Abstract
The Gross–Pitaevskii–Poisson equations that govern the evolution of self-gravitating Bose–Einstein condensates, possibly representing dark matter halos, experience a process of gravitational cooling and violent relaxation. We propose a heuristic parametrization of this complicated process in the spirit of Lynden-Bell’s theory of violent relaxation for collisionless stellar systems. We derive a generalized wave equation that was introduced phenomenologically in Chavanis (Eur Phys J Plus 132:248, 2017) involving a logarithmic nonlinearity associated with an effective temperature \(T_\mathrm{eff}\) and a damping term associated with a friction \(\xi \). These terms can be obtained from a maximum entropy production principle and are linked by a form of Einstein relation expressing the fluctuation-dissipation theorem. The wave equation satisfies an H-theorem for the Lynden-Bell entropy and relaxes towards a stable equilibrium state which is a maximum of entropy at fixed mass and energy. This equilibrium state represents the most probable state of a Bose–Einstein condensate dark matter halo. It generically has a core-halo structure. The quantum core prevents gravitational collapse and may solve the core-cusp problem. The isothermal halo leads to flat rotation curves in agreement with the observations. These results are consistent with the phenomenology of dark matter halos. Furthermore, as shown in a previous paper (Chavanis in Phys Rev D 100:123506, 2019), the maximization of entropy with respect to the core mass at fixed total mass and total energy determines a core mass–halo mass relation which agrees with the relation obtained in direct numerical simulations. We stress the importance of using a microcanonical description instead of a canonical one. We also explain how our formalism can be applied to the case of fermionic dark matter halos.
Graphic abstract
Similar content being viewed by others
Data Availability Statement
This manuscript has no associated data or the data will not be deposited. [Authors’ comment: Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.]
Notes
Numerical simulations of CDM show that DM halos have a universal density profile called the Navarro–Frenk–White (NFW) profile [27]. The density diverges approximately as \(r^{-1}\) (cusp) for \(r\rightarrow 0\) and decreases as \(r^{-3}\) for \(r\rightarrow +\infty \). The observational Burkert [28] profile also decreases as \(r^{-3}\) at large distances but tends to a constant (core) at the center.
Some authors have considered the case where the DM particle is a fermion like a massive neutrino (see the Introduction of Ref. [42] for a short review and an exhaustive list of references). In this model, gravitational collapse is prevented by the quantum pressure arising from the Pauli exclusion principle.
The condensation of integer spin particles was theoretically predicted by Bose [43] and Einstein [44, 45] in 1924 and observed for the first time in laboratory experiments of dilute alkali gases in 1995 [46,47,48]. The condensation occurs when the particles in the gas become correlated quantum mechanically, i.e., when the de Broglie thermal wavelength of a particle \(\lambda _\mathrm{dB}=\sqrt{2\pi \hbar ^2/mk_B T}\) turns out to be greater than the mean interparticle distance \(l=n^{-1/3}\) (i.e. \(n\lambda _\mathrm{dB}^3>1\)). The exact condensation temperature is given by
$$\begin{aligned} T_c=\frac{2\pi \hbar ^2n^{2/3}}{m k_B \zeta (3/2)^{2/3}} \end{aligned}$$(1)with \(\zeta (3/2)=2.612\ldots \) If we apply these results to ULAs with a mass \(m=2.92\times 10^{-22}\, \mathrm{eV/c^2}\) and consider a typical DM halo density \(\rho =7.02\times 10^{-3}\, M_{\odot }/\mathrm{pc}^3\) (medium spiral), we get \(T_c=4.82\times 10^{36}\, \mathrm{K}\). On the other hand, the typical temperature of the halo (obtained from the virial relation \(k_B T_\mathrm{eff}/m\sim v_h^2=GM_h/r_h\)) is \(T_\mathrm{eff}\sim 4.41\times 10^{-25}\, \mathrm{K}\) where we have taken \((k_B T_\mathrm{eff}/m)^{1/2}=108\, \mathrm{km/s}\). Therefore, we find that \(T_\mathrm{eff}\ll T_c\). This inequality shows that \(T_\mathrm{eff}\) is necessarily an out-of-equilibrium (effective) temperature otherwise the halo would be completely condensed. For \(m=1.10\times 10^{-3}\, \mathrm{eV/c^2}\) we get \(T_c=5.29\times 10^5\, \mathrm{K}\) and \(T_\mathrm{eff}=1.66\times 10^{-6}\, \mathrm{K}\) leading to similar conclusions. If we apply the self-gravitating BEC model to neutron stars with the idea that neutrons of mass \(m_n\) could form Cooper pairs and behave as bosons of mass \(m=2m_n\) [49] we find \(T_c=8.26\times 10^{11}\, \mathrm{K}\) (we have taken \(m_n=1.675\times 10^{-24}\, \mathrm{g}\) and \(\rho _0=3.54\times 10^{21}\, \mathrm{g/m^3}\)) which is comparable to the typical temperature T of neutron stars.
A repulsive self-interaction (\(a_s>0\)) stabilizes the quantum core. By contrast, an attractive self-interaction destabilizes the quantum core above a maximum mass \(M_\mathrm{max}=1.012\, \hbar /\sqrt{Gm|a_s|}\) first identified in [52].
The halo cannot be exactly isothermal otherwise it would have an infinite mass [7]. In reality, the density in the halo decreases as \(r^{-3}\), similarly to the NFW [27] and Burkert [28] profiles, or even as \(r^{-4}\) (see Appendix D of [79] and Appendix I of [80]), instead of \(r^{-2}\) corresponding to the isothermal sphere [7]. This extra-confinement may be due to incomplete relaxation, tidal effects, and stochastic perturbations as discussed in Appendix B of [81]. We stress that the halo is in a dynamical equilibrium (virialized) state but in an out-of-equilibrium thermodynamical equilibrium state (see footnote 4). As discussed in [75,76,77,78], the quasiparticles are responsible for a slow (secular) collisional evolution of the halo towards thermodynamical equilibrium on a timescale of the order of the Hubble time. By this process part of the halo condense (since \(T\ll T_c\)) and feeds the soliton. We shall not consider this collisional regime in the present paper.
Lynden-Bell [91], who was concerned with the study of elliptical galaxies, argued that these objects are described by the nondegenerate limit of his theory where his DF is similar to the Boltzmann distribution. However, his theory may also apply to fermionic DM halos where degeneracy effects may be important as suggested in [79, 80, 92]. The theory of violent relaxation explains how a collisionless self-gravitating system may reach an isothermal distribution on a very short timescale (of the order of the dynamical time \(t_D\)) without recourse to collisions or gravitational encounters that operate on a relaxation timescale \(t_R\sim (N/\ln N)t_D\) much larger than the age of the universe [7]. The theory of violent relaxation thus solves a notoriously important timescale problem in astrophysics [79, 80, 91, 92].
The connection between self-gravitating BECs and the Lynden-Bell theory of violent relaxation was mentioned in [81, 93] and in the Appendix of [84]. There also exist interesting analogies between the violent relaxation of stellar systems and the formation of two-dimensional large-scale vortices like Jupiter’s Great Red Spot [94, 95].
To describe a system of bosons in interaction we should in principle “second-quantize” the Schrödinger equation, which then becomes an equation for the evolution of field operators with appropriate commutation rules [see Eq. (4)]. However, in the limit of very large occupation numbers \(\mathcal{N}\gg 1\), the classical description of the Schrödinger wave equation equation is adequate.
In the case of spiral or elliptical galaxies containing \(N\sim 10^{12}\) stars, the relaxation time is much larger than the age of the universe [7]. In the case of bosonic DM halos, the relaxation time is reduced by Bose stimulation (the number of quasiparticles is \(N_\mathrm{eff}\sim 10^5\) [78]) but it remains relatively long, of the order of the Hubble time, like in the case of globular clusters containing \(N\sim 10^5\) stars [7].
We can also base the collisional kinetic theory of self-gravitating bosons on an equation similar to Eq. (5) for a classical field \(\psi _d(\mathbf{r},t)\) provided that we take fluctuations into account (i.e., \(\psi _d\) can be decomposed into \(\psi _d=\psi +\delta \psi \) where \(\psi \) is a smooth component and \(\delta \psi \) is a stochastic component arising from two-body collisions).
This is a notoriously difficult mathematical problem even at the classical level, i.e., for the Vlasov–Poisson equations. The relaxation of the Vlasov–Poisson equations towards a quasistationary state on the coarse-grained scale is related to the concepts of violent relaxation, phase mixing and nonlinear Landau damping [107].
See Appendix G of [93] for a general derivation of the quantum virial theorem applying to self-gravitating BECs from the hydrodynamic representation of the GPP equations.
See Appendix B of [104] for a detailed proof of this result.
This is true in a “smoothed-out” sense where the density profile is averaged on a scale larger than the de Broglie length. Indeed, the small-scale interferences produce time-dependent granules (or quasiparticles) of size \(\lambda _\mathrm{dB}\) that can induce a slow collisional evolution of the halo [75,76,77,78]. We shall not consider this collisional evolution here and remain at a purely collisionless level.
In principle, for a system of bosons, we would expect that \(f(\mathbf{r},\mathbf{v})\) is given by the Bose–Einstein or Rayleigh-Jeans DF. That would be the case if the statistical equilibrium state resulted from a collisional relaxation [78]. However, in the present context, the evolution of the system is collisionless and the DF is given by the Lynden-Bell DF which is similar to the Fermi–Dirac DF (see below). This is why Lin et al. [113] find that the DF of the halo is well-described by the Fermi–Dirac DF [91, 92] or by the fermionic King model [80, 96]. In the nondegenerate limit, these DFs reduce to the Boltzmann DF and to the classical King model.
As we shall see, the effective collision term accounts for the formation of the halo. In the halo, quantum effects are erased on the coarse-grained scale (they manifest themselves only on the fine-grained scale through the presence of granules [75,76,77,78]). By contrast, we keep quantum effects in the left hand side of the coarse-grained Wigner equation (advection term). This term accounts for the formation of the soliton in which quantum effects are dominant.
If DM is made of fermions, we have \(\eta _0\sim m^4/h^3\) (see footnote 34 in [80]). Therefore, degeneracy effects in the sense of Lynden-Bell are important. They lead to a quantum core in the form of a fermion ball. By contrast, the Heisenberg uncertainty principle (quantum potential) is negligible, or small, for fermions. For condensed bosons, this is the opposite. Indeed, \(\eta _0\) is in general very large (\(\eta _0\gg m^4/h^3\)) so that degeneracy effects in the sense of Lynden-Bell are usually negligible (\(\overline{f}\ll \eta _0\)). By contrast, the Heisenberg uncertainty principle (quantum potential) is important for bosons. It leads to a quantum core in the form of a soliton. Here, for the sake of generality, we shall take into account all these effects although some of them may be negligible depending on the situation.
The fact that the coarse-grained DF \(\overline{f}(\mathbf{r},\mathbf{v},t)\) relaxes towards the Lynden-Bell DF on a few dynamical times is interpreted by Kadomtsev and Pogutse [98] in terms of “collisions” between macroparticles with a large effective mass \(m_\mathrm{eff}\sim \eta _0 \epsilon _r^3\epsilon _v^3\) (these macroparticles are fundamentally different from the quasiparticles introduced by [75] in relation to the collisional evolution of FDM halos on a secular timescale). In the foregoing equations, \(\epsilon _r(t)\) and \(\epsilon _v(t)\) denote the position and velocity correlation scales of the fluctuations \(\delta f\) (see Appendix D2). These correlation scales are expected to decrease with time. If this decay is sufficiently rapid, it can slow down and even stop the collisionless relaxation before the Lynden-Bell DF is reached. This kinetic blocking can account for incomplete relaxation and solve the infinite mass problem of the Lynden-Bell DF (see [94, 96, 97, 101] and more specifically [117] for additional discussion about the concept of incomplete relaxation). This may explain why DM halos have a NFW or Burkert profile instead of a Lynden-Bell profile (see footnote 7).
Quantum effects (Heisenberg uncertainty principle) lead to a small modification of the Lynden-Bell DF of the order \(O(\hbar ^2)\) in the same manner that they lead to a small modification of the Boltzmann distribution for a system at thermal equilibrium [114].
For simplicity, we assume here that D is constant.
More generally, we can build up an infinite hierarchy of such equations by taking the successive moments of the DF.
The Lynden-Bell pressure and the friction term are in some sense related to the correlation function \(\overline{\delta \rho \nabla \delta \Phi }\) that emerges from the coarse-graining of the quantum Euler–Poisson equations (17)–(20), where \(\delta \rho \) and \(\delta \Phi \) denote the fluctuations about the coarse-grained (smooth) fields. However, the proper description of these correlations requires the analysis of Appendix D2 in phase space.
The Lynden-Bell effective potential (enthalpy) and the friction term are in some sense related to the correlation function \(\overline{\delta \Phi \delta \psi }\) that emerges from the coarse-graining of the Schrödinger–Poisson equations (10) and (11), where \(\delta \psi \) and \(\delta \Phi \) denote the fluctuations about the coarse-grained (smooth) fields. These fluctuations are due here to the violent variations of the gravitational potential, not to collisions.
The H-theorem and the relaxation towards an equilibrium state are due to the friction term \(\xi >0\) which provides a source of dissipation and implies the irreversibility of the generalized GPP equations (92) and (93). By contrast, the usual GPP equations (8) and (9) conserve the energy and are reversible. Their relaxation towards a quasistationary state is due to gravitational cooling and violent relaxation and can be understood only at a coarse-grained scale (see Sects. 2.3 and 2.4). It is in this sense that the generalized GPP equations (92) and (93) provide a parametrization of the ordinary GPP equations (8) and (9) taking into account the processes of gravitational cooling and violent relaxation.
This is also the Lynden-Bell entropy multiplied by \(-T_\mathrm{eff}\) (see Appendix B3). As a result, we can write \(F=E_\mathrm{tot}-T_\mathrm{eff}S\) up to constant additive terms.
This is because, in DM models, fermions have a much larger mass than bosons (see, e.g., [102]).
As discussed above, in the case of fermions, the Fermi–Dirac pressure \(P_\mathrm{FD}\) is equivalent to the Lynden-Bell pressure \(P_\mathrm{LB}\). However, for the sake of clarity (and generality) we shall treat these two terms separately.
This correction is usually negligible in the case of DM halos.
The Lynden-Bell exclusion principle is usually negligible for self-gravitating bosons (for which \(\eta _0\gg m^4/h^3\)) but it may be relevant for self-gravitating fermions (for which \(\eta _0\sim gm^4/h^3\)).
One can also account for the self-interaction of the fermions and introduce a self-interaction pressure in addition to the pressure due to the Pauli exclusion principle.
In terms of the wavefunction, the effective temperature is given by
$$\begin{aligned} \frac{3}{2}M T_\mathrm{eff}(t)= & {} E_\mathrm{tot}-\frac{\hbar ^2}{2m^2}\int |\nabla \psi |^2\, \mathrm{d}{} \mathbf{r}\nonumber \\&-\int V(|\psi |^2)\, \mathrm{d}{} \mathbf{r}-\frac{1}{2}\int |\psi |^2\Phi \, d\mathbf{r}. \end{aligned}$$(B33)The structure of this equation shows that a system described by the generalized GPP equations undergoes damped oscillations towards a virialized state. These damped oscillations are characteristic of the process of gravitational cooling and violent relaxation (see Sect. 2.3). In this sense, the generalized GPP equations provide a parametrization of the ordinary GPP equations.
These results are basically due to the fact that both \(E_\mathrm{tot}\) and M are conserved by the GPP equations [109]. A direct proof of the equivalence between the energy principle (minimum of energy at fixed mass) and the condition of dynamical stability (positivity of the squared pulsation \(\omega ^2\) of all modes) for the GPP equations is given in Appendix B of [104].
In an expanding universe, using the Zel’dovich approximation, the pressureless Euler equations can be reduced to the so-called Burgers equation [140]. On the other hand, the Navier–Stokes equation can be reduced to the viscous Burgers equation which corresponds to the adhesion model [141] of cosmology. This model is relevant provided that the viscosity is sufficiently small (\(\nu \rightarrow 0\)).
Sensible expressions of \(P(\rho )\) have been considered by Buchert et al. [142]. In particular, they singled out an equation of state of the form \(P(\rho )\propto \rho ^2\) that leads, under certain assumptions, to a viscous Burgers-like equation in cosmology. Interestingly, as discussed in [143], this expression is similar to the pressure created by self-interacting bosons in the BECDM model.
The Schrödinger equation is equivalent to the Wigner equation. In the Schrödinger equation, \(\psi (\mathbf{r},t)\) encodes both position and momentum information in a single position-space function. It is argued that, when \(\hbar _\mathrm{eff}\rightarrow 0\), the Vlasov–Poisson equations are recovered and that a finite value of \(\hbar _\mathrm{eff}\) provides a small-scale regularization of the dynamics. In that case, the Schrödinger–Poisson system has nothing to do with quantum mechanics since it aims at describing the evolution of classical collisionless matter under the influence of gravity.
For self-gravitating BECs described by the “true” Schrödinger equation, this identification is exact (see Sect. 3.4.1). Inversely, this identification may suggest an interpretation of the quantum potential \(P_{ij}^Q\) in terms of a classical kinetic theory.
The Lynden-Bell entropy is proportional to the logarithm of the number of microstates corresponding to a given macrostate. This is a measure of disorder. Therefore, the maximization of S under constraints determines the most probable state of the system, i.e., the macrostate that is the most represented at the “microscopic” level.
More generally, the coarse-grained DF must always be smaller than the maximum value of the fine-grained (or initial) DF.
Since the process of violent relaxation is very nonlinear, we cannot in principle use perturbation methods (see, however, [96,97,98,99,100,101] in a regime of “gentle” relaxation). We thus capitalize our ignorance and assume that the system evolves so as to maximize its rate of entropy production at fixed mass and energy. This thermodynamic principle is expected to determine the most probable evolution of the system.
We have some freedom on the determination of the diffusion coefficient since it is related to a constraint that is not explicitly known.
In the nondegenerate limit \(\overline{f}\ll \eta _0\), Eq. (D33) is similar to the classical Kramers equation. On the other hand, assuming \(D=\mathrm{cst}\), Eq. (D30) can be written as \(\beta (t)=3M/2K(t)=1/\langle T_\mathrm{kin}\rangle (t)\) where \(\langle T_\mathrm{kin}\rangle (t)\) is the average kinetic temperature of the system (see Appendix D3).
This result can also be directly obtained from the H-theorem by using Lyapunov’s direct method.
If the system reaches a state of statistical equilibrium through a collisional relaxation, \(p_{\alpha }\) should be given by the Bose–Einstein DF for bosons and by the Fermi–Dirac DF for fermions. However, if the evolution of the system is collisionless, like in the present situation, \(p_{\alpha }\) should be given by the Lynden-Bell DF, which is similar to the Fermi–Dirac DF, in all cases (fermions and bosons). In the nondegenerate limit, the Lynden-Bell DF is similar to the Maxwell-Boltzmann DF.
References
F. Zwicky, Helv. Phys. Acta 6, 110 (1933)
A.V. Kravtsov, S. Borgani, Ann. Rev. Astron. Astrophys. 50, 353 (2012)
V.C. Rubin, W.K. Ford, Astrophys. J. 159, 379 (1970)
V.C. Rubin, W.K. Ford, N. Thonnard, Astrophys. J. 238, 471 (1980)
A. Bosma, Astron. J. 86, 1791 (1981)
M. Persic, P. Salucci, F. Stel, Mon. Not. R. Astron. Soc. 281, 27 (1996)
J. Binney, S. Tremaine, Galactic Dynamics (Princeton Series in Astrophysics, Princeton, 1987)
A.G. Riess et al., Astron. J. 116, 1009 (1998)
S. Perlmutter et al., Astrophys. J. 517, 565 (1999)
P. de Bernardis et al., Nature 404, 995 (2000)
S. Hanany et al., Astrophys. J. 545, L5 (2000)
Planck Collaboration, Astron. Astrophys. 571, 66 (2014)
Planck Collaboration, Astron. Astrophys. 594, A13 (2016)
C. Wegg, O. Gerhard, M. Portail, Mon. Not. R. Astron. Soc. 463, 557 (2016)
J.B. Muñoz, E.D. Kovetz, L. Dai, M. Kamionkowski, Phys. Rev. Lett. 117, 091301 (2016)
A. Chudaykin, D. Gorbunov, I. Tkachev, Phys. Rev. D 94, 023528 (2016)
S.W. Randall, M. Markevitch, D. Clowe, A.H. Gonzalez, M. Bradac, Astrophys. J. 679, 1173 (2008)
M. Milgrom, Astrophys. J. 270, 365 (1983)
T. Harko, F.S.N. Lobo, Extension of f(R) Gravity: Curvature-Matter Couplings and Hybrid Metric-Palatini Theory (Cambridge University Press, Cambridge, 2018)
G. Jungman, M. Kamionkowski, K. Griest, Phys. Rep. 267, 195 (1996)
J.M. Overduin, P.S. Wesson, Phys. Rep. 402, 267 (2004)
P. Bode, J.P. Ostriker, N. Turok, Astrophys. J. 556, 93 (2001)
J.E. Kim, G. Carosi, Rev. Mod. Phys. 82, 557 (2010)
R.D. Peccei, H.R. Quinn, Phys. Rev. Lett. 38, 1440 (1977)
D. Marsh, Phys. Rep. 643, 1 (2016)
P.J.E. Peebles, The Large-Scale Structure of the Universe (Princeton University Press, Princeton, 1980)
J.F. Navarro, C.S. Frenk, S.D.M. White, Astrophys. J. 462, 563 (1996)
A. Burkert, Astrophys. J. 447, L25 (1995)
G. Kauffmann, S.D.M. White, B. Guiderdoni, Mon. Not. R. Astron. Soc. 264, 201 (1993)
A. Klypin, A.V. Kravtsov, O. Valenzuela, Astrophys. J. 522, 82 (1999)
M. Kamionkowski, A.R. Liddle, Phys. Rev. Lett. 84, 4525 (2000)
B. Moore, T. Quinn, F. Governato, J. Stadel, G. Lake, Mon. Not. R. Astron. Soc. 310, 1147 (1999)
M. Boylan-Kolchin, J.S. Bullock, M. Kaplinghat, Mon. Not. R. Astron. Soc. 415, L40 (2011)
J.S. Bullock, M. Boylan-Kolchin, Ann. Rev. Astron. Astrophys. 55, 343 (2017)
E. Romano-Díaz, I. Shlosman, Y. Hoffman, C. Heller, Astrophys. J. 685, L105 (2008)
A. Pontzen, F. Governato, Nature 506, 171 (2014)
J. Oñorbe et al., Mon. Not. R. Astron. Soc. 454, 2092 (2015)
D.N. Spergel, P.J. Steinhardt, Phys. Rev. Lett. 84, 3760 (2000)
M. Kaplinghat, S. Tulin, H.-B. Yu, Phys. Rev. Lett. 116, 041302 (2016)
E. Braaten, D. Kang, R. Laha, JHEP 11, 084 (2018)
S.D. McDermott, Phys. Rev. Lett. 120, 221806 (2018)
P.H. Chavanis, Eur. Phys. J. Plus 135, 290 (2020)
S.N. Bose, Z. Phys. 26, 178 (1924)
A. Einstein, Sitz. Preu. Akad. Wiss. 1, 261 (1924)
A. Einstein, Sitz. Preu. Akad. Wiss. 1, 3 (1925)
M.H. Anderson, J.R. Ensher, M.R. Matthews, C.E. Wieman, E.A. Cornell, Science 269, 198 (1995)
K.B. Davis, M.O. Mewes, M.R. Andrews, N.J. van Druten, D.S. Durfee, D.M. Kurn, W. Ketterle, Phys. Rev. Lett. 75, 3969 (1995)
C.C. Bradley, C.A. Sackett, J.J. Tollett, R.G. Hulet, Phys. Rev. Lett. 75, 1687 (1995)
P.H. Chavanis, T. Harko, Phys. Rev. D 86, 064011 (2012)
E. Madelung, Z. Phys. 40, 322 (1927)
P.H. Chavanis, Phys. Rev. D 102, 083531 (2020)
P.H. Chavanis, Phys. Rev. D 84, 043531 (2011)
J.W. Lee, EPJ Web Conf. 168, 06005 (2018)
A. Suárez, V.H. Robles, T. Matos, Astrophys. Space Sci. Proc. 38, 107 (2014)
T. Rindler-Daller, P.R. Shapiro, Astrophys. Space Sci. Proc. 38, 163 (2014)
P.H. Chavanis, Self-gravitating Bose-Einstein condensates, in Quantum Aspects of Black Holes. ed. by X. Calmet (Springer, Berlin, 2015)
E. Braaten, H. Zhang, Rev. Mod. Phys. 91, 041002 (2019)
J.C. Niemeyer, Prog. Part. Nucl. Phys. 113, 103787 (2020)
E. Ferreira, Ultra-Light Dark Matter. arXiv:2005.03254
M.Yu Khlopov, B.A. Malomed, Ya.B. Zeldovich, Mon. Not. R. astr. Soc. 215, 575 (1985)
M. Bianchi, D. Grasso, R. Ruffini, Astron. Astrophys. 231, 301 (1990)
W. Hu, R. Barkana, A. Gruzinov, Phys. Rev. Lett. 85, 1158 (2000)
P. Sikivie, Q. Yang, Phys. Rev. Lett. 103, 111301 (2009)
A. Suárez, T. Matos, Mon. Not. R. Astron. Soc. 416, 87 (2011)
P.H. Chavanis, Astron. Astrophys. 537, A127 (2012)
A. Suárez, P.H. Chavanis, Phys. Rev. D 92, 023510 (2015)
A. Suárez, P.H. Chavanis, Phys. Rev. D 98, 083529 (2018)
T. Harko, Eur. Phys. J. C 79, 787 (2019)
P.H. Chavanis, Universe 6, 226 (2020)
E. Seidel, W.M. Suen, Phys. Rev. Lett. 72, 2516 (1994)
F.S. Guzmán, L.A. Ureña-López, Phys. Rev. D 69, 124033 (2004)
F.S. Guzmán, L.A. Ureña-López, Astrophys. J. 645, 814 (2006)
H.Y. Schive, T. Chiueh, T. Broadhurst, Nat. Phys. 10, 496 (2014)
H.Y. Schive et al., Phys. Rev. Lett. 113, 261302 (2014)
L. Hui, J. Ostriker, S. Tremaine, E. Witten, Phys. Rev. D 95, 043541 (2017)
B. Bar-Or, J.B. Fouvry, S. Tremaine, Astrophys. J. 871, 28 (2019)
B. Bar-Or, J.B. Fouvry, S. Tremaine, Astrophys. J. 915, 27 (2021)
P.H. Chavanis, Eur. Phys. J. Plus 136, 703 (2021)
P.H. Chavanis, M. Lemou, F. Méhats, Phys. Rev. D 91, 063531 (2015)
P.H. Chavanis, M. Lemou, F. Méhats, Phys. Rev. D 92, 123527 (2015)
P.H. Chavanis, Phys. Rev. D 100, 083022 (2019)
B. Schwabe, J. Niemeyer, J. Engels, Phys. Rev. D 94, 043513 (2016)
P. Mocz et al., Mon. Not. R. Astron. Soc. 471, 4559 (2017)
P. Mocz, L. Lancaster, A. Fialkov, F. Becerra, P.H. Chavanis, Phys. Rev. D 97, 083519 (2018)
J. Veltmaat, J.C. Niemeyer, B. Schwabe, Phys. Rev. D 98, 043509 (2018)
P. Mocz et al., Phys. Rev. Lett. 123, 141301 (2019)
P. Mocz et al., Mon. Not. R. Astron. Soc. 494, 2027 (2020)
J. Veltmaat, B. Schwabe, J.C. Niemeyer, Phys. Rev. D 101, 083518 (2020)
A. Bernal, T. Matos, D. Núñez, Rev. Mex. A. A. 44, 149 (2008)
V.H. Robles, T. Matos, Mon. Not. R. Astron. Soc. 422, 282 (2012)
D. Lynden-Bell, Mon. Not. R. Astron. Soc. 136, 101 (1967)
P.H. Chavanis, J. Sommeria, Mon. Not. R. Astron. Soc. 296, 569 (1998)
P.H. Chavanis, Eur. Phys. J. Plus 132, 248 (2017)
P.H. Chavanis, J. Sommeria, R. Robert, Astrophys. J. 471, 385 (1996)
P.H. Chavanis, Statistical mechanics of two-dimensional vortices and stellar systems, in Dynamics and Thermodynamics of Systems with Long Range Interactions. Lecture Notes in Physics. ed. by T. Dauxois, S. Ruffo, E. Arimondo, M. Wilkens (Springer, Berlin, 2002)
P.H. Chavanis, Mon. Not. R. Astron. Soc. 300, 981 (1998)
P.H. Chavanis, Statistical mechanics of violent relaxation in stellar systems, in Multiscale Problems in Science and Technology. ed. by N. Antonić, C.J. van Duijn, W. Jäger, A. Mikelić (Springer, Berlin, 2002)
B.B. Kadomtsev, O.P. Pogutse, Phys. Rev. Lett. 25, 17 (1970)
G. Severne, M. Luwel, Astrophys. Space Sci. 72, 293 (1980)
P.H. Chavanis, Physica A 332, 89 (2004)
P.H. Chavanis, Kinetic theory of collisionless relaxation for systems with long-range interactions. arXiv:2112.13664
P.H. Chavanis, Phys. Rev. D 100, 123506 (2019)
P.H. Chavanis, Phys. Rev. D 101, 063532 (2020)
P.H. Chavanis, Phys. Rev. D 103, 123551 (2021)
A. Griffin, T. Nikuni, E. Zaremba, Bose-Condensed Gases at Finite Temperatures (Cambridge University Press, Cambridge, 2009)
P.-H. Chavanis, Physica A 391, 3680 (2012)
C. Mouhot, C. Villani, Acta Math. 207, 29 (2011)
P.H. Chavanis, L. Delfini, Phys. Rev. D 84, 043532 (2011)
D.D. Holm, J.E. Marsden, T. Ratiu, A. Weinstein, Phys. Rep. 123, 1 (1985)
R. Ruffini, S. Bonazzola, Phys. Rev. 187, 1767 (1969)
M. Membrado, A.F. Pacheco, J. Sanudo, Phys. Rev. A 39, 4207 (1989)
D. Marsh, A.R. Pop, Mon. Not. R. Astron. Soc. 451, 2479 (2015)
S.C. Lin, H.Y. Schive, S.K. Wong, T. Chiueh, Phys. Rev. D 97, 103523 (2018)
E. Wigner, Phys. Rev. 40, 749 (1932)
I.E. Moyal, Proc. Cambr. Philos. Soc. 45, 99 (1949)
K. Husimi, Proc. Phys. Math. Soc. Jpn. 22, 264 (1940)
P.H. Chavanis, Physica A 365, 102 (2006)
R. Ruffini, L. Stella, Astron. Astrophys. 119, 35 (1983)
I.R. King, Astron. J. 70, 376 (1965)
S. Chandrasekhar, An Introduction to the Study of Stellar Structure (Dover, New York, 1958)
P.H. Chavanis, Phys. Rev. E 84, 031101 (2011)
P.H. Chavanis, Phys. Rev. E 83, 031131 (2011)
P.H. Chavanis, C. Rosier, C. Sire, Phys. Rev. E 66, 036105 (2002)
P.H. Chavanis, M. Ribot, C. Rosier, C. Sire, Banach Cent. Publ. 66, 103 (2004)
P.H. Chavanis, Phys. Dark Univ. 22, 80 (2018)
P.H. Chavanis, Eur. Phys. J. Plus 132, 286 (2017)
P.H. Chavanis, in preparation
T. Padmanabhan, Phys. Rep. 188, 285 (1990)
J. Katz, Found. Phys. 33, 223 (2003)
P.H. Chavanis, Int. J. Mod. Phys. B 20, 3113 (2006)
A. Campa, T. Dauxois, D. Fanelli, S. Ruffo, Physics of Long-Range Interacting Systems (Oxford University Press, Oxford, 2014)
A. Campa, P.H. Chavanis, J. Stat. Mech. 06, 06001 (2010)
C.R. Argüelles, M.I. Díaz, A. Krut, R. Yunis, Mon. Not. R. Astron. Soc. 502, 4227 (2021)
L.H. Thomas, Proc. Cambr. Philos. Soc. 23, 542 (1927)
E. Fermi, Z. Physik 48, 73 (1928)
P.H. Chavanis, Predictive model of fermionic dark matter halos with a quantum core and an isothermal atmosphere. arXiv:2112.07726
A.A. Vlasov, Zh. Eksp. i Teor. Fiz. 8, 291 (1938)
J.H. Jeans, Mon. Not. R. Astron. Soc. 76, 70 (1915)
M. Hénon, Astron. Astrophys. 114, 211 (1982)
M. Vergassola, B. Dubrulle, U. Frisch, A. Noullez, Astron. Astrophys. 289, 325 (1994)
S.N. Gurbatov, A.I. Saichev, S.F. Shandarin, Mon. Not. R. Astron. Soc. 236, 385 (1989)
T. Buchert, A. Domínguez, J. Pérez-Mercander, Astron. Astrophys. 349, 343 (1999)
P.H. Chavanis, Phys. Rev. D 84, 063518 (2011)
L.M. Widrow, N. Kaiser, Astrophys. J. 416, L71 (1993)
C.J. Short, P. Coles, J. Cosmol. Astropart. Phys. 12, 012 (2006)
P.H. Chavanis, Entropy 21, 1006 (2019)
L. Onsager, Phys. Rev. 37, 405 (1931)
L. Onsager, Phys. Rev. 38, 2265 (1931)
P.H. Chavanis, Eur. Phys. J. B 62, 179 (2008)
L. Onsager, S. Machlup, Phys. Rev. 91, 1505 (1953)
P.H. Chavanis, Phys. Rev. E 68, 036108 (2003)
P.H. Chavanis, F. Bouchet, Astron. Astrophys. 430, 771 (2005)
P.H. Chavanis, P. Laurençot, M. Lemou, Physica A 341, 145 (2004)
L.D. Landau, E.M. Lifshitz, Fluid Mechanics (Pergamon Press, London, 1959)
R.G. Parr, W. Yang, Density Functional Theory of Atoms and Molecules (Oxford University Press, Oxford, 1989)
D.R. Hartree, Proc. Cambr. Philos. Soc. 24, 89 (1928)
D.R. Hartree, Proc. Cambr. Philos. Soc. 24, 111 (1928)
V. Fock, Z. Physik 61, 126 (1930)
D.R. Hartree, W. Hartree, Proc. R. Soc. A 150, 9 (1935)
J.C. Slater, Phys. Rev. 81, 385 (1951)
P.A.M. Dirac, Proc. Cambr. Philos. Soc. 26, 376 (1931)
W. Kohn, L.J. Sham, Phys. Rev. 140, 1133 (1965)
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix A: Properties of the Wigner distribution
In this Appendix, we check that the density \(\rho \), velocity \(\mathbf{u}\) and pressure tensor \(P_{ij}\) obtained from the Wigner DF (43) coincide with the expressions obtained from the Schrödinger equation (10).
To that purpose, let us first establish useful identities. Using Eqs. (15) and (16), we get
and
We also note that
We are now ready to compute the first moments of the Wigner DF
Integrating Eq. (A4) over \(\mathbf{v}\) and using Eq. (A3), we obtain
which coincide with Eq. (A1).
Multiplying Eq. (A4) by \(\mathbf{v}\) and integrating over \(\mathbf{v}\), we get
where we have used an integration by parts to obtain the last equality. Calculating the derivative of the term in brackets, and using Eq. (A3), we obtain
which coincide with Eq. (A2).
Multiplying Eq. (A4) by \(v_iv_j\), integrating over \(\mathbf{v}\), and proceeding as in Eq. (A6), we obtain
Calculating the second derivative of the term in brackets, and using Eq. (A3), we get
The pressure tensor can be written as
Using Eqs. (A7) and (A9), we obtain after simplification
The quantum pressure tensor obtained from the Schrödinger equation (10) can be written as [see Eq. (23)]
Using Eq. (A1), we get after simplification
Comparing Eqs. (A11) and (A13) we see that
Let us now check that the expression of the energy in the Wigner representation is the same as in the Schrödinger representation. In terms of the Wigner DF, the total energy is given by
According to Eqs. (A10) and (A14), it can be rewritten as
Using Eq. (G.22) of [93] establishing that \(\int P_{ii}^Q\, \mathrm{d}{} \mathbf{r}=2\Theta _Q\), we find that
which is in agreement with Eq. (28) obtained from the Schrödinger equation. From this equivalence, we can directly conclude that the Wigner–Poisson equations conserve the energy.
Finally, we derive the Wigner equation (44). Taking the time derivative of the Wigner DF (A4), we get
The next step is to substitute the Schrödinger equation (10) into the term in brackets of Eq. (A18). The contribution of the potential term in the Schrödinger equation is
The second equality can be checked directly by substituting Eq. (A4), integrating over \(\mathbf{v}'\), and using Eq. (A3). The contribution of the kinetic term in the Schrödinger equation is
Integrating by parts, we get
Making the same operations with the kinetic term of the complex conjugate Schrödinger equation and adding the two results, we obtain
The second equality can be checked directly by substituting Eq. (A4). Summing Eqs. (A19) and (A22), we obtain the Wigner equation (44).
Remark
When \(\hbar \rightarrow 0\), we can make the approximation \(\Phi (\mathbf{r}\pm \mathbf{y}/2,t)\simeq \Phi (\mathbf{r},t)\pm \nabla \Phi (\mathbf{r},t)\cdot \mathbf{y}/2\) in the integral of the Wigner equation (44) and we obtain
This returns the Vlasov equation (C1). The quantum correction to the Vlasov equation (i.e. the difference between the Vlasov and the Wigner equation) is of order \(\hbar ^2\). Similarly, let us assume that orbit crossing has not yet occurred and let us take the limit \(\hbar \rightarrow 0\) in the Wigner DF (43). Substituting Eq. (14) into Eq. (43), making the approximations \(\rho (\mathbf{r}\pm \mathbf{y}/2,t)\simeq \rho (\mathbf{r},t)\) and \(S(\mathbf{r}\pm \mathbf{y}/2,t)\simeq S(\mathbf{r},t)\pm \nabla S(\mathbf{r},t)\cdot \mathbf{y}/2\), and using Eq. (16), we get
This returns the single-speed solution from Eq. (C13). For finite \(\hbar /m\), the DF f is proportional to a Gaussian in \(\mathbf{v}\) of thickness \(\hbar /m\).
Appendix B: Generalized GPP equations
In this Appendix, we recall the basic properties of the generalized GPP equations introduced in [93]. We also discuss the difference between the canonical and microcanonical formulation.
1.1 1. Basic properties
Let us consider the generalized GPP equations
with a potential \(V(|\psi |^2)\). Introducing the enthalpy (see Appendix B5)
we can rewrite the generalized GP equation (B1) as
The stationary solutions of the generalized GPP equations (B1) and (B2) are of the form
where \(\phi (\mathbf{r})=\sqrt{\rho (\mathbf{r})}\) and E (Eigenenergy) are real. These quantities are determined by the Eigenvalue problem
Dividing Eq. (B6) by \(\phi \), we get
where Q is the quantum potential defined by Eq. (21).
Using the Madelung [50] transformation (see Sect. 2.2), the hydrodynamic equations corresponding to the generalized GPP equations (B1) and (B2) are
where the barotropic equation of state \(P(\rho )\) is determined by the relation (see Appendix B5)
For a general polytropic equation of state of the form
the enthalpy and the potential are given by
In particular, for the standard self-interacting BEC described by the polytropic equation of state
we have
For the Lynden-Bell pressure of zero-point energy
we have
For the Lynden-Bell isothermal equation of state
the enthalpy and the potential are given by
An equilibrium state of the quantum damped Euler equations (B9)–(B12) satisfies the condition of quantum hydrostatic equilibrium
This equation is equivalent to Eq. (B8). Indeed, dividing Eq. (B23) by \(\rho \), using Eq. (B13) and integrating, we obtain Eq. (B8), where E appears as a constant of integration.
1.2 2. Microcanonical model
We introduce the mass
and the energy
which includes the classical kinetic energy
the quantum kinetic energy
the internal energy
the gravitational energy
and the thermal energy
We also introduce the entropy (up to an additive constant)
The functionals (B26)–(B29) are justified in [93] and the functionals (B30) and (B31) are justified in Appendix D3.
In the microcanonical situation, the total energy \(E_\mathrm{tot}\) is fixed and the temperature \(T_\mathrm{eff}(t)\) evolves in time so as to satisfy the constraint from Eq. (B25) yielding
The generalized GPP equations (B1) and (B2) with the time-dependent temperature \(T_\mathrm{eff}(t)\) given by Eq. (B32)Footnote 37 conserve the energy (by construction) and satisfy an H-theorem for the entropy (B31). This can be proven as follows. From Eq. (B31), we have
Using the equation of continuity (B9) and integrating by parts, we get
On the other hand, taking the time derivative of Eq. (B32), and using Eqs. (B26)–(B29), we obtain
where we have used the identities of Appendix C of [93]. Using the equation of continuity (B9), the damped quantum Euler equation (B11), and recalling the identity of vector analysis \((\mathbf{u}\cdot \nabla )\mathbf{u}=\nabla (\mathbf{u}^2/2)-\mathbf{u}\times (\nabla \times \mathbf{u})\) and the fact that the flow is irrotational (\(\nabla \times \mathbf{u}=\mathbf{0}\)) since \(\mathbf{u}=\nabla S/m\), we get after simplification (using straightforward integrations by parts)
Combining Eqs. (B35) and (B37), and using Eq. (B21), we finally obtain
Therefore, the entropy increases monotonically. At equilibrium (where \(\dot{S}=0\)), we get \(\mathbf{u}=\mathbf{0}\) leading to the quantum equation of hydrostatic equilibrium (B23).
We can easily show that the generalized GPP equations in the microcanonical ensemble relax towards an equilibrium state that maximizes the entropy at fixed mass and energy. First of all, an extremum of entropy at fixed mass and energy is determined by the variational principle
where \(\alpha =\mu /mT_\mathrm{eff}\) is a Lagrange multiplier taking into account the conservation of mass [the conservation of energy is taken into account in Eq. (B32)]. Using
and
we obtain
which is equivalent to Eq. (B8) with \(\mu =E\). This shows that the Eigenenergy E coincides with the chemical potential \(\mu \). Taking the gradient of Eq. (B42) and using Eq. (B13), we recover the equation of quantum hydrostatic equilibrium (B23). Therefore, an equilibrium state of the generalized GPP equations is an extremum of entropy at fixed mass and energy. On the other hand, by proceeding as explained at the end of Appendix D2, we can show that only entropy maxima (not minima or saddle points) at fixed mass and energy are dynamically stable [127]. From these results, we conclude that the generalized GPP equations with a time-dependent effective temperature relax towards an equilibrium state which maximizes the entropy at fixed mass and energy:
Proceeding as in Appendix G of [93], we obtain the damped quantum virial theorem
where \(I=\int \rho r^2\, \mathrm{d}{} \mathbf{r}\) is the moment of inertia. According to Eq. (B25), we have \(\Theta _c+\Theta _\mathrm{th}+\Theta _Q=E_\mathrm{tot}-U-W\) so we can rewrite Eq. (B44) asFootnote 38
For a polytropic equation of state, the internal energy satisfies the identity [see Eqs. (B14) and (B15)]
and the damped quantum virial theorem becomes
We note that the term involving the internal energy vanishes for the particular index \(\gamma =5/3\) which corresponds to the Lynden-Bell pressure of zero-point energy (B18).
Remark
In the strong friction limit \(\xi \rightarrow +\infty \), we can neglect the inertial term (l.h.s.) in the damped quantum Euler equation (B11) and get
Substituting this relation into the continuity equation (B9) we obtain the quantum Smoluchowski–Poisson equations
Since \(|\mathbf{u}|=O(1/\xi )\) in the strong friction limit, we can neglect the classical kinetic energy \(\Theta _c\) in Eq. (B32). Therefore, the evolution of the effective temperature is given by
The H-theorem takes the form
and we have the same general properties as those discussed after Eq. (B38). On the other hand, the overdamped quantum virial theorem is given by
Using \(\Theta _\mathrm{th}+\Theta _Q=E_\mathrm{tot}-U-W\), we can rewrite Eq. (B53) as
For a polytropic equation of state, using Eq. (B46), the overdamped quantum virial theorem becomes
1.3 3. Canonical model
In the canonical ensemble, the temperature \(T_\mathrm{eff}\) is fixed and the appropriate thermodynamic potential is the free energy
Using Eqs. (B25) and (B31), we obtain (up to an additive constant)
where
is the internal energy associated with the thermal pressure. This returns the free energy introduced in [93].
The generalized GPP equations (B1) and (B2) with a fixed temperature \(T_\mathrm{eff}\) satisfy an H-theorem for the free energy (B57). Indeed, using calculations similar to the previous ones (see also Appendix D of [93]) we obtain
Therefore, the free energy decreases monotonically. At equilibrium (where \(\dot{F}=0\)), we get \(\mathbf{u}=\mathbf{0}\) leading to the equation of quantum hydrostatic equilibrium (B23).
We can easily show that the generalized GPP equations in the canonical ensemble relax towards an equilibrium state that minimizes the free energy at fixed mass. First of all, an extremum of free energy at fixed mass is determined by the variational principle
where \(\mu /m\) is a Lagrange multiplier taking into account the conservation of mass. Using calculations similar to the previous ones (see also section 3.4 of [93]) we obtain Eq. (B42) which is equivalent to Eq. (B8) with \(\mu =E\) or to the quantum equation of hydrostatic equilibrium (B23). Therefore, an equilibrium state of the generalized GPP equations is an extremum of free energy at fixed mass. On the other hand, by proceeding as explained at the end of Appendix D2 we can show that only minima (not maxima or saddle points) of free energy at fixed mass are dynamically stable [127]. From these results, we conclude that the generalized GPP equations with a constant effective temperature relax towards an equilibrium state which minimizes the free energy at fixed mass:
The damped quantum virial theorem is given by Eq. (B44) or, equivalently, by
where \(T_\mathrm{eff}\) is constant.
Remark
In the strong friction limit \(\xi \rightarrow +\infty \), we obtain the quantum Smoluchowski–Poisson equations (B49) and (B50) where \(T_\mathrm{eff}\) is constant. The H-theorem now reads
On the other hand, the overdamped quantum virial theorem is given by Eq. (B53) or, equivalently, by
where \(T_\mathrm{eff}\) is constant.
1.4 4. Numerical algorithm
A stationary solution of the ordinary GPP equations
is of the form \(\psi (\mathbf{r},t)=\phi (\mathbf{r})\mathrm{e}^{-iEt/\hbar }\), where \(\phi (\mathbf{r})=\sqrt{\rho (\mathbf{r})}\) and E (Eigenenergy) are real. These quantities are determined by the Eigenvalue problem
Dividing Eq. (B67) by \(\phi \), we get
where Q is the quantum potential defined by Eq. (21).
Using the Madelung transformation, the ordinary GPP equations (B65) and (B66) are equivalent to the hydrodynamic equations
A stationary solution of the quantum Euler–Poisson equations (B70)–(B73) satisfies the condition of quantum hydrostatic equilibrium
This equation is equivalent to Eq. (B69) as can be seen by taking the gradient of Eq. (B69) and using Eq. (B13).
On the other hand, a stationary solution of the ordinary GPP equations (B65) and (B66) is an extremum of energy
at fixed mass M. Indeed, writing the variational principle as
where \(\mu /m\) is a Lagrange multiplier (global chemical potential) taking into account the mass constraint, we get
which is equivalent to Eq. (B69) with \(E=\mu \), and to Eq. (B74). Furthermore, it can be shown that an equilibrium state is dynamically stable if, and only if, it is a minimum of energy at fixed mass [52].Footnote 39 Therefore, a stable stationary solution of the GPP equations is determined by the minimization problem
We stress that the ordinary GPP equations do not relax towards the stationary state that minimizes the energy at fixed mass (ground state) since these equations are reversible and the energy is conserved. They rather experience a process of gravitational cooling and violent relaxation (on a coarse-grained scale) towards a quasistationary state with a core-halo structure (see Sect. 2.3). The characterization of this core-halo state was the topic of this paper.
Independently from the physical problem treated in this paper, it is an interesting mathematical problem in itself to be able to construct stationary solutions of the ordinary GPP equations (B65) and (B66). However, it is difficult in practice to numerically solve the nonlinear Eigenvalue problem defined by Eqs. (B67) and (B68) and make sure that the solution is dynamically stable. Interestingly, the generalized GPP equations (B1) and (B2) with \(T_\mathrm{eff}=0\) provide a useful numerical algorithm to reach that goal. Indeed, we have shown in Appendix B3 that these equations satisfy an H-theorem and that they relax towards a stationary state that minimizes the free energy F defined by Eq. (B57) at fixed mass M. Since F with \(U_\mathrm{th}=0\) coincides with \(E_\mathrm{tot}\), this equilibrium state solves the minimization problem (B78). Therefore, it is a dynamically stable steady state of the ordinary GPP equations (B65) and (B66). Consequently, the generalized GPP equations (B1) and (B2) with \(T_\mathrm{eff}=0\) provide a useful numerical algorithm to construct stable stationary solutions of the ordinary GPP equations (B65) and (B66). By construction, the equilibrium solution reached by the generalized GPP equations (B1) and (B2) with \(T_\mathrm{eff}=0\), which are true relaxation equations, is guaranteed to be a stable stationary solution of the ordinary GPP equations (B65) and (B66).
Remark
Instead of solving the generalized GPP equations (B1) and (B2), we may equivalently solve the quantum damped Euler equations (B9)–(B12) or the simpler (diffusive) quantum Smoluchowski–Poisson equations (B49) and (B50) with \(T_\mathrm{eff}=0\) which also relax towards a stationary solution that satisfies the minimization problem (B78).
1.5 5. General identities for a cold gas
For a cold (\(T=0\)) gas, the first principle of thermodynamics
where u is the density of internal energy, s the density of entropy, \(\rho \) the mass density, and P the pressure reduces to
Introducing the enthalpy per particle
we get
Comparing Eq. (B81) with the Gibbs–Duhem relation at \(T=0\):
we see that the enthalpy \(h(\mathbf{r})\) is equal to the local chemical potential \(\mu (\mathbf{r})\) by unit of mass: \(h(\mathbf{r})=\mu (\mathbf{r})/m\). On the other hand, for a barotropic gas for which \(P=P(\rho )\), the foregoing equations can be written as
Comparing Eq. (B86) with Eqs. (B3) and (B13), we see that the potential \(V(\rho )\) represents the density of internal energy
We then have
The squared speed of sound is
Appendix C: Classical collisionless self-gravitating systems
In this Appendix and in the following one, we discuss certain aspects of the dynamical evolution of classical collisionless self-gravitating systems to facilitate the comparison with the dynamical evolution of quantum self-gravitating systems treated in the main text.
1.1 1. Vlasov equation
A classical collisionless self-gravitating system (such as a stellar system or such as CDM) is governed by the Vlasov–Poisson equations
where \(f=f(\mathbf{r},\mathbf{v},t)\) is the six-dimensional DF. The Vlasov equation is also known as the collisionless Boltzmann equation [7]. It states that, in the absence of encounters (“collisions”) between the particles, the density (DF) of a “fluid” particle is conserved when we follow its motion in phase space, i.e., \(Df/Dt=0\) where \(D/Dt=\partial /\partial t+\mathbf{v}\cdot \partial /\partial \mathbf{r}-\nabla \Phi \cdot \partial /\partial \mathbf{v}\) is the material derivative (Stokes operator). As a result, the Vlasov–Poisson equations conserve the energy \(E=(1/2)\int f v^2\, \mathrm{d}{} \mathbf{r}\mathrm{d}{} \mathbf{v}+(1/2)\int \rho \Phi \, \mathrm{d}{} \mathbf{r}\), the mass \(M=\int f \, \mathrm{d}{} \mathbf{r}\mathrm{d}{} \mathbf{v}\) and an infinite class of Casimir integrals of the form \(\int h(f)\, \mathrm{d}{} \mathbf{r}\mathrm{d}{} \mathbf{v}\) where h(f) is an arbitrary function of f [91, 94].
Remark
The Vlasov equation can be viewed as the expression of the Liouville theorem in the individual phase space. Under the circumstance in which stellar encounters can be ignored, each star can be idealized as an independent conservative system described by the Hamiltonian \(H=\mathbf{v}^2/2+\Phi (\mathbf{r},t)\) yielding the equations of motion \(\mathrm{d}{} \mathbf{r}/dt=\mathbf{v}\), \(d\mathbf{v}/dt=-\nabla \Phi \). The equation of continuity \(\partial f/\partial t+\nabla _6\cdot (f \mathbf{U}_6)=0\), where \(\nabla _6=(\partial _\mathbf{r},\partial _\mathbf{v})\) is a generalized nabla operator and \(\mathbf{U}_6=(\mathbf{v},-\nabla \Phi )\) a generalized velocity field, and the fact that the flow in phase space in incompressible, \(\nabla _6\cdot \mathbf{U}_6=0\), leads to the Liouville equation \(\partial f/\partial t+\mathbf{U}_6\cdot \nabla _6 f=0\), which is the Vlasov equation.
1.2 2. Hydrodynamics of the Vlasov equation: Jeans equations
From the Vlasov equation, we can derive a system of hydrodynamic equations called the Jeans equations.Footnote 40 By integrating the Vlasov equation (C1) over the velocity, we get the continuity equation (expressing the local mass conservation)
where we have introduced the local density
and the local velocity
Then, multiplying the Vlasov equation (C1) by \(\mathbf{v}\) and integrating over the velocity, we obtain the momentum equation
Introducing the difference \(\mathbf{w}=\mathbf{v}-\mathbf{u}\) between the velocity \(\mathbf{v}\) of a particle and the mean local velocity \(\mathbf{u}(\mathbf{r},t)\), and using the fact that \(\langle \mathbf{w}\rangle =\mathbf{0}\), we get
Therefore, the momentum equation (C6) takes the form
where we have introduced the pressure tensor (\(-P_{ij}\) is the stress tensor)
It can be written as
Using the continuity equation (C3), we obtain the identity
As a result, the momentum equation (C8) can be rewritten as
These equations are essentially those for a compressible fluid which is supported by pressure in the form of a velocity dispersion. These equations are not closed because the pressure tensor \(P_{ij}\) depends on the DF which is not explicitly known in general. Actually, we can build up an infinite hierarchy of equations by introducing higher and higher moments of the velocity. In general, there is no simple way to close this hierarchy of Jeans equations except in the single speed case (see Appendix C3). In more general cases, some approximations must be introduced.
1.3 3. Single-speed solution: pressureless Euler equations
The Vlasov–Poisson equations admit a particular solution of the form
This is called the single-speed solution because there is a single velocity attached to any given point \(\mathbf{r}\) at time t. It corresponds to the “dust model” where the pressure is zero because there is no thermal motion (at a given location all the particles have the same velocity). The density \(\rho (\mathbf{r},t)\) and the velocity \(\mathbf{u}(\mathbf{r},t)\) satisfy the pressureless Euler equations
These equations are exact. They can be deduced from the Jeans equations (C3) and (C12) by closing the hierarchy with the condition \(P_{ij}=0\) obtained by substituting Eq. (C13) into Eq. (C9). They correspond to very particular initial conditions where the particles at a given location all have the same velocity. However, there is a well-known difficulty with the solution (C13). Even if one starts with a DF of the form of Eq. (C13) then, after a finite time, the solution of the Vlasov–Poisson equations becomes multi-stream because of particle crossing. This leads to the formation of caustics (singularities) in the density field at shell-crossing. Therefore, Eq. (C13) ceases to be valid. This phenomenon renders the pressureless hydrodynamical description (C14)–(C16) useless beyond the first time of crossing when the fast particles cross the slow ones. Therefore, after shell-crossing, the pressureless Euler equations are not defined anymore and we must come back to the original Vlasov–Poisson equations, or to the Jeans equations, because we need to account for a velocity dispersion. Indeed, the velocity field becomes multi-valued even if, initially, it is single-valued. The pressureless Euler equations are only valid until shell crossing and they fail as soon as orbit crossing (multistreaming) occurs.
Different attempts to cure this problem have been proposed in order to continue using a hydrodynamical model.
(i) A first heuristic possibility to avoid multi-streaming is to introduce a viscosity term \(\nu \Delta \mathbf{u}\) in the momentum equation (C15) yielding the Navier–Stokes equation
In order for the diffusion term to have a smoothing effect only in the regions where particle-crossing is about to occur, the viscosity \(\nu \) should be small. More precisely, the limit \(\nu \rightarrow 0\) should be taken, which is different from setting \(\nu =0\).Footnote 41
(ii) A second heuristic possibility is to introduce a pressure term \((-1/\rho )\nabla P\) in the momentum equation (C15) yielding
where P is the fluid pressure, a local quantity given by a specified equation of state which takes into account velocity dispersion. This amounts to closing the hierarchy of Jeans equations with the isotropy ansatz \(P_{ij}=P(\rho ) \delta _{ij}\). In this manner, there is no shell-crossing singularities. The velocity dispersion giving rise to the pressure can be a consequence of the multi-streaming or it can be already present in the initial condition. We need \(P\rightarrow 0\) at large scales to recover the CDM model and \(P\ne 0\) at small scales to avoid singularities.Footnote 42
(iii) A third heuristic possibility, proposed by Widrow and Kaiser [144], is to replace the Vlasov equation by a wave equation having the form of a Schrödinger equation with an effective Planck constant \(\hbar _\mathrm{eff}\) controlling the spatial resolution.Footnote 43 Using the Madelung transformation, this prescription leads, instead of Eq. (C15), to a momentum equation of the form
with an effective “quantum” pressure tensor
Interestingly, the effective quantum pressure tensor \(P_{ij}^Q\) bares some resemblance with the Jeans pressure tensor \(P_{ij}\) of Eq. (C10) with the velocity average operators replaced by density gradients (a nonlocal quantity). This amounts to closing the hierarchy of Jeans equations with the condition \(P_{ij}=P_{ij}^Q\).Footnote 44 Therefore, although the system is classical, the procedure of Widrow and Kaiser [144] amounts to closing the Jeans equations by introducing an effective quantum potential of order \(O(\hbar _\mathrm{eff}^2/m^2)\). When \(\hbar _\mathrm{eff}/m\rightarrow 0\) the effective quantum pressure \(P_{ij}^Q\) tends to zero at large scales while \(P_{ij}^Q\ne 0\) at small scales (in line with the fact that quantum mechanics is negligible at large scales and important at small scales in BECDM), thereby preventing singularities. This is exactly what is required. The effective quantum pressure tensor acts as a regularizer of caustics and singularities in classical solutions. The quantum pressure also replaces the role of the viscosity in the adhesion model (see footnote 41) [145]. Therefore, the Schrödinger method can handle multiple streams in phase space.
(iv) A fourth possibility is to use the heuristic kinetic theory of violent relaxation developed by Chavanis et al. [94] (see also the Appendix of [84]). In that case, the pressureless Euler equation (C15) is replaced by the damped Euler equation
where \(P_\mathrm{LB}\) is the Lynden-Bell pressure (or a generalized pressure) and \(\xi \) is a friction coefficient accounting for nonlinear Landau damping.
In these different hydrodynamic models, we can go beyond the first time of crossing. Therefore, these hydrodynamic models are defined for all times. The solution of these hydrodynamic equations is expected to remain close to the solution of the Vlasov–Poisson equations for all times provided that \(\nu \rightarrow 0\), \(P\rightarrow 0\), and \(\hbar _\mathrm{eff}/m\rightarrow 0\) in these respective models. We note that the limits \(\nu ,P,\hbar _\mathrm{eff}/m\rightarrow 0\) are crucially different from taking \(P=\nu =\hbar _\mathrm{eff}/m=0\).
Remark
The Vlasov–Poisson equations (C1) and (C2) are valid for all times. Similarly, the Jeans equations (C3)–(C12), which are equivalent to the Vlasov equation, are valid for all times but they are not closed (we have to consider the whole hierarchy of equations). By contrast, the pressureless Euler–Poisson equations (C14)–(C16) are valid only until the first time of crossing. Before that time they coincide with the Vlasov–Poisson equations and after that time they break down. The modified Euler–Poisson equations (C17)–(C21) are valid for all times. If \(\nu \rightarrow 0\), \(P\rightarrow 0\) and \(\hbar _\mathrm{eff}\rightarrow 0\), they are expected to be close to the Vlasov–Poisson equations. In particular, Eqs. (C19)–(C20) are equivalent to the Schrödinger and Wigner equations, which are themselves equivalent to the Vlasov equation when \(\hbar _\mathrm{eff}\rightarrow 0\).
Appendix D: Violent relaxation of classical collisionless self-gravitating systems
The Vlasov–Poisson equations (C1) and (C2) describing classical collisionless stellar systems are known to experience a process of violent relaxation [91] caused by the rapid fluctuations of the strongly varying gravitational potential at the early stage of galaxy formation. While the fine-grained DF \({f}(\mathbf{r},\mathbf{v},t)\) always evolves in time, forming intermingled filaments in phase space at smaller and smaller scales, the coarse-grained DF \(\overline{f}(\mathbf{r},\mathbf{v},t)\), which smooths out this intricate filamentation, rapidly relaxes towards a quasistationary state \(\overline{f}_\mathrm{QSS}(\mathbf{r},\mathbf{v})\). This process takes place on a very short timescale of the order of the dynamical time \(t_D\). For \(t>t_D\), the evolution of \({f}(\mathbf{r},\mathbf{v},t)\) occurs on scales smaller than the coarse-graining mesh. This complicated dynamics is associated with phase mixing, violent relaxation and nonlinear Landau damping. As a result, the coarse-grained DF \(\overline{f}(\mathbf{r},\mathbf{v},t)\) does not satisfy the Vlasov equation. Phase-space correlations introduce an effective “collision” term \(\mathcal{C}[\overline{f}]\) on the right hand side of the coarse-grained Vlasov equation. This collision term drives the relaxation of the coarse-grained DF towards the quasistationary state. The determination of the quasistationary state \(\overline{f}_\mathrm{QSS}(\mathbf{r},\mathbf{v})\) and of the collision term \(\mathcal{C}[\overline{f}]\) is a problem of fundamental interest but also of great difficulty because of the very nonlinear nature of the process. Here, we tackle this problem through a thermodynamical approach. We use a MEP to determine the quasistationary state \(\overline{f}_\mathrm{QSS}(\mathbf{r},\mathbf{v})\) and we use a MEPP to determine the collision term \(\mathcal{C}(\overline{f})\). At a general level, the MEP determines the most probable equilibrium state of the system and the MEPP determines the most probable evolution of the system [146].
1.1 1. Maximum entropy principle
In a seminal paper, Lynden-Bell [91] argued that the coarse-grained DF \(\overline{f}(\mathbf{r},\mathbf{v},t)\) violently relaxes towards a quasistationary state \(\overline{f}_\mathrm{QSS}(\mathbf{r},\mathbf{v})\) which maximizes a suitable mixing entropy while taking into account all the constraints of the Vlasov equation.Footnote 45 In the two-level approximation of the theory, where the fine-grained DF \({f}(\mathbf{r},\mathbf{v},t)\) takes only two values \(f=\eta _0\) and \(f=0\),Footnote 46 the constraints reduce to the conservation of mass and energy and the Lynden-Bell entropy reads
It can be obtained from a combinatorial analysis taking into account the specificities of the Vlasov equation. In particular, the coarse-grained DF must satisfy the inequality \(\overline{f}(\mathbf{r},\mathbf{v},t)\le \eta _{0}\) arising from the incompressibility of the flow in phase space and the conservation of the DF on the fine-grained scale.Footnote 47 This constraint is similar to the Pauli exclusion principle in quantum mechanics (with another interpretation) and this is why the Lynden-Bell entropy (D1) resembles the Fermi–Dirac entropy of quantum mechanics. In this sense, the process of violent relaxation is similar in some respects to the collisional relaxation of self-gravitating fermionic particles. The Lynden-Bell DF corresponds to a fourth type of statistics corresponding to distinguishable particles experiencing an exclusion principle [91].
According to the MEP, the DF of the quasistationary state \(\overline{f}_\mathrm{QSS}\) maximizes the Lynden-Bell entropy (D1) at fixed mass
and energy
where K is the kinetic energy and W the potential (gravitational) energy. Introducing Lagrange multipliers and writing the variational principle under the form
we find that the extrema of entropy at fixed mass and energy correspond to the Lynden-Bell DF
where \(\epsilon ={v^{2}/2}+\Phi (\mathbf{r})\) is the energy of a particle by unit of mass. The Lagrange multipliers \(\beta =1/T_\mathrm{eff}\) and \(\alpha =\mu _\mathrm{eff}/T_\mathrm{eff}\) are the effective inverse temperature and the effective chemical potential (divided by the effective temperature). The Lynden-Bell DF (D5) which maximizes the Lynden-Bell entropy at fixed mass and energy is the most probable, or most mixed, state taking into account all the constraints of the Vlasov equation. The Lynden-Bell DF is similar to the Fermi–Dirac DF in quantum mechanics provided that we make the correspondence
where \(g=2s+1\) is the multiplicity of the quantum states. In particular, the Lynden-Bell DF (D5) satisfies the constraint \(\overline{f}_\mathrm{LB}(\mathbf{r},\mathbf{v})\le \eta _{0}\) which is similar to the Pauli exclusion principle in quantum mechanics. We note that the effective temperature \(T_\mathrm{eff}\) has not the dimension of a temperature. It should rather be interpreted as a velocity dispersion. However, we shall use this notation which is more transparent. We also note that the mass m of the particles does not appear in the Lynden-Bell theory since it is based on the Vlasov equation for collisionless systems which is independent of the mass of the particles. In this sense, we can say that the temperature in Lynden-Bell’s theory is proportional to the mass of the particles [91].
From the Lynden-Bell DF (D5), one can determine the density and the pressure through the equations
They can be rewritten as
where
denote the Fermi–Dirac integrals. Eliminating formally the gravitational potential \(\Phi (\mathbf{r})\) between Eqs. (D9) and (D10), we see that the equation of state is barotropic: \(P(\mathbf{r})=P[\rho (\mathbf{r})]\). Equations (D9) and (D10) determine the Lynden-Bell equation of state \(P_\mathrm{LB}(\rho )\) in parametric form. The Lynden-Bell equation of state is formally similar to the Fermi–Dirac equation of state. Using Eq. (D8), the kinetic energy can be written as
On the other hand, the Lynden-Bell DF (D5), and more generally any DF of the form \(f=f(\epsilon )\), implies the condition of hydrostatic equilibrium (see, e.g., [42] and the Remark of Appendix F)
In the completely degenerate limit \(\overline{f}_\mathrm{LB}\sim \eta _0\), the Lynden-Bell DF reduces to a step function
where H is the Heaviside function (\(H(x)=1\) if \(x<1\) and \(H(x)=0\) if \(x>1\)) and \(\epsilon _\mathrm{LB}\) is the Lynden-Bell energy, which is the counterpart of the Fermi energy. In that case, Eqs. (D7) and (D8) reduce to
where \(v_\mathrm{LB}(\mathbf{r})=\sqrt{2(\epsilon _\mathrm{LB}-\Phi (\mathbf{r}))}\) is the Lynden-Bell velocity. Equations (D15) and (D16) lead to the equation of state
This is a polytropic equation of state of index \(n=3/2\) like the one arising in the theory of nonrelativistic white dwarf stars at \(T=0\) corresponding to the ground state of the self-gravitating Fermi gas [120].
In the nondegenerate, or dilute, limit \(\overline{f}_\mathrm{LB}\ll \eta _{0}\), the Lynden-Bell entropy becomes similar to the Boltzmann entropy
the Lynden-Bell DF becomes similar to the Maxwell-Boltzmann DF
and the Lynden-Bell equation of state becomes similar to the equation of state of an isothermal gas
This equation of state has been studied in relation to isothermal stars [120] and to the statistical mechanics of “collisional” self-gravitating systems relaxing under the effect of gravitational encounters [128]. Remarkably, the Lynden-Bell theory of violent relaxation explains how a collisionless self-gravitating system can “thermalize” on a very short timescale, much shorter that the collisional relaxation time \(t_\mathrm{relax} \sim (N/\ln N)t_D\), without the need of “collisions” [91].
So far, we have assumed that the system is isolated so that it conserves the mass and the energy. This corresponds to the microcanonical ensemble. We now consider the canonical ensemble in which the temperature \(T_\mathrm{eff}=1/\beta \) is fixed instead of the energy. In that case, the equilibrium state is obtained by minimizing the free energy \(F=E-T_\mathrm{eff}S\) at fixed mass M or, equivalently, by maximizing the Massieu function \(J=S-\beta E\) at fixed mass M. The variational principle determining the extrema of free energy at fixed mass reads
Since \(\beta \) is fixed, this variational principle for the first variations of the thermodynamical potential is equivalent to Eq. (D4) and it returns the Lynden-Bell DF (D5). Therefore, the equilibrium states are the same in the microcanonical and canonical ensembles (the extrema of entropy at fixed mass and energy coincide with the extrema of free energy at fixed mass). However, their thermodynamical stability (related to the sign of the second variations of the thermodynamical potential) may be different in the microcanonical and canonical ensembles. This is the notion of ensembles inequivalence for systems with long-range interactions [128,129,130,131]. An equilibrium state is thermodynamically stable in the microcanonical ensemble if it is a maximum of entropy at fixed mass and energy. This corresponds to
for all perturbations \(\delta f\) that conserve mass and energy at first order. An equilibrium state is thermodynamically stable in the canonical ensemble if it is a minimum of free energy at fixed mass. This corresponds to the inequality of Eq. (D22) for all perturbations \(\delta f\) that conserve mass. We note that canonical stability implies microcanonical stability but the converse is wrong [132]. For example, it is shown in [102] that the core-halo solution with a negative specific heat is stable in the microcanonical ensemble while it is unstable in the canonical ensemble.
1.2 2. Maximum entropy production principle
We now consider the dynamical evolution of the coarse-grained DF \(\overline{f}(\mathbf{r},\mathbf{v},t)\). Writing \(f=\overline{f}+\delta f\) and \(\Phi =\overline{\Phi }+\delta \Phi \), where \(\delta f\) and \(\delta \Phi \) denote fluctuations about the coarse-grained fields, and taking the local average of the Vlasov equation (C1), we get
This equation shows that the correlations of the fluctuations of the gravitational potential and DF create an effective “collision” term \(\mathcal{C}[\overline{f}]\) in the r.h.s. of Eq. (D23). In Ref. [94] we obtained an explicit expression of this collision term by using heuristic arguments based on a MEPP.Footnote 48 We considered the general case where the fine-grained DF may take an arbitrary number of values. Below, we detail this procedure in the simpler case where the fined-grained DF takes only two values \(f=0\) and \(f=\eta _0\) as in the preceding section.
To apply the MEPP, we first write the relaxation equation for the coarse-grained DF under the form
where \(\mathbf{J}\) is the current to be determined. The form of Eq. (D24) ensures the conservation of mass provided that \(\mathbf{J}\) decreases sufficiently rapidly for large \(|\mathbf{v}|\). From Eqs. (D1), (D3) and (D24), we get
where we have used straightforward integrations by parts. Following the MEPP, we shall determine the optimal current \(\mathbf{J}\) which maximizes the rate of entropy production (D25) while satisfying the conservation of energy \(\dot{E}=0\). For this problem to have a solution, we shall also impose a limitation on the current \(|\mathbf{J}|\) characterized by a bound \(C(\mathbf{r},\mathbf{v},t)\) which exists but is not explicitly known, so that
It can be shown by a convexity argument that reaching the bound (D27) is always favorable for increasing \(\dot{S}\), so this constraint can be replaced by an equality. The variational problem can then be solved by introducing at each time t Lagrange multipliers \(\beta (t)\) and \(1/{\tilde{D}}(\mathbf{r},\mathbf{v},t)\) for the two constraints. The condition
yields an optimal current of the form
where we have set \(\tilde{D}=D(1-\overline{f}/\eta _0)\) to avoid divergences when \(\overline{f}\rightarrow \eta _0\).Footnote 49 The time evolution of the Lagrange multiplier \(\beta (t)\) is determined by the conservation of energy \(\dot{E}=0\), introducing Eq. (D29) into the constraint (D26). This yields
Note that the optimal current (D29) can be written as
where
is a time-dependent chemical potential which is uniform at equilibrium according to Eq. (D5). The relation from Eq. (D31) then corresponds to the linear thermodynamics of Onsager [147, 148] where the currents are proportional to the gradients of the thermodynamic potentials that are uniform at statistical equilibrium. Therefore, the MEPP can be viewed as a variational formulation of Onsager’s linear thermodynamics [149].Footnote 50
Introducing the optimal current (D29) into Eq. (D24), we obtain the relaxation equation
Morphologically, this relaxation equation has the form of a nonlinear Fokker–Planck equation or, more precisely, the form of a fermionic Kramers equation [149, 151]. The first term is a diffusion term and the second term is a friction term. The fermionic factor \(\overline{f}(1-\overline{f}/\eta _0)\) takes into account the Lynden-Bell exclusion principle. The function \(\beta (t)\) can be considered as a time dependent inverse temperature evolving with time so as to conserve the energy (microcanonical formulation).Footnote 51 The friction coefficient \(\xi \) satisfies a generalized Einstein relation: \(\xi =D\beta \). Note that D is not determined by the MEPP since it is related to the unknown bound \(C(\mathbf{r},\mathbf{v},t)\) in Eq. (D27).
It is straightforward to check that Eq. (D33) with the constraint (D30) satisfies an H-theorem for the Lynden-Bell entropy (D1). From Eq. (D25), we can write
The last integral vanishes due to the conservation of the energy (see Eq. (D26) with \(\dot{E}=0\)). Using Eq. (D29), we obtain
which is positive (\(\dot{S}\ge 0\)) provided that \(D>0\). This proves the H-theorem. At equilibrium, we have \(\dot{S}=0\) which implies \(\mathbf{J}=\mathbf{0}\). Then, according to Eq. (D29), we obtain
Integrating this equation with respect to \(\mathbf{v}\), we get
where \(A(\mathbf{r})\) is a constant of integration that may depend on \(\mathbf{r}\). Taking the gradient of the foregoing equation with respect to \(\mathbf{r}\), we find that
On the other hand, since \(\partial _t\overline{f}=0\) and \(\mathbf{J}=\mathbf{0}\), the advection term in Eq. (D24) cancels out:
Together with Eqs. (D36) and (D38), this implies the relation
This relation must be true for all \(\mathbf{v}\) so that
which can be integrated into
where \(\alpha \) is a constant. Substituting Eq. (D42) into Eq. (D37), we finally recover the Lynden-Bell DF (D5) with \(\beta =\lim _{t\rightarrow +\infty }\beta (t)\). Therefore, the stationary solutions of the generalized Fokker–Planck equation (D33) are the Lynden-Bell DFs which extremize the Lynden-Bell entropy at fixed energy and mass. In addition, one can show that an equilibrium state is linearly dynamically stable if, and only if, it is a (local) maximum of S at fixed M.Footnote 52 Indeed, considering the linear dynamical stability of a stationary solution of Eqs. (D30) and (D33), we can derive the general relation [151]
connecting the growth rate \(\lambda \) of the perturbation \(\delta f\sim \mathrm{e}^{\lambda t}\) to the second order variations of the free energy (Massieu function) \(J=S-\beta E\) and the second order variations of the rate of entropy production \(\delta ^{2}\dot{S}\). Since the product \(\lambda \delta ^{2}{J}\) is positive because \(\delta ^{2}\dot{S}\ge 0\) according to Eq. (D35), we conclude that a stationary solution of Eqs. (D30) and (D33) is linearly dynamically stable (\(\lambda <0\)) if, and only if, it is an entropy maximum at fixed mass and energy (\(\delta ^2 J<0\)). This aesthetic formula shows the equivalence between dynamical and thermodynamical stability for the generalized Fokker–Planck equation (D33) with the constraint from Eq. (D30). Therefore, this equation can only relax towards maxima of S, not towards minima or saddle points.
A relaxation equation of the form of Eq. (D24) appropriate to the canonical ensemble can be obtained by maximizing the rate of free energy dissipation \(\dot{J}=\dot{S}-\beta \dot{E}\) with the constraint (D27). The corresponding variational principle
again yields an optimal current of the form of Eq. (D29) now involving a constant inverse temperature \(\beta \). This equation satisfies an H-theorem for the Lynden-Bell free energy. From Eqs. (D25) and (D26) we have
Using Eq. (D29), we obtain
which is positive (\(\dot{J}\ge 0\)) provided that \(D>0\). Therefore, the free energy F decreases monotonically until an equilibrium state of the form (D5) is reached (H-theorem). In the canonical ensemble, we can show that \(2\lambda \delta ^{2}{J}=\delta ^{2}\dot{J}\ge 0\) [149]. Therefore, a stationary solution of the generalized Fokker–Planck equation (D33) with constant temperature is linearly stable if, and only if, it is a (local) minimum of free energy F at fixed mass. Therefore, this equation can only relax towards minima of F, not towards maxima or saddle points.
Remark
By using the MEPP (or the linear thermodynamics of Onsager), we have constructed a kinetic equation for the coarse-grained DF, Eq. (D33), which relaxes towards the Lynden-Bell DF (D5). This kinetic equation has the form of a fermionic Kramers equation. We stress that this thermodynamical approach is phenomenological in nature. It is also possible to derive a kinetic equation for the coarse-grained DF through a systematic procedure by developing a quasilinear theory of the Vlasov–Poisson equations [96,97,98,99,100,101]. This kinetic equation has the form of a fermionic Landau equation. Although it is satisfying to derive this equation from a systematic procedure, its domain of validity remains unclear (it may only describe a “gentle” relaxation). However, the fermionic Kramers equation can be obtained from the fermionic Landau equation by making a thermal bath approximation [96]. In that case, the diffusion coefficient (which is not determined by the MEPP) can be explicitly calculated. The fermionic Kramers equation with a time-dependent temperature and the fermionic Landau equation both satisfy an H-theorem for the Lynden-Bell entropy. We note, however, that the fermionic Landau equation conserves the energy locally while the fermionic Kramers equation with a time-dependent temperature conserves the energy globally. This difference affects the hydrodynamic equations derived from these kinetic equations: they involve either a viscosity (Landau) or a friction (Kramers) [127].Footnote 53 We finally note that it is possible to generalize the results of this Appendix to an arbitrary entropy of the form \(S=-\int C(\overline{f})\, \mathrm{d}{} \mathbf{r}\mathrm{d}{} \mathbf{v}\), where C is a convex function [100, 151]. This generalization may allow us to solve heuristically the problem of incomplete relaxation mentioned in Sect. 3.3 by considering entropic functionals different from the Lynden-Bell entropy that yield DFs with a finite mass. It can also allow us to deal with more complex situations than those considered here. In any case, the MEPP provides a numerical algorithm in the form of a generalized Fokker–Planck equation that can be used to construct dynamically stable stationary solutions of the Vlasov–Poisson equations, a nontrivial problem in itself [151]. Finally, to be complete, we note that a different kinetic approach of collisionless relaxation has been developed in [152] where the coarse-grained DF is defined as the convolution of the fine-grained DF by a Gaussian window.
1.3 3. Hydrodynamic equations
We now take the hydrodynamic moments of the coarse-grained Vlasov equation (D33). To make the results fully explicit, we consider the nondegenerate limit of the theory (the general case is treated in [94]). In that case, the coarse-grained Vlasov equation takes the form
which is similar to the classical Kramers equation. The evolution of the inverse effective temperature \(\beta (t)=1/T_\mathrm{eff}(t)\) is given by
Assuming that D is constant, and making an integration by parts, we get
where
is the total kinetic energy. Eq. (D49) can be rewritten as \(K(t)=\frac{3}{2}MT_\mathrm{eff}(t)\). This relation shows that the effective temperature \(T_\mathrm{eff}(t)=1/\beta (t)\) can be interpreted as the average kinetic temperature of the system, i.e., \(T_\mathrm{eff}(t)=\langle T_\mathrm{kin}(\mathbf{r},t)\rangle \) where \(\frac{3}{2}\rho T_\mathrm{kin}(\mathbf{r},t)=\int f \frac{v^2}{2}\, d\mathbf{v}\) is the local kinetic energy. One has \(E=\frac{3}{2}MT_\mathrm{eff}+W\).
Taking the hydrodynamic moments of the coarse-grained Vlasov equation (D47), we obtain a system of equations similar to the Jeans equations (see Appendix C2) but including dissipative effects:
where \(\xi =D\beta \). They are called the damped Jeans equations. On the other hand, the effective collision term in Eq. (D47) provides a source of relaxation which allows us to close the hierarchy of moment equations. Indeed, we can compute the pressure tensor in Eq. (D52) by making a LTE approximation
In that case, we get
This leads to a system of hydrodynamic equations of the form
called the damped Euler equations. Using the LTE approximation (D53), the total energy (D3) can be written as
In the microcanonical ensemble where the energy E is fixed, this equation determines the evolution of the effective temperature \(T_\mathrm{eff}(t)\). On the other hand, the Boltzmann-like entropy (D18) can be written (up to an additive constant) as
This justifies the expressions of E and S used in Appendix B2. One can then show (like in Appendix B2) that the hydrodynamic equations (D55)–(D57) with a time-dependent effective temperature \(T_\mathrm{eff}(t)\) determined by Eq. (D58) satisfy an H-theorem for the entropy S given by Eq. (D59). Analogously, in the canonical ensemble where the effective temperature \(T_\mathrm{eff}\) is fixed, one can show (like in Appendix B3) that the hydrodynamic equations (D55)–(D57) with a constant effective temperature \(T_\mathrm{eff}\) satisfy an H-theorem for the free energy \(F=E-T_\mathrm{eff}S\). One can also derive explicit expressions of the virial theorem like in Appendices B2 and B3 (one simply has to take \(P=Q=0\) in the equations of these Appendices).
Remark
We note that, in Eq. (D58), the effective temperature \(T_\mathrm{eff}\) represents the thermal kinetic energy \(\frac{3}{2}M T_\mathrm{eff}(t)=\int f \frac{w^2}{2}\, \mathrm{d}{} \mathbf{r}d\mathbf{v}\) where \(\mathbf{w}=\mathbf{v}-\mathbf{u}(\mathbf{r},t)\) is the fluctuating velocity while the exact relation (D49) indicates that it should represent the total kinetic energy \(\frac{3}{2}M T_\mathrm{eff}(t)=\int f \frac{v^2}{2}\, \mathrm{d}{} \mathbf{r}\mathrm{d}{} \mathbf{v}\) including the mean kinetic energy. This is an artefact of the LTE approximation (D53) which involves a uniform temperature \(T_\mathrm{eff}(t)\) instead of a space-dependent temperature \(T(\mathbf{r},t)\). We can solve this problem by replacing \(T_\mathrm{eff}(t)\) by \(T(\mathbf{r},t)\) in Eq. (D53) and introducing a hydrodynamic equation for the local temperature \(T(\mathbf{r},t)\) (second moment) as in Ref. [94] (see Appendix D4). However, the LTE approximation (D53) becomes exact close to equilibrium or in the strong friction limit \(\xi \rightarrow +\infty \) where \(|\mathbf{u}|=O(1/\xi )\). Indeed, in the strong friction limit, we have
Substituting this relation into the continuity equation (D55) we obtain the Smoluchowski–Poisson equations
Since \(|\mathbf{u}|=O(1/\xi )\), we can neglect the kinetic energy \(\Theta _c\) in Eq. (D58). Therefore, the evolution of the effective temperature \(T_\mathrm{eff}(t)\) is given by the energy constraint
In the canonical ensemble, Eqs. (D61) and (D62) are valid with fixed \(T_\mathrm{eff}\). One can then derive the H-theorems and the virial theorems like in the Remarks at the end of Appendices B2 and B3 (one simply has to take \(P=Q=0\) in the equations of these Appendices). We refer to the series of papers initiated in [123] for a detailed study of these equations.
1.4 4. Inhomogeneous temperature
In the previous section, we have assumed for simplicity that the temperature (velocity dispersion) is uniform. In a more elaborate model (see Ref. [94]), we can take into account an inhomogeneous temperature by considering the second moment of the coarse-grained Vlasov equation and closing the hierarchy of hydrodynamic equations with the LTE approximation (D53) with \(T_\mathrm{eff}(t)\) replaced by \(T(\mathbf{r},t)\). This yields (see Refs. [94, 153] for details and generalizations)
where \(T_\mathrm{eff}(t)=1/\beta (t)\) evolves according to Eq. (D49) so as to conserve the total energy
The pressure is \(P=\rho T\) and the thermal energy is \(\Theta _\mathrm{th}=\int f \frac{w^2}{2}\, \mathrm{d}{} \mathbf{r}\mathrm{d}{} \mathbf{v}=\frac{3}{2}\int \rho T\, \mathrm{d}{} \mathbf{r}\). These equations conserve the mass and the energy and satisfy an H-theorem for the Boltzmann-like entropy
As a result, they relax towards the equilibrium Boltzmann-like distribution with a uniform temperature \(T=T_\mathrm{eff}\). The equation for the entropy density \(\sigma =-\ln (\rho T^{-3/2})\) is
In the absence of dissipation (\(\xi =0\)) we recover the usual Euler equations [154] which conserve the entropy. In that case, \(\rho T^{-3/2}\) is constant along the fluid (Lagrangian) trajectories, corresponding to the adiabatic law \(P\rho ^{-5/3}=\mathrm{cst}\). The damped virial theorem is given by (see, e.g., Appendix G of [93])
Using the energy conservation equation (D68), it can be rewritten as
In the canonical ensemble, \(T_\mathrm{eff}\) is constant and the hydrodynamic equations satisfy an H-theorem for the free energy \(F=E-T_\mathrm{eff}S\).
Appendix E: The classical limit \(\hbar \rightarrow 0\)
In this Appendix, we discuss how the quantum equations studied in the present paper pass to the limit \(\hbar \rightarrow 0\).Footnote 54 Our discussion is essentially heuristic. A mathematically rigorous treatment of the classical limit \(\hbar \rightarrow 0\) is difficult but would certainly be very valuable.
1.1 1. Classical versus quantum descriptions
In the collisionless regime, classical self-gravitating systems are described by the Vlasov–Poisson equations (C1) and (C2). The Vlasov equation is equivalent to the infinite hierarchy of Jeans equations (see Appendix C2). This is true for all times and for arbitrary initial conditions. Let us now assume that we start from a single-speed initial condition. Then, as long as the DF remains single-speed, the DF is given by Eq. (C13), the pressure tensor vanishes (\(P_{ij}=0\)), and the Jeans equations reduce to the pressureless Euler equations (C14) and (C15). This is true until shell crossing, after which the pressureless Euler equations develop singularities and the DF becomes multi-streamed. Before shell crossing the Vlasov equation is equivalent to the pressureless Euler equations but after shell crossing the pressureless Euler equations are not valid anymore. In that case \(P_{ij}\ne 0\). Unfortunately, it is not possible to calculate the pressure tensor exactly so the Jeans equations are not closed. One has to come back to the Vlasov–Poisson equations. Some heuristic procedures to compute \(P_{ij}\) approximately to continue using a hydrodynamical approach are described in Appendix C3.
In the collisionless regime, quantum self-gravitating systems made of condensed bosons are described by the Schrödinger–Poisson equations (10) and (11). The Schrödinger equation is equivalent to the quantum Euler equations (17)–(19). These equations are also equivalent to the Wigner equation (44), whose exact solution is given by Eq. (43), and to the quantum Jeans equations (see Sect. 3.4.1). For condensed bosons, \(P_{ij}=P_{ij}^Q\), where \(P_{ij}^Q\) is given by Eq. (23), and the quantum Jeans equations are closed (they reduce to the quantum Euler equations (17)–(19)). This is true for all times and for arbitrary initial conditions.
1.2 2. Comparison between the Wigner equation and the Vlasov equation
When \(\hbar \rightarrow 0\), the Wigner equation (44) reduces to the Vlasov equation (C1) which corresponds to \(\hbar =0\) (see Appendix A). On the other hand, the Wigner equation is equivalent to the Schrödinger equation (10) and to the quantum Euler equations (17)–(19). Therefore, the solution of the Vlasov equation (\(\hbar =0\)) is expected to be well-approximated by the solution of the Wigner, Schrödinger and quantum Euler equations with \(\hbar \rightarrow 0\). In particular, \(f(\mathbf{r},\mathbf{v},t)\simeq f_W(\mathbf{r},\mathbf{v},t)\) where \(f_W\) is given by Eq. (43). This equivalence is valid for all times. Therefore, for what concerns the Wigner equation, the limit \(\hbar \rightarrow 0\) is equivalent to \(\hbar =0\).
1.3 3. Comparison between the quantum Euler equation and the pressureless Euler equations
When \(\hbar \rightarrow 0\), the quantum Euler equations (17)–(19) seem to reduce to the pressureless Euler equations (C14) and (C15) which correspond to \(P=\hbar =0\). However, this equivalence is only valid before shell crossing (\(t<t_*\)). After shell crossing (\(t>t_*\)), the quantum Euler equations are still valid while the pressureless Euler equations are not valid anymore (see Appendix C3). Therefore, for what concerns the quantum Euler equations (17)–(19), the limit \(\hbar \rightarrow 0\) is different from \(\hbar =0\). A small but finite value of \(\hbar \) allows us to extend the solutions of the hydrodynamic equations past \(t_*\) for all times.
1.4 4. Conclusion
The Vlasov equation is valid for all times. By contrast, the pressureless Euler equations are valid only for a single speed solution until the first time of crossing. Before that time, they are equivalent to the Vlasov equation but after that time they break down. The Schrödinger equation, the quantum Euler equations, and the Wigner equation are equivalent and they are valid for all times. When \(\hbar \rightarrow 0\) their solutions are expected to be close to the solution of the Vlasov equation.
In conclusion, the Vlasov equation and the Schrödinger, quantum Euler, and Wigner equations with \(\hbar \rightarrow 0\) are superior to the classical pressureless Euler equations. They take into account velocity dispersion whereas the classical pressureless fluid description does not. They can be used to describe multistreaming and caustics in the nonlinear regime, whereas the classical pressureless fluid equations break down in that regime. Indeed, the pressureless Euler equations develop shocks so they are not well-defined after the first shock. The velocity dispersion, or the quantum pressure, allow to regularize the dynamics and solve the problems of the classical pressureless hydrodynamic description.
Appendix F: The Vlasov–Bohm equation
Instead of using the rather complicated Wigner equation (44), one could consider the simpler Vlasov–Bohm equation
This equation is similar to the classical Vlasov equation (C1) except that it includes the quantum potential Q from Eq. (21). We stress, however, that this equation is heuristic and that it is not expected to give an exact description of the dynamics. The Vlasov–Bohm–Poisson equations conserve the energy \(E=(1/2)\int f v^2\, \mathrm{d}{} \mathbf{r}\mathrm{d}{} \mathbf{v}+(1/2)\int \rho \Phi \, \mathrm{d}{} \mathbf{r}+(1/m)\int \rho Q\, \mathrm{d}{} \mathbf{r}\) and an infinite class of Casimir integrals of the form \(\int h(f)\, \mathrm{d}{} \mathbf{r}\mathrm{d}{} \mathbf{v}\) where h(f) is an arbitrary function of f. Coarse-graining the Vlasov–Bohm equation like in Sect. 3.2, we obtain the fermionic Vlasov–Bohm-Kramers equation
instead of Eq. (46) [one can also obtain a fermionic Vlasov-Bohm–Landau equation instead of Eq. (49)]. This equation relaxes towards an equilibrium DF of the form
which can be viewed as a Lynden-Bell DF incorporating the quantum potential. Since Q depends on the density itself this equation is a complicated integral equation. It implies the condition of quantum hydrostatic equilibrium from Eq. (103) (see the Remark below).
We can then take the hydrodynamic moments of the coarse-grained Vlasov–Bohm–Kramers equation (F2) as in Sect. 3.4. Instead of Eq. (57), we get
in which the quantum potential appears explicity. Closing the hierarchy of quantum Jeans equations by making the LTE approximation from Eq. (64) to compute the pressure tensor, we obtain
which coincides with Eq. (73). Therefore, at the level of the hydrodynamic equations, and within the assumptions made in our approach, the Vlasov–Bohm equation yields the same results as the Wigner equation. The reason is that our approach neglects collective effects (encapsulated in the dielectric function) where differences between the Wigner equation and the Vlasov–Bohm equation occur [127]. This approximation may be justified during the very nonlinear process of violent relaxation.
Remark
For spherically symmetric systems, the stationary solutions of the Vlasov–Bohm equation (F1) are of the form \(f=f(\epsilon _Q)\) with \(\epsilon _Q=v^2/2+\Phi +Q/m\). We can easily show that this relation implies the condition of quantum hydrostatic equilibrium from Eq. (103). Indeed, defining the density and the pressure by \(\rho =\int f\, \mathrm{d}{} \mathbf{v}\) and \(P=\frac{1}{3}\int f v^2\, \mathrm{d}{} \mathbf{v}\), we get
yielding
The effective collision term on the right hand side of Eq. (F2) selects the form of the equilibrium DF \(f=f(\epsilon _Q)\) among all the possible stationary solutions of the Vlasov–Bohm equation. In the present case, the fermionic Kramers operator selects the Lynden-Bell DF from Eq. (F3).
Appendix G: Multistate systems
1.1 1. Hartree equations
We consider a system of N quantum particles in gravitational interaction. These particles may be fermions or bosons. Fundamentally, they are described by an N-body wave function \(\Psi (\mathbf{r}_1,...,\mathbf{r}_N,t)\) satisfying the exact N-body Schrödinger equation [155]
When N is large, this equation is completely untractable. The aim of statistical mechanics and kinetic theory is to obtain simpler models described by one-body equations. Following Hartree [156, 157] we shall make a mean field approximation and ignore correlations among the particles. We thus assume that the N-body wave function \(\Psi (\mathbf{r}_1,...,\mathbf{r}_N,t)\) can be written as a product of N one-body wave functions \(\psi _{\alpha }(\mathbf{r},t)\) so that
The mean field approximation is known to become exact for systems with long-range interactions (like self-gravitating systems) when \(N\rightarrow +\infty \) in a proper thermodynamic limit [131]. In that case, the evolution of the quantum system is described by N coupled mean field Schrödinger equations of the form
where \(p_{\alpha }\) is the occupation probability of state \(\alpha \) such that the normalization condition \(\sum _{\alpha }p_{\alpha }=1\) is fulfilled. We shall assume that \(p_{\alpha }\) is constant or that it varies slowly with \(\mathbf{r}\) and t. The total density is \(\rho =\sum _{\alpha } p_{\alpha } |\psi _{\alpha }|^2\).
In the mean field approximation, each particle evolves in a self-consistent gravitational potential \(\Phi \) that is produced by the particles themselves through the Poisson equation (G4). Equations (G3) and (G4) are called the Hartree (mean field) equations.
From the wavefunctions \(\psi _{\alpha }(\mathbf{r},t)\) of the multistate system, we can define the Wigner DF by
It satisfies the Wigner equation (44). The Wigner DF (G5) involves the density matrix \(\rho (\mathbf{r},\mathbf{r}',t)=\sum _{\alpha =1}^N p_{\alpha }\psi _{\alpha }(\mathbf{r},t)\psi _{\alpha }^*(\mathbf{r},t)\) which appears in the von Neumann equation. The Hartree, Wigner and von Neumann equations are equivalent.
In the case of fermions, the wave function \(\Psi \) is antisymmetric with respect to the exchange of any two variables \(\mathbf{r}_\alpha \) and \(\mathbf{r}_\beta \). The antisymmetry condition stems from the Pauli exclusion principle which establishes that two fermions cannot share the same position so that the probability density \(|\Psi |^2\) vanishes as \(\mathbf{r}_\alpha =\mathbf{r}_\beta \) (for a rigorous treatment, the spin of the fermions must be considered). A system of fermions is necessarily in a mixed quantum state to respect the Pauli exclusion principle. In the Hartree–Fock [158, 159] theory, the N-body wave function is expressed as
In other words, the wave function for many fermions is taken as a Slater determinant which is zero for two particles at the same position. This expression guarantees that the system obeys the Pauli exclusion principle. This leads to the Hartree–Fock equations corresponding to the Hartree Eqs. (G3) and (G4) with an additional exchange energy term introduced by Fock [158]. This exchange energy term can be simplified by using the Slater [160] approximation (see also Dirac [161] and Kohn–Sham [162]) which amounts to introducing a term \(C_S \rho ^{1/3}\psi _{\alpha }\) in the right hand side of Eq. (G3), where \(C_S\) is a positive constant in the gravitational case (it is negative in the electrostatic case). The Hartree–Fock equations neglect correlations. More general equations taking into account correlations (assuming that the fermions interact by microscopic forces in addition to the gravitational force) can be obtained through the density functional theory [155] initiated by Kohn and Sham [162].
A gas of bosons at \(T=0\) forms a BEC in which all the bosons are in the same quantum state described by a single wave function \(\psi (\mathbf{r},t)\). In the mean field approximation, the N-body wave function is expressed as
In that case, the evolution of the self-gravitating BEC is described by the Schrödinger–Poisson equations (10) and (11). This amounts to taking \(\alpha =1\) and \(p_{\alpha }=1\) in Eqs. (G3) and (G4). A system of bosons at \(T=0\) is in a pure quantum state.
Remark
Actually, even in the case of BECs, we can have a quantum superposition of modes. Indeed, the wave function \(\psi \) can be decomposed into Eigenfunctions of the Schrödinger–Poisson equations involving the fundamental state (\(n=0\)) and the excited states (\(n>0\)) as in Eq. (41). As a result, the multistate equations (G3) and (G4) are also relevant for bosons with this interpretation (in that case, \(p_\alpha \) is the probability of mode \(\alpha \) and N is the number of modes). The resulting equations can be called the Hartree (mean field) equations for bosons. In the case of fermions, the Hartree equations (G3) and (G4) are basically valid with \(p_{\alpha }=1/N\) if \(\alpha =1,...,N\) and \(p_{\alpha }=0\) otherwise. However, as explained above, they can also be viewed as an expansion over the modes \(\alpha \) of the system where \(p_{\alpha }\) represents the probability of mode \(\alpha \) and N represents the number of modes.Footnote 55 Below, we generalize the hydrodynamic representation of the SP equations to the case of a multistate system and show how a pressure term arises in the quantum Euler equations. This pressure term is similar to the one obtained from the approach of Sect. 3.
1.2 2. Madelung transformation
We can use the Madelung transformation to rewrite the Hartree–Fock equations (G3) and (G4) under the form of hydrodynamic equations for each state \(\alpha \) (we shall use the Slater approximation \(C_S \rho ^{1/3}\psi _{\alpha }\) to evaluate the exchange term for fermions). Let us write the wave function as
where \(\rho _{\alpha }(\mathbf{r},t)\) is the density and \(S_{\alpha }(\mathbf{r},t)\) is the action given by
Following Madelung, we introduce the velocity field
Substituting Eq. (G8) into Eqs. (G3) and (G4) and separating the real and the imaginary parts, we find that the Hartree–Fock equations are equivalent to hydrodynamic equations of the form
where
is the quantum potential and
is the Slater pressure. It corresponds to a polytrope of index \(n=3\) with a polytropic constant \(K_S=C_S/4m\). The Slater pressure is positive (\(P_\mathrm{Slater}>0\)) meaning that the exchange interaction is effectively repulsive in the gravitational case (it is attractive in the electrostatic case). Using the continuity equation (G11), we obtain the identity
The quantum Euler equation (G13) can then be rewritten as
The foregoing equations are valid for each state \(\alpha \). We now introduce the total density
and the total velocity
From Eqs. (G11), (G19) and (G20), we obtain the continuity equation
From Eqs. (G18), (G19) and (G20), we get
where \(\mathbf{w}_{\alpha }=\mathbf{u}_{\alpha }-\mathbf{u}\). Expanding the advection term in Eq. (G22) and using the identity
we are left with
Using the continuity equation (G21), we obtain the identity
The Euler equation (G24) can then be rewritten as
where
is the pressure tensor arising from the difference between the multistate velocities \(\mathbf{u}_{\alpha }\) and the total velocity \(\mathbf{u}\). It can also be written as
If we make the approximation
where Q is the quantum potential defined by Eq. (21), we obtain the hydrodynamic equations
These equations are not closed since the pressure tensor \(P_{ij}\) depends on the multistate variables. We note that the pressure tensor defined by Eq. (G27) is similar to the pressure tensor defined by Eq. (60) if we identify \(p_{\alpha }\rho _{\alpha }\) with the coarse-grained DF \(\overline{f}(\mathbf{r},\mathbf{v})\). We shall consider successively the case of bosons and fermions.
1.3 3. Bosons
In the case of bosons at \(T=0\) forming a BEC, we just have one pure state (\(\alpha =p_\alpha =1\)), and we trivially find that the pressure tensor defined by Eq. (G27) vanishes
We also have \(P_\mathrm{Slater}=0\) in that case (\(C_S=0\)). As a result, the hydrodynamic equations (G30)–(G32) reduce to Eqs. (17)–(20) as it should. However, if we write \(\psi \) as a superposition of modes [see Eq. (41)], and use Eqs. (G3) and (G4) with this interpretation (see the Remark at the end of Appendix G1), we can close Eqs. (G30)–(G32) by using the Lynden-Bell pressure from Eq. (65). This leads to Eqs. (72)–(74) and, finally, to Eq. (100). Note, however, that the friction term, which represents a form of nonlinear Landau damping, does not explicitly appear in the present formalism.
1.4 4. Fermions
In the case of fermions, the pressure tensor \(P_{ij}\) takes into account the Pauli exclusion principle but, without further assumption, it cannot be explicitly evaluated. Now, if we view \(\sum _\alpha \) as a sum over the modes with \(p_\alpha \) being the probability of mode \(\alpha \) (see the Remark at the end of Appendix G1), we can close Eqs. (G30)–(G32) by using the Lynden-Bell pressure from Eq. (65) which, in the case of fermions, coincides with the Fermi–Dirac pressure. This leads to the hydrodynamic equations
and, finally, to Eq. (158) with the same comment as above concerning the absence of friction term. We note that the Lynden-Bell (or Fermi–Dirac) pressure has a statistical origin (it depends on the specification of \(p_\alpha \)) while the Slater pressure has a purely quantum nature.
Rights and permissions
About this article
Cite this article
Chavanis, PH. A heuristic wave equation parameterizing BEC dark matter halos with a quantum core and an isothermal atmosphere. Eur. Phys. J. B 95, 48 (2022). https://doi.org/10.1140/epjb/s10051-022-00299-9
Received:
Accepted:
Published:
DOI: https://doi.org/10.1140/epjb/s10051-022-00299-9