WorksPhysics and Astronomy
i-SPin: An integrator for multicomponent Schrödinger-Poisson systems with self-interactions
Abstract
We provide an algorithm and a publicly available code to numerically evolve multicomponent Schrödinger-Poisson (SP) systems with a SO() symmetry, including attractive or repulsive self-interactions in addition to gravity. Focusing on the case where the SP system represents the non-relativistic limit of a massive vector field, non-gravitational self-interactions (in particular spin-spin interactions) introduce complexities related to mass and spin conservation which are not present in purely gravitational systems. We address them with an analytical solution for the `kick' step in the algorithm, where we are able to decouple the multicomponent system completely. Equipped with this analytical solution, the full field evolution is second order accurate, preserves spin and mass to machine precision, and is reversible. Our algorithm allows for an expanding universe relevant for cosmology, and the inclusion of external potentials relevant for laboratory settings.
Introduction
Physical systems described by Schrödinger(-like) equations are ubiquitous in many areas of physics, ranging from (ultra-)light dark matter in cosmology [11, 12, 13, 14] [1, 2, 3, 4, 5, 6, 7, 8, 9, 10], to terrestrial systems in non-linear optics [15, 16, 17, 18, 19, 20, 21, 22], water waves [23, 24, 25] and Bose-Einstein condensates (BEC) [26, 27, 28, 29, 30, 31, 32]. Apart from long-range interactions such as gravity in the astrophysical/cosmological context, or external potential in laboratory settings, the Schrödinger field can have point-like quartic self-interactions, with its evolution determined by time-dependent Gross–Pitaevskii(-like) equations.
For a single Schrödinger field, the self-interaction only depends on the number density, and there are many numerical integrators available in the literature. In the astrophysical context, see the algorithms in [33, 34, 35, 36, 37, 38, 39] [40] for simulating the dynamics of ultralight scalar dark matter with only gravitational interaction, and [41, 42] where quartic self-interaction was also included.
For a single -component Schrödinger field, however, there are additional self-interaction terms possible, that do not just depend on the number density. When the system respects an symmetry, it is possible to have isospin-isospin interactions, where the total isospin is the conserved charge associated with the symmetry. Such interactions mix the different field components, and the extension of the scalar algorithm towards multicomponent Schrödinger field is complex.1 We are interested in exploring the nonlinear dynamics associated with such interactions (alongside other interactions), which necessitates developing a numerical algorithm capable of simulating dynamics faithfully. We present a new algorithm that evolves a multicomponent Schrödinger-Poisson system with (iso-)spin-spin interaction, that preserves both isospin and mass to machine precision. We also extend the algorithm to allow for dynamics in an expanding universe.
Such multicomponent Schrödinger systems with these additional interactions, are present in many areas of physics. For 3-component systems representing vector dark matter, where the aforementioned isospin is the intrinsic spin, spin-spin type interactions can be naturally present. For instance, they are present in the low-energy effective theory of vector dark matter arising from an Abelian Higgs model (with a heavy Higgs field). These interactions can lead to interesting modifications to the phenomenology of vector dark matter formation [43–49], black-hole superradiance [50–53], and impact nonlinear small-scale structure in such dark matter [11, 12, 13] [54]. As discussed in [14] [55], they also remove the energy degeneracy between polarized vector solitons which could potentially impact their cosmological population. Studying such phenomenology, with an eye towards nonlinear spin dynamics, is one of our main motivations.
In a laboratory context, multicomponent/spinor BECs can also include spin-spin type of interactions [26, 27, 28] [56]. Our algorithm is equally applicable in this context and can accommodate external potentials such as harmonic traps. There already exists literature on simulating such spin-spin interactions for spin-1 BECs [57, 58], where the three spin multiplicity fields are mixed, and evolved together.2 Our algorithm on the other hand, evolves the field in a “Cartesian” basis where we are able to decouple the different field components. Because of this, we can easily evolve a general Schrödinger system with a symmetry.
The organization of this paper is as follows. In section 2 we begin by writing down the non-relativistic Schrödinger-Poisson system having both gravitational and point-like self-interactions, lay down the various conserved quantities, and also re-scale the system to work with dimensionless quantities. In section 3 we first review the split-step Fourier algorithm, generally employed for evolving a single Schrödinger field (scalar system). We then develop an algorithm to evolve vector Schrödinger-Poisson system, containing the point-like spin-spin interaction. In section 4 we provide some numerical tests to verify the convergence of our algorithm, its unitary and spin conserving property, along with its time reversibility feature. Then in section 5, we generalize our algorithm to allow for an expanding background as well as evolution of a general -component Schrödinger system with a symmetry. Finally in section 6, we summarize our work. A collection of appendices provide a derivation of the nonrelativistic action, fluid and spin conservation equations, and polarized soliton solutions in vector fields including both gravitational and non-gravitational interactions.
Spin-1 Schrödinger-Poisson system
We begin with a 3-component Schrödinger-Poisson system with symmetry with non-relativistic massive vector fields in mind. That is, the transformation (with ) of the Schrödinger field , leaves the action unchanged.3 On account of this, we have the following general action that includes both Newtonian gravity and point self-interactions
Here, the first two terms dictate the usual free field evolution (of each of the field component ), while the third and fourth terms account for the Gauss’ law for Newtonian gravity where only the mass density contributes to the Newtonian potential . Finally, the last term accounts for point interactions of the vector field , and takes the following form for quartic self-interaction
In terms of the number density and spin density , the spin-spin interaction becomes apparent:
This admits the following Schrödinger-Poisson system of equations
The above form of the potential in eq. (2) arises from the relativistic quartic potential
upon taking the non-relativistic limit of an effective theory of a self-interacting massive spin-1 field that is minimally coupled to gravity. See [14] [55] for details. For completeness, we also provide salient aspects of this derivation in appendix § 8.§ 1.
Such a quartic potential naturally arises in the low energy effective field theory of the Abelian Higgs model, when the heavy Higgs field is integrated out of the spectrum [14] [55].4 In this case , dictating attractive self-interaction. This is easily understood on account of the (heavy) scalar particle exchange. While the case of repulsive self-interaction is naturally realized with multiple spin-1 fields (Yang-Mills structure) with appropriate modifications to the form of the potential in (2) [14], in this paper we take a more phenomenological point of view and allow to also take negative values in (2) without including multiple spin-1 fields.
Extension to the FLRW universe can be made by replacing and in (4). Here is the scale factor and is the Hubble parameter. This extension is discussed further in § 5. In the same section, we also discuss more general -component Schrödinger-Poisson systems, i.e. , with a symmetry. In this case, the factors in front of the two terms in (2) can take arbitrary values.
Conserved quantities
The various conserved quantities associated with our non-relativistic system (1) are
(particle number and rest mass) (2.6)
(linear momentum) (2.7)
(energy) (2.8)
(spin angular momentum) (2.9)
(orbital angular momentum) (2.10)
In general, these conserved quantities find a natural extension for systems with symmetry. While for the scalar case () spin is trivially zero, for multicomponent systems () in general, ‘spin’ is the charge associated with the invariance. See § 5 ahead.
In appendix § 8.§ 2, we provide local conservation equations for mass and momentum conservation (multicomponent Madelung equations including self-interactions), as well as the continuity equation for spin.
Re-scaled system
In what follows, we shall work with dimensionless quantities. For this purpose, we rescale the fields, space and time in the following fashion
Here, serves as a measure of the characteristic time-scales present in the system. In terms of these quantities, the Schrödinger-Poisson system becomes
where is the re-scaled number density, and
In the description of the algorithm, as well as our numerical code, we use this scaled, dimensionless system of equations (7). The only choices to be made are the value of and the initial conditions for the field. From now on we shall remove all the tildes to de-clutter our presentation with the understanding that all quantities are dimensionless.
Time evolution
We first review the usual algorithm for scalar SP system (including self-interactions), that is employed in the literature. See [63] for a broad overview of different integrators used for nonlinear time-dependent Schrödinger type equations. Building upon some of the key concepts discussed from the scalar case, we will develop an algorithm for the vector case that includes point-like self-interactions in subsequent sections.
The presentation below is somewhat formal, and the reader interested in getting to the self-interacting vector algorithm can skip directly to § 3.2.1.
A review of the scalar system
For the scalar case, the self-interaction only depends on the number density (), leading to the following evolution of the Schrödinger field
= \mathcal{H}_{\mathrm{drift}} + \mathcal{H}_{\mathrm{kick}}
where we have defined and as ‘drift’ and ‘kick’ Hamiltonians respectively. Explicitly, the Hamiltonian density has a position basis representation , where the non-local drift and local (diagonal) kick Hamiltonian densities are
Here, is the Fourier transform matrix element and .5 With this representation, the formal solution of (3.1) is
Here, “T” stands for time-ordering and .6 The evolution operator satisfies the unitarity relationship
The unitary evolution operators related to the ‘drift’ and ‘kick’ parts of the Hamiltonian are
In terms of and , the formal solution in (10) becomes7
Note that and in are evaluated after the first half-drift is performed. This is crucial for accuracy of and to maintain time-reversal symmetry.
To see this reversibility in effect, let us first evolve the system forward for half a time step of by using and :
where is evaluated using the field obtained after the half-drift has been performed. In conjugation to this, let us now evolve the system backwards from to using time reversed operations:
Here too, is evaluated using the field after the half-drift has been performed. We can invert the above eq. (16) to get
and use it in (15) to get the full unitary evolution from to :
Now note that since each of the exponential operators is time reversible, i.e. , the above expression reduces to eq. (14), proving that the algorithm is indeed time reversible. Furthermore, the evolution accuracy is now and the error is . This can be seen by noting that the re-evaluation of the kick Hamiltonian after the drift step takes into account the leading order correction due to the time integral in (10). See [39] for some details.
It is worth noting that for the case under consideration, the kick operator only rotates the phase of the field at every point in space, while itself only depending upon the norm . As a result, the kick operator doesn’t change with time during the kick operation. That is, it is manifestly true that evaluated using the backward half-drifted field from , is equal to evaluated using the forward half drifted field from .
Algorithm summary
Starting with the field at time , it is evolved through a time step according to the differential equation
Then, it is evolved through a time step according to the differential equation
where and overbar represents spatial average.8 Finally, it is again evolved using the following differential equation through a time step
Note that for every differential evolution above, the corresponding integral evolution is exact. While the half drift steps result in accuracy in the set of drift-kick-drift operations, the successive computation of the kick (after the performance of drift) ensures reversibility. Every operation being unitary ensures unitary evolution throughout.
Vector system
For the vector case, the differential evolution takes a matrix form
In particular, note the non-diagonal piece, , in . The Hamiltonian density, written explicitly in the position basis, is
and as always, we have suppressed the time dependence of the fields. Here the latin indices run over the field components, with the (unitary) integral evolution for this system being
The unitary condition is the following
With these definitions at hand, we shall employ the previously discussed algorithm for scalars, with appropriate modifications appearing due to the non-diagonal kick in the vector SP system.
Algorithm summary
The algorithm proceeds as follows. Starting with the field components at time , they are ‘drifted’ through a time step according to
Then, every component is ‘kicked’ through a time step according to
where recall that , and . The operator is evaluated using the field ; its evaluation based on an analytic solution of the kick equation is the key step, which we discuss in detail in the next subsection. The explicit expression for is provided in (37). Once this operation is performed, the fields are again drifted through a time step
The half drift steps in the set of operations ensure accuracy, while successive computation of the kick ensures reversibility. Since every operation is unitary, the algorithm conserves total mass. It also conserves total spin.
Kick step
We want an exact solution for the kick operation , in order to ensure reversibility along with mass and spin conservation. This was trivial for the scalar and vector case with no spin-spin interactions since was a constant of motion throughout the kick step. In that case the solution for the kick step was the exponential of the kick Hamiltonian density. With self-interactions, due to the non-diagonal piece in , the evolution of each of the field component becomes convoluted via the mixing of the field components and non-linearity of the kick differential equation. In this section we work with the kick step governed by the following differential equation
Working with the vector system, we first quickly outline an approach where the solution to the above 3-level system is obtained by exponentiating a time-independent matrix. Upon adding and subtracting in the right hand side, we note that the off-diagonal term is the skew-symmetric Hermitian matrix , where is the spin density.9 It can be easily seen that both the spin density and the number density (hence also the gravitational potential ) are conserved throughout the kick step.10 Hence, the matrix is constant, rendering the following simple solution
where matrix exponentiation is understood, and has the following specific entries
where no summation is assumed, and is the magnitude of the spin density. Another complimentary way of evolving the system is to work in the spin basis where the field is decomposed into the three different spin multiplicities [11] (as opposed to working with Cartesian components as we did here). For example with where and , the evolution of due to the off-diagonal kick step becomes , where is also a constant skew-symmetric Hermitian matrix. This gives the desired solution . See [57] for this approach.
In the scheme above, the different field components are mixed due to (iso-)spin-spin coupling. As a result, the expression for for an -component system can become convoluted as increases (see § 5). Below, we continue with the vector system and present a new solution for the above kick differential equation, in which all the Cartesian field components are decoupled. Only the real and imaginary parts of each Cartesian component remain coupled. This decoupling yields a much simpler evolution for each component, and allows for a straightforward generalization to -component Schrödinger systems (given in § 5).
We begin with the ansatz in (28) to get non-linear differential equations for with only the non-diagonal piece in contributing to its evolution:
Then, upon multiplying the above with and vector summing, we get . This equation has a simple solution
where is the magnitude of (a constant of motion)11, and is its initial phase (at ). Owing to this analytical (and simple) evolution of the squared sum of the field components, we can now decouple the different field components by using it in (31). This renders the following linear differential equation for each of the field components
In terms of the real and imaginary parts of : and , this differential equation becomes
We have dropped the vector component index on (related to “” in ), since each now evolves the same way, differing only in their initial condition. We have also re-scaled the time as for convenience (with ), and are the (real) Pauli matrices.12 This is the equation for which we seek an analytical solution.13
We note that the time dependence of the matrix can be stripped off by going to a rotating frame. The corresponding transformation is achieved via the matrix . Then, with and , we get the following transformed equation with the corresponding trivial solution
Using and re-instating quantities in terms of the ‘actual’ time , we therefore have the following solution for
With this exact non-diagonal kick evolution, the real and imaginary parts of each of the field component are evolved using , accompanied by a phase translation due to the diagonal Hamiltonian density. In terms of the field itself, the evolution is
where we used . Including the drift evolution, the full vector field is therefore evolved according to steps (25), (26), and (27), with the above solution for . In Appendix § 8.§ 4, we provide the explicit expression for . This fetches our desired mass and spin conserving, time-reversible drift + kick + drift algorithm, to evolve the multicomponent/vector SP system. Since both the drift and kick evolutions are now exact, the overall accuracy of the system as compared to the true evolution in (23), like the scalar case, is .
Let us briefly comment on the operator in regards to the conservation of mass and spin density
While indicates that the spin density remains unchanged throughout the kick evolution, indicates that the mass density within each component is not conserved. This is reflective of the fact that there is no separate U(1) symmetry within each component of the vector field, and hence no mass conservation within each component. On the other hand, the total mass density is conserved. This may seem peculiar since . We note that since , and appearing in the operator are consistent with the initial (meaning and ), the total mass density is indeed conserved. In general, when this consistency doesn’t hold, is not conserving.
Courant–Friedrichs–Lewy condition
The Courant–Friedrichs–Lewy (CFL) condition ensures that the time step is sufficiently small in order to resolve the fastest process happening in the simulation. For the drift-kick-drift evolution of the multi-component Schrödinger system, this means resolving the fastest kick and drift processes every time step. The drift evolution is governed by the operator dictating that be at-least as small as in order to resolve one full rotation of the drift phase.14 For the diagonal kick step, the operation is governed by which requires that be at least as small as the smallest of the two quantities and . Finally, the non-diagonal kick dictates that epsilon be at least as small as in order to resolve both the and containing parts of the exponentials appearing in in the previous section. Since by definition, this condition is less constraining than the one appearing due to diagonal kick. With this, all of the above requirements together result in the following CFL condition:15
In our simulations we take , i.e. we resolve the fastest oscillation in the system by points within its full cycle.
With the above CFL condition, the time step can be driven to small values when sufficiently dense regions (with ) start to appear in the course of evolution of the Schrödinger field. Care should be taken in interpreting results in this regime since, in general, large variations in over sufficiently small spatial regions can take us out of the domain of our non-relativistic theory. Typically in the repulsive case, does not develop large enough variations to endanger our nonrelativistic approximations. However, the situation in more precarious in the attractive self-interaction case.
Consider a spatial region of size and total enclosed mass , with a roughly isotropic mass density around it. The total energy in the region is due to contributions from gradient pressure , gravity , and self-interaction . It can be seen that as (meaning ), there is a runaway possibility (when the attractive self-interaction is relevant) where the system can keep on lowering its energy by either focusing all this mass to ever-smaller regions, and/or by accumulating more mass from its surroundings in a given region. This process can lead to very large densities in small regions of space.16 At this point, relativistic corrections (including higher-order terms in the self-interaction which are not present in the system being simulated) cannot be justifiably ignored. Note that this discussion equally applies to a single component SP system (scalar) with point self-interactions, and is not particular to multicomponent systems only. For the demonstration of fidelity of our algorithm/code, we pick small enough so that in the CFL condition throughout the duration of the simulation.
Numerical tests and results
To test our algorithm, we ran several simulations including: (1) a single sitting soliton with different polarizations, (2) two, three or more soliton collisions, for and . For a given value of , the spherically symmetric soliton configurations were obtained by numerical shooting in (A.10), which were then put on the discrete lattice. See Appendix § 8.§ 3 for a brief discussion of vector solitons arising in a self-interacting massive spin-1 field.
For illustrating the robustness of our algorithm we present results from a set of both forward and time reversed test simulations of a three soliton collision scenario. In these simulations, we worked with a (periodic) grid of dimensionless length , with . All three solitons had a 95% radius , with two of them linearly polarized, and one circularly polarized.
We reserve the investigation of more involved scenarios such as many soliton collisions, emergence of solitons from random initial conditions etc., for a separate work.
Mass and Spin conservation
Since every step in the drift-kick-drift operation is both mass and spin conserving, the overall evolution is guaranteed to be unitary and spin conserving. We track the fractional change in the total mass and spin:
which is expected to be zero up to machine precision. Note that the fractional change in the spin along any direction only makes sense if it was nonzero to begin with.17 The plot at the bottom of Figure 1 show these two quantities for our chosen three soliton collision scenario (for ). Note that mass and spin are conserved to machine precision.18

Figure 1. Mass and spin conservation: Top panel shows snapshots of projected mass density at three instants , , and (upper panel). Lower panel are snapshots of magnitude of spin density at the same times. One of the solitons is initialized with maximal spin, whereas the other two have zero spin initially. The bottom plot shows quantitative measures of total spin (blue curve) and total mass (red curve) conservation; both are conserved to better than one part in . Animations of the simulation results can be seen on Youtube.
Reversibility
runaway scenario, and hence a breakdown of the nonrelativistic EFT.
Reversibility
In order to test reversibility of our algorithm, we define an asymmetry parameter
where is the total re-scaled mass (and is already conserved up to machine precision; see the previous subsection), the subscript “+” denotes the forward evolved field starting from some initial condition up to a final configuration , while the subscript “−” denotes the reverse evolved field, starting from the initial condition and time-reversed ( in the simulation). This gives a direct measure of the separation between the forward and backward trajectories.
In Figure 2 we show snapshots from the chosen three soliton collision scenario for repulsive self-interaction along with the asymmetry parameter. We take the distinctive unwinding of the final soliton to the initial 3 soliton state, as well as the smallness of the asymmetry parameter as evidence that our algorithm is indeed reversible.
Figure 2. Tracking reversibility: Top panel shows snapshots of projected mass density at three instants , , and (upper panel). The self-interaction was chosen to be repulsive (). Lower panel include snapshots from the backward evolution at the same instants. The unwinding of the final state to the initial state gives a qualitative proof of reversibility of our algorithm. In the bottom graph we show the asymmetry parameter , which provides a quantitative measure of reversibility.
Convergence and accuracy
Starting from a given initial condition and a time discretization step , we assume that the simulated field upto time , has the following series representation
Here is the true solution in the continuum limit , and (independent of ) are the error coefficients at different order in , with giving the leading order accumulated error. We then calculate the separation between three trajectories obtained for three different values of , and construct the following ratio
Here ‘...’ represents dependent terms that we neglect. We find , implying . Hence the simulated field solution at any given time is away from the true solution. Note that this is one order lower than the truncation error of at any time step.
We have also checked that the total energy of the system has the same accuracy as that for the scalar case.
Generalizations
Expanding Universe
With the inclusion of a scale factor relevant for standard cosmology, the equation of motion reads
Here we have rescaled the field as . We only outline the modifications needed in the kick step. With , the corresponding evolution is governed by
Upon assuming the ansatz , we get the same equation as (31) but with replaced by :
Note that and are conserved under the above evolution (in terms of ).19
Extension to -component Schrödinger system & external potentials
We now discuss the more general case of a -component Schrödinger system. Requiring invariance of the general action
the most general form of the quartic self-interaction potential is
where and .20 In the above action, is any general potential (including the gravitational potential).
Apart from total mass, linear and angular momentum, and energy, the total ‘isospin’ in this case is the charge associated with the internal symmetry. With as the skew-symmetric matrices/generators (where the integer index ), it takes the general form21
Similar to the case of vectors, one way to handle the kick step is to add and subtract in the above potential to give , which ultimately renders the following kick evolution for
In order to avoid the non-trivial exponentiation of the matrix , we can instead use our trick of decoupling the different Cartesian components. With this approach, the algorithm for evolving the -component system is exactly the same as developed in this paper. That is, the same steps as outlined in § 3.2.1, with and the factor of replaced by in (c.f. (37)). The only difference is that now we have fields to evolve instead of 3.
Summary
We have devised an algorithm for evolving multicomponent Schrödinger systems obeying symmetry and containing both gravitational and non-gravitational self-interactions, in particular (iso-)spin-spin interactions that mix the different field components. The crucial aspect of our algorithm is the kick step, arising due to all the interactions. Approximate extensions of existing scalar algorithms with self-interactions or multicomponent algorithms without (iso-)spin-spin interactions to our system can lead to a loss of spin, mass conservation and also reversibility. An analytic solution for the kick step containing (iso-)spin-spin interactions mitigates these issues.
We provided such an analytic solution for the kick step. We first showed that different field components can be decoupled in their evolution. This was made possible by noticing that the magnitude of is a constant of motion throughout the kick step, since both the number density and (iso-)spin density remain constant. Owing to this decoupling, we get a reduced 2-level system in which only the real and imaginary parts of each of the field components remain coupled, with the evolution of each component being the exactly the same. We are then able to find an exact analytical solution for this reduced coupled system. Equipped with this analytical solution, we developed a split-step Fourier algorithm that involves drift-kick-drift set of operations, in analogy to the scalar case. Our algorithm is time reversible, unitary and spin/iso-spin conserving (up to machine precision).
For concreteness, we introduced the above algorithm first for a 3-component system (vector/spin-1 field).22 To demonstrate the fidelity of our algorithm, we showed that for sample simulations (with gravity + attractive or repulsive self-interactions) that total mass and total spin are conserved up to machine precision, and the algorithm maintains time reversibility. The overall accuracy of the algorithm, as compared to the true (continuous) evolution, is and is therefore on par with the usually employed split-step Fourier algorithm for a single scalar field, or multiple scalar fields without the aforementioned spin-spin self-interaction.
Our algorithm is general and flexible enough for a wide range of applications in astrophysics, cosmology and condensed-matter physics. We have provided generalizations of the algorithm to include an expanding cosmological background, and external potentials in laboratory systems, and an arbitrary number of components (respecting symmetry). The computational cost of including self-interactions is only order unity larger than the case without them.
With this paper, we make the code (written in Python) for a multicomponent Schrödinger system with/without gravity + with/without self-interactions, publicly available at GitHub.
Acknowledgments
We would like to especially thank Wisha Wanichwecharungruang (Rice U.) for help in testing various aspects of an earlier version of the algorithm presented in this paper. We would also like to thank Dorian Amaral for helpful discussions surrounding the NumPy library in Python. We acknowledge Rohith Karur’s involvement and help in the initial stages of the project, and thank Andrew Long, Philip Mocz, Jonathan Thomas and Han Pu for discussions. MA and MJ are partly supported by a DOE grant DE-SC0021619.
Appendix
Non-relativistic limit
A (dark) massive spin-1 field minimally coupled to gravity and with non-gravitational self-interactions, is described by the following effective action:
Here , and the ‘’ represents the Standard Model Lagrangian and other possible dark sector(s). The parameter is the mass of the vector boson, and contains self-interactions of the vector field, arising on account of some ultraviolet physics. We shall be interested in the leading order (dimension 4) operator describing a quartic interaction, which takes the form
Without loss of generality, the spatial part of the (real-valued) vector field can be represented in terms of a complex vector as
where has dimensions of . Similarly, . We are interested in the non-relativistic behavior of the vector field where the spatial variation in the field is slow compared to the Compton scale . For capturing this non-relativistic behaviour, we insert the above form of and drop all the second time derivative terms acting on as well as terms containing the fast oscillating pieces of the form . We also restrict ourselves to Newtonian gravity.23 With these considerations, we arrive at the nonrelativistic action for the slowly varying part of the vector field and Newtonian gravity:
with
For the case of only gravitational interactions, the nonrelativistic limit was derived earlier in [10]. For massive spin-2 case, see [64]. For a generalization to the spin- case including spin-1 and spin-2 cases, see [11]. Going further, relativistic corrections to this multicomponent system should be investigated. See [65] for single-component/scalar case.
Fluid and Conservation Equations
We can also transform our multicomponent SP system eq. (4) into a set of three, coupled fluid equations (following the Madelung transform commonly used in SDM [66]). With the following field re-definition, , and the velocity , we have (for ):
where is the mass density of each component and
Note that while the mass (and particle number) of individual components is not conserved in presence of self-interactions, the total mass (and number) is conserved. That is, . Furthermore, the potential is present only when we include non-gravitational self-interactions, whereas is present with gravity alone.
For future reference, we provide the continuity equations for the number and spin densities also. Defining the general Schrödinger current tensor
the continuity equations are
Lowest energy soliton solutions
In order to obtain soliton states, we can extremize the energy functional at a fixed particle number , i.e. extremize the quantity where can be thought of as the chemical potential. This is equivalent to assuming the ansatz , where in order to get the lowest energy solutions, is a radially symmetric function and is a spatially independent (and normalized) polarization vector characterizing the polarization of the soliton along the direction. The label corresponds to the spin multiplicity, which takes the values 0 or 1 for linearly polarized and circularly polarized soliton respectively. For example, and . See [11, 14] for details. This renders the following time-independent Schrödinger-Poisson system

Figure 3. Radius vs mass (left panel), and energy vs mass (right panel) for vector solitons with self-interactions. The dotted black curve is the limit as (without any factors of appearing in the re-scaling of axes). The curves lying above and below the dotted black curve on the left panel, are for repulsive () and attractive () self-interactions respectively. While on the right panel the curves lying above and below the dotted black curve, are for attractive and repulsive self-interactions respectively.
The solutions to the above set of equations, for different values (and given and ), can be obtained by numerical shooting method. Note that there is a scaling symmetry in the system where different soliton solutions (with different values of ) for a given and , can be mapped to solitons with different and same and . That is, once a soliton solution is obtained for a , another soliton solution for a different , can be obtained by re-scaling the fields, space, and as , , , and .
Fig. 3 shows the two possible families of solitons, corresponding to linear (), and circular polarization (), for both attractive () and repulsive () interactions.24 Here and are the mass and energy (excluding rest mass) of the soliton, and is the radius containing 95% of the soliton mass. Solitons where attractive self-interaction balances the gradient pressure (with gravity being sub-dominant) are unstable [14] [67] and lie on the colored dashed curves.
To highlight the difference between the non self-interaction case () for which and , we have also plotted the respective curves in dashed gray. From the right panel in Fig. 3, note that solitons with the same mass have different energies for different polarizations (spin) when non-gravitational self-interactions are present.25 For a fixed and mass , linearly polarized soliton is the lower energy state when the interaction is attractive (). Whereas for repulsive self-interaction (), it is the circularly polarized soliton that is lower in energy.
Explicit expression for the evolution operator
With where is the spin density, , and as the sinc function, the 4 different components of are
solitons, with total spin ranging anywhere from 0 to , called fractionally polarized solitons [11].
References
- [11]M. Jain and M. A. Amin, “Polarized solitons in higher-spin wave dark matter,” Phys. Rev. D 105 no. 5, (2022) 056019, arXiv:2109.04892 [hep-th].DOI
- [12]M. A. Amin, M. Jain, R. Karur, and P. Mocz, “Small-scale structure in vector dark matter,” JCAP 08 no. 08, (2022) 014, arXiv:2203.11935 [astro-ph.CO].DOI
- [13]M. Gorghetto, E. Hardy, J. March-Russell, N. Song, and S. M. West, “Dark photon stars: formation and role as dark matter substructure,” JCAP 08 no. 08, (2022) 018, arXiv:2203.10100 [hep-ph].DOI
- [14]M. Jain, “Soliton stars in Yang-Mills-Higgs theories,” Phys. Rev. D 106 no. 8, (2022) 085011, arXiv:2205.03418 [hep-ph].
- [15]G. P. Agrawal, “Nonlinear fiber optics,” in Nonlinear Science at the Dawn of the 21st Century, P. L. Christiansen, M. P. Sørensen, and A. C. Scott, eds., pp. 195–211. Springer Berlin Heidelberg, Berlin, Heidelberg, 2000.
- [16]C. Menyuk, “Nonlinear pulse propagation in birefringent optical fibers,” IEEE Journal of Quantum electronics 23 no. 2, (1987) 174–176.DOI
- [17]D. N. Christodoulides and R. Joseph, “Vector solitons in birefringent nonlinear dispersive media,” Optics letters 13 no. 1, (1988) 53–55.pubmed.ncbi.nlm.nih.gov/19741978
- [18]D. Rand, I. Glesk, C.-S. Brès, D. A. Nolan, X. Chen, J. Koh, J. W. Fleischer, K. Steiglitz, and P. R. Prucnal, “Observation of temporal vector soliton propagation and collision in birefringent fiber,” Physical review letters 98 no. 5, (2007) 053902.pubmed.ncbi.nlm.nih.gov/17358859
- [19]Z.-Y. Sun, Y.-T. Gao, X. Yu, W.-J. Liu, and Y. Liu, “Bound vector solitons and soliton complexes for the coupled nonlinear schrödinger equations,” Physical Review E 80 no. 6, (2009) 066608.pubmed.ncbi.nlm.nih.gov/20365295
- [20]F. Baronio, A. Degasperis, M. Conforti, and S. Wabnitz, “Solutions of the vector nonlinear schrödinger equations: evidence for deterministic rogue waves,” Physical review letters 109 no. 4, (2012) 044102.pubmed.ncbi.nlm.nih.gov/23006091
- [21]S. V. Manakov, “On the theory of two-dimensional stationary self-focusing of electromagnetic waves,” Zhurnal Eksperimentalnoi i Teoreticheskoi Fiziki 65 (Aug., 1973) 505–516.
- [22]C. Menyuk, “Nonlinear pulse propagation in birefringent optical fibers,” IEEE Journal of Quantum Electronics 23 no. 2, (1987) 174–176.DOI
- [23]D. J. Benney and A. C. Newell, “The propagation of nonlinear wave envelopes,” Journal of Mathematics and Physics 46 no. 1-4, (1967) 133–139. https://onlinelibrary.wiley.com/doi/abs/10.1002/sapm1967461133.DOI
- [24]G. J. Roskes, “Some nonlinear multiphase interactions,” Studies in Applied Mathematics 55 no. 3, (1976) 231–238. https://onlinelibrary.wiley.com/doi/abs/10.1002/sapm1976553231.DOI
- [25]K. Dysthe, H. E. Krogstad, and P. Müller, “Oceanic rogue waves,” Annual Review of Fluid Mechanics 40 no. 1, (2008) 287–310. https://doi.org/10.1146/annurev.fluid.40.111406.102203.
- [26]C. K. Law, H. Pu, and N. P. Bigelow, “Quantum spins mixing in spinor bose-einstein condensates,” Phys. Rev. Lett. 81 (Dec, 1998) 5257–5261. https://link.aps.org/doi/10.1103/PhysRevLett.81.5257.DOI
- [27]X. Liu, H. Pu, B. Xiong, W. M. Liu, and J. Gong, “Formation and transformation of vector solitons in two-species bose-einstein condensates with a tunable interaction,” Phys. Rev. A 79 (Jan, 2009) 013423. https://link.aps.org/doi/10.1103/PhysRevA.79.013423.DOI
- [28]H. Hu, B. Ramachandhran, H. Pu, and X.-J. Liu, “Spin-orbit coupled weakly interacting bose-einstein condensates in harmonic traps,” Phys. Rev. Lett. 108 (Jan, 2012) 010402. https://link.aps.org/doi/10.1103/PhysRevLett.108.010402.DOI
- [29]F. Dalfovo, S. Giorgini, L. P. Pitaevskii, and S. Stringari, “Theory of Bose-Einstein condensation in trapped gases,” Rev. Mod. Phys. 71 (1999) 463–512, arXiv:cond-mat/9806038.DOI
- [30]K. E. Strecker, G. B. Partridge, A. G. Truscott, and R. G. Hulet, “Formation and propagation of matter-wave soliton trains,” Nat 417 no. 6885, (May, 2002) 150–153, arXiv:cond-mat/0204532 [cond-mat.stat-mech].DOI
- [31]J. H. V. Nguyen, P. Dyke, D. Luo, B. A. Malomed, and R. G. Hulet, “Collisions of matter-wave solitons,” Nature Physics 10 no. 12, (Nov, 2014) 918–922. http://dx.doi.org/10.1038/nphys3135.
- [32]D. Luo, Y. Jin, J. Nguyen, B. Malomed, O. Marchukov, V. Yurovsky, V. Dunjko, M. Olshanii, and R. Hulet, “Creation and characterization of matter-wave breathers,” Physical Review Letters 125 no. 18, (Oct, 2020) . http://dx.doi.org/10.1103/PhysRevLett.125.183902.
- [33]V. Springel, “The Cosmological simulation code GADGET-2,” Mon. Not. Roy. Astron. Soc. 364 (2005) 1105–1134, arXiv:astro-ph/0505010.DOI
- [34]H.-Y. Schive, Y.-C. Tsai, and T. Chiueh, “GAMER: a GPU-Accelerated Adaptive Mesh Refinement Code for Astrophysics,” Astrophys. J. Suppl. 186 (2010) 457–484, arXiv:0907.3390 [astro-ph.IM].DOI
- [35]P. Mocz and S. Succi, “Numerical solution of the nonlinear Schrödinger equation using smoothed-particle hydrodynamics,” Phys. Rev. E 91 no. 5, (2015) 053304, arXiv:1503.03869 [physics.comp-ph].DOI
- [36]B. Schwabe, J. C. Niemeyer, and J. F. Engels, “Simulations of solitonic core mergers in ultralight axion dark matter cosmologies,” Phys. Rev. D 94 no. 4, (2016) 043513, arXiv:1606.05151 [astro-ph.CO].DOI
- [37]P. Mocz, M. Vogelsberger, V. H. Robles, J. Zavala, M. Boylan-Kolchin, A. Fialkov, and L. Hernquist, “Galaxy formation with BECDM – I. Turbulence and relaxation of idealized haloes,” Mon. Not. Roy. Astron. Soc. 471 no. 4, (2017) 4559–4570, arXiv:1705.05845 [astro-ph.CO].DOI
- [38]J. Zhang, Y.-L. S. Tsai, J.-L. Kuo, K. Cheung, and M.-C. Chu, “Ultralight Axion Dark Matter and Its Impact on Dark Halo Structure in N-body Simulations,” Astrophys. J. 853 no. 1, (2018) 51, arXiv:1611.00892 [astro-ph.CO].DOI
- [39]F. Edwards, E. Kendall, S. Hotchkiss, and R. Easther, “PyUltraLight: A Pseudo-Spectral Solver
