betaFeedback

WorksPhysics and Astronomy

Stability of quantised vortices in two-component condensates

Abstract

Multiply quantised vortices (MQVs) within single component Bose-Einstein condensates are unstable and decay rapidly. We show that MQVs can be stabilised by adding a small number of atoms of a second species to the vortex cores, and that these atoms remain in the vortex core as the system evolves. A consequence of the stabilisation is that nearby co-rotating vortices can orbit in the opposite sense to their individual rotations when enough of the second species is present. This has implications concerning the imaging of vortices, as well as quantum turbulence and vortex nucleation in two-component condensates, such as those involving mixtures of 87^{87}Rb and 133^{133}Cs.

Introduction

The most striking property of quantum fluids (superfluid 4He{}^{4}\mathrm{He} and 3He{}^{3}\mathrm{He}, atomic Bose-Einstein condensates, polariton condensates, etc) is the existence of a macroscopic complex wavefunction Ψ(x,t)\Psi(\mathbf{x},t), where x\mathbf{x} is the position and tt is time. If we write the wavefunction as Ψ=∣Ψ∣eiΦ\Psi=|\Psi|e^{i\Phi} in terms of its amplitude ∣Ψ∣|\Psi| and phase Φ\Phi, then the density nn and the velocity v\mathbf{v} of the superfluid are [1]:

n(x,t)=∣Ψ(x,t)∣2,v(x,t)=ℏM∇Φ(x,t),(1)n(\mathbf{x},t)=|\Psi(\mathbf{x},t)|^2,\qquad\mathbf{v}(\mathbf{x},t)=\frac{\hbar}{M}\nabla\Phi(\mathbf{x},t), \tag*{(1)}

where ℏ\hbar is the reduced Planck’s constant and MM is the mass of the relevant boson. The wavefunction imposes constraints on the rotational motion. Whereas rotation in ordinary (classical) fluids takes the form of eddies of arbitrary size and circulation, rotation in superfluids is quantised, occurring around special points (in two-dimensions) or lines (in three-dimensions) called quantised vortices. At these special points the density nn is exactly zero, hence the phase Φ\Phi is not defined: quantised vortices are therefore phase defects. Within a superfluid, the circulation Γ\Gamma of the velocity along a closed path CC is either zero if the region inside CC does not include any vortex, or an integer multiple ℓ\ell of the quantum of circulation κ=2πℏ/M\kappa=2\pi\hbar/M [1],

Γ=∮Cv⋅dr=ℓκ,(2)\Gamma=\oint_C \mathbf{v}\cdot\mathrm{d}\mathbf{r}=\ell\kappa, \tag*{(2)}

if the region inside CC includes a vortex of quantisation ℓ\ell (since Ψ\Psi must be continuous, ℓ\ell must take on integer values). This condition strongly constrains the velocity field around the vortex axis. Vortices with ∣ℓ∣=1|\ell|=1 are commonly referred to as singly quantised vortices, SQVs, whereas ∣ℓ∣>1|\ell|>1 are called multiply quantised vortices, MQVs (ℓ\ell is commonly called the “charge” of the vortex). Since the angular momentum and the kinetic energy carried by a vortex are proportional to ℓ\ell and ℓ2\ell^2 respectively, MQVs are energetically unstable [3, 4] [2] in simple homogeneous superfluids [5]; any MQV with charge ℓ\ell will rapidly decay into ℓ\ell SQVs carrying the same total angular momentum. This decay is the result of an energetic instability in the spectrum of the vortex, i.e. the vortex possesses a normal mode with negative energy. In the presence of a dissipation mechanism, the amplitude of this mode will grow eventually causing the central vortex to split up. In large enough systems, the surrounding bath of sound waves can also act as a reservoir for the vortex to dissipate energy into, allowing MQVs to split even when the governing equations are energy conserving. In this case, the spectrum of the vortex contains an unstable mode with complex frequency which grows spontaneously, see e.g. [6, 7, 8, 9, 10, 11].

The possibility of stabilizing inherently unstable MQVs has been explored by using inhomogeneous trapping potentials to force the spatial distribution of superfluid matter [2] and by manipulating the superfluid via electro-magnetic field fluctuations [12]. A third possibility is based on two-component condensates [13, 14]. These systems can consist of different atoms [15], different isotopes of the same atom [16] or even different hyperfine states of the same isotope [17, 18]. It has been suggested that in such systems [19, 20, 21, 22, 23, 24] the overlapping second component can slow down or even prevent decay of MQVs [5] of the first component. As we shall see, spatial separation is relevant to our problem. Two distinct modes of spatial separation exist [25, 26]: separation arising from the geometry of the trapping potentials, and separation resulting from the immiscibility of the two components [27], a property characterised by the inter-component interactions. Kuopanportti et al. [5] suggested how to stabilise MQVs in a tightly confined two-dimensional 87Rb{}^{87}\mathrm{Rb} and 41K{}^{41}\mathrm{K} mixture confined within a rotating harmonic trap. They successfully sta bilised ℓ=2\ell= 2 MQVs with both attractive and repulsive inter-component interactions by a clever distribution of the secondary component. In each case, the secondary component was distributed in such a way that the energetic favourability of MQV decay was not sufficient to overcome the inter-component interactions: the MQVs were effectively ‘propped up’ and could not decay. Unfortunately, the scope of this technique is limited by the tightness of the trapping potential: the vortex core was comparable in size to the size of the entire condensate.

In this work, we propose a method of stabilising MQVs which exploits the spatial separation of mixtures with highly repulsive inter-component interactions. We shall show that, by letting the second component concentrate inside the vortex core of the first component, the energetic advantage of vortex decay can be offset. To best facilitate stabilisation of MQVs, it is paramount to choose a condensate mixture such that both components are self-repulsive and the inter-component repulsion is large. A mixture of 87Rb{}^{87}\mathrm{Rb} and 133Cs{}^{133}\mathrm{Cs} atoms meets both requirements and is realisable in the laboratory [15]. Rather than a conventional harmonic trap, we consider the problem in a potential trap with a wide, flat bottom and steep walls, which allows for condensates that are homogeneous in the bulk [28, 29]. The advantage of such large circular bucket potentials is that the vortex core profiles remain unchanged as they move about the condensate, unlike what happens in a tightly bound system, making our results independent of the trap’s details.

The system

Governing equations

The equations of motion of a trapped two-dimensional two-component condensate are,

iℏ∂tΨ1=[K^1+V(x)+G1∣Ψ1∣2+G12∣Ψ2∣2−μ~1]Ψ1,i\hbar\partial_t\Psi_1 = \left[\hat{K}_1 + V(\mathbf{x}) + G_1|\Psi_1|^2 + G_{12}|\Psi_2|^2 - \tilde{\mu}_1\right]\Psi_1,
iℏ∂tΨ2=[K^2+V(x)+G2∣Ψ2∣2+G12∣Ψ1∣2−μ~2]Ψ2.(3)i\hbar\partial_t\Psi_2 = \left[\hat{K}_2 + V(\mathbf{x}) + G_2|\Psi_2|^2 + G_{12}|\Psi_1|^2 - \tilde{\mu}_2\right]\Psi_2. \tag*{(3)}

where Ψj\Psi_j are the macroscopic wavefunctions for the components j=1,2j = 1, 2, μ~j\tilde{\mu}_j are the chemical potentials, K^j=−ℏ22Mj∇2\hat{K}_j = -\frac{\hbar^2}{2M_j}\nabla^2 are the kinetic energy operators, GjG_j are the 2D self-interaction parameters and G12G_{12} is the 2D inter-species interaction parameter. The 2D trapping potential V(x)V(\mathbf{x}) (where x=(x,y)\mathbf{x} = (x,y)) is taken to be a cylindrical box-trap of radius rBr_B and height V0V_0. It is convenient to assume the circularly symmetric form,

V(x)=V(r)=V01+(V0−1)ea(rB−r),(4)V(\mathbf{x}) = V(r) = \frac{V_0}{1+(V_0-1)e^{a(r_B-r)}}, \tag*{(4)}

where r2=x2+y2r^2 = x^2+y^2. Assuming that V0V_0 is larger than both chemical potentials, the two species are essentially confined in the region r<rBr < r_B; within this region, V(r)V(r) is negligible. The quantity aa is a smoothing parameter. We assume the system is sufficiently low temperature that thermal noise and dissipation can be neglected. The assumption of an effectively two-dimensional system can be satisfied in practice by imposing a strong confinement of both components in the vertical direction, e.g. a harmonic potential with frequency ωz≫μ~j/ℏ\omega_z \gg\tilde{\mu}_j/\hbar, such that the dynamics in the extra dimension are frozen out (see Appendix A).

The system of equations (3) can be derived from the following action,

S=∫dt d2x[∑j(iℏΨj∗Ψ˙j−Hj)−HI],(5)\mathcal{S} = \int dt\,d^2x\left[\sum_j\left(i\hbar\Psi_j^*\dot{\Psi}_j-\mathcal{H}_j\right)-\mathcal{H}_I\right], \tag*{(5)}

where the free energy density for each component jj and inter-species interaction energy density are respectively defined as,

Hj=Ψj∗[K^j+V(x)−μ~j]Ψj+Gj2∣Ψj∣4,\mathcal{H}_j = \Psi_j^*\left[\hat{K}_j+V(\mathbf{x})-\tilde{\mu}_j\right]\Psi_j+\frac{G_j}{2}|\Psi_j|^4,
HI=G12∣Ψ1∣2∣Ψ2∣2.(6)\mathcal{H}_I = G_{12}|\Psi_1|^2|\Psi_2|^2. \tag*{(6)}

Since independent phase rotations of Ψ1\Psi_1 and Ψ2\Psi_2 are symmetries of S\mathcal{S}, we obtain two conserved currents

∂tnj+∇⋅(njvj)=0,(7)\partial_t n_j+\nabla\cdot(n_j\mathbf{v}_j)=0, \tag*{(7)}

where nj=∣Ψj∣2n_j=|\Psi_j|^2 is the number density of component jj and vj=ℏMj∇arg⁡(Ψj)\mathbf{v}_j=\frac{\hbar}{M_j}\nabla\arg(\Psi_j) is its velocity. In a closed system, one can integrate over space to find that the number of atoms in each species, Nj=∫njd2xN_j=\int n_jd^2x, is conserved by the evolution dictated by (3). Hereafter, we fix the number of atoms in the first component N1N_1 and treat η=N2/N1\eta=N_2/N_1 as our variable.

An approximate expression for the ground state solution of equations (3) is provided by the Thomas-Fermi approximation: neglecting the kinetic energy terms, within the trap (where V(r)≃0V(r)\simeq0) the steady state solutions of (3) satisfy,

G1n1+G12n2−μ~1=0,G2n2+G12n1−μ~2=0.(8)G_1n_1+G_{12}n_2-\tilde{\mu}_1=0,\qquad G_2n_2+G_{12}n_1-\tilde{\mu}_2=0. \tag*{(8)}

Since the two components are immiscible for G12>G1G2G_{12}>\sqrt{G_1G_2} [27], we can assume that component 2 is confined to the region A2=πr02A_2=\pi r_0^2 and component 1 occupies the remaining space A1=π(rB2−r02)A_1=\pi(r_B^2-r_0^2). We then have n1=N1/A1n_1=N_1/A_1 and n2=N2/A2n_2=N_2/A_2 so that the chemical potentials are approximately,

μ~1=G1N1A1,μ~2=ηG2N1A2.(9)\tilde{\mu}_1=\frac{G_1N_1}{A_1},\qquad\tilde{\mu}_2=\frac{\eta G_2N_1}{A_2}. \tag*{(9)}

In this work, we consider systems with a vortex in Ψ1\Psi_1 and no vortex in Ψ2\Psi_2. In the stationary state, solutions of (3) are of the form,

Ψ1(r,θ)=n1(r)eiℓθ,Ψ2(r)=n2(r).(10)\Psi_1(r,\theta)=\sqrt{n_1(r)}e^{i\ell\theta},\qquad\Psi_2(r)=\sqrt{n_2(r)}. \tag*{(10)}

where ℓ\ell is the winding number of the vortex. The corresponding velocity field is v1=ℏℓM1re^θ\mathbf{v}_1 = \frac{\hbar\ell}{M_1 r}\hat{\mathbf{e}}_\theta where e^θ\hat{\mathbf{e}}_\theta is the unit vector in the azimuthal direction. We focus on the case ℓ=2\ell= 2, however, we expect our conclusions to extend to vortices of higher winding numbers.

Dimensionless variables

In the rest of the paper we express the governing equations (3) in dimensionless form using the following characteristic units of length, time and density,

ξ1=ℏμ~1,0M1,τ1=ℏμ~1,0,n0=μ~1,0G1,(11)\xi_1 = \frac{\hbar}{\sqrt{\tilde{\mu}_{1,0}M_1}}, \qquad\tau_1 = \frac{\hbar}{\tilde{\mu}_{1,0}}, \qquad n_0 = \frac{\tilde{\mu}_{1,0}}{G_1}, \tag*{(11)}

where μ~1,0=μ~1(η=0)\tilde{\mu}_{1,0} = \tilde{\mu}_1(\eta= 0) is the chemical potential of species 1 in the vortex state when species 2 is absent. In other words, we perform the following rescalings,

rξ1→r,tτ1→t,Vμ~1,0→V,Ψjn0→Ψj,(12)\frac{r}{\xi_1} \to r, \qquad\frac{t}{\tau_1} \to t, \qquad\frac{V}{\tilde{\mu}_{1,0}} \to V, \qquad\frac{\Psi_j}{\sqrt{n_0}} \to\Psi_j, \tag*{(12)}

so that (3) becomes,

i∂tΨ1=[−12∇2+V(x)+∣Ψ1∣2+g12∣Ψ2∣2−μ1]Ψ1,i\partial_t\Psi_1 = \left[-\frac{1}{2}\nabla^2 + V(\mathbf{x}) + |\Psi_1|^2 + g_{12}|\Psi_2|^2 - \mu_1\right]\Psi_1,
i∂tΨ2=[−12m2∇2+V(x)+g2∣Ψ2∣2+g12∣Ψ1∣2−μ2]Ψ2,(13)i\partial_t\Psi_2 = \left[-\frac{1}{2m_2}\nabla^2 + V(\mathbf{x}) + g_2|\Psi_2|^2 + g_{12}|\Psi_1|^2 - \mu_2\right]\Psi_2, \tag*{(13)}

where we have defined the new dimensionless parameters,

μj=μ~jμ~1,0,g2=G2G1,g12=G12G1,m2=M2M1.(14)\mu_j = \frac{\tilde{\mu}_j}{\tilde{\mu}_{1,0}}, \qquad g_2 = \frac{G_2}{G_1}, \qquad g_{12} = \frac{G_{12}}{G_1}, \qquad m_2 = \frac{M_2}{M_1}. \tag*{(14)}

Note the true total number of atoms can be recovered from Nj=∫d3x∣Ψj∣2N_j = \int d^3x|\Psi_j|^2 in these units by multiplying by a factor of ℏ2/M1G1=d1/8πa1\hbar^2/M_1G_1 = d_1/\sqrt{8\pi a_1}, where d1d_1 is the width of species 1 in the vertical direction and a1a_1 is its scattering length (see Appendix A).

Since ξ1ℓ\xi_1\ell (where ξ1\xi_1 is the healing length of the condensate 1 for η=0\eta= 0) is the characteristic core size of an ℓ\ell-vortex, in the new units the vortex core now has radius ∼ℓ\sim\ell. Furthermore, since the density of condensate 1 approaches n0n_0 far from the vortex axis where n2→0n_2 \to0, in the new units the bulk value of n1n_1 will approach μ1\mu_1.

Parameters

In these units, the problem depends only on m2m_2, g2g_2, g12g_{12}, the parameters in the trapping potential and the chemical potentials μj\mu_j. The latter are related to the total number of atoms in the ground state (see next section). The s–wave scattering lengths for the 87Rb{}^{87}\mathrm{Rb} and 133Cs{}^{133}\mathrm{Cs} mixture [30] are a1=100a0a_1 = 100a_0, a2=280a0a_2 = 280a_0 and a12=650a0a_{12} = 650a_0 [26] where a0a_0 is Bohr radius and hereafter the subscripts 1 and 2 refer respectively to the 87Rb{}^{87}\mathrm{Rb} and 133Cs{}^{133}\mathrm{Cs} components. In Appendix A, we show that when both species are subject to the same strong vertical confinement, the dimensionless interaction parameters are g2=2.04g_2 = 2.04 and g12=5.65g_{12} = 5.65. The mass ratio is simply m2=133/87m_2 = 133/87. For the trapping potential, we fix a=5a = 5 and V0=5V_0 = 5 so that there is essentially a hard wall located at r=rBr = r_B. For the trap radius, we pick rB=40r_B = 40 so that the trap is not too small to prohibit the instabilities under consideration but not so large that the problem becomes computationally impractical.

Stationary solutions

To solve for the stationary profiles n1,2n_{1,2}, we evolve (13) in imaginary time τ=it\tau= it. This amounts to making the replacement i∂t→−∂τi\partial_t \to-\partial_\tau in (3). Starting from a sensible initial condition, the population of the higher energy modes will then decrease exponentially with increasing τ\tau, leaving essentially only the ground state at late times. Since the anticipated densities are independent of the azimuthal coordinates, the speed of the computation is increased by working in polar coordinates. We therefore solve the following coupled set of equations,

∂τY1=(∇r22−ℓ22r2−V(r)+μ1−∣Y1∣2−g12∣Y2∣2)Y1,\partial_\tau Y_1 = \left(\frac{\nabla_r^2}{2} - \frac{\ell^2}{2r^2} - V(r) + \mu_1 - |Y_1|^2 - g_{12}|Y_2|^2\right)Y_1,
∂τY2=(∇r22m2−V(r)+μ2−g2∣Y2∣2−g12∣Y1∣2)Y2,(15)\partial_\tau Y_2 = \left(\frac{\nabla_r^2}{2m_2} - V(r) + \mu_2 - g_2|Y_2|^2 - g_{12}|Y_1|^2\right)Y_2, \tag*{(15)}

where ∇r2=∂r2+1r∂r\nabla_r^2 = \partial_r^2 + \frac{1}{r}\partial_r and Yj=∣Ψj∣=njY_j = |\Psi_j| = \sqrt{n_j}. We work on a radial grid r=[Δr,rB+5]r = [\Delta r,r_B + 5] with Δr=0.05625\Delta r = 0.05625 such that there are 800 radial points and use an imaginary time step Δτ=4×10−4\Delta\tau= 4 \times10^{-4}. The evolution is performed using a 4th order Runge-Kutta algorithm with the radial derivatives approximated using finite difference stencils.

The first step is to fix a value for N1N_1 against which we can compare the effect of increasing η\eta. This value is determined by solving (15) for η=0\eta= 0, using the fact that μ1(η=0)=1\mu_1(\eta= 0) = 1, and integrating over the resulting profile to get N1=∫d2xn1(r,η=0)N_1 = \int d^2x n_1(r,\eta= 0). For rB=40r_B = 40 and ℓ=2\ell= 2, we have N1≃4824N_1 \simeq4824. Again, since this is the atom number in dimensionless variables, one must multiply by the dimensionless ratio d1/8πa1d_1/\sqrt{8\pi a_1} to find the true total number of atoms.

The second step is to choose an initial condition when the second component is added. We start by considering case where no vortex is present (ℓ=0\ell= 0) with component 2 localised in the centre of the system and component 1 surrounding it. A subtlety of the problem is that if we want to keep N1,2N_{1,2} fixed, the chemical potentials must vary to accommodate these constraints. A priori, we do not know the values of μ1,2\mu_{1,2} and must determine them during the imaginary time evolution. We start by guessing their values using the Thomas-Fermi approximation (9), which in dimensionless units gives સ μ1=N1/π(rB2−r02)\mu_1=N_1/\pi(r_B^2-r_0^2) and μ2=ηg2N1/πr02\mu_2=\eta g_2N_1/\pi r_0^2. Since we consider η\eta to be very small, we take 1 as our initial guess for the value of μ1\mu_1 and invert the first relation to find the value of r0r_0. The second relation provides a guess for μ2\mu_2. The densities are then n1(r0<r<rB)=μ1n_1(r_0<r<r_B)=\mu_1 and n2(r>r0)=μ2/g2n_2(r>r_0)=\mu_2/g_2. We evolve these profiles forward in imaginary time, renormalising Ψ1,2\Psi_{1,2} after each time step to keep the values of N1,2N_{1,2} fixed. After a long time, the system will have relaxed to its lowest energy state, however, there will be residual time dependence due to the fact that the chemical potentials were chosen incorrectly. We can use this time dependence to infer the shift in μj\mu_j using Yj(τ+Δτ)=e−δμjΔτYj(τ)Y_j(\tau+\Delta\tau)=e^{-\delta\mu_j\Delta\tau}Y_j(\tau), so that the true chemical potentials are μ1=1+δμ1\mu_1=1+\delta\mu_1 and μ2=ηg2N1/πr02+δμ2\mu_2=\eta g_2N_1/\pi r_0^2+\delta\mu_2 respectively. The evolution is terminated once the changes δμ1\delta\mu_1 and δμ2\delta\mu_2 over a time step are both less than 10−610^{-6}. This is suitably small given that μj∼O(1)\mu_j\sim O(1) for both species and is computationally feasible when solving for a large range of η\eta values.

In the third step, we imprint a vortex by setting ℓ≠0\ell\ne0 in (15), taking the ground states obtained from step 2 as the initial condition. We evolve again in imaginary time (using the same termination criterion) to find the stationary vortex solutions and update the chemical potentials accordingly.

In Figure 1, we present the resulting density profiles n1,2n_{1,2} as a function of rr for a vortex with ℓ=2\ell=2. We see that component 2 stays inside the core of the vortex in component 1, and its effect is to widen the core. It is remarkable that even the small values of η\eta which we consider here have such a significant effect. This widening becomes more significant as η\eta increases. We also display the variation of the chemical potentials with η\eta in the inset of the right panel. μ1\mu_1 increases with η\eta by a very small amount due to the repulsive effect of species 2 inside the vortex core. This forces more of the N1N_1 particles into the bulk of the condensate, increasing the mean value of the density in this region. μ2\mu_2 increases much more obviously with η\eta since N2N_2 is increasing and therefore the density of species 2 must go up accordingly. Note however that μ2\mu_2 does not vanish as η→0\eta\to0. This is because there is a contribution of the curvature of n2n_2 to the chemical potential. Indeed, solving the equation in (15) for Y2Y_2 in limit r→0r\to0 gives the relation μ2=(g2n2−n2′′/2m2n2)∣r=0\mu_2=(g_2n_2-n_2''/2m_2n_2)|_{r=0}. This does not go to zero as n2→0n_2\to0 since the ratio n2′′/n2n_2''/n_2 stays finite in this limit.

Figure 1. Density profiles for species 1 (left) and species 2 (right) for various η\eta values (indicated in the figure legend on the right). The vortex axis is r=0r=0 is on the left and the trap’s boundary is at r=rB=40r=r_B=40. The other parameters are ℓ=2\ell=2, and N1≃4824N_1 \simeq4824 (the dimensionless number of atoms of species 1 when η=0\eta=0). The inset on the left panel shows a close up of the vortex axis. Component 2 occupies the vortex core where the density of component 1 goes to zero. As the proportion of component 2 increases, the vortex core region widens. The inset on the right panel shows how the chemical potentials vary with η\eta.

Linear analysis

Having found the stationary vortex solutions to (13), we determine their stability by analysing the normal modes of the system. It is well-known that MQVs in one-component condensates spontaneously decay into clusters of SQVs and that this instability can be attributed to an unstable mode of the linearised equations of motion [7]. We now outline how this analysis can be extended to two-component condensates and demonstrate that, for sufficiently high values of η\eta, this mode can be stabilised.

When the equations of motion (13) are linearized, since neither n1,2n_{1,2} nor v1,2v_{1,2} depend on θ\theta, each mm component evolves independently of the others. Due to the symmetries uj∗(m)=vj(−m)u_j^{*}(m)=v_j(-m) and vj∗(m)=uj(−m)v_j^{*}(m)=u_j(-m), we can restrict our attention to m≥0m\geq0 modes without loss of generality. The fluctuations ∣U⟩=(u1,u2,v1,v2)T|U\rangle=(u_1,u_2,v_1,v_2)^{\mathrm{T}} obey the following coupled set of equations,

i∂t∣U⟩=L^∣U⟩.(16)i\partial_t|U\rangle=\hat{L}|U\rangle. \tag*{(16)}

Bogoliubov method

Let δψj\delta\psi_j be small fluctuations about Ψj\Psi_j and write the following Ansatz,

\begin{bmatrix} \delta\psi_1\\ \delta\psi_2\\ \delta\psi_1^*\\ \delta\psi_2^* \end{bmatrix} = \sum_{m=-\infty}^{\infty}e^{im\theta} \begin{bmatrix} u_1e^{i\ell\theta}\\ u_2\\ v_1e^{-i\ell\theta}\\ v_2 \end{bmatrix}, \tag*{(17)}\end{bmatrix}

with uju_j and vjv_j functions of rr, tt and mm, where mm is the azimuthal wavenumber.

The operator L^\hat{L} is,

L^=[D1+εn1εεD2εg2n2−n1−ε−D1−ε−ε−g2n2−ε−D2],(18)\hat{L}= \begin{bmatrix} D_1^+ & \varepsilon& n_1 & \varepsilon\\ \varepsilon& D_2 & \varepsilon& g_2n_2\\ -n_1 & -\varepsilon& -D_1 & -\varepsilon\\ -\varepsilon& -g_2n_2 & -\varepsilon& -D_2 \end{bmatrix}, \tag*{(18)}

where,

D1±=−12[∇r2−(ℓ±m)2r2]+V+2n1+g12n2−μ1,D2=−12M2[∇r2−m2r2]+V+2g2n2+g12n1−μ2.(19)\begin{aligned} D_1^\pm&= -\frac{1}{2}\left[\nabla_r^2-\frac{(\ell\pm m)^2}{r^2}\right]+V+2n_1+g_{12}n_2-\mu_1,\\ D_2 &= -\frac{1}{2M_2}\left[\nabla_r^2-\frac{m^2}{r^2}\right]+V+2g_2n_2+g_{12}n_1-\mu_2. \tag*{(19)} \end{aligned}

and the coupling between fluctuations of both species is determined by ϵ=g12n1n2\epsilon=g_{12}\sqrt{n_1n_2}. Note that for immiscible fluids, the region of overlap of the two components is small. ϵ\epsilon will be small in this region and zero everywhere else. The fields uju_j and vjv_j are related to the components of the density field by,

δnj,m=nj(uj+vj)(20)\delta n_{j,m}=\sqrt{n_j}(u_j+v_j) \tag*{(20)}

The equations in (17) can be derived from the action SBdG=∫dtd2xLBdG\mathcal{S}_{\mathrm{BdG}}=\int dt d^2x\mathcal{L}_{\mathrm{BdG}} where,

LBdG=iu1∗u˙1−iv1∗v˙1+iu2∗u˙2−iv2∗v˙2−HBdG,HBdG=HBdG,1+HBdG,2+HBdG,int,HBdG,1=u1∗D1+u1+v1∗D1−v1+n1(u1∗v1+u1v1∗),HBdG,2=u2∗D2u2+v2∗D2v2+g2n2(u2∗v2+u2v2∗),HBdG,int=ϵ[(u1∗+v1∗)(u2+v2)+c.c.].(21)\begin{aligned} \mathcal{L}_{\mathrm{BdG}}&=iu_1^*\dot{u}_1-iv_1^*\dot{v}_1+iu_2^*\dot{u}_2-iv_2^*\dot{v}_2-\mathcal{H}_{\mathrm{BdG}},\\ \mathcal{H}_{\mathrm{BdG}}&=\mathcal{H}_{\mathrm{BdG},1}+\mathcal{H}_{\mathrm{BdG},2}+\mathcal{H}_{\mathrm{BdG},\mathrm{int}},\\ \mathcal{H}_{\mathrm{BdG},1}&=u_1^*D_1^+u_1+v_1^*D_1^-v_1+n_1(u_1^*v_1+u_1v_1^*),\\ \mathcal{H}_{\mathrm{BdG},2}&=u_2^*D_2u_2+v_2^*D_2v_2+g_2n_2(u_2^*v_2+u_2v_2^*),\\ \mathcal{H}_{\mathrm{BdG},\mathrm{int}}&=\epsilon[(u_1^*+v_1^*)(u_2+v_2)+\mathrm{c.c.}]. \tag*{(21)} \end{aligned}

SBdG\mathcal{S}_{\mathrm{BdG}} is invariant when each of the fields uju_j and vjv_j simultaneously undergo the same phase rotation. The corresponding conserved quantity is the Bogoliubov norm,

NBdG=⟨U∣γ0∣U⟩≡∑j∫d2x(∣uj∣2−∣vj∣2),(22)\mathcal{N}_{\mathrm{BdG}}=\langle U|\gamma^0|U\rangle\equiv\sum_j\int d^2x(|u_j|^2-|v_j|^2), \tag*{(22)}

where γ0=diag⁡(1,1,−1,−1)\gamma^0=\operatorname{diag}(1,1,-1,-1) is the zeroth gamma matrix. Notice that due to the presence of the cross-terms with coefficient ϵ\epsilon in (21), excitations of species 1 can be converted into excitations of species 2 and vice versa. However, the total number of excitations, as measured by NBdG\mathcal{N}_{\mathrm{BdG}}, is conserved.

Since the background is stationary, we can further decompose the fluctuations into the separate frequency components,

∣U(r,t)⟩=∫−∞∞dω2πaω(t)∣U~(r;ω)⟩,(23)|U(\boldsymbol{r},t)\rangle=\int_{-\infty}^{\infty}\frac{d\omega}{2\pi}a_\omega(t)|\widetilde{U}(\boldsymbol{r};\omega)\rangle, \tag*{(23)}

where aωa_\omega and ∣U~⟩|\widetilde{U}\rangle respectively satisfy,

i∂taω=ωaω,ω∣U~⟩=L^∣U~⟩.(24)i\partial_t a_\omega=\omega a_\omega,\qquad\omega|\widetilde{U}\rangle=\hat{L}|\widetilde{U}\rangle. \tag*{(24)}

The first equation is solved by aω∝e−iωta_\omega\propto e^{-i\omega t} up to an arbitrary constant, whilst the second must be solved numerically. The Hamiltonian for stationary modes is related to the Bogoliubov norm via HBdG=∫d2xHBdG=ωNBdG\mathcal{H}_{\mathrm{BdG}}=\int d^2x\mathcal{H}_{\mathrm{BdG}}=\omega\mathcal{N}_{\mathrm{BdG}}. Therefore a positive frequency mode with negative norm (or a negative frequency mode with positive norm) has negative energy. The sign of the energy plays an important role in the discussion of the instability.

To solve the second equation in (24) computationally, we write the radial derivative terms using finite difference stencils on a discrete rr-grid then find the eigenvalues and eigenfunctions of the resulting matrix L^\hat{L}. The eigenfunctions are normalised so that ∣Re⁡[NBdG]∣=1|\operatorname{Re}[\mathcal{N}_{\mathrm{BdG}}]| = 1, which gives an unambiguous normalisation condition for all solutions obtained. We place a Dirichlet boundary condition at large rr, although this does not play a significant role as it is applied in a region where V(r)V(r) is large, hence, modes with energy lower than the barrier height quickly decay to zero inside this region. A discussion of the boundary conditions at r=0r = 0 can be found in Appendix B.

In Figure 2, we present the eigenvalues ω\omega as a function of η\eta for ℓ=2\ell= 2 and rB=40r_B = 40. There are two distinct types of excitation. Bulk excitations (shown in black) predominantly occupy the region outside of the vortex core and are associated with collective excitations of the condensate, or phonons at low wavenumbers. Indeed, on panel (c), it can be seen that these modes have a large oscillatory component in the large rr region. They have positive energy (HBdG>0H_{\mathrm{BdG}} > 0) and are largely unaffected by increasing η\eta, which influences only the vortex core. There is a second class of excitation which we refer to as vortex modes. These have negative energy (HBdG<0H_{\mathrm{BdG}} < 0) at low η\eta and are mainly localised around the vortex axis, as illustrated in panel (c). In panel (a), we see that there is only one such mode for the ℓ=2\ell= 2 vortex represented by the red line. Since this mode occupies the same region as component 2, it is strongly affected when η\eta is increased. Indeed, the frequency of the vortex mode on panel (a) can be seen to decrease rapidly even for relatively small amounts of component 2. In the presence of a dissipation mechanism, these modes will be excited since they lower the total energy of the system, i.e. they are an energetic instability.

Figure 2. On the left we display the eigenvalues ω\omega of (17) as a function of η\eta for m=ℓ=2m=\ell=2 and rB=40r_B=40, with the real part in panel (a) and the imaginary part in panel (b). For clarity, both panels (a) and (b) are split in two regions for small and large values of η\eta. The modes whose frequency is approximately independent of η\eta (solid black horizontal lines) are bulk excitations (phonons) which are not affected by component 2 in the vortex core. The mode whose frequency decreases with η\eta (solid red) is an excitation of the vortex core and has negative norm for all frequencies. When the vortex mode crosses a bulk excitation mode, the two modes couple with each other to produce a pair of complex conjugate modes (solid green) with zero norm and complex frequency ω\omega. At a critical value ηc\eta_c (marked as a solid black dot), the frequency of the vortex mode (red line) changes sign. On the right, we display some example mode functions for the vortex mode. In panel (c), we pick an η\eta value slightly to the left of the third instability bubble on panel (b). The solid black (resp. red) curve corresponds to the bulk (resp. vortex) excitations of component 1 with the frequency indicated. The dashed curves of the same colour correspond to component 2. In panel (d), we pick the η\eta value at the peak of the third instability bubble and display the real and imaginary parts of the growing mode. Comparing panels (c) and (d), we see that the vortex mode enters mainly the real component of the unstable eigenmode whilst the bulk excitation is captured by the imaginary part.

When Re⁡[ω]\operatorname{Re}[\omega] of the vortex mode crosses that of a bulk excitation, the two couple to produce a complex conjugate pair of modes with vanishing Bogoliubov norm. This leads to the appearance of isolated bubbles in Im⁡[ω]\operatorname{Im}[\omega] in panel (b). The mode with Im⁡[ω]>0\operatorname{Im}[\omega] > 0 represents a dynamical instability, that is, the central vortex spontaneously sheds its energy into a sound wave in the bulk even without a dissipation mechanism. This does not violate energy conservation since the total energy of the mode encoding this behaviour is zero (HBdG=0H_{\mathrm{BdG}} = 0). The complex conjugate mode describes the time reversed process, i.e. absorption of a sound wave by the central vortex. In panel (d), we show how the waveforms of the two modes in panel (c) combine to produce the real and imaginary parts of the instability.

Note that for large enough rBr_B (larger than considered here) i.e. in the limit that finite size effects become negligible, the bubbles in Im⁡[ω]\operatorname{Im}[\omega] eventually merge into a continuum [9]. In that case, there is a dynamical instability at η=0\eta= 0 for all larger trap sizes and, in particular, in the infinite system limit.

We also show in Figure 3 the effect of increasing η\eta on the waveform of the vortex mode around the vortex axis. The main feature is that the amplitude of the mode at r=0r = 0 is suppressed as the number of atoms in species 2 is increased. This will be addressed in the next section.

Figure 3. Variation of the vortex mode waveform with η\eta. In particular we display the modulus of uj+vju_j+v_j, which is the combination appearing in formula (20) for density fluctuations. The main effect of the second component is to reduce the amplitude of excitations of species 1 on the vortex axis. The shift in the maximum of ∣u1+v1∣|u_1+v_1| between η=2×10−4\eta=2\times10^{-4} and 3×10−43\times10^{-4} is responsible for HBdG,1\mathcal{H}_{\mathrm{BdG},1} becoming positive on Figure 4, since the waveform moves into a region of positive energy density.

For large η\eta, the vortex mode has a frequency lower than all positive frequency bulk excitations. At a critical η\eta value, which we call ηc\eta_c (the black dot in Figure 2(a)), the frequency of the vortex mode changes sign. Since the Bogoliubov norm of this mode is negative, the energy of the vortex mode is positive for η>ηc\eta> \eta_c and it will no longer get excited when a dissipation mechanism is included. Furthermore, since the negative frequency bulk excitations have negative norm, they have positive energy and the vortex mode cannot mix with these modes to produce a dynamical instability with zero energy. Above ηc\eta_c, the system is stabilised by the presence of the second component. The next section will explore the mechanism underpinning this stabilisation whilst Section IV will demonstrate stability via a full numerical simulation of (13).

From Figure 2(a), we find the value ηc=5.933×10−4\eta_c = 5.933 \times10^{-4}. Since the value obtained will in general depend on the resolution used in the numerics, we estimate that the infinite-resolution limit would yield the value ηc=5.936×10−4\eta_c = 5.936 \times10^{-4} (details in Appendix C). Note that this value is also specific to an ℓ=2\ell= 2 vortex in a trap of a radius rB=40r_B = 40 and will in general depend on both these parameters. Due to computational cost, we do not study this dependence here.

A computation of the mode energy in (21) gives insight into the behaviour around ηc\eta_c. In Figure 4, we integrate the (vortex mode) energy density components in (21) over space and plot as a function of η\eta. The total energy HBdGH_{\mathrm{BdG}} is negative at low η\eta except inside finite windows where it is vanishing due to the dynamical instability. At the edge of these windows, the amplitudes of the normalised mode diverges, as reflected in the components of HBdGH_{\mathrm{BdG}}, but does so in a way that keeps the total energy finite. The energy of species 2 fluctuations HBdG,2H_{\mathrm{BdG},2} is always positive whilst the interaction energy HBdG,intH_{\mathrm{BdG,int}} between fluctuations of each species is always negative. The energy of species 1 fluctuations HBdG,1H_{\mathrm{BdG},1} is initially negative but becomes positive around η=2.33×10−4\eta= 2.33 \times10^{-4}. Correspondingly, the maximum amplitude of the vortex mode waveform shifts outwards into a region where it has positive energy density (see Figure 3). Since HBdG,intH_{\mathrm{BdG,int}} is close to the value of −2HBdG,2-2H_{\mathrm{BdG},2} the total mode energy is still negative even when HBdG,1=0H_{\mathrm{BdG},1} = 0. Interestingly, in this regime, the energetic instability of the vortex mode is a result of the interaction between fluctuations of the two species, rather than being an inherent property of the vortex in species 1 alone. Increasing η\eta further, the total energy changes sign at ηc\eta_c. This coincides with HBdG,1H_{\mathrm{BdG},1} exceeding the value of HBdG,2H_{\mathrm{BdG},2}.

Figure 4. Vortex mode energy densities in (21) integrated over space. The total energy smoothly increases from negative to positive, except in finite windows where it vanishes due to the dynamical instability. The change in sign of the total energy is associated with the energy of fluctuations in species 1 exceeding that of species 2.

WKB method

To interpret why the vortex mode is shifted to lower frequency with increasing η\eta, we apply a WKB method. This method has been demonstrated to give insight into the physical origin of vortex instabilities [10] as well as providing reasonable estimates of the unstable mode frequencies [11]. Following the procedure outlined in [11], we first write a WKB Ansatz,

\begin{pmatrix} u_j \\ v_j \end{pmatrix} = \begin{pmatrix} A_j(r) \\ B_j(r) \end{pmatrix} e^{i\int p(r)\,\mathrm{d}r}, \tag*{(25)}\end{pmatrix}

where pp is the local component of the wavevector in the radial direction and we assume ∣p2∣≫∣∂rp∣|p^2| \gg|\partial_r p|, ∣pAj∣≫∣∂rAj∣|p A_j| \gg|\partial_r A_j| and similarly for BjB_j. Taking only the leading order contributions and eliminating the four amplitudes, (24) becomes,

D1(k1)D2(k2)=ε2k12k22,(26)\mathcal{D}_1(k_1)\mathcal{D}_2(k_2)=\varepsilon^2 k_1^2 k_2^2, \tag*{(26)}

where,

D1(k1)=Ω2−n1k12−k144,\mathcal{D}_1(k_1)=\Omega^2-n_1 k_1^2-\frac{k_1^4}{4},
D2(k2)=ω2−g2n2k22−k244,(27)\mathcal{D}_2(k_2)=\omega^2-g_2 n_2 k_2^2-\frac{k_2^4}{4}, \tag*{(27)}

and Ω=ω−mℓ/r2\Omega=\omega-m\ell/r^2 is the frequency in the frame rotating with the vortex. We have also defined a pair of effective wavevectors by,

k12=p2+m~12r2,M2k22=p2+m~22r2,(28)k_1^2=p^2+\frac{\widetilde{m}_1^2}{r^2},\qquad M_2 k_2^2=p^2+\frac{\widetilde{m}_2^2}{r^2}, \tag*{(28)}

where the effective azimuthal numbers are,

m~12=m2+ℓ2+2r2(n1+g12n2−μ1),\widetilde{m}_1^2=m^2+\ell^2+2r^2(n_1+g_{12}n_2-\mu_1),
m~22=m2+2M2r2(g2n2+g12n1−μ2).(29)\widetilde{m}_2^2=m^2+2M_2r^2(g_2n_2+g_{12}n_1-\mu_2). \tag*{(29)}

Hence, we can identify the condition in (26) as two coupled dispersion relations respectively describing excitations of components 1 and 2 interacting through a small overlap region where ε\varepsilon is non-zero. As a first approximation, we study the non-interacting case with ε→0\varepsilon\to0. We are particularly interested in the fluctuations of component 1 (since this is where the instability arises) which approximately satisfy D1(k1)≃0\mathcal{D}_1(k_1)\simeq0. This condition has two sets of solutions for pp which are decoupled from each other. The first type are evanescent over the whole system whereas the second can propagate. We are only interested in the second type. These are determined by,

p2+W+=0,W+=−2n12+Ω2+2n1+m~12r2.(30)p^2+W_{+}=0,\qquad W_{+}=-2\sqrt{n_1^2+\Omega^2+2n_1+\frac{\widetilde{m}_1^2}{r^2}}. \tag*{(30)}

Points where p=0p=0 correspond to turning points, i.e. radial locations where an incoming wave instantaneously comes to rest before reversing its direction and propagating back out [11]. We can therefore view W+W_{+} as an effective potential; when W+<0W_{+}<0 these waves are propagating whilst for W+>0W_{+}>0 they are evanescent.

To see the effect of adding a second component, one can expand the m~12\widetilde{m}_1^2 term in W+W_{+}:

W+=−2n12+Ω2+4n1−2μ1+m2+ℓ2r2+2g12n2.(31)W_{+}=-2\sqrt{n_1^2+\Omega^2+4n_1-2\mu_1+\frac{m^2+\ell^2}{r^2}}+2g_{12}n_2. \tag*{(31)}

In the absence of the second component, this potential has a cavity located about r=0r=0 where waves with correct frequency can become trapped as bound states. These states approximately satisfy cos⁡I≃0\cos I\simeq0 [11], where

I=∫cavp drI=\int_{\mathrm{cav}} p\,\mathrm{d}r

is the phase integral inside the cavity in the vortex core (see Appendix D for a further discussion). When a second species is present inside the vortex core, the final term in (31) creates a bulge in the potential around r=0r=0 (as shown in Figure 5) thereby decreasing the amount of space the mode has to fit in the cavity. This bulge inhibits wave propagation in the immediate vicinity of the vortex axis, explaining the feature of Figure 3 where the amplitude of the waveform drops on the vortex axis with increasing η\eta. Conventional wisdom gained from the 1D Schrödinger equation in quantum mechanics would dictate that this should raise the frequency of the bound state, however, this is not the case here as we now illustrate.

Plot of the potential $W_{+}$ for three values of $\eta$

Figure 5. The potential W+W_{+} from (31) with m=ℓ=2m=\ell=2 for an example frequency ω=0.75\omega=0.75. In the WKB approximation, excitations of component 1 propagate when W+<0W_{+}<0. Excitations of the vortex are bound states inside the cavity centred on the axis and correspond to the central vortex splitting into a cluster. A second component in the vortex core induces a bulge in the cavity, decreasing the amount of space for vortex excitations. As the bulge grows in size, the vortex excitation decreases its frequency until, for large enough η\eta, there are no positive frequency modes which fit inside the cavity.

To solve the resonance condition in the cavity, one should compute I(ω)I(\omega) for all frequencies which probe the cavity before finding the one which solves cos⁡I=0\cos I = 0. For small η\eta, we show in Appendix D that the integral near the bottom of the cavity (for m=ℓ=2m = \ell= 2) can be expanded as,

I(ω)=π(ω0−ω)∣2g12n2,0′′∣1/2+O[(ω0−ω)2],(32)I(\omega) = \frac{\pi(\omega_0-\omega)}{\lvert2g_{12}n''_{2,0}\rvert^{1/2}} + O[(\omega_0-\omega)^2], \tag*{(32)}

where ω0\omega_0 is the maximum frequency which probes the cavity and n2,0′′n''_{2,0} is the curvature of the density of component 2 on the axis. This expression demonstrates that II decreases as ω\omega increases, i.e. ∂ωI<0\partial_\omega I < 0. More generally, the reason for this difference with the standard Schrödinger picture (in which ∂ωI>0\partial_\omega I > 0) is that the vortex mode in the WKB picture is on the lower branch of the dispersion relation where Ω<0\Omega< 0 (which is related to the negative norm of the mode). On this branch, the root with p>0p > 0 has negative group velocity ∂pω<0\partial_p\omega< 0, i.e. decreasing ω\omega increases the value of pp. Therefore, since increasing η\eta will decrease pp through the final term in (31), the frequency of the bound state decreases to keep II fixed at the value satisfying the resonance condition.

Although the WKB method correctly predicts the decrease of the vortex mode frequency with increasing ω\omega, it fails at providing the quantitative value. This is to be expected since the approximation works best for high frequencies in regions where the background functions change gradually. Indeed, in [11], the WKB estimates are practically indistinguishable from numerical calculations for bulk excitations, whilst for the vortex mode (which occupies a region where the background density quickly changes) the WKB estimate differs notably from the numerics. When η≠0\eta\ne0, the frequency is lower than the η=0\eta= 0 case and the bulge in W+W_+ makes background variations more significant. Both of these features decrease the quantitative power of the WKB approximation.

Numerics

Having gained an understanding of the instability mechanism from the linear theory, we now perform a series of fully non-linear simulations of (13). The first goal will be to demonstrate explicitly the stabilisation mechanism, showing that for high η\eta values the ℓ=2\ell= 2 vortex fails to split. We then move on to discuss a by-product of the stabilisation mechanism: the counter-rotating orbital motion of a pair of co-rotating vortices.

One component

We first evolve (3) for η=0\eta= 0, i.e. when the second species is absent. For our spatial grid, we take x∈[−(rB+5),rB+5]x \in[-(r_B + 5), r_B + 5] and similarly for yy with N=512N = 512 spatial points. We take Δt=5×10−3\Delta t = 5 \times10^{-3} and evolve using a split-step Fourier method for 2×1052 \times10^5 steps in time, saving every 200th frame. For the initial condition, we seed the stationary vortex profile for ℓ=2\ell= 2 and rB=40r_B = 40 with the unstable mode of amplitude ∣aω(t=0)∣=10−3\lvert a_\omega(t = 0)\rvert= 10^{-3}. We note that the corresponding resolution in the radial direction is 256 points, which differs from the value of 800 used in Section II D. The reduced resolution slightly shifts the location of the instability bubbles in Figure 2. Hence, for consistency, we repeat the procedure in Section II D and III A for Δτ=5×10−3\Delta\tau= 5 \times10^{-3} and 256 radial grid points when producing the initial condition, since we want to precisely seed the mode which is unstable in our non-linear simulations. A comparison of the normal mode spectra at different resolutions can be found in Appendix C.

The results are shown in Figure 6. Panels (a-c) show the density at early, intermediate and late times, with constant density contour superimposed in red/white, and panels (d-f) show the phase at the same times. In panels (a) and (d), the vortex is close to its stationary profile whilst in panels (b) and (e), a deformation of the central vortex can be seen. In panels (c) and (f), the central vortex has split into two ℓ=1\ell= 1 vortices. The density contours indicate that this splitting is accompanied by the production of a bulk excitation, namely, a sound wave with m=2m = 2. At sufficiently late times, the vortices move back into the centre due to finite system effects studied in [10]. Since our focus is on the splitting mechanism, we do not discuss this behaviour further here. In the presence of dissipation, the initial splitting described here continues and the vortices separate widely, as we show in Section IV D.

Figure 6. Splitting of a vortex with ℓ=2\ell= 2 when η=0\eta= 0. The density is displayed on panels (a-c) at three different times. The constant density contours are set at 0.95 (red outer contour) and 0.05 (white inner contours). These help indicate the vortex splitting and the production of an m=2m = 2 sound wave in the bulk of the condensate. Panels (d-f) indicate the phase at the same times. In panel (g), we show the variation of the modulus of the m=0m = 0 and m=2m = 2 Fourier components of the density with time. The m=2m = 2 mode grows exponentially before saturating and decaying due to finite size effects [10]. The vertical black lines indicate the times t=600,825t = 600, 825. We display various components of the energy in panels (h-j), showing that the vortex splitting is associated with a decrease (increase) in incompressible (compressible/quantum) energy.

In panel (g), we perform a Fourier decomposition of the density n=(2π)−1∫dθeimθnmn = (2\pi)^{-1}\int d\theta e^{im\theta}n_m and display modulus of the Fourier components m=0,2m = 0, 2 at the location r=1.37r = 1.37. This location corresponds to the peak of the radial waveform nm=2(r)n_{m=2}(r). At first, the m=2m = 2 mode grows exponentially in line with the prediction of Figure 2. Around t=800t = 800, the m=2m = 2 mode reaches a maximum amplitude and starts to decay as the vortices move back into the centre. Near the maximum amplitude of nm=2n_{m=2}, we also observe the non-linear effect of the unstable mode on the m=0m = 0 component. This is simply the backreaction of the growing mode onto the background density profile, which can be physically understood as the condensate filling in the centre as the vortices move apart.

The instability is associated with a decrease of incompressible (vortex) energy and an increase in compressible (sound wave) energy. The energy of the system can be decomposed as,

E=Ekin+Eq+Epot+Eint,Ea=∫d2x Ea(33)E = E_{\rm kin} + E_q + E_{\rm pot} + E_{\rm int}, \qquad E_a = \int d^2x\,\mathcal{E}_a \tag*{(33)}
Ekin=12n∣∇Φ∣2,Eq=12∣∇n∣2,\mathcal{E}_{\rm kin} = \frac{1}{2}n|\nabla\Phi|^2, \qquad\mathcal{E}_q = \frac{1}{2}|\nabla\sqrt{n}|^2,

where the kinetic energy EkinE_{\rm kin} is associated with a non-vanishing velocity field and the quantum energy EqE_q results from density variations. Epot=Vn\mathcal{E}_{\rm pot}=Vn and Eint=n2/2\mathcal{E}_{\rm int}=n^2/2 are the usual potential and interactions energy densities respectively. The kinetic energy can be further split into compressible and incompressible parts by defining u=n∇Φ\mathbf{u}=\sqrt{n}\nabla\Phi and writing u=uc+ui\mathbf{u}=\mathbf{u}_c+\mathbf{u}_i, where the incompressible part obeys ∇⋅ui=0\nabla\cdot\mathbf{u}_i=0. Operationally, the compressible part can be found by performing a Fourier transform F\mathcal{F}, projecting u\mathbf{u} onto the wavevector k\mathbf{k} and computing the inverse Fourier transform F−1\mathcal{F}^{-1}. That is, uc=F−1[k(k⋅Fu)/∥k∥2]\mathbf{u}_c=\mathcal{F}^{-1}[\mathbf{k}(\mathbf{k}\cdot\mathcal{F}\mathbf{u})/\|\mathbf{k}\|^2] and ui=u−uc\mathbf{u}_i=\mathbf{u}-\mathbf{u}_c. Since uc\mathbf{u}_c and ui\mathbf{u}_i have zero overlap, the kinetic energy can be written Ekin=Ec+EiE_{\rm kin}=E_c+E_i, with Ec=∫d2x ucE_c=\int d^2x\,\mathbf{u}_c and similarly for EiE_i. We display EqE_q, EiE_i and EcE_c in panels (h), (i) and (j) respectively. The incompressible energy decreases as the vortices move apart (into a lower energy configuration) and the compressible and quantum energies increase in response. The two vertical black lines in the rightmost panels correspond to the times in panels (b) and (c) respectively.

Two components

To assess the effect of adding a small amount of component 2 inside the vortex core of component 1, we repeat the simulation of the last section for various values of η\eta. In particular, we start from the stationary profiles for condensate 1 and 2 and perturb the system by inserting the unstable mode with an amplitude 10−310^{-3}. The values of η\eta chosen correspond to the peaks of Im⁡[ω]\operatorname{Im}[\omega] in the stability windows at the chosen resolution.

In Figure 7, we display the densities of species 1 and species 2 for four values of η\eta at the same times depicted as the η=0\eta=0 case in Figure 1. As η\eta increases, we clearly see that the vortices have separated less at a given instant of time when compared with the single component system, thereby demonstrating the stabilising effect of the second species. Since the minute separation becomes less and less transparent from the density plots with increasing η\eta, we show on Figure 8 the growth of the m=2m=2 mode in the vortex core. Here, we clearly see that both the oscillation frequency and growth rate decrease with increasing η\eta. The values of Re⁡[ω]\operatorname{Re}[\omega] and Im⁡[ω]\operatorname{Im}[\omega] correspond to those predicted by the linear analysis in Figure 2 (see Appendix C for an explicit comparison).

Figure 7. The densities n1,2=∣Ψ1,2∣2n_{1,2} = \lvert\Psi_{1,2}\rvert^2 for various η\eta at the same times shown in Figure 1. On each panel, the density of component 1 is shown on the left (corresponding colorbar bottom left) and component 2 on the right (bottom right). The constant density contours for component 1 are set at 0.9 (red outer contour) and 0.005 (white inner contours) and help indicate the vortex splitting. As η\eta is increased, the central vortex in component 1 is less deformed at a given time, whilst the density of component 2 increases, becoming more and more visible on the colorscale.

Figure 8. Growth of the m=2m = 2 unstable mode for various η\eta values. For each η\eta, we look at the time dependence of the mode at the spatial location where n1,m=2n_{1,m=2} has maximum amplitude (which is around r∼1.6r \sim1.6–1.91.9). As η\eta increases, the oscillation frequency and growth rate decreases in accordance with the linear prediction of Fig. 2.

Orbit reversal

In general, the unstable mode in the spectrum of the MQV is associated with its splitting into a cluster of SQVs, whose distribution is determined by the mm-fold symmetry of the unstable mode. In single component systems, the cluster rotates in the direction of the net winding of the system, since each vortex essentially moves under the collective influence of all the others. The story is not so simple when a second component is added. The orbital frequency of the resulting cluster is given by Re⁡[ω]/m\operatorname{Re}[\omega]/m, with Re⁡[ω]\operatorname{Re}[\omega] the oscillation frequency of the unstable mode. Hence, the frequency inversion for η>ηc\eta>\eta_c seen in Figure 2 is associated with the reversal of the cluster’s orbital motion. We demonstrate this now with a numerical simulation.

We choose a series of η\eta values where the vortex is dynamically stable. Note here that dynamical stability below ηc\eta_c results from finite size effects for certain ranges of rBr_B, whilst above ηc\eta_c it persists for all trap sizes, as shown by Figure 2. Using the same resolution as the previous section, we now seed the stationary vortex profile with the vortex mode of amplitude ∣aω(t=0)∣=2|a_\omega(t=0)|=2. Due to the large amplitude, in all cases we observe a sudden burst of sound from the central vortex as the system relaxes through non-linear interactions. After this initial burst, we observe a stably orbiting vortex pair in the centre of the trap surrounded by small noise. On Figure 9, we display snapshots of the density profiles of the two species at three different times. The red arrows indicate the direction of the orbital motion. For the two leftmost columns, the orbital motion is counter-clockwise, corresponding to Re⁡[ω]>0\operatorname{Re}[\omega] > 0 for the vortex mode below ηc\eta_c. The atoms of species 2 are shared between the two ℓ=1\ell= 1 vortices in the pair and track their motion. Conversely, the orbital motion is clockwise on the two rightmost columns, corresponding to Re⁡[ω]<0\operatorname{Re}[\omega] < 0 above ηc\eta_c.

Figure 9. Vortex pair motion for various η\eta values at three different times. In each of the four columns, the left plot depicts ∣Ψ1∣2|\Psi_1|^2 (colorbar bottom left) and the right plot ∣Ψ2∣2|\Psi_2|^2 (colorbar bottom right). Each row corresponds to the time indicated on the left. The direction of the vortices is indicated with a red arrow and one of the vortices is marked with a white cross to help illustrate the motion. In panels (a1-3) the pair complete less than half a full orbit in the counter-clockwise sense between each snapshot. In panels (b1-3), the motion is also counter-clockwise with lower orbital frequency. In panels (c1-3), the motion is clockwise with low orbital frequency and in panels (d1-3) the motion is clockwise and slightly faster. The change in direction of the orbit is associated with a low density region between the vortex pair in component 1 and a high density in the same region in component 2.

Dissipation

In the presence of a dissipation mechanism, an MQV in a one-component condensate will still decay even if the vortex mode does not couple to any bulk excitations. The reason is that the vortex mode has negative energy, hence, its excitation lowers the total energy of the system, as required by the dissipation. The same is not necessarily true in a two-component system as we now show, since the vortex mode has a positive energy above ηc\eta_c.

For large η\eta we find that a region of depleted density in component 1 persists between the two separated vortices and is occupied by component 2. We interpret this as the result of the waveforms in Figure 3. For η=0\eta= 0, the maximum value of the waveform on the vortex axis non-linearly excites the m=0m = 0 mode as the vortices separate, causing atoms to fill in the space between them. For large η\eta, the waveform decreases approaching r=0r = 0, suppressing the non-linear excitation of m=0m = 0 in that region. Physically, the presence of component 2 on the vortex axis prevents excitations in component 1 propagating there, which means there is nothing to increase the value of the density of component 1 at r=0r = 0 as the vortices separate.

In Figure 10, we present results of three simulations performed with damping. The initial conditions in panels (a), (b) and (c) are, respectively, the final frames of the simulations shown in Figure 7(a), Figure 9(a) and Figure 9(d), corresponding to the same η\eta values indicated at the top of the figures. These respectively correspond to the lowest fraction of η\eta simulated which are unstable (panel (a)) and stable (panel (b)) when γ1=0\gamma_1 = 0 and the largest η\eta value simulated above ηc\eta_c (panel (c)). As η\eta is small, damping applied to atoms of the second species will quickly reduce N2N_2 to zero. Hence, we only apply damping to the first component. This is achieved by replacing i∂t\mathrm{i}\partial_t by (i−γ1)∂t(\mathrm{i}-\gamma_1)\partial_t in (13), taking the value γ1=5×10−2\gamma_1 = 5 \times10^{-2}.

Figure 10. Vortex pair dynamics with damping for various η\eta. In panels (a1-3), atoms of the second species stay confined to the SQV vortex cores when the MQV decays. In panels (b1-3) the vortices spiral out in an η\eta independent manner, with the species 2 atoms tracking the motion of the vortex cores. In panels (c1-3), η\eta is above the critical fraction, hence, the system lowers its energy by recombining into a single, central MQV.

In Figure 10(a1-3), the initial condition is a dynamically unstable vortex mode with low amplitude. In the three snapshots, we observe the central MQV splitting into a pair of SQVs rotating counter-clockwise with equal amounts of second component in the two cores. During the splitting, the second species remains confined to the vortex cores, rather than being dispersed across the system.

In panels (b1-3), the initial condition is two nearby SQVs surrounded by noise. The noise is quickly damped out and the vortices spiral out in the counter-clockwise sense with decreasing orbital frequency. In this regime, the vortices orbit in accordance with the regular point-vortex dynamics [1] which dictates the orbital frequency is approximately Ω=2/s2\Omega= 2/s^{2}, where ss is the separation between vortices, with the atoms of species 2 playing a negligible role in the dynamics. In accordance with this expectation, the dynamics observed for t>300t > 300 at η=3.4×10−5\eta= 3.4 \times10^{-5} (i.e. after the last frame shown on panel (a)) also follow the point-vortex behaviour. From this, we conclude that the modified dynamics in two component condensates is absent when species 2 atoms are split into distinct parcels which occupy the separate vortex cores. In other words, the stabilisation is a feature of a single parcel of species 2 atoms shared across the vortices.

To support this conclusion, in panels (c1-3), we take such a configuration as our initial condition. The corresponding vortex mode at this η\eta value has positive energy, hence, the separation of vortices (whilst retaining a single parcel of species 2 atoms) is associated with an energy gain relative to the stationary MQV. As the system evolves, the vortices orbit in the clockwise sense and damping reduces the vortex separation to lower the energy of the system. At late times, we recover a stationary, central MQV. This demonstrates that MQVs seeded with species 2 atoms above a critical fraction can be inherently stable objects even when a dissipation mechanism is present.

Discussion

The results presented here show that, in a two-component condensate confined in a bucket trap in the immiscible regime, multiply-quantised vortices (MQVs) in the first component can be stabilised when a relatively small number of atoms of the second component collect in the vortex core. We have found that if the fraction η\eta of atoms of the second component is small, the growth rate of the dynamically unstable mode which causes vortex splitting is smaller than when the second component is absent. We found there exists a critical fraction ηc\eta_c above which vortex decay is energetically unfavorable. Hence, for η>ηc\eta> \eta_c, MQVs are stabilised even in the presence of a dissipation mechanism since it would cost energy to separate them. A consequence of the stabilisation mechanism is that nearby co-rotating vortices can orbit in the opposite sense to their individual rotations when enough atoms of the second component are shared across the cores.

The parameters which we have used in this model were chosen such that an experiment would be reasonably feasible. The greatest challenge in performing this experiment would be creating a trapping potential profile that directly matches what we have used here; however, since the first realisation of a uniform trapping potential [31], advancements have been made in generating homogeneous condensates [32, 33], with traps of the size of ≈100 μm\approx100\,\mu\mathrm{m}. We also expect this phenomenon to persist in harmonic traps since the instability is a feature of the vortex core rather than the details of the trapping potential (see e.g. [9] for an example where the normal mode spectra of MQVs in harmonic and box-like trapping potentials are qualitatively the same in one component condensates).

Our results stand in contrast to those in [34], where a miscible, two-component condensate (with equal mass and self-interaction parameters) was shown to be more unstable than vortices in one component condensates. This is a result of the region where spin waves have negative energy density extending outside the vortex core. It is an interesting parallel then that in the opposite case, where the mixtures are immiscible, MQVs can be made more or completely stable.

Our results further demonstrate that even in dynamic scenarios, the atoms of the second component remain attached to the vortex cores rather than becoming scattered about the trap. Therefore the proposed method of MQV stabilisation could provide a solution to the problem of destructive imaging. The most common method to image a condensate is absorption imaging, which consists of illuminating the cold gas with a laser and measuring the shadow cast by absorption [35]. Although this method provides a detailed image of the condensate, irradiating the system drastically changes its energy and dynamics, and in the worst case destroys the condensate [36]. It is also possible to remove a small fraction of the atoms of a specific component of a condensate from a trap and to image them independently [37]. If one stabilised vortex cores with a second component, the use of a specific microwave pulse could outcouple only secondary component atoms, resulting in an image describing vortex core locations without affecting the dynamics of the primary component. As this stabilisation has proven robust, vortex cores undergoing rapid motion could be reliably imaged and tracked.

An added consequence of the stabilisation is that

MQVs (and also SQVs) are endowed with an extra degree of freedom relative to one component condensates where an excited vortex can only radiate sound. Here, the extra degree of freedom is the excitations of the second component which would allow for more complicated interactions in systems containing more than one vortex. This situation has some analogy with superfluid 3He{}^{3}\mathrm{He}, where vortices may have a double vortex core whose rotational symmetry about the vortex axis is broken [38]. By creating a natural ribbon, the double vortex structure would help the definition of superfluid helicity [39].

Another interesting question relates to the dynamics of vortices in 3D condensates. If atoms of the second component are unevenly distributed along a pair of nearby vortex filaments, different segments of the vortex filament could feasibly orbit in opposite directions. This may induce vortex reconnections, with potential consequences for superfluid turbulence in two-component mixtures.

A further possibility we raise involves the nucleation of vortices containing atoms of the second component. In a system of species 1 atoms in which vortices are absent, a small amount of species 2 could be added close to the edge of the trapping potential where the density drops to zero. Rotating the condensate would lead to vortex nucleation close to the edge of the trap, and it is feasible such vortices would carry atoms of the second species. Furthermore, as we have seen, for large enough proportions of atoms in species 2, MQVs can be energetically preferable to SQVs, raising the intriguing possibility of nucleating vortices with large winding numbers.

Finally, since the vortex mode has vanishing oscillation frequency at the critical fraction ηc\eta_c, it would be interesting to investigate the non-linear dynamics in the vicinity of this value to see if static (i.e. non-orbiting) configurations of vortices can be constructed.

Acknowledgments

We are grateful to Tom Billam and Connor F. Swales for discussions. For the purpose of open access, the author has applied a Creative Commons Attribution (CC BY) licence to any Author Accepted Manuscript version arising. This work was supported in part by the STFC Quantum Technology Grants ST/T005858/1 (SP & RG), ST/T006900/1 (CFB) and Kings College London (AG). RG also acknowledges support from the Perimeter Institute. Research at Perimeter Institute is supported by the Government of Canada through the Department of Innovation, Science and Economic Development Canada and by the Province of Ontario through the Ministry of Research, Innovation and Science.

Dimensional reduction

The action for the two component system in three-dimensional (3D) space is,

S3D=∫dtd2x[∑j(iℏΨj3D∗Ψ˙j3D−Hj3D)−HI3D].\mathcal{S}^{\mathrm{3D}}=\int dt d^{2}x\left[\sum_{j}\left(i\hbar\Psi_{j}^{\mathrm{3D}*}\dot{\Psi}_{j}^{\mathrm{3D}}-\mathcal{H}_{j}^{\mathrm{3D}}\right)-\mathcal{H}_{I}^{\mathrm{3D}}\right].

with,

Hj3D=Ψj3D∗[K^j3D+Vj3D(x)−μ~j3D]Ψj3D+Gj3D2∣Ψj3D∣4,\mathcal{H}_{j}^{\mathrm{3D}}=\Psi_{j}^{\mathrm{3D}*}\left[\hat{K}_{j}^{\mathrm{3D}}+V_{j}^{\mathrm{3D}}(x)-\tilde{\mu}_{j}^{\mathrm{3D}}\right]\Psi_{j}^{\mathrm{3D}}+\frac{G_{j}^{\mathrm{3D}}}{2}|\Psi_{j}^{\mathrm{3D}}|^{4},
HI=G123D∣Ψ13D∣2∣Ψ23D∣2.\mathcal{H}_{I}=G_{12}^{\mathrm{3D}}|\Psi_{1}^{\mathrm{3D}}|^{2}|\Psi_{2}^{\mathrm{3D}}|^{2}.

We now assume a tight confinement in the vertical (zz) direction and seek an action governing the dimensionally reduced dynamics in the (x,y)(x,y) plane. To this end, we write,

Vj3D(x,y,z)=V(x,y)+12Mjωj2z2,μ~j3D=μ~j+ℏωj2.V_{j}^{\mathrm{3D}}(x,y,z)=V(x,y)+\frac{1}{2}M_{j}\omega_{j}^{2}z^{2},\qquad\tilde{\mu}_{j}^{\mathrm{3D}}=\tilde{\mu}_{j}+\frac{\hbar\omega_{j}}{2}.

where V(x,y)V(x,y) and μ~j\tilde{\mu}_{j} are the 2D bucket trap and chemical potentials in the main text and the zz dependent piece of Vj3DV_{j}^{\mathrm{3D}} is a harmonic trap of frequency ωj\omega_{j}. The break down of the chemical potential in this way amounts to assuming the zz dependent piece of the wavefunction is the ground state if we also assume μ~j≪ℏωj\tilde{\mu}_{j}\ll\hbar\omega_{j} and Ψj3D(x,y,z,t)=Ψj2D(x,y,t)fj(z)\Psi_{j}^{\mathrm{3D}}(x,y,z,t)=\Psi_{j}^{\mathrm{2D}}(x,y,t)f_{j}(z). The 3D action can then be split as S3D=Sz+S\mathcal{S}^{\mathrm{3D}}=S_{z}+S where Sz∼O(ℏωj)S_{z}\sim O(\hbar\omega_{j}) and S∼O(μ~j)S\sim O(\tilde{\mu}_{j}). The vertical part is,

Sz=∫dtd2x∣Ψj2D∣2∫dz∑j(ℏωj2∣fj∣2−ℏ22Mj∣∂zfj∣2−12Mjωj2z2∣fj∣2).S_{z}=\int dt d^{2}x|\Psi_{j}^{\mathrm{2D}}|^{2}\int dz\sum_{j}\left(\frac{\hbar\omega_{j}}{2}|f_{j}|^{2}-\frac{\hbar^{2}}{2M_{j}}|\partial_{z}f_{j}|^{2}-\frac{1}{2}M_{j}\omega_{j}^{2}z^{2}|f_{j}|^{2}\right).

whose variation yields,

ℏ22Mj∂z2fj+(ℏωj2−12Mjωj2z2)fj=0,(34)\frac{\hbar^{2}}{2M_{j}}\partial_{z}^{2}f_{j}+\left(\frac{\hbar\omega_{j}}{2}-\frac{1}{2}M_{j}\omega_{j}^{2}z^{2}\right)f_{j}=0, \tag*{(34)}

which has solutions,

fj(z)=e−z2/2dj2,dj2=ℏMjωj.(35)f_{j}(z)=e^{-z^{2}/2d_{j}^{2}},\qquad d_{j}^{2}=\frac{\hbar}{M_{j}\omega_{j}}. \tag*{(35)}

The fjf_{j} can then be inserted into the 2D action and the zz integrals evaluated. Terms involving ∣Ψj2D∣2|\Psi_{j}^{\mathrm{2D}}|^{2} (including the derivative term), ∣Ψj2D∣4|\Psi_{j}^{\mathrm{2D}}|^{4} and ∣Ψ12DΨ22D∣2|\Psi_{1}^{\mathrm{2D}}\Psi_{2}^{\mathrm{2D}}|^{2} are proportional to the following zz integrals,

∫dzfj2=djπ,∫dzfj4=djπ/2,(36)\int dz f_{j}^{2}=d_{j}\sqrt{\pi},\qquad\int dz f_{j}^{4}=d_{j}\sqrt{\pi/2}, \tag*{(36)}

and,

∫dzf12f22=d12π,d12−2=d1−2+d2−2.(37)\int dz f_{1}^{2}f_{2}^{2}=d_{12}\sqrt{\pi},\qquad d_{12}^{-2}=d_{1}^{-2}+d_{2}^{-2}. \tag*{(37)}

Finally, we obtain the action in (5) once we make the identifications,

Ψj=(djπ)12Ψj2D,(38)\Psi_{j}=(d_{j}\sqrt{\pi})^{\frac{1}{2}}\Psi_{j}^{\mathrm{2D}}, \tag*{(38)}

and,

Gj=Gj3D2πdj,G12=d12G123Dπd1d2.(39)G_{j}=\frac{G_{j}^{\mathrm{3D}}}{\sqrt{2\pi}d_{j}},\qquad G_{12}=\frac{d_{12}G_{12}^{\mathrm{3D}}}{\sqrt{\pi}d_{1}d_{2}}. \tag*{(39)}

The interaction parameters are related to the scattering lengths aja_{j} through [30],

Gj3D=4πℏ2ajMj,G123D=2πℏ2a12M12(40)G_{j}^{\mathrm{3D}}=\frac{4\pi\hbar^{2}a_{j}}{M_{j}},\qquad G_{12}^{\mathrm{3D}}=\frac{2\pi\hbar^{2}a_{12}}{M_{12}} \tag*{(40)}

where M12=M1M2/(M1+M2)M_{12}=M_{1}M_{2}/(M_{1}+M_{2}) is the reduced mass of the two species. The dimensionless parameters defined in Section II C become,

g2=d1M1a2d2M2a1,g12=d12M1a122d2M12a1,(41)g_{2}=\frac{d_{1}M_{1}a_{2}}{d_{2}M_{2}a_{1}},\qquad g_{12}=\frac{d_{12}M_{1}a_{12}}{\sqrt{2}d_{2}M_{12}a_{1}}, \tag*{(41)}

or in terms of the harmonic trapping frequencies,

g2=M1ω2a2M2ω1a1,g12=1+M1/M21+M1ω1/M2ω2a122a1.(42)g_{2}=\sqrt{\frac{M_{1}\omega_{2}a_{2}}{M_{2}\omega_{1}a_{1}}},\qquad g_{12}=\frac{1+M_{1}/M_{2}}{\sqrt{1+M_{1}\omega_{1}/M_{2}\omega_{2}}}\frac{a_{12}}{\sqrt{2}a_{1}}. \tag*{(42)}

The vertical confinement applied to the two species is the same provided M1ω12=M2ω22M_{1}\omega_{1}^{2}=M_{2}\omega_{2}^{2}. In this case, we find,

g2=(M1M2)3/4a2a1,g12=1+M1/M21+(M1/M2)1/2a122a1.g_{2}=\left(\frac{M_{1}}{M_{2}}\right)^{3/4}\frac{a_{2}}{a_{1}},\qquad g_{12}=\frac{1+M_{1}/M_{2}}{\sqrt{1+(M_{1}/M_{2})^{1/2}}}\frac{a_{12}}{\sqrt{2}a_{1}}.

The s–wave scattering lengths for the 87Rb^{87}\mathrm{Rb} and 133Cs^{133}\mathrm{Cs} mixture are a1=100a0a_{1}=100a_{0}, a2=280a0a_{2}=280a_{0} and a12=650a0a_{12}=650a_{0} [26] where a0a_{0} is Bohr radius [30]. Using M1/M2=87/133M_{1}/M_{2}=87/133, we obtain the values,

g2=2.04,g12=5.65,(43)g_{2}=2.04,\qquad g_{12}=5.65, \tag*{(43)}

which are the values quoted in the main text.

Boundary conditions

As r→0r\to0, we have n1→0n_{1}\to0 and ϵ→0\epsilon\to0. The fields u1u_{1} and v1v_{1} each obey the Schrödinger equation, whose regular solutions are of the form,

u1∼Jm+ℓ(α+r),v1∼Jm−ℓ(α−r)(44)u_{1}\sim J_{m+\ell}(\alpha_{+}r),\qquad v_{1}\sim J_{m-\ell}(\alpha_{-}r) \tag*{(44)}

with α±2=2(1−g12n2±ω)\alpha_{\pm}^{2}=2(1-g_{12}n_{2}\pm\omega). Thus, the boundary conditions are,

m=ℓ:u1(r=0)=∂rv1∣r=0=0,m=−ℓ:∂ru1∣r=0=v1(r=0)=0,∣m∣≠ℓ:u1(r=0)=v1(r=0)=0.(45)\begin{aligned} m=\ell&: \quad u_{1}(r=0)=\partial_{r}v_{1}|_{r=0}=0,\\ m=-\ell&: \quad\partial_{r}u_{1}|_{r=0}=v_{1}(r=0)=0,\\ |m|\ne\ell&: \quad u_{1}(r=0)=v_{1}(r=0)=0. \tag*{(45)} \end{aligned}

The fields u2u_2 and v2v_2 remained coupled to one-another, although they do decouple from u1u_1 and v1v_1. To find their asymptotic form, we can expand in power series about r=0r=0 and apply the method of Frobenius. Writing,

u2=∑n=0∞bn+rn+p+,v2=∑n=0∞bˉn−rn+p−,(46)u_2 = \sum_{n=0}^{\infty} b_n^+ r^{n+p^+}, \qquad v_2 = \sum_{n=0}^{\infty} \bar{b}_n^- r^{n+p^-}, \tag*{(46)}

the equations of motion become,

∑n=0∞[(n+p+)2−m2]bn+rn+p++∑n=2∞[β+bn−2+−2g2n2bn−2+]rn+p+=0,(47)\sum_{n=0}^{\infty} \left[(n+p^+)^2-m^2\right]b_n^+ r^{n+p^+} +\sum_{n=2}^{\infty}\left[\beta^+b_{n-2}^+-2g_2n_2b_{n-2}^+\right]r^{n+p^+}=0, \tag*{(47)}
∑n=0∞[(n+p−)2−m2]bˉn−rn+p−+∑n=2∞[β−bˉn−2−−2g2n2bˉn−2−]rn+p−=0,\sum_{n=0}^{\infty} \left[(n+p^-)^2-m^2\right]\bar{b}_n^- r^{n+p^-} +\sum_{n=2}^{\infty}\left[\beta^-\bar{b}_{n-2}^--2g_2n_2\bar{b}_{n-2}^-\right]r^{n+p^-}=0,

with β±=2(μ2−2g2n2±ω)\beta^\pm=2(\mu_2-2g_2n_2\pm\omega). The first term in the series yields the values of p±p^\pm. The regular solutions are each given by p±=∣m∣p^\pm=|m|. Hence the asymptotics at the origin are u2∼r∣m∣u_2\sim r^{|m|} and v2∼r∣m∣v_2\sim r^{|m|} leading to the boundary conditions,

m=0:∂ru2∣r=0=∂rv2∣r=0=0,m=0:\quad\left.\partial_r u_2\right|_{r=0}=\left.\partial_r v_2\right|_{r=0}=0,
m≠0:u2(r=0)=v2(r=0)=0.(48)m\ne0:\quad u_2(r=0)=v_2(r=0)=0. \tag*{(48)}

Resolution

From the procedure outlined in Section II D, we obtain the chemical potentials μj0=μj+Δμj\mu_j^0=\mu_j+\Delta\mu_j, where μj\mu_j is the true value that one would obtain in the continuum limit. By checking for a range of resolutions, we observe that the error Δμj\Delta\mu_j scales linearly with ϵτ=Δτ\epsilon_\tau=\Delta\tau and ϵr=Δr2\epsilon_r=\Delta r^2. This error will offset the normal mode frequencies calculated in Figure 2, and therefore the value of ηc\eta_c obtained from looking where the vortex mode has zero frequency.

We account for this offset using the following linear perturbation theory argument. We start with the resolution used to produce Figure 2, ϵτ0=4×10−4\epsilon_\tau^0=4\times10^{-4} and ϵr0=(45/800)2\epsilon_r^0=(45/800)^2, and expand the vortex mode frequency in the vicinity of the zero crossing,

ω=ω0+∂ηω0(η−ηc0),(49)\omega=\omega^0+\partial_\eta\omega^0(\eta-\eta_c^0), \tag*{(49)}

where ηc0\eta_c^0 is our prediction for the critical η\eta value at this resolution and ω0=0\omega^0=0 by definition. The gradient ∂ηω0\partial_\eta\omega^0 is measured from the Figure 2. Now consider a change in the resolution. The expansion above to leading order becomes,

ω=ω0+δω+∂ηω0(η−ηc0),(50)\omega=\omega^0+\delta\omega+\partial_\eta\omega^0(\eta-\eta_c^0), \tag*{(50)}

where δω\delta\omega is the shift in the value of ω\omega at ηc0\eta_c^0. Solving ω=0\omega=0 gives the new value of ηc\eta_c,

ηc=ηc0−δω∂ηω0.(51)\eta_c=\eta_c^0-\frac{\delta\omega}{\partial_\eta\omega^0}. \tag*{(51)}

Now, to compute ηc\eta_c at infinite resolution, we need to estimate δω\delta\omega as we take ϵτ\epsilon_\tau and ϵr\epsilon_r to zero. To do this, we use the fact that ω\omega depends linearly on ϵτ\epsilon_\tau and ϵr\epsilon_r and expand the value of ω\omega at ηc0\eta_c^0,

ω(ηc0)=ωτ0(ϵτ−ϵτ0)+ωr0(ϵr−ϵr0),(52)\omega(\eta_c^0)=\omega_\tau^0(\epsilon_\tau-\epsilon_\tau^0)+\omega_r^0(\epsilon_r-\epsilon_r^0), \tag*{(52)}

with,

ωτ0≡∂ω∂ϵτ∣ϵτ0,ϵr0,ωr0≡∂ω∂ϵr∣ϵτ0,ϵr0.(53)\omega_\tau^0\equiv\left.\frac{\partial\omega}{\partial\epsilon_\tau}\right|_{\epsilon_\tau^0,\epsilon_r^0},\qquad \omega_r^0\equiv\left.\frac{\partial\omega}{\partial\epsilon_r}\right|_{\epsilon_\tau^0,\epsilon_r^0}. \tag*{(53)}

We can estimate these gradient terms by independently varying the resolutions away slightly from their initial values. Once ωr0\omega_r^0 and ωτ0\omega_\tau^0 are known, the frequency shift at infinite resolution is given by,

δω=−ωτ0ϵτ0−ωr0ϵr0(54)\delta\omega=-\omega_\tau^0\epsilon_\tau^0-\omega_r^0\epsilon_r^0 \tag*{(54)}

The expression for ηc\eta_c in the continuum limit is therefore,

ηc=ηc0+ωτ0ϵτ0+ωr0ϵr0∂ηω0.(55)\eta_c=\eta_c^0+\frac{\omega_\tau^0\epsilon_\tau^0+\omega_r^0\epsilon_r^0}{\partial_\eta\omega^0}. \tag*{(55)}

Using ηc0=5.933×10−4\eta_c^0=5.933\times10^{-4}, we estimate the true value to be ηc=5.936×10−4\eta_c=5.936\times10^{-4}.

When performing the non-linear simulations of Section IV, it was not computationally feasible to use the same resolution used for Figure 2. In Figure 11, we compare the spectrum for Δτ=4×10−4\Delta\tau=4\times10^{-4} and 800 radial points (red curves) with the same for Δτ=5×10−3\Delta\tau=5\times10^{-3} and 256 radial points (black lines). Bulk excitations (horizontal lines) are insensitive to the change since they occupy a region where there are a large number of grid points. The main difference is in the vortex mode frequency, since this mode is covered by far fewer grid points at lower resolution. This leads to a slight displacement of the instability bubbles in the plot of Im⁡[ω]\operatorname{Im}[\omega]. We also show with black dots the values of Re⁡[ω]\operatorname{Re}[\omega] and Im⁡[ω]\operatorname{Im}[\omega] extracted from the full numerics, namely Figure 8. We observe good agreement for all η\eta values, although the largest η\eta point is displaced slightly above its true value on the plot of Im⁡[ω]\operatorname{Im}[\omega]. We interpret this as being due to finite size effects similar to those studied in [10], which onset earlier for this particular mode and make it more difficult to extract the growth rate.

Figure 11. Normal mode spectra for resolutions Δτ=4×10−4\Delta\tau= 4 \times10^{-4} and 800 radial points (red) used in Figure 2 and Δτ=5×10−3\Delta\tau= 5 \times10^{-3} and 256 radial points (black) comparable to that used in § 1.3.

Small rr integral

Here we evaluate the phase integral inside the vortex core in the small rr approximation. The integral is

I=∫r−r+−W+ dr.(56)I=\int_{r_-}^{r_+}\sqrt{-W_+}\,dr. \tag*{(56)}

For W+(0)>0W_{+}(0)>0, r+r_{+} are the smallest rr values for which W+=0W_{+}=0 whilst for W+(0)<0W_{+}(0)<0, r−=0r_{-}=0 and r+r_{+} is the smallest rr value for which W+=0W_{+}=0. Close to the axis, we can expand W+W_{+} treating rr as small

W+=(m−ℓ)2r2−2(1−g12n2,0−ω)+g12n2,0′′r2+(4n1,0′′′′+2n2,0′′′′)r4/4!+O(r6),(57)W_{+} = \frac{(m-\ell)^2}{r^2} - 2(1-g_{12}n_{2,0}-\omega)+g_{12}n_{2,0}^{\prime\prime}r^2 + (4n_{1,0}^{\prime\prime\prime\prime}+2n_{2,0}^{\prime\prime\prime\prime})r^4/4!+O(r^6), \tag*{(57)}

where we have expanded n1,2n_{1,2} up to and including quartic order terms. Since the small rr behaviour of component 1 is n1∼r2ℓn_1\sim r^{2\ell}, we assume at this stage we have an ℓ=2\ell=2 vortex so that n0,1′′′′≠0n_{0,1}^{\prime\prime\prime\prime}\ne0. To perform the calculation for ℓ≠2\ell\ne2, one would need to retain larger terms in the rr expansion. The unstable mode for the doubly quantised vortex occurs for m=ℓ=2m=\ell=2, hence we can write W+W_{+} in the form,

W+=2ω−2U−αr2+βr4,U=1−g12n2,0(58)W_{+}=2\omega-2U-\alpha r^2+\beta r^4,\qquad U=1-g_{12}n_{2,0} \tag*{(58)}
α=−g12n2,0′′,β=(4n1,0′′′′+2n2,0′′′′)/4!\alpha=-g_{12}n_{2,0}^{\prime\prime},\qquad\beta=(4n_{1,0}^{\prime\prime\prime\prime}+2n_{2,0}^{\prime\prime\prime\prime})/4!

where α\alpha and β\beta are positive constants. For small enough η\eta the minimum of W+W_{+} will be located in the region where this expansion is valid. The minimum is r0=α/2βr_0=\sqrt{\alpha/2\beta} and the value of the potential there is W0=2(ω−ω0)W_0=2(\omega-\omega_0) where ω0=U+α2/8β\omega_0=U+\alpha^2/8\beta is the maximum frequency which propagates in the cavity and is positive for small η\eta. Now we evaluate the cavity integral for ω\omega close to ω0\omega_0. For this we expand W+W_{+} this time around r0r_0,

W+=W0+2α(r−r0)2.(59)W_{+}=W_0+2\alpha(r-r_0)^2. \tag*{(59)}

In this approximation r±=r0±−W0/2αr_{\pm}=r_0\pm\sqrt{-W_0/2\alpha}. We can then compute II,

I=∫r−r+dr−W0−2α(r−r0)2=−W02α∫−11dz1−z2=−πW022α,(60)\begin{aligned} I=\int_{r_-}^{r_+}dr\sqrt{-W_0-2\alpha(r-r_0)^2} \\ &=\frac{-W_0}{\sqrt{2\alpha}}\int_{-1}^{1}dz\sqrt{1-z^2}=-\frac{\pi W_0}{2\sqrt{2\alpha}}, \tag*{(60)} \end{aligned}

which can be written as,

I(ω)=π(ω0−ω)∣2g12n2,0′′∣1/2+O[(ω0−ω)2],(61)I(\omega)=\frac{\pi(\omega_0-\omega)}{|2g_{12}n_{2,0}^{\prime\prime}|^{1/2}}+O[(\omega_0-\omega)^2], \tag*{(61)}

as stated in (32) of the main text. It is also instructive to write ω0\omega_0 as,

ω0=1−g12(n0,2+n0,2′′r02/4),(62)\omega_0=1-g_{12}(n_{0,2}+n_{0,2}^{\prime\prime}r_0^2/4), \tag*{(62)}

where the term in parentheses is slightly smaller than the value of n2n_2 at r0r_0. Since n2n_2 increases with η\eta, we deduce that ω0\omega_0 decreases as η\eta is increased from 0.

References

  1. [3]S. K. Adhikari and P. Muruganandam, Phys. Lett. A 301, 333 (2002).
  2. [4]R. Srinivasan, Pramana 66, 3 (2006).
  3. [5]P. Kuopanportti, N. V. Orlova, and M. V. Milošević, Phys. Rev. A 91, 043605 (2015).
  4. [6]H. Pu, C. K. Law, J. H. Eberly, and N. P. Bigelow, Phys. Rev. A 59, 1533 (1999).
  5. [7]Y. Shin, M. S. Saba, M. Vengalattore, T. A. Pasquini, C. Sanner, A. E. Leanhardt, M. Prentiss, D. E. Pritchard, and W. Ketterle, Phys. Rev. Lett. 93, 160406 (2004).
  6. [8]T. Isoshima, M. Okano, H. Yasuda, K. Kasa, J. A. M. Huhtamäki, M. Kumakura, and Y. Takahashi, Phys. Rev. Lett. 99, 200403 (2007).
  7. [9]L. Giacomelli and I. Carusotto, Phys. Rev. Research 2, 033139 (2020).
  8. [10]S. Patrick, A. Geelmuyden, S. Erne, C. F. Barenghi, and S. Weinfurtner, Phys. Rev. Research 4, 043104 (2022).
  9. [11]S. Patrick, A. Geelmuyden, S. Erne, C. F. Barenghi, and S. Weinfurtner, Phys. Rev. Research 4, 033117 (2022).
  10. [12]J. Qin, G. Dong, and B. A. Malomed, Phys. Rev. A 94, 053611 (2016).
  11. [13]C. J. Myatt, E. A. Burt, R. W. Ghrist, E. A. Cornell, and C. E. Wieman, Phys. Rev. Lett. 78, 586 (1997).
  12. [14]D. S. Hall, M. R. Matthews, J. R. Ensher, C. E. Wieman, and E. Cornell, Phys. Rev. Lett. 81, 1539 (1998).
  13. [15]D. J. McCarron, H. W. Cho, D. L. Jenkin, M. P. Köppingер, and S. L. Cornish, Phys. Rev. A 84, 011603 (2011).
  14. [16]S. B. Papp, J. M. Pino, and C. E. Wieman, Phys. Rev. Lett. 101, 040402 (2008).
  15. [17]J. Kronjäger, C. Becker, M. Brinkmann, R. Walser, P. Navez, K. Bongs, and K. Sengstock, Phys. Rev. A 72, 063619 (2005).
  16. [18]S. Tojo, Y. Taguchi, Y. Masuyama, T. Hayashi, H. Saito, and T. Hirano, Phys. Rev. A 82, 033609 (2010).
  17. [19]P. Kuopanportti, S. Bandyopadhyay, A. Roy, and D. Angom, Phys. Rev. A 100, 033615 (2019).
  18. [20]P. Mason, J. Phys. B: Atomic, Molecular and Optical Physics 50, 165002 (2017).
  19. [21]X. Zhang, L. Wen, C. Dai, R. Dong, H. Jiang, H. Chang, and S. Zhang, Sci. Rep. 6, 1 (2016).
  20. [22]B. Dong, L. Wang, G. Chen, W. Han, S. Zhang, and X. Zhang, Ann. Phys. 373, 178 (2016).
  21. [23]X. Ma, R. Driben, B. A. Malomed, T. Meier, and S. Schumacher, Sci. Rep. 6, 34847 (2016).
  22. [24]Y. Kawaguchi and T. Ohmi, Phys. Rev. A 70, 043610 (2004).
  23. [25]P. Ao and S. T. Chui, Phys. Rev. A 58, 4836 (1998).
  24. [26]T. Ho and V. B. Shenoy, Phys. Rev. Lett. 77, 3276 (1996).
  25. [27]E. Timmermans, Phys. Rev. Lett. 81, 5718 (1998).
  26. [28]C. Wang, C. Gao, C. Jian, and H. Zhai, Phys. Rev. Lett. 105, 160403 (2010).
  27. [29]C. Wu, I. Mondragon-Shem, and X. Zhou, Chin. Phys. Lett. 28, 097102 (2011).
  28. [30]R. W. Pattinson, T. P. Billam, S. A. Gardiner, D. J. McCarron, H. W. Cho, S. L. Cornish, N. G. Parker, and N. P. Proukakis, Phys. Rev. A 87, 013625 (2013).
  29. [31]N. Navon, A. L. Gaunt, R. P. Smith, and Z. Hadzibabic, Nature 539, 72 (2016).
  30. [32]S. P. Johnstone, A. J. Groszek, P. T. Starkey, C. J. Billington, T. P. Simula, and K. Helmerson, Science 364, 1267 (2019).
  31. [33]G. Gauthier, M. T. Reeves, X. Yu, A. S. Bradley, M. A. Baker, T. A. Bell, H. Rubinstein-Dunlop, M. J. Davis, and T. W. Neely, Science 364, 1264 (2019).
  32. [34]A. Berti, L. Giacomelli, and I. Carusotto, arXiv preprint arXiv:2212.07337 (2022).DOI
  33. [35]B. P. Anderson and M. A. Kasevich, Phys. Rev. A 59, R938 (1999).
  34. [36]T. Pyragius, arXiv preprint arXiv:1209.3408 (2012).arxiv.org/abs/1209.3408
  35. [37]S. Serafini, M. Barbiero, M. Debortoli, S. Donadello, F. Larcher, F. Dalfovo, G. Lamporesi, and G. Ferrari, Phys. Rev. Lett. 115, 170402 (2015).
  36. [38]E. V. Thuneberg, Phys. Rev. Lett. 56, 359 (1986).
  37. [39]H. Salman, Proc. Roy. Soc. A 473, 20160853 (2017).

Paper details

vlnc:W_29y2m3ka
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents