions are the only spin-dependent energy term. In the experiment, the Zeeman-state populations of the gas were measured, and indeed one found the expected ratio of populations in the states.

The Zeeman state populations on their own do not confirm that the gas has adopted a state with nonzero magnetization; for this, one needs to establish that the Zeeman populations are coherent with one another, and that the phase relations among them lead to a non-zero magnetic moment. For example, compare the two spin wavefunctions


The first of these, , is the state magnetized along the axis. The second of these, , differing from the first only in a phase factor on one of the probability amplitudes, is a polar state with zero magnetization. Without measurements of the relative phase between Zeeman-state probability amplitudes, one cannot discern these states.

Two separate experiments confirmed that these relative phases are indeed those that maximize the magnetization in the case of ferromagnetic interactions. The Chapman group utilized spin-mixing dynamics, which we discuss in Sec. Spinor Bose-Einstein gases, to cause the magnetic state of the gas to reveal itself [47]. After the gas had come presumably to equilibrium, the experimenters applied a briefly pulsed quadratic Zeeman energy shift. This pulse knocked the state away from its equilibrium state and initiated a tell-tale oscillation of the Zeeman-state populations, the dynamics of which were in excellent agreement with numerical calculations and indicative of the initial state being ferromagnetic as expected.

More direct evidence of the ground-state structure of the Rb gas comes from direct measurements of the gas magnetization. The imaging methods used to measure such magnetization are discussed in Sec. LABEL:sec:imaging. In the work of Ref. [48], such measurements confirmed that a ferromagnetic spinor gas prepared with no initial spin coherence whatsoever will, upon being cooled to quantum degeneracy and allowed to evolve for long times, become spontaneously magnetized. For a positive quadratic Zeeman energy (), the magnetization was found to lie in the transverse plane, while for (realized by the application of microwave fields) the magnetization tended to orient along the magnetic field axis (Fig. 4).

Figure 4.: Direct imaging of the formation of ferromagnetic order in a Rb spinor gas. An unpolarized gas was prepared above the Bose-Einstein condensation transition temperature. The gas was then cooled by evaporation, and allowed to equilibrate, for the time indicated on the figure, before its vector column magnetization was measured by magnetization-sensitive phase-contrast imaging. The magnetization for each single sample probed is represented in two images stacked vertically; the top shows the transverse magnetization and the bottom the longitudinal magnetization. The color scheme used for all these images is shown. The gas is shown to develop ferromagnetically ordered regions of increasing size over time. The magnetization realized with is isotropically distributed. The magnetization realized for tends to lie in the transverse plane, while for it tends to point longitudinally. Figure adapted from Ref. [48].

The case of antiferromagnetic interactions is exemplified by the spinor condensate of sodium. The first experiments on spinor condensates, performed with such a gas, confirmed the increased energetic stability of the polar state by placing the trapped gas in a magnetic field gradient [1]. Under such a gradient, a gas within a net zero longitudinal magnetization would tend to become polarized, lowering its energy by adopting the strong-field seeking state ( in this case) at the high-field end of the trap, and the weak-field seeking state () at the low-field end. However, such an arrangement maximizes the spin-dependent interaction of the antiferromagnetic spinor gas. A lower energy configuration is achieved by reducing the magnetization of the gas at the cloud center. Depending on the sign of the quadratic Zeeman shift, the favored non-magnetic state is a polar state () with the director pointing either transversely (for ) or axially (). Measurements of the one-dimensional spin-state distribution of the gas (along the direction of the field gradient) showed the expected magnetization pattern, in quantitative agreement with expectations.


It is clear, however, that these mean-field product states cannot be the true ground state of the interacting spinor gas. Consider specifically the case of antiferromagnetic interactions, the ones we suspect, based on solid-state magnetism, are going to be the most interesting. Mean-field theory predicts the state with all atoms in the Zeeman state to be a ground state; for a gas of atoms, we may write this state in the Fock-state basis as where we enumerate the atoms in the states sequentially. However, the many-body Hamiltonian contains a term describing spin-mixing collisions, meaning that this mean-field state is coupled, for example, to the orthogonal state . The true ground-state of the system must contain a superposition of (at least) these two many-body states.

A similar scenario applies in the interacting scalar Bose gas. The mean-field ground state for a condensate in a uniform container is one with all atoms in the zero-momentum state. However, the Hamiltonian includes terms describing elastic collisions that conserve the total momentum of the colliding atom pair. One such term describes a collision in which two atoms are excited out of the zero-momentum state and promoted to states of non-zero counter-propagating momenta . Thus, the true many-body ground state must include also a population of atoms in non-zero momentum states. Yet, the number of such momentum-excited atom pairs remains small owing to the energy penalty for creating such excitations. The total number of excited pairs comprises the quantum depletion, which remains small in the case of weak interactions [49].

Is there a similar energy penalty for spin-mixing collisions, which might suppress the population of atoms outside the spin state and constrain the quantum depletion atop the mean-field state? Consider the state generated from our initial mean-field ansatz by a slight geometric rotation, i.e. all atoms in the polar state that is aligned with an axis just slightly tilted from the axis. Such a state also contains a superposition of the and states, and yet the energy of this rotated state is no higher than the initial state. Thus we suspect, without rigorous proof yet, that the spin-mixing interaction will mix new spin configurations into the mean-field initial state without an energy penalty, different from the situation with a weakly interacting scalar Bose gas. If this suspicion is correct, then the quantum depletion of the mean-field state will be massive, and mean-field state would no longer be a good approximation to the many-body ground state.

This vague argument was cast more rigorously and elegantly by Law, Pu and Bigelow [50]. One realizes that the exact many-body Hamiltonian, containing a sum of spin-dependent interactions between all pairs of atoms in the gas, treats all atoms symmetrically. We might expect, therefore, that the Hamiltonian can be written not just in the basis of spin operators acting on each individual atom within the gas, but also in terms of collective spin operators that describe the joint spin state of all atoms in the gas. This expectation turns out to be correct. Defining the collective spin operator as , we find the spin-dependent many-body Hamiltonian for a spinor gas of atoms, each assumed to be in the same state of motion, can be written as


where is the average density of the gas.

In the case of antiferromagnetic interactions (), the energy is minimized by minimizing . Such minimization is achieved by the state of zero collective spin . This state can be pictured as being composed of pairs of atoms (letting be even), each of which is in the spin-singlet state, written for a pair of particles as


using the Fock-state basis introduced above. The many-body state is a product state of such pairs, fully symmetrized under particle exchange.

This many-body state has several distinct features. First, it is a state that does not break the rotational symmetry of the Hamiltonian. Indeed, one may write this state as the superposition of mean-field polar states, integrated over all alignment axes [51]. Second, this ground state is unique up to an overall phase factor. Third, the state is not a standard Bose-Einstein condensate, which would be defined, according to the criterion of Penrose and Onsager, as a state for which there is just a single, single-particle state that is macroscopically occupied [52]. Finally, this state has measurable features that distinguish it from the mean-field ground state. Consider that for an -atom mean-field polar ground state, i.e. all atoms in the