Itinerant magnetism in spin-orbit coupled Bose gases

0.68 MB·English

by  D. L. Campbell

Itinerant magnetism in spin-orbit coupled Bose gases
D.L.Campbell,R.M.Price,A.Putra,A.Valdés-Curiel,D.Trypogeorgos,andI.B.Spielman
1Joint Quantum Institute, University of Maryland, College Park, Maryland, 20742, USA
5
1
0
2 Phases of matter are conventionally characterized by order parameters describ-

n
a
J ing the type and degree of order in a system. For example, crystals consist of

3
2
spatially ordered arrays of atoms, an order that is lost as the crystal melts. Like-

]
s
a wise in ferromagnets, the magnetic moments of the constituent particles align
g
-
t only below the Curie temperature, T . These two examples reflect two classes of
n C
a
u
phase transitions: the melting of a crystal is a first-order phase transition (the
q
.
t
a
crystalline order vanishes abruptly) and the onset of magnetism is a second-
m
-
d order phase transition (the magnetization increases continuously from zero as
n
o
c the temperature falls below T ). Such magnetism is robust in systems with lo-
C
[

1 calized magnetic particles, and yet rare in model itinerant systems1,2 where the
v
4
8 particles are free to move about. Here for the first time, we explore the itiner-
9
5
0 ant magnetic phases present in a spin-1 spin-orbit coupled atomic Bose gas3–6;
.
1
0
in this system, itinerant ferromagnetic order is stabilized by the spin-orbit cou-
5
1
:
v pling, vanishing in its absence. We first located a second-order phase transition
i
X
r that continuously stiffens until, at a tricritical point, it transforms into a first-
a
order transition (with observed width as small as h×4 Hz). We then studied
the long-lived metastable states associated with the first-order transition. These
1
measurements are all in agreement with theory.
Most magnetic systems are composed of localized particles such as electrons7, atomic
nuclei8, and ultracold atoms in optical lattices9–11, each with a magnetic moment µ. By
contrast, itinerant magnetism1 describes systems where the magnetic particles – here ultra-
cold neutral atoms – can themselves move freely, and for which magnetism is generally weak.
Our spin-orbit coupled Bose-Einstein condensates (BECs) constitute a magnetically ordered
itinerant system in which the atoms’ kinetic energy explicitly drives a phase transition be-
tween two different ordered phases4. The coupling between spin and momentum afforded by
spin-orbit coupling (SOC) provides a new route for stabilizing ferromagnetism in itinerant
systems, both for fermions12 and here for bosons5,6.
PhasetransitionscangenerallybedescribedintermsofafreeenergyG(M )–including
z
the total internal energy along with thermodynamic contributions – that is minimized for
an equilibrium system. Here the magnetization M = hSˆ i/(cid:126) is an order parameter which
z z
changes abruptly as our system undergoes a phase transition, where Sˆ is the spin. Figure 1c
showstypicalfreeenergies: afirst-orderphasetransition(toppanel)occurswhenthenumber
of local minima in G(M ) stays fixed, but the identity of the global minima changes; and a
z
second-order phase transition (bottom panel) occurs when degenerate global minima merge
or separate. These defining features are true both for T > 0 thermal and T = 0 quantum
phase transitions. Here we realize f = 1 BECs with SOC and study the magnetic order and
associated quantum phase transitions present in their phase diagram.
2
For spin-1/2 systems (i.e, total angular momentum, f = 1/2) like electrons, ferromag-
netic order can be represented in terms of a magnetization vector M = hSˆi/(cid:126). This is rooted
in the fact that the three components of the spin operator Sˆ transform vectorially under
rotation. More specifically, any Hamiltonian describing a two level system may be expressed
as H = (cid:126)Ω +Ω ·Sˆ, the sum of a scalar (rank-0 tensor) and a vector (rank-1 tensor) contri-
0 1
bution. The former, described by Ω gives an overall energy shift, and the latter takes the
0
form of the linear Zeeman effect from an effective magnetic field proportional to Ω . Going
1
beyond this, fully representing a spin-1 (i.e., total angular momentum f = 1) Hamiltonian
with angular momentum Fˆ rather than spin Sˆ, requires an additional five-component rank-2
tensor operator – the quadrupole tensor – and therefore there exist “magnetization” order
parameters that are not simply associated with any spatial direction13–15.
PioneeringstudiesinGaAsquantumwells16,17 showedthatmaterialsystemswithequal
contributions of Rashba and Dresselhaus SOC described by the term 2(cid:126)k k Fˆ /m, subject
R x z
ˆ
to a transverse magnetic field with Zeeman term Ω F , can equivalently be described as a
1 x
spatiallyperiodiceffectivemagneticfield. Ourspin-orbitcoupledspin-1atomicsystems3,18–24
have this form of SOC and can therefore be described by the magnetic Hamiltonian
(cid:126)2k2
Hˆ = +Ω (x)·Fˆ +Ω Fˆ(2), (1)
2m 1 2 zz
describing atoms with mass m and momentum (cid:126)k interacting with an effective Zeeman
magnetic field Ω (x)/Ω = cos(2k x)e −sin(2k x)e helically precessing in the e -e plane
1 1 R x R y x y
with spatial period π/k set by the SOC strength; and an additional Zeeman-like tensor
R
coupling with strength Ω . Here, Fˆ(2)/(cid:126) = Fˆ2/(cid:126)2 − 2/3 is an element of the quadrupole
2 zz z
3
tensor operator.
The competing contributions – from kinetic and magnetic ordering energies – make
ours an archetype system for studying robust itinerant magnetic order and understanding
the associated phase transitions, of which both first- and second- order are present (Fig. 1b).
For infinitesimal Ω , the tensor field favors either: for Ω > 0, a polar BEC (m = 0:
1 2 F
unmagnetized, M = 0), or for Ω < 0, a ferromagnetic BEC (m = +1 or −1: magnetized,
z 2 F
|M | = 1); thesephases are separatedbyafirst-orderphasetransitionatΩ = 0. Incontrast,
z 2
for large Ω the system forms a spin helix BEC (with local magnetization antiparallel to Ω :
1 1
unmagnetized, M = 0). This order increases the system’s kinetic energy, leading to the
z
second-order phase transition shown in Fig. 1b. These two extreme limits continuously
connect at the point (Ω∗,Ω∗), the green star in Fig. 1b, where the small-Ω first-order
1 2 1
phase transition gives way to the large-Ω second-order transition, and together these regions
1
constitute a curve of critical points {(ΩC,ΩC)}.
1 2
As shown in Fig. 1a, we realized this situation by illuminating 87Rb BECs in the
total angular momentum f = 1 ground state manifold with a pair of counter propagating
“Raman” lasers that coherently coupled the manifold’s three m states. As we first showed3
F
usingeffectivef = 1/2systems, thisintroducesbothaspin-orbitandaZeemantermintothe
BEC’sHamiltonian, equivalenttoEq.(1). TheRamanlaserwavelengthλdefinesthenatural
units for our system: the single-photon recoil energy E = (cid:126)2k 2/2m, and momentum
R R
(cid:126)k = 2π(cid:126)/λ setting the SOC strength. Here the quadratic Zeeman shift from a large
R
bias magnetic field B e split the low-field degeneracy of the |m = −1i ↔ |m = 0i and
0 z F F
4
|m = 0i ↔ |m = +1i transitions, and we independently Raman coupled these state-
F F
pairs with equal strength Ω . As described in the Methods Summary, we dynamically tuned
1
the quadrupole tensor field strength Ω by simultaneously adjusting the Raman frequency
2
differences. Withoutthis technique, only theupper half-plane ofthe phasediagram (Fig.1b)
would be accessible: containing only an unmagnetized phase, therefore lacking any phase
transitions.
In each experiment, we first prepared BECs at a desired point in the phase dia-
gram, possibly having crossed the phase transition during preparation; a combination of
trap dynamics25,26, collisions, and evaporation27 kept the system in or near (local) thermal
equilibrium. We then made magnetization measurements directly from the spin resolved
momentum distribution obtained using the time-of-flight techniques described in Ref. 26.
Our experiment first focused on thermodynamic phase transitions. We made vertical
(horizontal) scans through the phase diagram by initializing the system in the unmagnetized
phase at a desired value of Ω (Ω ) with Ω (cid:38) 0 (Ω (cid:46) 10E ), and then ramping Ω
1 2 2 1 R 2
(Ω ) through the transition region. (As discussed in the methods summary our nominally
1
horizontal Ω -scans followed slightly curved trajectories through the phase diagram, such as
1
the red dashed curve in Fig. 2c). Following such ramps, domains with both ±M can rapidly
z
form, and we therefore focus on the tensor magnetization M = hFˆ(2)i/(cid:126) + 2/3 which is
zz zz
sensitive to this local magnetic order.
Usinghorizontalscans, wecrossedthroughthesecond-orderphasetransition(Ω < Ω∗)
2 2
5
where the free energy evolves continuously from having one minimum (with M = 0, for
zz
large Ω ) to having two degenerate minima (with M > 0, for smaller Ω ). As shown
1 zz 1
in Fig. 2a, M continuously decreases, reaching its saturation value as Ω exceeds ΩC.
zz 1 1
Repeating this processes for Ω∗ < Ω < 0, we found a sharp first-order transition. In each
2 2
case, data is plotted along with theory with no adjustable parameters. Using data of this
type for a range of Ω and fitting to numeric solutions of Eq. (1), we obtained the critical
2
points plotted in Fig. 2c. Because horizontal cuts through the phase diagram are nearly
tangent to the transition curve for small Ω , this produced large uncertainties in ΩC for the
2 1
first-order phase transition.
Westudiedthefirst-orderphasetransitionwithgreaterprecisionbyrampingΩ through
2
the transition at fixed Ω (Fig. 2b) and found near perfect agreement with theory. For all
1
the experimentally measured critical points, see Fig. 2c top, separating the unmagnetized
and ferromagnetic phase, we also measured the corresponding transition widths. Defined as
the required interval for the curve to fall from 50% to 20% of its full range, this width ∆
decreases sharply at Ω∗, marking the crossover between second- and first-order phase tran-
1
sitions (see Fig. 2c bottom). In these data, the width of the first-order transition becomes
astonishingly narrow: as small as 0.0011(3)E = h×4(1)Hz at Ω = 0.41(1).
R 1
We observed that scans crossing the second-order transition typically required 50 ms
to equilibrate, while for scans crossing the first-order transitions we allowed as long as 1.5 s
for equilibration. Systems taken through a first-order phase transition can remain in long-
lived metastable states. Here a metastable state with M = 0 persists in the ferromagnetic
z
6
phase, and a pair of metastable states with M 6= 0 persists in the unmagnetized phase.
z
We began our study of this metastability by quenching through the first-order transition at
Ω = 0.74(8)E with differing rates from 0.5 to 0.2 E /s, as shown in Fig. 3. We observed
1 R R
the transition width continuously decreases with decreasing ramp rate (inset to Fig. 3),
consistent with slow relaxation from a metastable initial state.
We explored the full regime of metastability by initializing BECs in each of the |m =
F
0,±1i states, at fixed Ω , then rapidly ramping Ω (t) from zero to its final value while
2 1
remaining adiabatic when M changed rapidly (but still fast enough so that metastable
z
states survive, (cid:46) 200E /s). For points near the first-order phase transition three metastable
R
states exist (Fig. 4); near the second-order transition this count decreases, giving the two
local minima which merge to a single minimum beyond the second-order transition.
We experimentally identified the number of metastable states by using M and its
z
higher moments, having started in each of the three |m = 0,±1i initial states. A small
F
variance indicates the final states are clustered together – associated with a single global
minimum in the free energy G(M ) – and it increases when metastable or degenerate ground
z
states are present. We distinguished systems with two degenerate magnetization states
(M ≈ ±1) from those with three states by the same method, since when M ≈ ±1, the
z z
variance of |M | is small, and it distinguishably increases as a third metastable state appears
z
with M = 0. In this way we fully mapped the system’s metastable states in agreement with
z
theory, as shown in Fig. 4.
7
In conclusion, we accurately measured the two-parameter phase diagram of a spin-1
BEC,containingaferromagneticphaseandanunmagnetizedphase, continuouslyconnecting
a polar spinor BEC to a spin-helix BEC. The ferromagnetic phase in this itinerant system
is stabilized by SOC, and vanishes as the SOC strength (cid:126)k goes to zero. Our observation
R
of controlled quench dynamics through a first-order phase transition opens the door for
realizing Kibble-Zurek physics28,29 in this system, where the relevant parameters can be
controlled at the individual Hz level. The quadrupole tensor field ∝ Fˆ(2) studied here is the
zz
q = 0 component of the rank-2 spherical tensor operator Fˆ(2), with q ∈ {±2,±1,0}. The
q
physics of this system would be further enriched by the addition of the remaining four tensor
fields: this is straightforward using a combination of radio-frequency and microwave fields30.
Furthermore, the first-order phase transition revealed in this work is further modified by the
addition of spin-dependent interactions: very weak in 87Rb, giving effects below our current
ability to resolve. Our numerical studies show that a number of new symmetry-broken spinor
phases emerge along the line that defines the first-order transition.
Methods Summary
System Preparation We created N ≈ 4×105 atoms 87Rb BECs in the ground electronic
state |f = 1i manifold31, confined in the locally harmonic trapping potential formed at the
intersection of two 1,064 nm laser beams propagating along e and e giving trap frequencies
x y
of (ω ,ω ,ω )/2π = (33(2),33(2),145(5)) Hz. The quadratic contribution to the B =
x y z 0
35.468(1) G magnetic field’s ≈ h×25 MHz Zeeman shift lifts the degeneracy between the
8
|f =1,m =−1i ↔ |f =1,m =0i and |f =1,m =0i ↔ |f =1,m =1i transitions, by
F F F F
(cid:15) = h × 90.417(1) kHz. We denote the energy differences between these states as (cid:126)δω ,
−1,0
(cid:126)δω , and (cid:126)δω .
0,+1 −1,+1
Frequency selective Raman coupling We Raman-coupled the three m states using a
F
pair of λ = 790.024(5) nm laser beams counter-propagating along e . The beam traveling
x
along +e had frequency components ω+ and ω+ , while the beam traveling along −e
x −1 +1 x
contained the single frequency ω−. These frequencies were chosen such that the differences
δω ≈ ω− −ω+ and δω ≈ ω− −ω+ independently Raman coupled the |m = −1,0i
−1,0 −1 0,+1 +1 F
and |m = 0,+1i state-pairs, respectively. Furthermore we selected 2ω− − (ω+ + ω+ ) =
F −1 +1
δω such that, after making the rotating wave approximation (RWA) the |m = ±1i
−1,+1 F
states were energetically shifted by the same Ω(0) = [(δω − δω ) + (ω+ − ω+ )]/2
2 −1,0 0,+1 −1 +1
energy from |m = 0i, thereby yielding the frequency-tuned tensor energy shift depicted
F
in Fig. 1a. In addition, Ω in Eq. (1) differs from Ω(0) by a small shift ∝ Ω2/(cid:15) resulting
2 2 1
from off-resonant coupling to transitions detuned by 2(cid:15), which we computed directly using
Floquet theory [see Eq. (6) in Methods]. This tensor contribution Ω Fˆ(2) to the Hamiltonian
2 zz
might equivalently be introduced by a quadratic Zeeman shift alone, giving Ω ∝ B2 > 0.
2 0
1. Stoner, E. Atomic moments in ferromagnetic alloys with non- ferromagnetic elements.
Philos. Mag. 15, 1018–1034 (1933).
2. Sanner, C. et al. Correlations and Pair Formation in a Repulsively Interacting Fermi
Gas. Phys. Rev. Lett. 108, 240404 (2012).
9
3. Lin, Y.-J., Jiménez-García, K. & Spielman, I. B. Spin-orbit-coupled Bose-Einstein con-
densates. Nature 471, 83–86 (2011).
4. Lan, Z. & Öhberg, P. Raman-dressed spin-1 spin-orbit-coupled quantum gas. Phys. Rev.
A 89, 023630 (2014).
5. Radić, J., Natu, S. S. & Galitski, V. Stoner ferromagnetism in a thermal pseudospin-1/2
Bose gas. Phys. Rev. Lett. 113, 185302 (2014).
6. Hickey, C. & Paramekanti, A. Thermal Phase Transitions of Strongly Correlated Bosons
with Spin-Orbit Coupling. Phys. Rev. Lett. 113, 265302 (2014).
7. Aharoni, A. Introduction to the Theory of Ferromagnetism (Oxford University Press,
2011).
8. Abragam, A. Nuclear ferromagnetism and ant ferromagnetism. Contemporary Physics
33, 305–312 (1992).
9. Kronjäger, J., Becker, C., Soltan-Panahi, P., Bongs, K. & Sengstock, K. Spontaneous
pattern formation in an antiferromagnetic quantum gas. Phys. Rev. Lett. 105, 090402
(2010).
10. Simon, J. et al. Quantum simulation of antiferromagnetic spin chains in an optical
lattice. Nature 472, 307–312 (2011).
11. Hart, R. A. et al. Observation of antiferromagnetic correlations in the Hubbard model
with ultracold atoms (2014).
10