Introduction
We consider the three-dimensional (3-D) attractive compressible Euler–Riesz equations (CEREs)
{ ∂ t ρ + div M = 0 , ∂ t M + div ( M ⊗ M ρ ) + ∇ p ( ρ ) + ρ ∇ W α = 0 , (1) \begin{cases}
\partial_t\rho+\operatorname{div}\mathbf{M}=0,\\
\partial_t\mathbf{M}+\operatorname{div}\left(\frac{\mathbf{M}\otimes\mathbf{M}}{\rho}\right)+\nabla p(\rho)+\rho\nabla W_\alpha=0, \tag*{(1)}
\end{cases} { ∂ t ρ + div M = 0 , ∂ t M + div ( ρ M ⊗ M ) + ∇ p ( ρ ) + ρ ∇ W α = 0 , (1)
for ( t , x ) ∈ R + × R 3 (t,\mathbf{x})\in\mathbb{R}_{+}\times\mathbb{R}^{3} ( t , x ) ∈ R + × R 3 , where ρ : R + × R 3 → R + \rho:\mathbb{R}_{+}\times\mathbb{R}^{3}\to\mathbb{R}_{+} ρ : R + × R 3 → R + is the density, M : R + × R 3 → R 3 \mathbf{M}:\mathbb{R}_{+}\times\mathbb{R}^{3}\to\mathbb{R}^{3} M : R + × R 3 → R 3 is the momentum, and p = p ( ρ ) p=p(\rho) p = p ( ρ ) is a barotropic pressure law. For the relative entropy analysis, we restrict to α ∈ ( 0 , 2 ) \alpha\in(0,2) α ∈ ( 0 , 2 ) and set W α = Φ α ∗ ρ W_\alpha=\Phi_\alpha*\rho W α = Φ α ∗ ρ , where
∇ W α ( x ) = ( ∇ Φ α ∗ ρ ) ( x ) \nabla W_\alpha(x)=(\nabla\Phi_\alpha*\rho)(x) ∇ W α ( x ) = ( ∇ Φ α ∗ ρ ) ( x )
and
Φ α ( x ) = − ∣ x ∣ − α α . \Phi_\alpha(x)=-\frac{|x|^{-\alpha}}{\alpha}. Φ α ( x ) = − α ∣ x ∣ − α .
Riesz potentials provide a natural scale of long-range interactions appearing in plasma physics, stellar and galactic dynamics, condensed-matter models, random-matrix theory, and collective phenomena such as swarming [4 , 8 , 9 , 36 ] . In 3-D, α = 1 \alpha=1 α = 1 corresponds to the Newtonian (Coulomb) potential and W 1 W_1 W 1 satisfies the Poisson equation, up to normalization. In this case, (1.1) reduces to the compressible Euler–Poisson equations (CEPEs). For α ≠ 1 \alpha\ne1 α = 1 , the potential satisfies the fractional elliptic equation
( − Δ ) 3 − α 2 W α = − c α ρ (-\Delta)^{\frac{3-\alpha}{2}}W_\alpha=-c_\alpha\rho ( − Δ ) 2 3 − α W α = − c α ρ
in the sense of distributions, where c α > 0 c_\alpha>0 c α > 0 is a normalization constant. Such equations can also be applied in astrophysical models of fractional gravity. For example, Jeans instability for CEREs was studied in [43 ] . The lack of a local Poisson formulation and the differing singular structures of sub-Coulomb and super-Coulomb potentials present significant analytical challenges, particularly in establishing compactness and obtaining uniform estimates for nonlocal interaction terms.
Choi-Jeong [17 ] established the local classical well-posedness for the CEREs with either attractive or repulsive interactions, and for the pressureless system in the repulsive case; they also obtained sufficient conditions for finite-time breakdown in isentropic and isothermal flows. Global classical well-posedness results are perturbative and use different mechanisms. For isentropic flows, global solutions have been constructed for small initial density and velocity close to a uniform state [21 , 23 ] . Other results show that damping can stabilize pressureless repulsive CEREs near a constant background [18 ] , or exploit the dispersive structure of the 3-D repulsive CEREs for small irrotational perturbations of a uniform equilibrium [20 ] . More recently, for the pressureless damped CEREs with repulsive interactions, Chi-Shou-Xu [16 ] established the global well-posedness for small perturbations of a positive constant state in a critical Besov framework and obtained decay rates.
For large initial data, most recently, Carrillo-Charles-Chen-Yuan [6 ] constructed global spherically symmetric finite-energy weak solutions of the multidimensional CEREs for both attractive and repulsive interactions. The existence of weak solutions and the nonlinear stability and instability of steady states were subsequently extended to general pressure laws; see [7 ] . Unlike the Coulomb potential, the radial Riesz potential has no local Poisson representation, and its singular behavior near the origin must therefore be controlled carefully. For the polytropic case, developing uniform higher-integrability estimates and using compensated compactness yield the strong convergence of the approximate solutions and exclude the mass concentration at the origin, even for attractive super-Coulomb potentials. Passing to the limit also requires the strong compactness of the nonlocal interaction term.
For the gravitational CEPEs, the global existence of finite-energy weak solutions for large spherically symmetric initial data was first established by Chen-He-Wang-Yuan [13 ] for polytropic stars and subsequently extended by Chen et al. [14 ] to general pressure laws. A different class of global solutions with large data was constructed by Hadžić-Jang [32 ] , who established the existence of compactly supported physical-vacuum solutions in 3-D, without imposing symmetry assumptions, that remain close to expanding affine motions. Concerning the existence and stability of non-rotating stellar equilibria, Rein [44 ] constructed barotropic steady states as constrained energy minimizers and established their conditional nonlinear stability under general perturbations. For general equations of state, Lin-Zeng [40 ] established a turning point principle for the linear stability of non-rotating stars and determined the number of unstable modes along the mass-radius curve. Building on the sharp linear stability criterion, Lin-Wang-Zhu [39 ] subsequently proved nonlinear orbital stability for the CEPEs. Further results concern the instability and expansion of self-gravitating stars. Deng-Liu-Yang-Yao [24 ] established the linear growth of the support for positive energy solutions with γ > 4 3 \gamma> \frac{4}{3} γ > 3 4 . Jang [33 , 34 ] proved the nonlinear instability of Lane-Emden stars for γ = 6 5 \gamma= \frac{6}{5} γ = 5 6 and γ ∈ ( 6 5 , 4 3 ) \gamma\in\left(\frac{6}{5}, \frac{4}{3}\right) γ ∈ ( 5 6 , 3 4 ) . More recently, Cheng et al. [15 ] constructed expanding solutions near unstable Lane-Emden equilibria.
Of particular relevance here are Yan Guo’s contributions to both the plasma and self-gravitating CEPEs. For the 3-D electron-fluid model, Guo [26 ] constructed global smooth irrotational flows with sufficiently small initial velocity. Guo-Pausader [28 ] subsequently proved global smoothness for small-amplitude irrotational ion dynamics. In the gravitational setting, Guo-Hadžić-Jang showed that, for every γ ∈ ( 1 , 4 3 ) \gamma\in\left(1, \frac{4}{3}\right) γ ∈ ( 1 , 3 4 ) , there exists an infinite-dimensional family of spherically symmetric, compactly supported physical vacuum solutions undergoing continued gravitational collapse; see also [29 ] . They also rigorously constructed the isothermal Larson-Penston self-similar collapse profile [27 ] . Guo-Hadžić-Jang-Schrecker [30 ] constructed real-analytic Yahil-type polytropic profiles on the full self-similar radial domain for every γ ∈ ( 1 , 4 3 ) \gamma\in\left(1, \frac{4}{3}\right) γ ∈ ( 1 , 3 4 ) . Their recent work proves nonlinear stability of the Larson-Penston family under radially symmetric perturbations [31 ] .
The variational construction of rotating Euler-Poisson stars was initiated by Auchmuty-Beals [3 ] , who established the existence of axisymmetric equilibria as constrained energy minimizers with prescribed mass and distribution of squared specific angular momentum. Subsequently, Li [37 ] proved the existence of uniformly rotating stellar equilibria with prescribed mass for sufficiently small angular velocities. Chanillo–Li [10 ] established a priori bounds on the diameters and the number of connected components of uniformly rotating white dwarf equilibria. Luo-Smoller [42 ] later constructed energy minimizers in a broader class of axisymmetric configurations and, under the assumption of uniqueness up to translations along the rotation axis, established their conditional nonlinear orbital stability with respect to global weak solutions. Further existence results for general equations of state, including models of rotating and non-rotating white dwarfs and rotating supermassive stars, were obtained in [41 ] . Concerning linear stability, Lin-Wang [38 ] derived a sharp stability criterion for axisymmetric perturbations in the Rayleigh-stable regime and established linear instability when the Rayleigh stability condition fails. For further developments on the existence, structure, and stability of rotating stellar equilibria, we refer to [11 , 25 , 35 , 45 , 46 , 47 , 48 , 49 ] and the references therein.
In [42 ] , Luo-Smoller employed the relative entropy method to establish finite-time stability around steady solutions of the CEPEs with W l o c 1 , ∞ W^{1,\infty}_{\mathrm{loc}} W loc 1 , ∞ -regularity, including rotating stellar equilibria. For the CEREs on the torus or in smooth bounded domains, Alves et al. [1 ] established weak-strong uniqueness through a relative entropy argument, under the assumption that the density of the strong solution is uniformly bounded above and bounded away from vacuum. Related advances in the treatment of singular nonlocal interactions have been achieved through modulated energy methods. Choi-Jung [19 ] established modulated interaction energy estimates for singular potentials. For the repulsive CEREs, Alves-Grafakos-Tzavaras [2 ] developed bilinear fractional integral estimates, obtaining higher space-time integrability of finite-energy solutions together with stability estimates. More recently, Carrillo et al. [7 ] established relative entropy estimates around compactly supported, non-rotating polytropic equilibria. Their approach relies on finite-energy bounds and, importantly, does not require a strictly positive lower bound on the density of the steady state.
The proof combines the finite-energy entropy framework developed in [6 ] with the relative entropy method and the weighted Riesz potential estimates established in [7 ] . Motivated by the analysis in [41 , 42 ] , we extend the finite-time stability estimates to rotating stellar equilibria governed by general pressure laws. The main analytical difficulties and the methods used to overcome them are outlined below:
For the compactly supported equilibrium ρ ˉ \bar{\rho} ρ ˉ , we retain the weighted term involving V ˉ : = V ρ ˉ \bar{V}:=V_{\bar{\rho}} V ˉ := V ρ ˉ in d 1 d_1 d 1 to obtain direct control of the far-field mass. An analogous weighted term is used for non-rotating CEPEs in [39 ] . Under Assumption 2.1, we allow different power-law behaviors of the pressure at low and high densities. With f = ρ − ρ ˉ f=\rho-\bar{\rho} f = ρ − ρ ˉ and Γ = { ρ ˉ > 0 } \Gamma=\{\bar{\rho}>0\} Γ = { ρ ˉ > 0 } , (35) gives
h ( ρ ) − h ( ρ ˉ ) + V ˉ ( ρ − ρ ˉ ) = D h ( ρ ∣ ρ ˉ ) on Γ . h(\rho)-h(\bar{\rho})+\bar{V}(\rho-\bar{\rho})=D_h(\rho\mid\bar{\rho}) \qquad\text{on }\Gamma. h ( ρ ) − h ( ρ ˉ ) + V ˉ ( ρ − ρ ˉ ) = D h ( ρ ∣ ρ ˉ ) on Γ.
On the vacuum region Γ c \Gamma^{c} Γ c , the corresponding integrand reduces to h ( ρ ) + V ˉ ρ h(\rho)+\bar{V}\rho h ( ρ ) + V ˉ ρ . The positivity property (29) provides control of the contribution from the vacuum region, while the far-field estimate (31) controls the mass spreading towards spatial infinity. More precisely, Lemma 4.1 provides control of the integrals of ∣ f ∣ 2 |f|^2 ∣ f ∣ 2 and ∣ f ∣ γ / 2 |f|^{\gamma/2} ∣ f ∣ γ /2 over Γ ∩ { ∣ f ∣ ≤ 1 } \Gamma\cap\{|f|\leq1\} Γ ∩ { ∣ f ∣ ≤ 1 } and Γ ∩ { ∣ f ∣ > 1 } \Gamma\cap\{|f|>1\} Γ ∩ { ∣ f ∣ > 1 } , respectively, as well as those of ρ γ \rho^{\gamma} ρ γ and ρ γ / 2 \rho^{\gamma/2} ρ γ /2 over Γ ∩ { ρ ≤ 1 } \Gamma\cap\{\rho\leq1\} Γ ∩ { ρ ≤ 1 } and Γ c ∩ { ρ > 1 } \Gamma^{c}\cap\{\rho>1\} Γ c ∩ { ρ > 1 } , respectively. It also controls the weighted vacuum term
∫ Γ c V ˉ ρ d x . \int_{\Gamma^{c}}\bar{V}\rho\,\mathrm{d}x. ∫ Γ c V ˉ ρ d x .
Combining these estimates with Lemma 4.2, we conclude that
ρ − ρ ˉ ∈ L s ( R 3 ) for any s ∈ ( 1 , γ ] . \rho-\bar{\rho}\in L^{s}(\mathbb{R}^{3}) \qquad\text{for any }s\in(1,\gamma]. ρ − ρ ˉ ∈ L s ( R 3 ) for any s ∈ ( 1 , γ ] .
The rotational energy K L ( ρ ) \mathcal{K}_{L}(\rho) K L ( ρ ) depends nonlinearly and nonlocally on the density through the cylindrical mass function m ρ m_{\rho} m ρ . Its first variation at the equilibrium density ρ ˉ \bar{\rho} ρ ˉ is represented by Ψ ρ ˉ \Psi_{\bar{\rho}} Ψ ρ ˉ . Under the hypotheses of Lemma 4.3, the condition L ′ ≥ 0 L'\geq0 L ′ ≥ 0 yields the exact
nonnegative remainder
d 2 ( ρ , ρ ˉ ) = ∫ 0 1 ( 1 − λ ) ∫ 0 ∞ L ′ ( m ρ λ ( r ) ) m f ( r ) 2 r 3 d r d λ ≥ 0 , ρ λ = ρ ˉ + λ f . d_2(\rho,\bar{\rho})=\int_0^1(1-\lambda)\int_0^\infty\frac{L'(m_{\rho_\lambda}(r))m_f(r)^2}{r^3}\,\mathrm{d}r\,\mathrm{d}\lambda\ge0,\qquad\rho_\lambda=\bar{\rho}+\lambda f. d 2 ( ρ , ρ ˉ ) = ∫ 0 1 ( 1 − λ ) ∫ 0 ∞ r 3 L ′ ( m ρ λ ( r )) m f ( r ) 2 d r d λ ≥ 0 , ρ λ = ρ ˉ + λ f .
The angular-momentum compatibility condition further yields (3.13), which relates the azimuthal kinetic energy to the rotational energy functional K L ( ρ ) \mathcal{K}_L(\rho) K L ( ρ ) .
A key observation is that combining this identity with the decompositions involving d 1 d_1 d 1 and d 2 d_2 d 2 leads to an exact cancellation of the two linear terms containing Ψ ρ ˉ \Psi_{\bar{\rho}} Ψ ρ ˉ . Consequently, Proposition 4.5 yields the effective energy identity (4.18), in which the linear term involving ρ − ρ ˉ \rho-\bar{\rho} ρ − ρ ˉ carries the weight
G ˉ = μ + W ˉ α . \bar{G}=\mu+\bar{W}_\alpha. G ˉ = μ + W ˉ α .
This cancellation is essential for obtaining the relative energy estimates in the presence of rotation and nonlocal gravitational interactions.
(iii) After the rotational cancellation, the source term in Lemma 6.1 contains
F = ∇ Φ α ∗ ( ρ − ρ ˉ ) . F=\nabla\Phi_\alpha*(\rho-\bar{\rho}). F = ∇ Φ α ∗ ( ρ − ρ ˉ ) .
The explicit kinetic part of E L \mathcal{E}_L E L controls only M r M^r M r and M z M^z M z , since the azimuthal kinetic energy has been absorbed into K L ( ρ ) \mathcal{K}_L(\rho) K L ( ρ ) . The L s L^s L s control obtained above permits the use of Lemma 5.1; axisymmetry then gives
F ⋅ e θ = 0 , M ⋅ F = ( M r e r + M z e z ) ⋅ F . F\cdot e_\theta=0,\qquad\mathbf{M}\cdot F=(M^r e_r+M^z e_z)\cdot F. F ⋅ e θ = 0 , M ⋅ F = ( M r e r + M z e z ) ⋅ F .
Thus, the nonlocal Riesz interaction involves only the momentum components controlled by the relative energy, thereby overcoming the apparent lack of control over the azimuthal component.
The remaining nonlocal contribution is controlled by Proposition 5.2, which provides the requisite weighted Riesz estimates in two regimes. Under condition (5.4), we have
∫ R 3 ρ ∣ F ∣ 2 d x ≤ C ( d 1 + d 1 χ ) , χ = 7 − 2 α 3 . \int_{\mathbb{R}^3}\rho\lvert F\rvert^2\,\mathrm{d}x\le C(d_1+d_1^\chi),\qquad\chi=\frac{7-2\alpha}{3}. ∫ R 3 ρ ∣ F ∣ 2 d x ≤ C ( d 1 + d 1 χ ) , χ = 3 7 − 2 α .
Alternatively, if (5.5) holds with p ˉ > 3 2 − α \bar{p}>\frac{3}{2-\alpha} p ˉ > 2 − α 3 , then
∫ R 3 ρ ∣ F ∣ 2 d x ≤ C ( 1 + ∥ ρ ∥ L p ˉ 2 ) d 1 . \int_{\mathbb{R}^3}\rho\lvert F\rvert^2\,\mathrm{d}x\le C(1+\lVert\rho\rVert_{L^{\bar{p}}}^2)d_1. ∫ R 3 ρ ∣ F ∣ 2 d x ≤ C ( 1 + ∥ ρ ∥ L p ˉ 2 ) d 1 .
In the latter regime, estimate (6.6) ensures that the coefficient on the right-hand side remains uniformly bounded on [ 0 , T ] [0,T] [ 0 , T ] . Together, these estimates provide the necessary control of the nonlocal interaction term and allow the relative energy argument to be closed.
(iv) The relative total energy satisfies
H ( ρ , M ∣ ρ ˉ , M ˉ ) = E L + Q α , Q α = 1 2 ∫ R 3 ( ρ − ρ ˉ ) Φ α ∗ ( ρ − ρ ˉ ) d x ≤ 0. \mathcal{H}(\rho,\mathbf{M}\mid\bar{\rho},\bar{\mathbf{M}})=\mathcal{E}_L+Q_\alpha,\qquad Q_\alpha=\frac{1}{2}\int_{\mathbb{R}^3}(\rho-\bar{\rho})\Phi_\alpha*(\rho-\bar{\rho})\,\mathrm{d}x\le0. H ( ρ , M ∣ ρ ˉ , M ˉ ) = E L + Q α , Q α = 2 1 ∫ R 3 ( ρ − ρ ˉ ) Φ α ∗ ( ρ − ρ ˉ ) d x ≤ 0.
In Case 1, Proposition 2.3 yields (2.9), which, combined with the total energy inequality, gives (6.17). Consequently,
E L χ ≤ B 0 χ − 1 E L , \mathcal{E}_L^\chi\le B_0^{\chi-1}\mathcal{E}_L, E L χ ≤ B 0 χ − 1 E L ,
allowing the nonlinear term in (6.19) to be controlled linearly by E L \mathcal{E}_L E L .
In Case 2, (6.19) already provides a linear bound in E L \mathcal{E}_L E L .
Finally, the flux condition (6.5) justifies passing to the limit as the spatial cutoff is removed in the entropy inequality. Combining the preceding estimates yields (6.22), from which the desired stability estimate (6.8) follows by Grönwall’s lemma.
The Chandrasekhar pressure law (8.1) satisfies Assumption 2.1. Indeed, the asymptotic behaviors (86) –(87) yield
γ 1 = 5 3 , γ 2 = 4 3 , γ = 4 3 . \gamma_1 = \frac{5}{3}, \qquad\gamma_2 = \frac{4}{3}, \qquad\gamma= \frac{4}{3}. γ 1 = 3 5 , γ 2 = 3 4 , γ = 3 4 .
For the Newtonian gravitational potential (α = 1 \alpha= 1 α = 1 ), the high-density exponent γ 2 = 4 3 \gamma_2 = \frac{4}{3} γ 2 = 3 4 coincides with the mass-critical exponent. Consequently, Case 1 of Proposition 5.2 is not applicable, since
4 3 < 9 5 . \frac{4}{3} < \frac{9}{5}. 3 4 < 5 9 .
However, Case 2 applies because
4 3 > 6 5 , \frac{4}{3} > \frac{6}{5}, 3 4 > 5 6 ,
provided that (8.5) holds for some p ‾ > 3 \overline{p} > 3 p > 3 . Thus, under the remaining hypotheses of Theorem 6.2, Corollary 8.1 yields the conditional stability estimate (8.6) around any prescribed rotating white dwarf equilibrium satisfying Definition 3.1.
The rest of the paper is organized as follows: In Section 2, we develop the necessary estimates for general pressure laws, state the fixed-mass coercivity condition, and define finite-energy entropy solutions. Section 3 presents rotating Euler-Riesz stars and angular momentum compatibility. In Section 4, we prove the coercivity of d 1 d_1 d 1 , the nonnegativity of d 2 d_2 d 2 , and the centrifugal cancellation. In Section 5, we prove that the gradient of the perturbed Riesz potential has no azimuthal component and establish the weighted Riesz estimates in the two admissible regimes. In Section 6, we derive the distributional relative entropy inequality and prove Theorem 6.2. Section 7 presents the relative energy decomposition. Finally, in Section 8, we verify the assumptions for the Chandrasekhar pressure law and prove Corollary 8.1.
Preliminaries
In this section, we develop the necessary estimates for general pressure laws, state the fixed-mass coercivity condition, and define finite-energy entropy solutions.
Internal Energy and Mechanical Entropy
Let p = p ( ρ ) : [ 0 , ∞ ) → [ 0 , ∞ ) p = p(\rho) : [0,\infty) \to[0,\infty) p = p ( ρ ) : [ 0 , ∞ ) → [ 0 , ∞ ) be a barotropic pressure law. Assuming that p ( s ) s 2 \frac{p(s)}{s^2} s 2 p ( s ) is integrable near s = 0 s = 0 s = 0 , define the specific internal energy e e e and the internal energy density h h h by
e ( ρ ) : = ∫ 0 ρ p ( s ) s 2 d s , h ( ρ ) : = ρ e ( ρ ) , h ( 0 ) : = 0. (2) e(\rho) := \int_0^\rho\frac{p(s)}{s^2}\,\mathrm{d}s, \qquad h(\rho) := \rho e(\rho), \qquad h(0) := 0. \tag*{(2)} e ( ρ ) := ∫ 0 ρ s 2 p ( s ) d s , h ( ρ ) := ρ e ( ρ ) , h ( 0 ) := 0. (2)
The required integrability follows from Assumption 2.1 below.
For ρ > 0 \rho> 0 ρ > 0 , direct differentiation yields
ρ h ′ ( ρ ) − h ( ρ ) = p ( ρ ) , h ′ ′ ( ρ ) = p ′ ( ρ ) ρ . \rho h'(\rho) - h(\rho) = p(\rho), \qquad h''(\rho) = \frac{p'(\rho)}{\rho}. ρ h ′ ( ρ ) − h ( ρ ) = p ( ρ ) , h ′′ ( ρ ) = ρ p ′ ( ρ ) .
In particular, the condition p ′ ( ρ ) > 0 p'(\rho) > 0 p ′ ( ρ ) > 0 implies that h h h is strictly convex on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) .
The corresponding mechanical entropy and entropy flux are given by
η ( ρ , M ) = ∣ M ∣ 2 2 ρ + h ( ρ ) , q ( ρ , M ) = M ( ∣ M ∣ 2 2 ρ 2 + h ′ ( ρ ) ) . \eta(\rho,\mathbf{M}) = \frac{|\mathbf{M}|^2}{2\rho} + h(\rho), \qquad q(\rho,\mathbf{M}) = \mathbf{M}\left(\frac{|\mathbf{M}|^2}{2\rho^2} + h'(\rho)\right). η ( ρ , M ) = 2 ρ ∣ M ∣ 2 + h ( ρ ) , q ( ρ , M ) = M ( 2 ρ 2 ∣ M ∣ 2 + h ′ ( ρ ) ) .
For smooth solutions of (1.1) with positive density, the mechanical entropy balance takes the form
∂ t η ( ρ , M ) + div q ( ρ , M ) = − M ⋅ ∇ W α . \partial_t \eta(\rho,\mathbf{M}) + \operatorname{div} q(\rho,\mathbf{M}) = -\mathbf{M} \cdot\nabla W_\alpha. ∂ t η ( ρ , M ) + div q ( ρ , M ) = − M ⋅ ∇ W α .
We impose the following structural assumptions on the pressure law, allowing different power-law behaviours at low and high densities.
Assumption 2.1 (General pressure law). The pressure function p p p satisfies the following conditions:
(i) p ∈ C ( [ 0 , ∞ ) ) ∩ C 1 ( ( 0 , ∞ ) ) p \in C([0,\infty)) \cap C^1((0,\infty)) p ∈ C ([ 0 , ∞ )) ∩ C 1 (( 0 , ∞ )) , p ( 0 ) = 0 p(0) = 0 p ( 0 ) = 0 , and p ′ ( ρ ) > 0 p'(\rho) > 0 p ′ ( ρ ) > 0 for ρ > 0 \rho> 0 ρ > 0 .
(ii) There exist exponents γ 1 ∈ ( 1 , 2 ] \gamma_1 \in(1,2] γ 1 ∈ ( 1 , 2 ] and γ 2 ∈ ( 1 , ∞ ) \gamma_2 \in(1,\infty) γ 2 ∈ ( 1 , ∞ ) , constants 0 < ρ ∗ < 1 < ρ ∗ 0 < \rho_* < 1 < \rho^* 0 < ρ ∗ < 1 < ρ ∗ , and positive constants c 0 c_0 c 0 , C 0 C_0 C 0 , c ~ 0 \widetilde{c}_0 c 0 , c ∞ c_\infty c ∞ , C ∞ C_\infty C ∞ such that
c 0 ρ γ 1 ≤ p ( ρ ) ≤ C 0 ρ γ 1 , p ′ ( ρ ) ≥ c ~ 0 ρ γ 1 − 1 when 0 < ρ ≤ ρ ∗ , (3) c_0\rho^{\gamma_1} \le p(\rho) \le C_0\rho^{\gamma_1}, \qquad p'(\rho) \ge\widetilde{c}_0\rho^{\gamma_1-1} \qquad\text{when } 0 < \rho\le\rho_*, \tag*{(3)} c 0 ρ γ 1 ≤ p ( ρ ) ≤ C 0 ρ γ 1 , p ′ ( ρ ) ≥ c 0 ρ γ 1 − 1 when 0 < ρ ≤ ρ ∗ , (3)
c ∞ ρ γ 2 ≤ p ( ρ ) ≤ C ∞ ρ γ 2 when ρ ≥ ρ ∗ . (4) c_\infty\rho^{\gamma_2} \le p(\rho) \le C_\infty\rho^{\gamma_2} \qquad\text{when } \rho\ge\rho^*. \tag*{(4)} c ∞ ρ γ 2 ≤ p ( ρ ) ≤ C ∞ ρ γ 2 when ρ ≥ ρ ∗ . (4)
Define
γ : = min { γ 1 , γ 2 } . \gamma:= \min\{\gamma_1,\gamma_2\}. γ := min { γ 1 , γ 2 } .
Lemma 2.2 (Properties of the internel energy density). Under Assumption 2.1, the internal energy density h h h defined in (2.1) is strictly convex on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . Moreover, there exist positive constants c 0 ′ c'_0 c 0 ′ , C 0 ′ C'_0 C 0 ′ , c ∞ ′ c'_\infty c ∞ ′ , C ∞ ′ C'_\infty C ∞ ′ such that
c 0 ′ ρ γ 1 ≤ h ( ρ ) ≤ C 0 ′ ρ γ 1 , 0 ≤ h ′ ( ρ ) ≤ C 0 ′ ρ γ 1 − 1 when 0 ≤ ρ ≤ 1 , (5) c'_0\rho^{\gamma_1} \le h(\rho) \le C'_0\rho^{\gamma_1}, \qquad0 \le h'(\rho) \le C'_0\rho^{\gamma_1-1} \qquad\text{when } 0 \le\rho\le1, \tag*{(5)} c 0 ′ ρ γ 1 ≤ h ( ρ ) ≤ C 0 ′ ρ γ 1 , 0 ≤ h ′ ( ρ ) ≤ C 0 ′ ρ γ 1 − 1 when 0 ≤ ρ ≤ 1 , (5) c ∞ ′ ρ γ 2 ≤ h ( ρ ) ≤ C ∞ ′ ρ γ 2 , 0 ≤ h ′ ( ρ ) ≤ C ∞ ′ ρ γ 2 − 1 when ρ ≥ 1. (6) c'_\infty\rho^{\gamma_2} \le h(\rho) \le C'_\infty\rho^{\gamma_2}, \qquad0 \le h'(\rho) \le C'_\infty\rho^{\gamma_2-1} \qquad\text{when } \rho\ge1. \tag*{(6)} c ∞ ′ ρ γ 2 ≤ h ( ρ ) ≤ C ∞ ′ ρ γ 2 , 0 ≤ h ′ ( ρ ) ≤ C ∞ ′ ρ γ 2 − 1 when ρ ≥ 1. (6) In addition, for any R > 0 R > 0 R > 0 ,
inf 0 < s ≤ R h ′ ′ ( s ) = inf 0 < s ≤ R p ′ ( s ) s > 0. (7) \inf_{0 < s \le R} h''(s) = \inf_{0 < s \le R} \frac{p'(s)}{s} > 0. \tag*{(7)} 0 < s ≤ R inf h ′′ ( s ) = 0 < s ≤ R inf s p ′ ( s ) > 0. (7)
Proof. Since p ( 0 ) = 0 p(0) = 0 p ( 0 ) = 0 and p ′ ( ρ ) > 0 p'(\rho) > 0 p ′ ( ρ ) > 0 for ρ > 0 \rho> 0 ρ > 0 , we see that p ( ρ ) > 0 p(\rho) > 0 p ( ρ ) > 0 for ρ > 0 \rho> 0 ρ > 0 . Consequently, e e e , h h h , and h ′ h' h ′ are strictly positive on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) .
We first establish the estimates at low densities. For 0 < ρ ≤ ρ ∗ 0 < \rho\le\rho_* 0 < ρ ≤ ρ ∗ , (2.2) yields
c 0 γ 1 − 1 ρ γ 1 − 1 ≤ e ( ρ ) = ∫ 0 ρ p ( s ) s 2 d s ≤ C 0 γ 1 − 1 ρ γ 1 − 1 . \frac{c_0}{\gamma_1-1}\rho^{\gamma_1-1} \le e(\rho) = \int_0^\rho\frac{p(s)}{s^2}\,\mathrm{d}s \le\frac{C_0}{\gamma_1-1}\rho^{\gamma_1-1}. γ 1 − 1 c 0 ρ γ 1 − 1 ≤ e ( ρ ) = ∫ 0 ρ s 2 p ( s ) d s ≤ γ 1 − 1 C 0 ρ γ 1 − 1 .
Using h ( ρ ) = ρ e ( ρ ) h(\rho) = \rho e(\rho) h ( ρ ) = ρ e ( ρ ) and h ′ ( ρ ) = e ( ρ ) + p ( ρ ) ρ h'(\rho) = e(\rho) + \frac{p(\rho)}{\rho} h ′ ( ρ ) = e ( ρ ) + ρ p ( ρ ) , we obtain
c ρ γ 1 ≤ h ( ρ ) ≤ C ρ γ 1 , 0 < h ′ ( ρ ) ≤ C ρ γ 1 − 1 for 0 < ρ ≤ ρ ∗ , c\rho^{\gamma_1} \le h(\rho) \le C\rho^{\gamma_1}, \qquad0 < h'(\rho) \le C\rho^{\gamma_1-1} \qquad\text{for } 0 < \rho\le\rho_*, c ρ γ 1 ≤ h ( ρ ) ≤ C ρ γ 1 , 0 < h ′ ( ρ ) ≤ C ρ γ 1 − 1 for 0 < ρ ≤ ρ ∗ ,
with positive constants c c c and C C C .
Since γ 1 > 1 \gamma_1 > 1 γ 1 > 1 , it follows that h ′ ( ρ ) → 0 h'(\rho) \to0 h ′ ( ρ ) → 0 as ρ ↓ 0 \rho\downarrow0 ρ ↓ 0 . We therefore extend h ′ h' h ′ continuously to the origin by setting h ′ ( 0 ) = 0 h'(0) = 0 h ′ ( 0 ) = 0 . On the compact interval [ ρ ∗ , 1 ] [\rho_*,1] [ ρ ∗ , 1 ] , the functions h ( ρ ) ρ γ 1 \frac{h(\rho)}{\rho^{\gamma_1}} ρ γ 1 h ( ρ ) and h ′ ( ρ ) ρ γ 1 − 1 \frac{h'(\rho)}{\rho^{\gamma_1-1}} ρ γ 1 − 1 h ′ ( ρ ) are continuous and strictly positive. Combining their bounds on this interval with the preceding estimates proves (2.4).
We next consider the high-density regime. For ρ ≥ ρ ∗ \rho\ge\rho^* ρ ≥ ρ ∗ , write
e ( ρ ) = e ( ρ ∗ ) + ∫ ρ ∗ ρ p ( s ) s 2 d s . e(\rho) = e(\rho^*) + \int_{\rho^*}^{\rho} \frac{p(s)}{s^2}\,\mathrm{d}s. e ( ρ ) = e ( ρ ∗ ) + ∫ ρ ∗ ρ s 2 p ( s ) d s .
By (2.3),
e ( ρ ∗ ) + c ∞ γ 2 − 1 ( ρ γ 2 − 1 − ( ρ ∗ ) γ 2 − 1 ) ≤ e ( ρ ) ≤ e ( ρ ∗ ) + C ∞ γ 2 − 1 ( ρ γ 2 − 1 − ( ρ ∗ ) γ 2 − 1 ) . e(\rho^*) + \frac{c_\infty}{\gamma_2-1}\left(\rho^{\gamma_2-1}-(\rho^*)^{\gamma_2-1}\right) \le e(\rho) \le e(\rho^*) + \frac{C_\infty}{\gamma_2-1}\left(\rho^{\gamma_2-1}-(\rho^*)^{\gamma_2-1}\right). e ( ρ ∗ ) + γ 2 − 1 c ∞ ( ρ γ 2 − 1 − ( ρ ∗ ) γ 2 − 1 ) ≤ e ( ρ ) ≤ e ( ρ ∗ ) + γ 2 − 1 C ∞ ( ρ γ 2 − 1 − ( ρ ∗ ) γ 2 − 1 ) .
Since γ 2 > 1 \gamma_2 > 1 γ 2 > 1 , there exist R 0 ≥ ρ ∗ R_0 \ge\rho^* R 0 ≥ ρ ∗ and positive constants c , C c,C c , C such that
c ρ γ 2 − 1 ≤ e ( ρ ) ≤ C ρ γ 2 − 1 when ρ ≥ R 0 . c\rho^{\gamma_2-1} \le e(\rho) \le C\rho^{\gamma_2-1} \qquad\text{when } \rho\ge R_0. c ρ γ 2 − 1 ≤ e ( ρ ) ≤ C ρ γ 2 − 1 when ρ ≥ R 0 .
Together with h ( ρ ) = ρ e ( ρ ) h(\rho) = \rho e(\rho) h ( ρ ) = ρ e ( ρ ) and h ′ ( ρ ) = e ( ρ ) + p ( ρ ) ρ h'(\rho) = e(\rho) + \frac{p(\rho)}{\rho} h ′ ( ρ ) = e ( ρ ) + ρ p ( ρ ) , this gives
c ρ γ 2 ≤ h ( ρ ) ≤ C ρ γ 2 , 0 < h ′ ( ρ ) ≤ C ρ γ 2 − 1 for ρ ≥ R 0 . c\rho^{\gamma_2} \le h(\rho) \le C\rho^{\gamma_2}, \qquad0 < h'(\rho) \le C\rho^{\gamma_2-1} \qquad\text{for } \rho\ge R_0. c ρ γ 2 ≤ h ( ρ ) ≤ C ρ γ 2 , 0 < h ′ ( ρ ) ≤ C ρ γ 2 − 1 for ρ ≥ R 0 .
On the compact interval [ 1 , R 0 ] [1,R_0] [ 1 , R 0 ] , the functions h ( ρ ) ρ γ 2 \frac{h(\rho)}{\rho^{\gamma_2}} ρ γ 2 h ( ρ ) and h ′ ( ρ ) ρ γ 2 − 1 \frac{h'(\rho)}{\rho^{\gamma_2-1}} ρ γ 2 − 1 h ′ ( ρ ) are continuous and strictly positive. Adjusting the constants therefore establishes (2.5) for all ρ ≥ 1 \rho\ge1 ρ ≥ 1 .
Finally, the identity
h ′ ′ ( ρ ) = p ′ ( ρ ) ρ > 0 when ρ > 0 , h''(\rho) = \frac{p'(\rho)}{\rho} > 0 \qquad\text{when } \rho> 0, h ′′ ( ρ ) = ρ p ′ ( ρ ) > 0 when ρ > 0 ,
shows that h h h is strictly convex.
To establish (2.6), observe that, for 0 < s ≤ ρ ∗ 0 < s \le\rho_* 0 < s ≤ ρ ∗ ,
h ′ ′ ( s ) = p ′ ( s ) s ≥ c ˉ 0 s γ 1 − 2 ≥ c ˉ 0 ρ ∗ γ 1 − 2 > 0 , h''(s) = \frac{p'(s)}{s} \ge\bar{c}_0s^{\gamma_1-2} \ge\bar{c}_0\rho_*^{\gamma_1-2} > 0, h ′′ ( s ) = s p ′ ( s ) ≥ c ˉ 0 s γ 1 − 2 ≥ c ˉ 0 ρ ∗ γ 1 − 2 > 0 ,
where the last inequality uses γ 1 ≤ 2 \gamma_1 \le2 γ 1 ≤ 2 .
If R > ρ ∗ R > \rho_* R > ρ ∗ , the continuity and strict positivity of p ′ ( s ) s \frac{p'(s)}{s} s p ′ ( s ) on [ ρ ∗ , R ] [\rho_*, R] [ ρ ∗ , R ] imply that
min ρ ∗ ≤ s ≤ R p ′ ( s ) s > 0. \min_{\rho_* \le s \le R} \frac{p'(s)}{s} > 0. ρ ∗ ≤ s ≤ R min s p ′ ( s ) > 0.
Combining this bound with the low-density estimate gives
inf 0 ≤ s ≤ R h ′ ′ ( s ) > 0. \inf_{0 \le s \le R} h''(s) > 0. 0 ≤ s ≤ R inf h ′′ ( s ) > 0.
The case R ≤ ρ ∗ R \le\rho_* R ≤ ρ ∗ follows directly from the low-density bound. This completes the proof. □
Coercivity Condition
For a nonnegative density ρ \rho ρ with prescribed total mass M M M , define the attractive interaction energy by
P α ( ρ ) : = − 1 2 ∫ R 3 ρ ( Φ α ∗ ρ ) d x = 1 2 α ∬ R 3 × R 3 ρ ( x ) ρ ( y ) ∣ x − y ∣ α d x d y . \mathcal{P}_{\alpha}(\rho) := -\frac{1}{2}\int_{\mathbb{R}^3}\rho(\Phi_{\alpha} * \rho)\,\mathrm{d}x = \frac{1}{2\alpha}\iint_{\mathbb{R}^3\times\mathbb{R}^3}\frac{\rho(x)\rho(y)}{|x-y|^\alpha}\,\mathrm{d}x\,\mathrm{d}y. P α ( ρ ) := − 2 1 ∫ R 3 ρ ( Φ α ∗ ρ ) d x = 2 α 1 ∬ R 3 × R 3 ∣ x − y ∣ α ρ ( x ) ρ ( y ) d x d y .
The corresponding total energy is
H ( ρ , M ) = ∫ R 3 ( ∣ M ∣ 2 2 ρ + h ( ρ ) ) d x − P α ( ρ ) . (8) \mathcal{H}(\rho,\mathbf{M}) = \int_{\mathbb{R}^3}\left(\frac{|\mathbf{M}|^2}{2\rho}+h(\rho)\right)\,\mathrm{d}x-\mathcal{P}_{\alpha}(\rho). \tag*{(8)} H ( ρ , M ) = ∫ R 3 ( 2 ρ ∣ M ∣ 2 + h ( ρ ) ) d x − P α ( ρ ) . (8)
We assume that the solutions under consideration satisfy the total energy inequality
H ( ρ ( t ) , M ( t ) ) ≤ H ( ρ 0 , M 0 ) for a.e. t ∈ ( 0 , T ) . (9) \mathcal{H}(\rho(t),\mathbf{M}(t)) \le\mathcal{H}(\rho_0,\mathbf{M}_0)\quad\text{for a.e. }t\in(0,T). \tag*{(9)} H ( ρ ( t ) , M ( t )) ≤ H ( ρ 0 , M 0 ) for a.e. t ∈ ( 0 , T ) . (9)
To obtain effective control of the internal and kinetic energies, we impose the following coercivity condition at the prescribed mass M M M : there exist constants ϑ M ∈ ( 0 , 1 ) \vartheta_M \in(0,1) ϑ M ∈ ( 0 , 1 ) and C M < ∞ C_M < \infty C M < ∞ such that
P α ( ϱ ) ≤ ϑ M ∫ R 3 h ( ϱ ) d x + C M (10) \mathcal{P}_{\alpha}(\varrho) \le\vartheta_M\int_{\mathbb{R}^3}h(\varrho)\,\mathrm{d}x+C_M \tag*{(10)} P α ( ϱ ) ≤ ϑ M ∫ R 3 h ( ϱ ) d x + C M (10)
for every nonnegative density ϱ \varrho ϱ satisfying
∫ R 3 ϱ d x = M , h ( ϱ ) ∈ L 1 ( R 3 ) . \int_{\mathbb{R}^3}\varrho\,\mathrm{d}x=M,\qquad h(\varrho)\in L^1(\mathbb{R}^3). ∫ R 3 ϱ d x = M , h ( ϱ ) ∈ L 1 ( R 3 ) .
The next proposition provides sufficient conditions for (2.9), distinguishing the mass-subcritical and mass-critical regimes.
Let C α ∗ C_\alpha^* C α ∗ denote the optimal constant in the interpolation form of the Hardy–Littlewood–Sobolev inequality
∬ R 3 × R 3 ϱ ( x ) ϱ ( y ) ∣ x − y ∣ α d x d y ≤ C α ∗ ∥ ϱ ∥ L 1 3 − α 3 ∥ ϱ ∥ L 3 + α 3 3 + α 3 . (11) \iint_{\mathbb{R}^3\times\mathbb{R}^3}\frac{\varrho(x)\varrho(y)}{|x-y|^\alpha}\,\mathrm{d}x\,\mathrm{d}y \le C_\alpha^*\|\varrho\|_{L^1}^{\frac{3-\alpha}{3}}\|\varrho\|_{L^{\frac{3+\alpha}{3}}}^{\frac{3+\alpha}{3}}. \tag*{(11)} ∬ R 3 × R 3 ∣ x − y ∣ α ϱ ( x ) ϱ ( y ) d x d y ≤ C α ∗ ∥ ϱ ∥ L 1 3 3 − α ∥ ϱ ∥ L 3 3 + α 3 3 + α . (11)
Proposition 2.3 (Sufficient conditions for coercivity). Let
q c : = 3 + α 3 . q_c := \frac{3+\alpha}{3}. q c := 3 3 + α . Under Assumption 2.1, the following statements hold:
(i) If γ 2 > q c \gamma_2 > q_c γ 2 > q c , then (2.9) holds for every prescribed mass $M > 0.
(ii) Suppose that γ 2 = q c \gamma_2 = q_c γ 2 = q c and
K ∞ : = lim inf ρ → ∞ h ( ρ ) ρ q c ∈ ( 0 , ∞ ) . K_\infty:= \liminf_{\rho\to\infty}\frac{h(\rho)}{\rho^{q_c}}\in(0,\infty). K ∞ := ρ → ∞ lim inf ρ q c h ( ρ ) ∈ ( 0 , ∞ ) . Then (2.9) holds whenever
C α ∗ 2 α M 3 − α 3 < K ∞ . \frac{C_\alpha^*}{2\alpha}M^{\frac{3-\alpha}{3}} < K_\infty. 2 α C α ∗ M 3 3 − α < K ∞ .
Proof. Set
A M : = C α ∗ 2 α M 3 − α 3 . A_M := \frac{C_\alpha^*}{2\alpha}M^{\frac{3-\alpha}{3}}. A M := 2 α C α ∗ M 3 3 − α .
By the pressure estimates (5) –(6) , there exists C > 0 C > 0 C > 0 such that
ρ γ 2 ≤ C ( h ( ρ ) + ρ ) for ρ ≥ 0. \rho^{\gamma_2} \le C\left(h(\rho)+\rho\right) \qquad\text{for } \rho\ge0. ρ γ 2 ≤ C ( h ( ρ ) + ρ ) for ρ ≥ 0.
Hence every admissible density ϱ \varrho ϱ belongs to L 1 ( R 3 ) ∩ L γ 2 ( R 3 ) L^1(\mathbb{R}^3) \cap L^{\gamma_2}(\mathbb{R}^3) L 1 ( R 3 ) ∩ L γ 2 ( R 3 ) .
Since γ 2 ≥ q c \gamma_2 \ge q_c γ 2 ≥ q c in both cases, interpolation gives ϱ ∈ L q c ( R 3 ) \varrho\in L^{q_c}(\mathbb{R}^3) ϱ ∈ L q c ( R 3 ) . The Hardy–Littlewood–Sobolev estimate (11) therefore yields
P α ( ϱ ) ≤ A M ∥ ϱ ∥ L q c q c . (12) \mathcal{P}_\alpha(\varrho) \le A_M\lVert\varrho\rVert_{L^{q_c}}^{q_c}. \tag*{(12)} P α ( ϱ ) ≤ A M ∥ ϱ ∥ L q c q c . (12)
Case 1. γ 2 > q c \gamma_2 > q_c γ 2 > q c : Since 1 < q c < γ 2 1 < q_c < \gamma_2 1 < q c < γ 2 , for every δ > 0 \delta> 0 δ > 0 there exists C δ > 0 C_\delta> 0 C δ > 0 such that
s q c ≤ δ s γ 2 + C δ s for s ≥ 0. s^{q_c} \le\delta s^{\gamma_2} + C_\delta s \qquad\text{for } s \ge0. s q c ≤ δ s γ 2 + C δ s for s ≥ 0.
Integrating this inequality and using the prescribed mass gives
∥ ϱ ∥ L q c q c ≤ δ ∥ ϱ ∥ L γ 2 γ 2 + C δ M . \lVert\varrho\rVert_{L^{q_c}}^{q_c} \le\delta\lVert\varrho\rVert_{L^{\gamma_2}}^{\gamma_2} + C_\delta M. ∥ ϱ ∥ L q c q c ≤ δ ∥ ϱ ∥ L γ 2 γ 2 + C δ M .
The pressure estimates then imply
∥ ϱ ∥ L q c q c ≤ C δ ∫ R 3 h ( ϱ ) d x + C δ , M . \lVert\varrho\rVert_{L^{q_c}}^{q_c} \le C\delta\int_{\mathbb{R}^3} h(\varrho)\,\mathrm{d}x + C_{\delta,M}. ∥ ϱ ∥ L q c q c ≤ C δ ∫ R 3 h ( ϱ ) d x + C δ , M .
Combining this with (12) , we obtain
P α ( ϱ ) ≤ C A M δ ∫ R 3 h ( ϱ ) d x + C δ , M . \mathcal{P}_\alpha(\varrho) \le CA_M\delta\int_{\mathbb{R}^3} h(\varrho)\,\mathrm{d}x + C_{\delta,M}. P α ( ϱ ) ≤ C A M δ ∫ R 3 h ( ϱ ) d x + C δ , M .
Choosing δ > 0 \delta> 0 δ > 0 sufficiently small that
ϑ M : = C A M δ < 1 \vartheta_M := CA_M\delta< 1 ϑ M := C A M δ < 1
establishes (10) for every M > 0 M > 0 M > 0 .
Case 2. γ 2 = q c \gamma_2 = q_c γ 2 = q c : Assume that A M < K ∞ A_M < K_\infty A M < K ∞ . Choose
0 < ε < K ∞ − A M . 0 < \varepsilon< K_\infty- A_M. 0 < ε < K ∞ − A M .
By the definition of K ∞ K_\infty K ∞ , there exists R ε > 0 R_\varepsilon> 0 R ε > 0 such that
h ( ρ ) ≥ ( K ∞ − ε ) ρ q c for ρ ≥ R ε . h(\rho) \ge(K_\infty-\varepsilon)\rho^{q_c} \qquad\text{for } \rho\ge R_\varepsilon. h ( ρ ) ≥ ( K ∞ − ε ) ρ q c for ρ ≥ R ε .
Since h ≥ 0 h \ge0 h ≥ 0 and
ρ q c ≤ R ε q c − 1 ρ for 0 ≤ ρ ≤ R ε , \rho^{q_c} \le R_\varepsilon^{q_c-1}\rho\qquad\text{for } 0 \le\rho\le R_\varepsilon, ρ q c ≤ R ε q c − 1 ρ for 0 ≤ ρ ≤ R ε ,
we obtain the estimate
h ( ρ ) ≥ ( K ∞ − ε ) ρ q c − C ε ρ for ρ ≥ 0. h(\rho) \ge(K_\infty-\varepsilon)\rho^{q_c} - C_\varepsilon\rho\qquad\text{for } \rho\ge0. h ( ρ ) ≥ ( K ∞ − ε ) ρ q c − C ε ρ for ρ ≥ 0.
Integrating and using the mass constraint, we conclude that
∥ ϱ ∥ L q c q c ≤ 1 K ∞ − ε ( ∫ R 3 h ( ϱ ) d x + C ε M ) . \lVert\varrho\rVert_{L^{q_c}}^{q_c} \le\frac{1}{K_\infty-\varepsilon}\left(\int_{\mathbb{R}^3}h(\varrho)\,\mathrm{d}x+C_\varepsilon M\right). ∥ ϱ ∥ L q c q c ≤ K ∞ − ε 1 ( ∫ R 3 h ( ϱ ) d x + C ε M ) .
Substituting this into (12) gives
P α ( ϱ ) ≤ A M K ∞ − ε ∫ R 3 h ( ϱ ) d x + C M . \mathcal{P}_\alpha(\varrho) \le\frac{A_M}{K_\infty-\varepsilon}\int_{\mathbb{R}^3}h(\varrho)\,\mathrm{d}x+C_M. P α ( ϱ ) ≤ K ∞ − ε A M ∫ R 3 h ( ϱ ) d x + C M .
Since
ϑ M : = A M K ∞ − ε < 1 , \vartheta_M := \frac{A_M}{K_\infty-\varepsilon} < 1, ϑ M := K ∞ − ε A M < 1 ,
the coercivity estimate (10) follows also in the mass-critical regime. □
Finite-Energy Entropy Solutions
We now introduce the notion of finite-energy entropy solutions to (1.1).
Definition 2.4 (Finite-energy entropy solutions). A measurable pair ( ρ , M ) (\rho,\mathbf{M}) ( ρ , M ) on [ 0 , T ] × R 3 [0,T]\times\mathbb{R}^{3} [ 0 , T ] × R 3 is called a finite-energy entropy solution of (1.1) with initial data ( ρ 0 , M 0 ) (\rho_{0},\mathbf{M}_{0}) ( ρ 0 , M 0 ) if
ρ ≥ 0 , ρ ∈ L ∞ ( 0 , T ; L 1 ( R 3 ) ) , (13) \rho\ge0,\qquad\rho\in L^{\infty}(0,T;L^{1}(\mathbb{R}^{3})), \tag*{(13)} ρ ≥ 0 , ρ ∈ L ∞ ( 0 , T ; L 1 ( R 3 )) , (13) h ( ρ ) ∈ L ∞ ( 0 , T ; L 1 ( R 3 ) ) , M ρ ∈ L ∞ ( 0 , T ; L 2 ( R 3 ) ) . (14) h(\rho) \in L^{\infty}(0,T;L^{1}(\mathbb{R}^{3})),\qquad\frac{\mathbf{M}}{\sqrt{\rho}} \in L^{\infty}(0,T;L^{2}(\mathbb{R}^{3})). \tag*{(14)} h ( ρ ) ∈ L ∞ ( 0 , T ; L 1 ( R 3 )) , ρ M ∈ L ∞ ( 0 , T ; L 2 ( R 3 )) . (14) Here M = 0 \mathbf{M}=0 M = 0 almost everywhere on { ρ = 0 } \{\rho=0\} { ρ = 0 } , and we adopt the convention
M ρ : = 0 on { ρ = 0 } . \frac{\mathbf{M}}{\sqrt{\rho}}:=0\qquad\text{on }\{\rho=0\}. ρ M := 0 on { ρ = 0 } . All quantities appearing in the weak formulations below are assumed to be locally integrable. In addition, the following conditions are satisfied:
(i) Continuity equation: For every ζ ∈ W c 1 , ∞ ( [ 0 , T ) × R 3 ) \zeta\in W_{\mathrm{c}}^{1,\infty}([0,T)\times\mathbb{R}^{3}) ζ ∈ W c 1 , ∞ ([ 0 , T ) × R 3 ) ,
∫ 0 T ∫ R 3 ( ρ ∂ t ζ + M ⋅ ∇ ζ ) d x d t + ∫ R 3 ρ 0 ( x ) ζ ( 0 , x ) d x = 0. (15) \int_{0}^{T}\int_{\mathbb{R}^{3}}\left(\rho\,\partial_{t}\zeta+\mathbf{M}\cdot\nabla\zeta\right)\,\mathrm{d}x\,\mathrm{d}t+\int_{\mathbb{R}^{3}}\rho_{0}(x)\zeta(0,x)\,\mathrm{d}x=0. \tag*{(15)} ∫ 0 T ∫ R 3 ( ρ ∂ t ζ + M ⋅ ∇ ζ ) d x d t + ∫ R 3 ρ 0 ( x ) ζ ( 0 , x ) d x = 0. (15) (ii) Momentum equations: For every ψ ∈ C c 1 ( [ 0 , T ) × R 3 ; R 3 ) \boldsymbol{\psi}\in C_{\mathrm{c}}^{1}([0,T)\times\mathbb{R}^{3};\mathbb{R}^{3}) ψ ∈ C c 1 ([ 0 , T ) × R 3 ; R 3 ) ,
∫ 0 T ∫ R 3 ( M ⋅ ∂ t ψ + M ⊗ M ρ : ∇ ψ + p ( ρ ) div ψ ) d x d t + ∫ R 3 M 0 ⋅ ψ ( 0 , x ) d x = ∫ 0 T ∫ R 3 ρ ∇ W α ⋅ ψ d x d t . (16) \begin{aligned}
&\int_{0}^{T}\int_{\mathbb{R}^{3}}\left(\mathbf{M}\cdot\partial_{t}\boldsymbol{\psi}+\frac{\mathbf{M}\otimes\mathbf{M}}{\rho}:\nabla\boldsymbol{\psi}+p(\rho)\operatorname{div}\boldsymbol{\psi}\right)\,\mathrm{d}x\,\mathrm{d}t+\int_{\mathbb{R}^{3}}\mathbf{M}_{0}\cdot\boldsymbol{\psi}(0,x)\,\mathrm{d}x\\
&=\int_{0}^{T}\int_{\mathbb{R}^{3}}\rho\nabla W_{\alpha}\cdot\boldsymbol{\psi}\,\mathrm{d}x\,\mathrm{d}t. \tag*{(16)}
\end{aligned} ∫ 0 T ∫ R 3 ( M ⋅ ∂ t ψ + ρ M ⊗ M : ∇ ψ + p ( ρ ) div ψ ) d x d t + ∫ R 3 M 0 ⋅ ψ ( 0 , x ) d x = ∫ 0 T ∫ R 3 ρ ∇ W α ⋅ ψ d x d t . (16) (iii) Mechanical entropy inequality: For every nonnegative ψ ∈ W c 1 , ∞ ( [ 0 , T ) × R 3 ) \psi\in W_{\mathrm{c}}^{1,\infty}([0,T)\times\mathbb{R}^{3}) ψ ∈ W c 1 , ∞ ([ 0 , T ) × R 3 ) ,
∫ 0 T ∫ R 3 ( η ( ρ , M ) ∂ t ψ + q ( ρ , M ) ⋅ ∇ ψ ) d x d t + ∫ R 3 η ( ρ 0 , M 0 ) ψ ( 0 , x ) d x ≥ ∫ 0 T ∫ R 3 ψ M ⋅ ∇ W α d x d t . (17) \int_{0}^{T}\int_{\mathbb{R}^{3}}\left(\eta(\rho,\mathbf{M})\,\partial_{t}\psi+\mathbf{q}(\rho,\mathbf{M})\cdot\nabla\psi\right)\,\mathrm{d}x\,\mathrm{d}t+\int_{\mathbb{R}^{3}}\eta(\rho_{0},\mathbf{M}_{0})\psi(0,x)\,\mathrm{d}x
\ge\int_{0}^{T}\int_{\mathbb{R}^{3}}\psi\,\mathbf{M}\cdot\nabla W_{\alpha}\,\mathrm{d}x\,\mathrm{d}t. \tag*{(17)} ∫ 0 T ∫ R 3 ( η ( ρ , M ) ∂ t ψ + q ( ρ , M ) ⋅ ∇ ψ ) d x d t + ∫ R 3 η ( ρ 0 , M 0 ) ψ ( 0 , x ) d x ≥ ∫ 0 T ∫ R 3 ψ M ⋅ ∇ W α d x d t . (17) (iv) Total energy inequality: The total energy H ( ρ ( t ) , M ( t ) ) \mathcal{H}(\rho(t),\mathbf{M}(t)) H ( ρ ( t ) , M ( t )) defined in (2.7) is finite for almost every t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) and satisfies
H ( ρ ( t ) , M ( t ) ) ≤ H ( ρ 0 , M 0 ) for a.e. t ∈ ( 0 , T ) , \mathcal{H}(\rho(t),\mathbf{M}(t))\le\mathcal{H}(\rho_{0},\mathbf{M}_{0})\qquad\text{for a.e. }t\in(0,T), H ( ρ ( t ) , M ( t )) ≤ H ( ρ 0 , M 0 ) for a.e. t ∈ ( 0 , T ) , as stated in (2.8).
Rotating Equilibria
Write x = ( x 1 , x 2 , z ) \mathbf{x}=(x_{1},x_{2},z) x = ( x 1 , x 2 , z ) ,
r = r ( x ) = x 1 2 + x 2 2 , e r = 1 r ( x 1 , x 2 , 0 ) , e θ = 1 r ( − x 2 , x 1 , 0 ) , e z = ( 0 , 0 , 1 ) . r=r(\mathbf{x})=\sqrt{x_{1}^{2}+x_{2}^{2}},\qquad\mathbf{e}_{r}=\frac{1}{r}(x_{1},x_{2},0),\qquad\mathbf{e}_{\theta}=\frac{1}{r}(-x_{2},x_{1},0),\qquad\mathbf{e}_{z}=(0,0,1). r = r ( x ) = x 1 2 + x 2 2 , e r = r 1 ( x 1 , x 2 , 0 ) , e θ = r 1 ( − x 2 , x 1 , 0 ) , e z = ( 0 , 0 , 1 ) .
A scalar function is axisymmetric if it depends only on ( r , z ) (r,z) ( r , z ) . A pair ( ρ , M ) (\rho,\mathbf{M}) ( ρ , M ) is axisymmetric if ρ = ρ ( r , z ) \rho=\rho(r,z) ρ = ρ ( r , z ) and, away from the rotation axis,
M = M r ( r , z ) e r + M θ ( r , z ) e θ + M z ( r , z ) e z . \mathbf{M}=M^{r}(r,z)\mathbf{e}_{r}+M^{\theta}(r,z)\mathbf{e}_{\theta}+M^{z}(r,z)\mathbf{e}_{z}. M = M r ( r , z ) e r + M θ ( r , z ) e θ + M z ( r , z ) e z .
For an axisymmetric density ϱ \varrho ϱ , define the mass inside the cylinder of radius r r r by
m ϱ ( r ) = ∫ { r ( y ) ≤ r } ϱ ( y ) d y = 2 π ∫ 0 r ∫ R ϱ ( s , z ) s d z d s . (18) m_{\varrho}(r)=\int_{\{r(\mathbf{y})\le r\}}\varrho(\mathbf{y})\,\mathrm{d}\mathbf{y}=2\pi\int_{0}^{r}\int_{\mathbb{R}}\varrho(s,z)s\,\mathrm{d}z\,\mathrm{d}s. \tag*{(18)} m ϱ ( r ) = ∫ { r ( y ) ≤ r } ϱ ( y ) d y = 2 π ∫ 0 r ∫ R ϱ ( s , z ) s d z d s . (18)
Fix a total mass M > 0 M>0 M > 0 . Let J : [ 0 , M ] → R J:[0,M]\to\mathbb{R} J : [ 0 , M ] → R be a prescribed angular momentum profile and set
L ( m ) : = J ( m ) 2 . (19) L(m):=J(m)^{2}. \tag*{(19)} L ( m ) := J ( m ) 2 . (19)
Throughout the paper, we assume
L ∈ C 1 ( [ 0 , M ] ) , L ( 0 ) = 0 , L ′ ( m ) ≥ 0. (20) L \in C^{1}([0,M]), \qquad L(0)=0, \qquad L'(m)\ge0. \tag*{(20)} L ∈ C 1 ([ 0 , M ]) , L ( 0 ) = 0 , L ′ ( m ) ≥ 0. (20)
The rotational kinetic energy is
K L ( ϱ ) = 1 2 ∫ R 3 ϱ ( x ) L ( m ϱ ( r ( x ) ) ) r ( x ) 2 d x . (21) K_L(\varrho)=\frac{1}{2}\int_{\mathbb{R}^{3}}\varrho(x)\frac{L(m_{\varrho}(r(x)))}{r(x)^2}\,dx. \tag*{(21)} K L ( ϱ ) = 2 1 ∫ R 3 ϱ ( x ) r ( x ) 2 L ( m ϱ ( r ( x ))) d x . (21)
For an axisymmetric density ϱ ˉ \bar{\varrho} ϱ ˉ , define the associated centrifugal potential by
Ψ ‾ ϱ ˉ ( r ) : = ∫ r ∞ L ( m ϱ ˉ ( s ) ) s 3 d s for r > 0 , (22) \overline{\Psi}_{\bar{\varrho}}(r):=\int_{r}^{\infty}\frac{L(m_{\bar{\varrho}}(s))}{s^3}\,ds \qquad\text{for }r>0, \tag*{(22)} Ψ ϱ ˉ ( r ) := ∫ r ∞ s 3 L ( m ϱ ˉ ( s )) d s for r > 0 , (22)
whenever the integral is finite. For brevity, we write Ψ ‾ : = Ψ ‾ ϱ ˉ \overline{\Psi}:=\overline{\Psi}_{\bar{\varrho}} Ψ := Ψ ϱ ˉ .
Definition 3.1 (Rotating Euler-Riesz stellar density). A nonnegative density
ϱ ˉ ∈ L 1 ( R 3 ) ∩ L ∞ ( R 3 ) \bar{\varrho}\in L^{1}(\mathbb{R}^{3})\cap L^{\infty}(\mathbb{R}^{3}) ϱ ˉ ∈ L 1 ( R 3 ) ∩ L ∞ ( R 3 ) is called a compactly supported rotating Euler-Riesz stellar density of mass M M M if the following conditions are satisfied:
(i) Mass, symmetry, and finite rotational energy: The density ϱ ˉ \bar{\varrho} ϱ ˉ is axisymmetric,
∫ R 3 ϱ ˉ d x = M , K L ( ϱ ˉ ) < ∞ , \int_{\mathbb{R}^{3}}\bar{\varrho}\,dx=M,\qquad K_L(\bar{\varrho})<\infty, ∫ R 3 ϱ ˉ d x = M , K L ( ϱ ˉ ) < ∞ , and its positivity set Γ : = { x ∈ R 3 : ϱ ˉ ( x ) > 0 } \Gamma:=\{x\in\mathbb{R}^{3}:\bar{\varrho}(x)>0\} Γ := { x ∈ R 3 : ϱ ˉ ( x ) > 0 } is bounded.
(ii) Euler-Lagrange equilibrium conditions: There exists a constant μ > 0 \mu>0 μ > 0 such that, with
W ‾ α : = Φ α ∗ ϱ ˉ , Ψ ‾ : = Ψ ‾ ϱ ˉ , \overline{W}_{\alpha}:=\Phi_{\alpha}*\bar{\varrho},\qquad\overline{\Psi}:=\overline{\Psi}_{\bar{\varrho}}, W α := Φ α ∗ ϱ ˉ , Ψ := Ψ ϱ ˉ , the following conditions hold:
h ′ ( ϱ ˉ ) + Ψ ‾ + W ‾ α + μ = 0 a.e. on Γ , (23) h'(\bar{\varrho})+\overline{\Psi}+\overline{W}_{\alpha}+\mu=0 \qquad\text{a.e. on }\Gamma, \tag*{(23)} h ′ ( ϱ ˉ ) + Ψ + W α + μ = 0 a.e. on Γ , (23) Ψ ‾ + W ‾ α + μ ≥ 0 a.e. on Γ c . (24) \overline{\Psi}+\overline{W}_{\alpha}+\mu\ge0 \qquad\text{a.e. on }\Gamma^{c}. \tag*{(24)} Ψ + W α + μ ≥ 0 a.e. on Γ c . (24) (iii) Stationary Euler-Riesz equations: Define the azimuthal velocity field by
u ˉ ( x ) : = J ( m ϱ ˉ ( r ( x ) ) ) r ( x ) e θ , M ‾ : = ϱ ˉ u ˉ , (25) \bar{u}(x):=\frac{J(m_{\bar{\varrho}}(r(x)))}{r(x)}e_{\theta},\qquad\overline{M}:=\bar{\varrho}\,\bar{u}, \tag*{(25)} u ˉ ( x ) := r ( x ) J ( m ϱ ˉ ( r ( x ))) e θ , M := ϱ ˉ u ˉ , (25) where r ( x ) r(x) r ( x ) denotes the cylindrical radius. The pair ( ϱ ˉ , M ‾ ) (\bar{\varrho},\overline{M}) ( ϱ ˉ , M ) is required to be a stationary weak solution of (1.1).
Lemma 3.2. Let ϱ ˉ \bar{\varrho} ϱ ˉ be the rotating star defined in Definition 3.1. If ϱ ∈ L + 1 ( R 3 ) \varrho\in L^{1}_{+}(\mathbb{R}^{3}) ϱ ∈ L + 1 ( R 3 ) is axisymmetric, has mass M M M , and K L ( ϱ ) < ∞ K_L(\varrho)<\infty K L ( ϱ ) < ∞ , then
Ψ ‾ ∣ ϱ − ϱ ˉ ∣ ∈ L 1 ( R 3 ) . (26) \overline{\Psi}|\varrho-\bar{\varrho}|\in L^{1}(\mathbb{R}^{3}). \tag*{(26)} Ψ ∣ ϱ − ϱ ˉ ∣ ∈ L 1 ( R 3 ) . (26) Moreover,
∫ R 3 ϱ Ψ ‾ d x ≤ K L ( ϱ ) + C ϱ ˉ , L , (27) \int_{\mathbb{R}^{3}}\varrho\,\overline{\Psi}\,dx\le K_L(\varrho)+C_{\bar{\varrho},L}, \tag*{(27)} ∫ R 3 ϱ Ψ d x ≤ K L ( ϱ ) + C ϱ ˉ , L , (27) where C ϱ ˉ , L < ∞ C_{\bar{\varrho},L}<\infty C ϱ ˉ , L < ∞ depends only on ϱ ˉ \bar{\varrho} ϱ ˉ and L L L .
Proof. Set
F ( q ) : = ∫ 0 q L ( s ) d s . F(q):=\int_{0}^{q}L(s)\,ds. F ( q ) := ∫ 0 q L ( s ) d s .
Since L L L is nondecreasing, F F F is convex. The supporting-line inequality at m ϱ ˉ ( r ) m_{\bar{\varrho}}(r) m ϱ ˉ ( r ) gives
m ϱ ( r ) L ( m ϱ ˉ ( r ) ) ≤ F ( m ϱ ( r ) ) − F ( m ϱ ˉ ( r ) ) + m ϱ ˉ ( r ) L ( m ϱ ˉ ( r ) ) . m_{\varrho}(r)L(m_{\bar{\varrho}}(r))\le F(m_{\varrho}(r))-F(m_{\bar{\varrho}}(r))+m_{\bar{\varrho}}(r)L(m_{\bar{\varrho}}(r)). m ϱ ( r ) L ( m ϱ ˉ ( r )) ≤ F ( m ϱ ( r )) − F ( m ϱ ˉ ( r )) + m ϱ ˉ ( r ) L ( m ϱ ˉ ( r )) .
By the definition of Ψ ‾ \overline{\Psi} Ψ and Tonelli’s theorem, it follows that
∫ R 3 ϱ Ψ ‾ d x ≤ ∫ 0 ∞ F ( m ϱ ( r ) ) r 3 d r + C ϱ ˉ , L , \int_{\mathbb{R}^{3}}\varrho\,\overline{\Psi}\,dx\le\int_{0}^{\infty}\frac{F(m_{\varrho}(r))}{r^{3}}\,dr+C_{\bar{\varrho},L}, ∫ R 3 ϱ Ψ d x ≤ ∫ 0 ∞ r 3 F ( m ϱ ( r )) d r + C ϱ ˉ , L ,
where
C ρ ˉ , L : = ∫ 0 ∞ m ρ ˉ ( r ) L ( m ρ ˉ ( r ) ) − F ( m ρ ˉ ( r ) ) r 3 d r . C_{\bar{\rho},L}:=\int_{0}^{\infty}\frac{m_{\bar{\rho}}(r)L(m_{\bar{\rho}}(r))-F(m_{\bar{\rho}}(r))}{r^{3}}\,\mathrm{d}r. C ρ ˉ , L := ∫ 0 ∞ r 3 m ρ ˉ ( r ) L ( m ρ ˉ ( r )) − F ( m ρ ˉ ( r )) d r .
We first verify that C ρ ˉ , L C_{\bar{\rho},L} C ρ ˉ , L is finite. Since L ′ ≥ 0 L'\ge0 L ′ ≥ 0 ,
0 ≤ q L ( q ) − F ( q ) = ∫ 0 q s L ′ ( s ) d s ≤ 1 2 ∥ L ′ ′ ∥ L ∞ ( 0 , M ) q 2 for 0 ≤ q ≤ M . 0\le qL(q)-F(q)=\int_{0}^{q}sL'(s)\,\mathrm{d}s\le\frac{1}{2}\lVert L''\rVert_{L^\infty(0,M)}q^{2}\qquad\text{for }0\le q\le M. 0 ≤ q L ( q ) − F ( q ) = ∫ 0 q s L ′ ( s ) d s ≤ 2 1 ∥ L ′′ ∥ L ∞ ( 0 , M ) q 2 for 0 ≤ q ≤ M .
Because ρ ˉ \bar{\rho} ρ ˉ is bounded and compactly supported, its cylindrical mass satisfies
m ρ ˉ ( r ) = O ( r 2 ) as r ↓ 0 , m ρ ˉ ( r ) = M for all sufficiently large r . m_{\bar{\rho}}(r)=O(r^{2})\quad\text{as }r\downarrow0,\qquad m_{\bar{\rho}}(r)=M\quad\text{for all sufficiently large }r. m ρ ˉ ( r ) = O ( r 2 ) as r ↓ 0 , m ρ ˉ ( r ) = M for all sufficiently large r .
Consequently, the integrand defining C ρ ˉ , L C_{\bar{\rho},L} C ρ ˉ , L is O ( r ) O(r) O ( r ) near the origin and O ( r − 3 ) O(r^{-3}) O ( r − 3 ) at infinity. Then
C ρ ˉ , L < ∞ . C_{\bar{\rho},L}<\infty. C ρ ˉ , L < ∞.
We next relate the first term to the rotational energy. By the definition of K L \mathcal{K}_{L} K L ,
K L ( ρ ) = 1 2 ∫ 0 ∞ L ( m ρ ( r ) ) m ρ ′ ( r ) r 2 d r . \mathcal{K}_{L}(\rho)=\frac{1}{2}\int_{0}^{\infty}\frac{L(m_{\rho}(r))m_{\rho}'(r)}{r^{2}}\,\mathrm{d}r. K L ( ρ ) = 2 1 ∫ 0 ∞ r 2 L ( m ρ ( r )) m ρ ′ ( r ) d r .
Since
d d r F ( m ρ ( r ) ) = L ( m ρ ( r ) ) m ρ ′ ( r ) for a.e. r > 0 , \frac{\mathrm{d}}{\mathrm{d}r}F(m_{\rho}(r))=L(m_{\rho}(r))m_{\rho}'(r)\qquad\text{for a.e. }r>0, d r d F ( m ρ ( r )) = L ( m ρ ( r )) m ρ ′ ( r ) for a.e. r > 0 ,
we have
K L ( ρ ) = 1 2 ∫ 0 ∞ 1 r 2 d d r F ( m ρ ( r ) ) d r . \mathcal{K}_{L}(\rho)=\frac{1}{2}\int_{0}^{\infty}\frac{1}{r^{2}}\frac{\mathrm{d}}{\mathrm{d}r}F(m_{\rho}(r))\,\mathrm{d}r. K L ( ρ ) = 2 1 ∫ 0 ∞ r 2 1 d r d F ( m ρ ( r )) d r .
To justify integration by parts, we verify that the boundary terms vanish. Since F ( 0 ) = 0 F(0)=0 F ( 0 ) = 0 and L L L is nonnegative and nondecreasing,
0 ≤ F ( m ρ ( r ) ) r 2 = 1 r 2 ∫ 0 r L ( m ρ ( s ) ) m ρ ′ ( s ) d s ≤ ∫ 0 r L ( m ρ ( s ) ) m ρ ′ ( s ) s 2 d s ⟶ 0 0\le\frac{F(m_{\rho}(r))}{r^{2}}=\frac{1}{r^{2}}\int_{0}^{r}L(m_{\rho}(s))m_{\rho}'(s)\,\mathrm{d}s\le\int_{0}^{r}\frac{L(m_{\rho}(s))m_{\rho}'(s)}{s^{2}}\,\mathrm{d}s\longrightarrow0 0 ≤ r 2 F ( m ρ ( r )) = r 2 1 ∫ 0 r L ( m ρ ( s )) m ρ ′ ( s ) d s ≤ ∫ 0 r s 2 L ( m ρ ( s )) m ρ ′ ( s ) d s ⟶ 0
as r → 0 r\to0 r → 0 , where the convergence follows from K L ( ρ ) < ∞ \mathcal{K}_{L}(\rho)<\infty K L ( ρ ) < ∞ .
At infinity, since m ρ ( r ) ≤ M m_{\rho}(r)\le M m ρ ( r ) ≤ M ,
0 ≤ F ( m ρ ( r ) ) r 2 ≤ F ( M ) r 2 ⟶ 0 as r → ∞ . 0\le\frac{F(m_{\rho}(r))}{r^{2}}\le\frac{F(M)}{r^{2}}\longrightarrow0\qquad\text{as }r\to\infty. 0 ≤ r 2 F ( m ρ ( r )) ≤ r 2 F ( M ) ⟶ 0 as r → ∞.
Integration by parts therefore yields
K L ( ρ ) = ∫ 0 ∞ F ( m ρ ( r ) ) r 3 d r . \mathcal{K}_{L}(\rho)=\int_{0}^{\infty}\frac{F(m_{\rho}(r))}{r^{3}}\,\mathrm{d}r. K L ( ρ ) = ∫ 0 ∞ r 3 F ( m ρ ( r )) d r .
Combining this identity with the convexity estimate gives
∫ R 3 ρ Ψ ˉ d x ≤ K L ( ρ ) + C ρ ˉ , L , \int_{\mathbb{R}^{3}}\rho\bar{\Psi}\,\mathrm{d}x\le\mathcal{K}_{L}(\rho)+C_{\bar{\rho},L}, ∫ R 3 ρ Ψ ˉ d x ≤ K L ( ρ ) + C ρ ˉ , L ,
which proves (3.10).
Finally, applying the same estimate with ρ = ρ ˉ \rho=\bar{\rho} ρ = ρ ˉ and using Ψ ˉ ≥ 0 \bar{\Psi}\ge0 Ψ ˉ ≥ 0 , we obtain
∫ R 3 ∣ ρ − ρ ˉ ∣ Ψ ˉ d x ≤ ∫ R 3 ( ρ + ρ ˉ ) Ψ ˉ d x ≤ K L ( ρ ) + K L ( ρ ˉ ) + 2 C ρ ˉ , L . \int_{\mathbb{R}^{3}}|\rho-\bar{\rho}|\bar{\Psi}\,\mathrm{d}x\le\int_{\mathbb{R}^{3}}(\rho+\bar{\rho})\bar{\Psi}\,\mathrm{d}x\le\mathcal{K}_{L}(\rho)+\mathcal{K}_{L}(\bar{\rho})+2C_{\bar{\rho},L}. ∫ R 3 ∣ ρ − ρ ˉ ∣ Ψ ˉ d x ≤ ∫ R 3 ( ρ + ρ ˉ ) Ψ ˉ d x ≤ K L ( ρ ) + K L ( ρ ˉ ) + 2 C ρ ˉ , L .
This establishes (3.9). □
For a time-dependent axisymmetric solution ( ρ , M ) (\rho,\mathbf{M}) ( ρ , M ) , decompose the momentum in cylindrical coordinates as
M = M r e r + M θ e θ + M z e z . (28) \mathbf{M}=M^{r}\mathbf{e}_{r}+M^{\theta}\mathbf{e}_{\theta}+M^{z}\mathbf{e}_{z}. \tag*{(28)} M = M r e r + M θ e θ + M z e z . (28)
Definition 3.3 (Angular momentum compatibility). An axisymmetric solution ( ρ , M ) (\rho,\mathbf{M}) ( ρ , M ) is said to be compatible with J J J if
M θ ( t , x ) = ρ ( t , x ) J ( m ρ ( t ) ( r ( x ) ) ) r ( x ) M^\theta(t,x)=\rho(t,x)\frac{J(m_{\rho(t)}(r(x)))}{r(x)} M θ ( t , x ) = ρ ( t , x ) r ( x ) J ( m ρ ( t ) ( r ( x ))) for a.e. ( t , x ) ∈ ( 0 , T ) × R 3 (t,x)\in(0,T)\times\mathbb{R}^3 ( t , x ) ∈ ( 0 , T ) × R 3 . {#eq:3.12}
where r ( x ) r(x) r ( x ) denotes the cylindrical radius.
Under this compatibility condition, the azimuthal kinetic energy is represented by the rotational energy functional:
1 2 ∫ R 3 ∣ M θ ∣ 2 ρ d x = K L ( ρ ) \frac{1}{2}\int_{\mathbb{R}^3}\frac{|M^\theta|^2}{\rho}\,dx=\mathcal{K}_L(\rho) 2 1 ∫ R 3 ρ ∣ M θ ∣ 2 d x = K L ( ρ )
for a.e. t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) . {#eq:3.13}
For subsequent estimates, introduce the effective equilibrium potentials
V ‾ ( x ) : = Ψ ‾ ( r ( x ) ) + W ‾ α ( x ) + μ , G ‾ ( x ) : = W ‾ α ( x ) + μ . \overline{V}(x):=\overline{\Psi}(r(x))+\overline{W}_\alpha(x)+\mu,\qquad\overline{G}(x):=\overline{W}_\alpha(x)+\mu. V ( x ) := Ψ ( r ( x )) + W α ( x ) + μ , G ( x ) := W α ( x ) + μ .
By the Euler–Lagrange conditions (23) –(24) ,
V ‾ = − h ′ ( ρ ‾ ) a.e. on Γ , V ‾ ≥ 0 a.e. on Γ c , (29) \overline{V}=-h'(\overline{\rho})\quad\text{a.e. on }\Gamma,\qquad\overline{V}\geq0\quad\text{a.e. on }\Gamma^c, \tag*{(29)} V = − h ′ ( ρ ) a.e. on Γ , V ≥ 0 a.e. on Γ c , (29)
where Γ = { ρ ‾ > 0 } \Gamma=\{\overline{\rho}>0\} Γ = { ρ > 0 } .
Since ρ ‾ ∈ L 1 ( R 3 ) ∩ L ∞ ( R 3 ) \overline{\rho}\in L^1(\mathbb{R}^3)\cap L^\infty(\mathbb{R}^3) ρ ∈ L 1 ( R 3 ) ∩ L ∞ ( R 3 ) is compactly supported and 0 < α < 2 0<\alpha<2 0 < α < 2 , the equilibrium interaction potential satisfies
W ‾ α ∈ W 1 , ∞ ( R 3 ) , G ‾ ∈ W 1 , ∞ ( R 3 ) , ∇ G ‾ = ∇ W ‾ α . (30) \overline{W}_\alpha\in W^{1,\infty}(\mathbb{R}^3),\qquad\overline{G}\in W^{1,\infty}(\mathbb{R}^3),\qquad\nabla\overline{G}=\nabla\overline{W}_\alpha. \tag*{(30)} W α ∈ W 1 , ∞ ( R 3 ) , G ∈ W 1 , ∞ ( R 3 ) , ∇ G = ∇ W α . (30)
Indeed, the kernels ∣ x ∣ − α |x|^{-\alpha} ∣ x ∣ − α and ∣ x ∣ − α − 1 |x|^{-\alpha-1} ∣ x ∣ − α − 1 are locally integrable in three dimensions when α < 2 \alpha<2 α < 2 . Splitting the convolution into the regions ∣ x − y ∣ ≤ 1 |x-y|\leq1 ∣ x − y ∣ ≤ 1 and ∣ x − y ∣ > 1 |x-y|>1 ∣ x − y ∣ > 1 , we obtain
∣ W ‾ α ( x ) ∣ ≤ C ∥ ρ ‾ ∥ L ∞ ∫ ∣ z ∣ ≤ 1 ∣ z ∣ − α d z + C ∥ ρ ‾ ∥ L 1 , |\overline{W}_\alpha(x)|\leq C\|\overline{\rho}\|_{L^\infty}\int_{|z|\leq1}|z|^{-\alpha}\,dz+C\|\overline{\rho}\|_{L^1}, ∣ W α ( x ) ∣ ≤ C ∥ ρ ∥ L ∞ ∫ ∣ z ∣ ≤ 1 ∣ z ∣ − α d z + C ∥ ρ ∥ L 1 ,
∣ ∇ W ‾ α ( x ) ∣ ≤ C ∥ ρ ‾ ∥ L ∞ ∫ ∣ z ∣ ≤ 1 ∣ z ∣ − α − 1 d z + C ∥ ρ ‾ ∥ L 1 . |\nabla\overline{W}_\alpha(x)|\leq C\|\overline{\rho}\|_{L^\infty}\int_{|z|\leq1}|z|^{-\alpha-1}\,dz+C\|\overline{\rho}\|_{L^1}. ∣∇ W α ( x ) ∣ ≤ C ∥ ρ ∥ L ∞ ∫ ∣ z ∣ ≤ 1 ∣ z ∣ − α − 1 d z + C ∥ ρ ∥ L 1 .
Both bounds are uniform in x ∈ R 3 x\in\mathbb{R}^3 x ∈ R 3 , proving (30) .
Moreover, compact support and integrability of ρ ‾ \overline{\rho} ρ imply
W ‾ α ( x ) ⟶ 0 as ∣ x ∣ ⟶ ∞ . \overline{W}_\alpha(x)\longrightarrow0\qquad\text{as }|x|\longrightarrow\infty. W α ( x ) ⟶ 0 as ∣ x ∣ ⟶ ∞.
Since Ψ ‾ ≥ 0 \overline{\Psi}\geq0 Ψ ≥ 0 and μ > 0 \mu>0 μ > 0 , there exists R 0 > 0 R_0>0 R 0 > 0 such that
V ‾ ( x ) ≥ μ 2 for ∣ x ∣ ≥ R 0 . (31) \overline{V}(x)\geq\frac{\mu}{2}\qquad\text{for }|x|\geq R_0. \tag*{(31)} V ( x ) ≥ 2 μ for ∣ x ∣ ≥ R 0 . (31)
Positive Rotational Relative Entropy
Let ρ \rho ρ be a nonnegative axisymmetric density with the same total mass M M M as ρ ‾ \overline{\rho} ρ . Following [41 , 42 ] , we introduce the relative energy functionals
d 1 ( ρ , ρ ‾ ) : = ∫ R 3 ( h ( ρ ) − h ( ρ ‾ ) + V ‾ ( ρ − ρ ‾ ) ) d x , (32) d_1(\rho,\overline{\rho}):=\int_{\mathbb{R}^3}\left(h(\rho)-h(\overline{\rho})+\overline{V}(\rho-\overline{\rho})\right)\,dx, \tag*{(32)} d 1 ( ρ , ρ ) := ∫ R 3 ( h ( ρ ) − h ( ρ ) + V ( ρ − ρ ) ) d x , (32)
d 2 ( ρ , ρ ‾ ) : = K L ( ρ ) − K L ( ρ ‾ ) − ∫ R 3 Ψ ‾ ( ρ − ρ ‾ ) d x . (33) d_2(\rho,\overline{\rho}):=\mathcal{K}_L(\rho)-\mathcal{K}_L(\overline{\rho})-\int_{\mathbb{R}^3}\overline{\Psi}(\rho-\overline{\rho})\,dx. \tag*{(33)} d 2 ( ρ , ρ ) := K L ( ρ ) − K L ( ρ ) − ∫ R 3 Ψ ( ρ − ρ ) d x . (33)
Throughout this section, we assume that ∫ R 3 h ( ρ ) d x < ∞ \int_{\mathbb{R}^3}h(\rho)\,dx<\infty ∫ R 3 h ( ρ ) d x < ∞ and K L ( ρ ) < ∞ \mathcal{K}_L(\rho)<\infty K L ( ρ ) < ∞ . By Lemma 3.2 , Ψ ‾ ∣ ρ − ρ ‾ ∣ ∈ L 1 ( R 3 ) \overline{\Psi}|\rho-\overline{\rho}|\in L^1(\mathbb{R}^3) Ψ ∣ ρ − ρ ∣ ∈ L 1 ( R 3 ) , so d 2 ( ρ , ρ ‾ ) d_2(\rho,\overline{\rho}) d 2 ( ρ , ρ ) is well defined and finite.
Coercivity of the Pressure Term
Lemma 4.1 (Coercivity of d 1 d_1 d 1 ). Under Assumption 2.1, there exists c > 0 c>0 c > 0 , depending only on h h h and ρ ˉ \bar{\rho} ρ ˉ , such that
d 1 ( ρ , ρ ˉ ) ≥ c ( ∫ Γ ∩ { ∣ f ∣ ≤ 1 } ∣ f ∣ 2 d x + ∫ Γ ∩ { ∣ f ∣ > 1 } ∣ f ∣ γ 2 d x + ∫ Γ c ∩ { ρ ≤ 1 } ρ γ 1 d x + ∫ Γ c ∩ { ρ > 1 } ρ γ 2 d x + ∫ Γ c V ˉ ρ d x ) , (34) d_1(\rho,\bar{\rho})\geq c\left(\int_{\Gamma\cap\{|f|\leq1\}}|f|^2\,\mathrm{d}x+\int_{\Gamma\cap\{|f|>1\}}|f|^{\gamma_2}\,\mathrm{d}x+\int_{\Gamma^c\cap\{\rho\leq1\}}\rho^{\gamma_1}\,\mathrm{d}x+\int_{\Gamma^c\cap\{\rho>1\}}\rho^{\gamma_2}\,\mathrm{d}x+\int_{\Gamma^c}\bar{V}\rho\,\mathrm{d}x\right), \tag*{(34)} d 1 ( ρ , ρ ˉ ) ≥ c ( ∫ Γ ∩ { ∣ f ∣ ≤ 1 } ∣ f ∣ 2 d x + ∫ Γ ∩ { ∣ f ∣ > 1 } ∣ f ∣ γ 2 d x + ∫ Γ c ∩ { ρ ≤ 1 } ρ γ 1 d x + ∫ Γ c ∩ { ρ > 1 } ρ γ 2 d x + ∫ Γ c V ˉ ρ d x ) , (34) where f : = ρ − ρ ˉ f:=\rho-\bar{\rho} f := ρ − ρ ˉ and Γ : = { ρ ˉ > 0 } \Gamma:=\{\bar{\rho}>0\} Γ := { ρ ˉ > 0 } .
Proof. For a , b ≥ 0 a,b\geq0 a , b ≥ 0 , define the relative internal energy
D h ( a ∣ b ) : = h ( a ) − h ( b ) − h ′ ( b ) ( a − b ) . D_h(a\mid b):=h(a)-h(b)-h'(b)(a-b). D h ( a ∣ b ) := h ( a ) − h ( b ) − h ′ ( b ) ( a − b ) .
By the Euler–Lagrange identity (29) , we have
h ( ρ ) − h ( ρ ˉ ) + V ˉ ( ρ − ρ ˉ ) = D h ( ρ ∣ ρ ˉ ) a.e. on Γ . (35) h(\rho)-h(\bar{\rho})+\bar{V}(\rho-\bar{\rho})=D_h(\rho\mid\bar{\rho})\qquad\text{a.e. on }\Gamma. \tag*{(35)} h ( ρ ) − h ( ρ ˉ ) + V ˉ ( ρ − ρ ˉ ) = D h ( ρ ∣ ρ ˉ ) a.e. on Γ. (35)
Set B : = ∥ ρ ˉ ∥ L ∞ B:=\|\bar{\rho}\|_{L^\infty} B := ∥ ρ ˉ ∥ L ∞ . We first consider the region Γ ∩ { ∣ f ∣ ≤ 1 } \Gamma\cap\{|f|\leq1\} Γ ∩ { ∣ f ∣ ≤ 1 } . By Taylor’s formula and (2.6),
D h ( ρ ∣ ρ ˉ ) = f 2 ∫ 0 1 ( 1 − s ) h ′ ′ ( ρ ˉ + s f ) d s ≥ 1 2 inf 0 < z ≤ B + 1 h ′ ′ ( z ) ∣ f ∣ 2 ≥ c ∣ f ∣ 2 . D_h(\rho\mid\bar{\rho})=f^2\int_0^1(1-s)h''(\bar{\rho}+sf)\,\mathrm{d}s\geq\frac{1}{2}\inf_{0<z\leq B+1}h''(z)|f|^2\geq c|f|^2. D h ( ρ ∣ ρ ˉ ) = f 2 ∫ 0 1 ( 1 − s ) h ′′ ( ρ ˉ + s f ) d s ≥ 2 1 0 < z ≤ B + 1 inf h ′′ ( z ) ∣ f ∣ 2 ≥ c ∣ f ∣ 2 .
Here the integral representation is understood by continuity when the interpolation segment reaches zero.
Next, suppose that ∣ f ∣ > 1 |f|>1 ∣ f ∣ > 1 . Since 0 ≤ ρ ˉ ≤ B 0\leq\bar{\rho}\leq B 0 ≤ ρ ˉ ≤ B , the functions h h h and h ′ h' h ′ are bounded on [ 0 , B ] [0,B] [ 0 , B ] . Moreover, (2.5) implies that h ( a ) ≥ c a γ 2 h(a)\geq ca^{\gamma_2} h ( a ) ≥ c a γ 2 for sufficiently large a a a . Consequently, there exist R > B + 1 R>B+1 R > B + 1 and c > 0 c>0 c > 0 such that
D h ( a ∣ b ) ≥ c a γ 2 ≥ c ∣ a − b ∣ γ 2 D_h(a\mid b)\geq ca^{\gamma_2}\geq c|a-b|^{\gamma_2} D h ( a ∣ b ) ≥ c a γ 2 ≥ c ∣ a − b ∣ γ 2
whenever a ≥ R a\geq R a ≥ R and 0 ≤ b ≤ B 0\leq b\leq B 0 ≤ b ≤ B .
On the compact set
A R : = { ( a , b ) ∈ [ 0 , R ] × [ 0 , B ] : ∣ a − b ∣ ≥ 1 } , A_R:=\{(a,b)\in[0,R]\times[0,B]:|a-b|\geq1\}, A R := {( a , b ) ∈ [ 0 , R ] × [ 0 , B ] : ∣ a − b ∣ ≥ 1 } ,
the function
( a , b ) ⟼ D h ( a ∣ b ) ∣ a − b ∣ γ 2 (a,b)\longmapsto\frac{D_h(a\mid b)}{|a-b|^{\gamma_2}} ( a , b ) ⟼ ∣ a − b ∣ γ 2 D h ( a ∣ b )
is continuous and strictly positive, by the strict convexity of h h h . It therefore has a positive minimum on A R A_R A R . Combining the two regions gives
D h ( ρ ∣ ρ ˉ ) ≥ c ∣ f ∣ γ 2 on Γ ∩ { ∣ f ∣ > 1 } . D_h(\rho\mid\bar{\rho})\geq c|f|^{\gamma_2}\qquad\text{on }\Gamma\cap\{|f|>1\}. D h ( ρ ∣ ρ ˉ ) ≥ c ∣ f ∣ γ 2 on Γ ∩ { ∣ f ∣ > 1 } .
On the vacuum region Γ c \Gamma^c Γ c , we have ρ ˉ = 0 \bar{\rho}=0 ρ ˉ = 0 and V ˉ ≥ 0 \bar{V}\geq0 V ˉ ≥ 0 by (3.14). Since h ( 0 ) = h ′ ( 0 ) = 0 h(0)=h'(0)=0 h ( 0 ) = h ′ ( 0 ) = 0 ,
h ( ρ ) − h ( ρ ˉ ) + V ˉ ( ρ − ρ ˉ ) = h ( ρ ) + V ˉ ρ . h(\rho)-h(\bar{\rho})+\bar{V}(\rho-\bar{\rho})=h(\rho)+\bar{V}\rho. h ( ρ ) − h ( ρ ˉ ) + V ˉ ( ρ − ρ ˉ ) = h ( ρ ) + V ˉ ρ .
The pressure estimates (2.4)–(2.5) therefore yield
h ( ρ ) + V ˉ ρ ≥ c ( ρ γ 1 1 { ρ ≤ 1 } + ρ γ 2 1 { ρ > 1 } + V ˉ ρ ) . h(\rho)+\bar{V}\rho\geq c\left(\rho^{\gamma_1}\mathbf{1}_{\{\rho\leq1\}}+\rho^{\gamma_2}\mathbf{1}_{\{\rho>1\}}+\bar{V}\rho\right). h ( ρ ) + V ˉ ρ ≥ c ( ρ γ 1 1 { ρ ≤ 1 } + ρ γ 2 1 { ρ > 1 } + V ˉ ρ ) .
Integrating these estimates over Γ \Gamma Γ and Γ c \Gamma^c Γ c proves (34) . □
Lemma 4.2 (Density norms controlled by d 1 d_1 d 1 ). Let s ∈ ( 1 , γ ] s\in(1,\gamma] s ∈ ( 1 , γ ] for γ = min { γ 1 , γ 2 } \gamma=\min\{\gamma_1,\gamma_2\} γ = min { γ 1 , γ 2 } and write d 1 : = d 1 ( ρ , ρ ˉ ) d_1:=d_1(\rho,\bar{\rho}) d 1 := d 1 ( ρ , ρ ˉ ) . Then
∥ ρ − ρ ˉ ∥ L s ( R 3 ) ≤ C s ( d 1 1 2 + d 1 1 γ 1 + d 1 1 γ 2 ) . (36) \|\rho-\bar{\rho}\|_{L^s(\mathbb{R}^3)}\leq C_s\left(d_1^{\frac{1}{2}}+d_1^{\frac{1}{\gamma_1}}+d_1^{\frac{1}{\gamma_2}}\right). \tag*{(36)} ∥ ρ − ρ ˉ ∥ L s ( R 3 ) ≤ C s ( d 1 2 1 + d 1 γ 1 1 + d 1 γ 2 1 ) . (36) In particular, if 0 ≤ d 1 ≤ 1 0\leq d_1\leq1 0 ≤ d 1 ≤ 1 , then
∥ ρ − ρ ˉ ∥ L s ( R 3 ) ≤ C s d 1 1 2 . (37) \|\rho-\bar{\rho}\|_{L^s(\mathbb{R}^3)}\leq C_sd_1^{\frac{1}{2}}. \tag*{(37)} ∥ ρ − ρ ˉ ∥ L s ( R 3 ) ≤ C s d 1 2 1 . (37)
Proof. Set f : = ρ − ρ ˉ f := \rho- \bar{\rho} f := ρ − ρ ˉ . Since
1 < s ≤ γ ≤ γ 1 ≤ 2 , s ≤ γ 2 , 1 < s \le\gamma\le\gamma_1 \le2,\qquad s \le\gamma_2, 1 < s ≤ γ ≤ γ 1 ≤ 2 , s ≤ γ 2 ,
we estimate f f f separately on the regions appearing in (4.3).
First, since Γ \Gamma Γ is bounded, Hölder’s inequality and the quadratic part of (4.3) give
∥ f ∥ L s ( Γ ∩ { ∣ f ∣ ≤ 1 } ) ≤ ∣ Γ ∣ 1 s − 1 2 ∥ f ∥ L 2 ( Γ ∩ { ∣ f ∣ ≤ 1 } ) ≤ C d 1 1 2 . \|f\|_{L^s(\Gamma\cap\{|f|\le1\})} \le|\Gamma|^{\frac{1}{s}-\frac{1}{2}}\|f\|_{L^2(\Gamma\cap\{|f|\le1\})} \le Cd_1^{\frac{1}{2}}. ∥ f ∥ L s ( Γ ∩ { ∣ f ∣ ≤ 1 }) ≤ ∣Γ ∣ s 1 − 2 1 ∥ f ∥ L 2 ( Γ ∩ { ∣ f ∣ ≤ 1 }) ≤ C d 1 2 1 .
On Γ ∩ { ∣ f ∣ > 1 } \Gamma\cap\{|f|>1\} Γ ∩ { ∣ f ∣ > 1 } , the inequality s ≤ γ 2 s\le\gamma_2 s ≤ γ 2 implies ∣ f ∣ s ≤ ∣ f ∣ γ 2 |f|^s\le|f|^{\gamma_2} ∣ f ∣ s ≤ ∣ f ∣ γ 2 . Then
∥ f ∥ L s ( Γ ∩ { ∣ f ∣ > 1 } ) ≤ C d 1 1 s . \|f\|_{L^s(\Gamma\cap\{|f|>1\})} \le Cd_1^{\frac{1}{s}}. ∥ f ∥ L s ( Γ ∩ { ∣ f ∣ > 1 }) ≤ C d 1 s 1 .
We next consider the vacuum region Γ c \Gamma^c Γ c , where f = ρ f=\rho f = ρ . On Γ c ∩ { ρ > 1 } \Gamma^c\cap\{\rho>1\} Γ c ∩ { ρ > 1 } ,
∥ ρ ∥ L s ( Γ c ∩ { ρ > 1 } ) s = ∫ Γ c ∩ { ρ > 1 } ρ s d x ≤ ∫ Γ c ∩ { ρ > 1 } ρ γ 2 d x ≤ C d 1 . \|\rho\|_{L^s(\Gamma^c\cap\{\rho>1\})}^s
= \int_{\Gamma^c\cap\{\rho>1\}}\rho^s\,dx
\le\int_{\Gamma^c\cap\{\rho>1\}}\rho^{\gamma_2}\,dx
\le Cd_1. ∥ ρ ∥ L s ( Γ c ∩ { ρ > 1 }) s = ∫ Γ c ∩ { ρ > 1 } ρ s d x ≤ ∫ Γ c ∩ { ρ > 1 } ρ γ 2 d x ≤ C d 1 .
Therefore,
∥ ρ ∥ L s ( Γ c ∩ { ρ > 1 } ) ≤ C d 1 1 s . \|\rho\|_{L^s(\Gamma^c\cap\{\rho>1\})} \le Cd_1^{\frac{1}{s}}. ∥ ρ ∥ L s ( Γ c ∩ { ρ > 1 }) ≤ C d 1 s 1 .
It remains to estimate the low-density region Γ c ∩ { ρ ≤ 1 } \Gamma^c\cap\{\rho\le1\} Γ c ∩ { ρ ≤ 1 } . Let R 0 R_0 R 0 be as in (3.16). On the bounded set B R 0 B_{R_0} B R 0 , Hölder’s inequality and s ≤ γ 1 s\le\gamma_1 s ≤ γ 1 give
∥ ρ ∥ L s ( Γ c ∩ { ρ ≤ 1 } ∩ B R 0 ) ≤ ∣ B R 0 ∣ 1 s − 1 γ 1 ∥ ρ ∥ L γ 1 ( Γ c ∩ { ρ ≤ 1 } ∩ B R 0 ) ≤ C d 1 1 γ 1 . \|\rho\|_{L^s(\Gamma^c\cap\{\rho\le1\}\cap B_{R_0})}
\le|B_{R_0}|^{\frac{1}{s}-\frac{1}{\gamma_1}}
\|\rho\|_{L^{\gamma_1}(\Gamma^c\cap\{\rho\le1\}\cap B_{R_0})}
\le Cd_1^{\frac{1}{\gamma_1}}. ∥ ρ ∥ L s ( Γ c ∩ { ρ ≤ 1 } ∩ B R 0 ) ≤ ∣ B R 0 ∣ s 1 − γ 1 1 ∥ ρ ∥ L γ 1 ( Γ c ∩ { ρ ≤ 1 } ∩ B R 0 ) ≤ C d 1 γ 1 1 .
Finally, set
A : = Γ c ∩ { ρ ≤ 1 } ∩ B R 0 c . A := \Gamma^c\cap\{\rho\le1\}\cap B_{R_0}^c. A := Γ c ∩ { ρ ≤ 1 } ∩ B R 0 c .
By (3.16) and (4.3),
∥ ρ ∥ L 1 ( A ) ≤ C d 1 , ∥ ρ ∥ L γ 1 ( A ) ≤ C d 1 1 γ 1 . \|\rho\|_{L^1(A)}\le Cd_1,\qquad\|\rho\|_{L^{\gamma_1}(A)}\le Cd_1^{\frac{1}{\gamma_1}}. ∥ ρ ∥ L 1 ( A ) ≤ C d 1 , ∥ ρ ∥ L γ 1 ( A ) ≤ C d 1 γ 1 1 .
For 1 < s < γ 1 1<s<\gamma_1 1 < s < γ 1 , choose θ ∈ ( 0 , 1 ) \theta\in(0,1) θ ∈ ( 0 , 1 ) such that
1 s = θ + 1 − θ γ 1 . \frac{1}{s}=\theta+\frac{1-\theta}{\gamma_1}. s 1 = θ + γ 1 1 − θ .
Interpolation yields
∥ ρ ∥ L s ( A ) ≤ ∥ ρ ∥ L 1 ( A ) θ ∥ ρ ∥ L γ 1 ( A ) 1 − θ ≤ C d 1 θ + 1 − θ γ 1 = C d 1 1 s . \|\rho\|_{L^s(A)}
\le\|\rho\|_{L^1(A)}^\theta\|\rho\|_{L^{\gamma_1}(A)}^{1-\theta}
\le Cd_1^{\theta+\frac{1-\theta}{\gamma_1}}
=Cd_1^{\frac{1}{s}}. ∥ ρ ∥ L s ( A ) ≤ ∥ ρ ∥ L 1 ( A ) θ ∥ ρ ∥ L γ 1 ( A ) 1 − θ ≤ C d 1 θ + γ 1 1 − θ = C d 1 s 1 .
For s = γ 1 s=\gamma_1 s = γ 1 , the same estimate follows directly from (4.3).
Combining the estimates on all regions proves (4.5).
If 0 ≤ d 1 ≤ 1 0\le d_1\le1 0 ≤ d 1 ≤ 1 , then d 1 1 γ 1 ≤ d 1 1 2 d_1^{\frac{1}{\gamma_1}}\le d_1^{\frac{1}{2}} d 1 γ 1 1 ≤ d 1 2 1 and d 1 1 s ≤ d 1 1 2 d_1^{\frac{1}{s}}\le d_1^{\frac{1}{2}} d 1 s 1 ≤ d 1 2 1 , because s ≤ γ 1 ≤ 2 s\le\gamma_1\le2 s ≤ γ 1 ≤ 2 . Therefore, (4.5) reduces to (4.6). □
Convexity of the Rotational Energy
We impose the following condition at the rotation axis:
lim r ↓ 0 max { L ( m ρ ( r ) ) , L ( m ρ ˉ ( r ) ) } ∣ m ρ ( r ) − m ρ ˉ ( r ) ∣ r 2 = 0. (38) \lim_{r\downarrow0}\frac{\max\{L(m_\rho(r)),L(m_{\bar{\rho}}(r))\}|m_\rho(r)-m_{\bar{\rho}}(r)|}{r^2}=0. \tag*{(38)} r ↓ 0 lim r 2 max { L ( m ρ ( r )) , L ( m ρ ˉ ( r ))} ∣ m ρ ( r ) − m ρ ˉ ( r ) ∣ = 0. (38)
This condition ensures that the boundary terms arising from the first variation of the rotational energy vanish at the axis. The following argument is motivated by [42 Lemma 3.1] .
Lemma 4.3 (Positivity of the rotational remainder). Let ρ ˉ \bar{\rho} ρ ˉ be a rotating Euler-Riesz stellar density satisfying Definition 3.1. Suppose that ρ ∈ L + 1 ( R 3 ) \rho\in L_+^1(\mathbb{R}^3) ρ ∈ L + 1 ( R 3 ) is axisymmetric, has the same mass M M M as ρ ˉ \bar{\rho} ρ ˉ , and satisfies
K L ( ρ ) < ∞ . K_L(\rho)<\infty. K L ( ρ ) < ∞. Assume further that (3.3) and (4.7) hold. Then
d 2 ( ρ , ρ ˉ ) ≥ 0. (39) d_2(\rho,\bar{\rho})\ge0. \tag*{(39)} d 2 ( ρ , ρ ˉ ) ≥ 0. (39) :::`
More precisely, setting
f : = ρ − ρ ˉ , ρ λ : = ρ ˉ + λ f , m f : = m ρ − m ρ ˉ , f := \rho-\bar{\rho}, \qquad\rho_\lambda:= \bar{\rho}+\lambda f, \qquad m_f := m_\rho-m_{\bar{\rho}}, f := ρ − ρ ˉ , ρ λ := ρ ˉ + λ f , m f := m ρ − m ρ ˉ ,
then
d 2 ( ρ , ρ ˉ ) = ∫ 0 1 ( 1 − λ ) ∫ 0 ∞ L ′ ( m ρ λ ( r ) ) m f ( r ) 2 r 3 d r d λ . (40) d_2(\rho,\bar{\rho})=\int_0^1(1-\lambda)\int_0^\infty\frac{L'(m_{\rho_\lambda}(r))m_f(r)^2}{r^3}\,dr\,d\lambda. \tag*{(40)} d 2 ( ρ , ρ ˉ ) = ∫ 0 1 ( 1 − λ ) ∫ 0 ∞ r 3 L ′ ( m ρ λ ( r )) m f ( r ) 2 d r d λ . (40)
Proof. We divide the proof into five steps.
1. Cylindrical mass representation: For every nonnegative axisymmetric density ϱ ∈ L 1 ( R 3 ) \varrho\in L^1(\mathbb{R}^3) ϱ ∈ L 1 ( R 3 ) of mass M M M , the cylindrical mass m ϱ m_\varrho m ϱ is absolutely continuous and satisfies
m ϱ ′ ( r ) = 2 π r ∫ R ϱ ( r , z ) d z for a.e. r > 0. (41) m'_\varrho(r)=2\pi r\int_{\mathbb{R}}\varrho(r,z)\,dz \qquad\text{for a.e. }r>0. \tag*{(41)} m ϱ ′ ( r ) = 2 π r ∫ R ϱ ( r , z ) d z for a.e. r > 0. (41)
In particular,
m ϱ ( 0 ) = 0 , 0 ≤ m ϱ ( r ) ≤ M , lim r → ∞ m ϱ ( r ) = M . m_\varrho(0)=0,\qquad0\le m_\varrho(r)\le M,\qquad\lim_{r\to\infty}m_\varrho(r)=M. m ϱ ( 0 ) = 0 , 0 ≤ m ϱ ( r ) ≤ M , r → ∞ lim m ϱ ( r ) = M .
Using cylindrical coordinates and (41) , we obtain
K L ( ϱ ) = 1 2 ∫ 0 ∞ ∫ R ϱ ( r , z ) L ( m ϱ ( r ) ) r 2 2 π r d z d r = 1 2 ∫ 0 ∞ L ( m ϱ ( r ) ) m ϱ ′ ( r ) r 2 d r . \mathcal{K}_L(\varrho)=\frac{1}{2}\int_0^\infty\int_{\mathbb{R}}\varrho(r,z)\frac{L(m_\varrho(r))}{r^2}\,2\pi r\,dz\,dr=\frac{1}{2}\int_0^\infty\frac{L(m_\varrho(r))m'_\varrho(r)}{r^2}\,dr. K L ( ϱ ) = 2 1 ∫ 0 ∞ ∫ R ϱ ( r , z ) r 2 L ( m ϱ ( r )) 2 π r d z d r = 2 1 ∫ 0 ∞ r 2 L ( m ϱ ( r )) m ϱ ′ ( r ) d r .
2. First variation on a truncated interval: By linearity of the cylindrical mass,
m λ : = m ρ λ = ( 1 − λ ) m ρ ˉ + λ m ρ , m_\lambda:=m_{\rho_\lambda}=(1-\lambda)m_{\bar{\rho}}+\lambda m_\rho, m λ := m ρ λ = ( 1 − λ ) m ρ ˉ + λ m ρ ,
and hence
∂ λ m λ = m f , ∂ λ m λ ′ = m f ′ . \partial_\lambda m_\lambda=m_f,\qquad\partial_\lambda m'_\lambda=m'_f. ∂ λ m λ = m f , ∂ λ m λ ′ = m f ′ .
Fix 0 < ε < R < ∞ 0<\varepsilon<R<\infty 0 < ε < R < ∞ and define
K ε , R ( λ ) : = 1 2 ∫ ε R L ( m λ ( r ) ) m λ ′ ( r ) r 2 d r . \mathcal{K}_{\varepsilon,R}(\lambda):=\frac{1}{2}\int_\varepsilon^R\frac{L(m_\lambda(r))m'_\lambda(r)}{r^2}\,dr. K ε , R ( λ ) := 2 1 ∫ ε R r 2 L ( m λ ( r )) m λ ′ ( r ) d r .
Since
∣ m f ∣ ≤ M , ∣ m f ′ ∣ ≤ m ρ ′ + m ρ ˉ ′ , 0 ≤ m λ ′ ≤ m ρ ′ + m ρ ˉ ′ , |m_f|\le M,\qquad|m'_f|\le m'_\rho+m'_{\bar{\rho}},\qquad0\le m'_\lambda\le m'_\rho+m'_{\bar{\rho}}, ∣ m f ∣ ≤ M , ∣ m f ′ ∣ ≤ m ρ ′ + m ρ ˉ ′ , 0 ≤ m λ ′ ≤ m ρ ′ + m ρ ˉ ′ ,
and L , L ′ L,L' L , L ′ are bounded on [ 0 , M ] [0,M] [ 0 , M ] , differentiation under the integral sign is justified on [ ε , R ] [\varepsilon,R] [ ε , R ] .
Moreover,
∂ λ ( L ( m λ ) m λ ′ ) = L ′ ( m λ ) m f m λ ′ + L ( m λ ) m f ′ = d d r ( L ( m λ ) m f ) . \partial_\lambda\bigl(L(m_\lambda)m'_\lambda\bigr)=L'(m_\lambda)m_fm'_\lambda+L(m_\lambda)m'_f=\frac{d}{dr}\bigl(L(m_\lambda)m_f\bigr). ∂ λ ( L ( m λ ) m λ ′ ) = L ′ ( m λ ) m f m λ ′ + L ( m λ ) m f ′ = d r d ( L ( m λ ) m f ) .
Integrating first in λ \lambda λ and then by parts in r r r , we obtain
K ε , R ( 1 ) − K ε , R ( 0 ) = 1 2 ∫ 0 1 [ L ( m λ ( r ) ) m f ( r ) r 2 ] r = ε r = R d λ + ∫ 0 1 ∫ ε R L ( m λ ( r ) ) m f ( r ) r 3 d r d λ . (42) \mathcal{K}_{\varepsilon,R}(1)-\mathcal{K}_{\varepsilon,R}(0)=\frac{1}{2}\int_0^1\left[\frac{L(m_\lambda(r))m_f(r)}{r^2}\right]_{r=\varepsilon}^{r=R}\,d\lambda+\int_0^1\int_\varepsilon^R\frac{L(m_\lambda(r))m_f(r)}{r^3}\,dr\,d\lambda. \tag*{(42)} K ε , R ( 1 ) − K ε , R ( 0 ) = 2 1 ∫ 0 1 [ r 2 L ( m λ ( r )) m f ( r ) ] r = ε r = R d λ + ∫ 0 1 ∫ ε R r 3 L ( m λ ( r )) m f ( r ) d r d λ . (42)
3. Vanishing of the boundary terms: Since the integrands defining the rotational energies are nonnegative, monotone convergence gives
K ε , R ( 1 ) ⟶ K L ( ρ ) , K ε , R ( 0 ) ⟶ K L ( ρ ˉ ) as ε → 0 and R → ∞ . \mathcal{K}_{\varepsilon,R}(1)\longrightarrow\mathcal{K}_L(\rho),\qquad\mathcal{K}_{\varepsilon,R}(0)\longrightarrow\mathcal{K}_L(\bar{\rho})\qquad\text{as }\varepsilon\to0\text{ and }R\to\infty. K ε , R ( 1 ) ⟶ K L ( ρ ) , K ε , R ( 0 ) ⟶ K L ( ρ ˉ ) as ε → 0 and R → ∞.
At infinity, the bounds 0 ≤ m λ ( R ) ≤ M 0\le m_\lambda(R)\le M 0 ≤ m λ ( R ) ≤ M and ∣ m f ( R ) ∣ ≤ M |m_f(R)|\le M ∣ m f ( R ) ∣ ≤ M imply
sup λ ∈ [ 0 , 1 ] ∣ L ( m λ ( R ) ) m f ( R ) R 2 ∣ ≤ M ∥ L ∥ L ∞ ( 0 , M ) R 2 ⟶ 0. \sup_{\lambda\in[0,1]}\left|\frac{L(m_\lambda(R))m_f(R)}{R^2}\right|\le\frac{M\lVert L\rVert_{L^\infty(0,M)}}{R^2}\longrightarrow0. λ ∈ [ 0 , 1 ] sup R 2 L ( m λ ( R )) m f ( R ) ≤ R 2 M ∥ L ∥ L ∞ ( 0 , M ) ⟶ 0.
At the rotation axis, m λ ( ε ) m_\lambda(\varepsilon) m λ ( ε ) lies between m ρ ( ε ) m_\rho(\varepsilon) m ρ ( ε ) and m ρ ˉ ( ε ) m_{\bar{\rho}}(\varepsilon) m ρ ˉ ( ε ) . Since L L L is nonnegative and nondecreasing,
L ( m λ ( ε ) ) ≤ max { L ( m ρ ( ε ) ) , L ( m ρ ˉ ( ε ) ) } . L(m_\lambda(\varepsilon))\le\max\{L(m_\rho(\varepsilon)),L(m_{\bar{\rho}}(\varepsilon))\}. L ( m λ ( ε )) ≤ max { L ( m ρ ( ε )) , L ( m ρ ˉ ( ε ))} .
Consequently, (38) gives
sup λ ∈ [ 0 , 1 ] ∣ L ( m λ ( ε ) ) m f ( ε ) ε 2 ∣ ≤ max { L ( m ρ ( ε ) ) , L ( m ρ ˉ ( ε ) ) } ∣ m f ( ε ) ∣ ε 2 ⟶ 0. \sup_{\lambda\in[0,1]}\left|\frac{L(m_\lambda(\varepsilon))m_f(\varepsilon)}{\varepsilon^2}\right|\le\frac{\max\{L(m_\rho(\varepsilon)),L(m_{\bar{\rho}}(\varepsilon))\}|m_f(\varepsilon)|}{\varepsilon^2}\longrightarrow0. λ ∈ [ 0 , 1 ] sup ε 2 L ( m λ ( ε )) m f ( ε ) ≤ ε 2 max { L ( m ρ ( ε )) , L ( m ρ ˉ ( ε ))} ∣ m f ( ε ) ∣ ⟶ 0.
Thus the boundary contribution in (42) vanishes in the limit.
4. Identification of the first variation: By (41) ,
∣ m f ′ ( r ) ∣ ≤ 2 π r ∫ R ∣ f ( r , z ) ∣ d z . \lvert m_f'(r)\rvert\le2\pi r \int_{\mathbb{R}} \lvert f(r,z)\rvert\, \mathrm{d}z. ∣ m f ′ ( r )∣ ≤ 2 π r ∫ R ∣ f ( r , z )∣ d z .
Therefore, Lemma 3.2 yields
∫ 0 ∞ ∣ m f ′ ( r ) ∣ ∣ Ψ ‾ ( r ) ∣ d r ≤ ∫ R 3 ∣ f ( x ) ∣ ∣ Ψ ‾ ( r ( x ) ) ∣ d x < ∞ . (43) \int_0^\infty\lvert m_f'(r)\rvert\lvert\overline{\Psi}(r)\rvert\,\mathrm{d}r \le\int_{\mathbb{R}^3} \lvert f(x)\rvert\lvert\overline{\Psi}(r(x))\rvert\,\mathrm{d}x < \infty. \tag*{(43)} ∫ 0 ∞ ∣ m f ′ ( r )∣ ∣ Ψ ( r )∣ d r ≤ ∫ R 3 ∣ f ( x )∣ ∣ Ψ ( r ( x ))∣ d x < ∞. (43)
Set
A ( s ) : = L ( m ρ ‾ ( s ) ) s 3 , Ψ ‾ ( r ) = ∫ r ∞ A ( s ) d s . A(s) := \frac{L(m_{\overline{\rho}}(s))}{s^3}, \qquad\overline{\Psi}(r) = \int_r^\infty A(s)\,\mathrm{d}s. A ( s ) := s 3 L ( m ρ ( s )) , Ψ ( r ) = ∫ r ∞ A ( s ) d s .
Since m f ( 0 ) = 0 m_f(0)=0 m f ( 0 ) = 0 and m f m_f m f is absolutely continuous,
∣ m f ( s ) ∣ ≤ ∫ 0 s ∣ m f ′ ( r ) ∣ d r . \lvert m_f(s)\rvert\le\int_0^s \lvert m_f'(r)\rvert\,\mathrm{d}r. ∣ m f ( s )∣ ≤ ∫ 0 s ∣ m f ′ ( r )∣ d r .
Tonelli’s theorem and (43) therefore give
∫ 0 ∞ A ( s ) ∣ m f ( s ) ∣ d s ≤ ∫ 0 ∞ A ( s ) ∫ 0 s ∣ m f ′ ( r ) ∣ d r d s = ∫ 0 ∞ ∣ m f ′ ( r ) ∣ ∫ r ∞ A ( s ) d s d r < ∞ . \int_0^\infty A(s)\lvert m_f(s)\rvert\,\mathrm{d}s \le\int_0^\infty A(s)\int_0^s \lvert m_f'(r)\rvert\,\mathrm{d}r\,\mathrm{d}s = \int_0^\infty\lvert m_f'(r)\rvert\int_r^\infty A(s)\,\mathrm{d}s\,\mathrm{d}r < \infty. ∫ 0 ∞ A ( s ) ∣ m f ( s )∣ d s ≤ ∫ 0 ∞ A ( s ) ∫ 0 s ∣ m f ′ ( r )∣ d r d s = ∫ 0 ∞ ∣ m f ′ ( r )∣ ∫ r ∞ A ( s ) d s d r < ∞.
Fubini’s theorem is thus applicable, and we obtain
∫ R 3 f ( x ) Ψ ‾ ( r ( x ) ) d x = ∫ 0 ∞ m f ′ ( r ) Ψ ‾ ( r ) d r = ∫ 0 ∞ L ( m ρ ‾ ( r ) ) m f ( r ) r 3 d r . (44) \int_{\mathbb{R}^3} f(x)\overline{\Psi}(r(x))\,\mathrm{d}x = \int_0^\infty m_f'(r)\overline{\Psi}(r)\,\mathrm{d}r = \int_0^\infty\frac{L(m_{\overline{\rho}}(r))m_f(r)}{r^3}\,\mathrm{d}r. \tag*{(44)} ∫ R 3 f ( x ) Ψ ( r ( x )) d x = ∫ 0 ∞ m f ′ ( r ) Ψ ( r ) d r = ∫ 0 ∞ r 3 L ( m ρ ( r )) m f ( r ) d r . (44)
5. Convexity and the exact remainder: Subtracting the truncated integral corresponding to (44) from (42) , we obtain
K ε , R ( 1 ) − K ε , R ( 0 ) − ∫ ε R L ( m ρ ‾ ( r ) ) m f ( r ) r 3 d r = 1 2 ∫ 0 1 [ L ( m λ ( r ) ) m f ( r ) r 2 ] r = ε r = R d λ + ∫ 0 1 ∫ ε R ( L ( m λ ( r ) ) − L ( m ρ ‾ ( r ) ) ) m f ( r ) r 3 d r d λ . (45) \begin{aligned}
K_{\varepsilon,R}(1)-K_{\varepsilon,R}(0)-\int_\varepsilon^R \frac{L(m_{\overline{\rho}}(r))m_f(r)}{r^3}\,\mathrm{d}r
&= \frac{1}{2}\int_0^1 \left[\frac{L(m_\lambda(r))m_f(r)}{r^2}\right]_{r=\varepsilon}^{r=R}\,\mathrm{d}\lambda\\
&\quad+ \int_0^1\int_\varepsilon^R \frac{\left(L(m_\lambda(r))-L(m_{\overline{\rho}}(r))\right)m_f(r)}{r^3}\,\mathrm{d}r\,\mathrm{d}\lambda. \tag*{(45)}
\end{aligned} K ε , R ( 1 ) − K ε , R ( 0 ) − ∫ ε R r 3 L ( m ρ ( r )) m f ( r ) d r = 2 1 ∫ 0 1 [ r 2 L ( m λ ( r )) m f ( r ) ] r = ε r = R d λ + ∫ 0 1 ∫ ε R r 3 ( L ( m λ ( r )) − L ( m ρ ( r )) ) m f ( r ) d r d λ . (45)
Since m λ − m ρ ‾ = λ m f m_\lambda-m_{\overline{\rho}}=\lambda m_f m λ − m ρ = λ m f and L L L is nondecreasing,
( L ( m λ ) − L ( m ρ ‾ ) ) m f ≥ 0. (46) \left(L(m_\lambda)-L(m_{\overline{\rho}})\right)m_f \ge0. \tag*{(46)} ( L ( m λ ) − L ( m ρ ) ) m f ≥ 0. (46)
The left-hand side of (45) converges to d 2 ( ρ , ρ ‾ ) d_2(\rho,\overline{\rho}) d 2 ( ρ , ρ ) by Steps 3–4, while the boundary term vanishes by Step 3. The monotone convergence theorem therefore yields
d 2 ( ρ , ρ ‾ ) = ∫ 0 1 ∫ 0 ∞ ( L ( m λ ( r ) ) − L ( m ρ ‾ ( r ) ) ) m f ( r ) r 3 d r d λ . (47) d_2(\rho,\overline{\rho})=\int_0^1\int_0^\infty\frac{\left(L(m_\lambda(r))-L(m_{\overline{\rho}}(r))\right)m_f(r)}{r^3}\,\mathrm{d}r\,\mathrm{d}\lambda. \tag*{(47)} d 2 ( ρ , ρ ) = ∫ 0 1 ∫ 0 ∞ r 3 ( L ( m λ ( r )) − L ( m ρ ( r )) ) m f ( r ) d r d λ . (47)
Introduce the primitive
A ( q ) : = ∫ 0 q L ( s ) d s . \mathcal{A}(q):=\int_0^q L(s)\,\mathrm{d}s. A ( q ) := ∫ 0 q L ( s ) d s .
Then
A ′ ( q ) = L ( q ) , A ′ ′ ( q ) = L ′ ( q ) ≥ 0 , \mathcal{A}'(q)=L(q), \qquad\mathcal{A}''(q)=L'(q)\ge0, A ′ ( q ) = L ( q ) , A ′′ ( q ) = L ′ ( q ) ≥ 0 ,
so A \mathcal{A} A is convex on [ 0 , M ] [0,M] [ 0 , M ] .
For each fixed r > 0 r>0 r > 0 , the fundamental theorem of calculus gives
∫ 0 1 ( L ( m λ ) − L ( m ρ ‾ ) ) m f d λ = A ( m ρ ) − A ( m ρ ‾ ) − L ( m ρ ‾ ) m f . \int_0^1 \left(L(m_\lambda)-L(m_{\overline{\rho}})\right)m_f\,\mathrm{d}\lambda= \mathcal{A}(m_\rho)-\mathcal{A}(m_{\overline{\rho}})-L(m_{\overline{\rho}})m_f. ∫ 0 1 ( L ( m λ ) − L ( m ρ ) ) m f d λ = A ( m ρ ) − A ( m ρ ) − L ( m ρ ) m f .
Consequently,
d 2 ( ρ , ρ ‾ ) = ∫ 0 ∞ 1 r 3 ( A ( m ρ ( r ) ) − A ( m ρ ‾ ( r ) ) − A ′ ( m ρ ‾ ( r ) ) m f ( r ) ) d r . d_2(\rho,\overline{\rho})=\int_0^\infty\frac{1}{r^3}\left(\mathcal{A}(m_\rho(r))-\mathcal{A}(m_{\overline{\rho}}(r))-\mathcal{A}'(m_{\overline{\rho}}(r))m_f(r)\right)\,\mathrm{d}r. d 2 ( ρ , ρ ) = ∫ 0 ∞ r 3 1 ( A ( m ρ ( r )) − A ( m ρ ( r )) − A ′ ( m ρ ( r )) m f ( r ) ) d r .
The integrand is nonnegative by convexity of A \mathcal{A} A , proving (39) .
Finally,
L ( m λ ) − L ( m ρ ˉ ) = ∫ 0 λ L ′ ( m σ ) m f d σ . L(m_\lambda)-L(m_{\bar{\rho}})=\int_{0}^{\lambda}L'(m_\sigma)m_f\,d\sigma. L ( m λ ) − L ( m ρ ˉ ) = ∫ 0 λ L ′ ( m σ ) m f d σ .
Substituting into (4.16) and applying Tonelli’s theorem gives
d 2 ( ρ , ρ ˉ ) = ∫ 0 1 ∫ 0 ∞ ∫ 0 λ L ′ ( m σ ( r ) ) m f ( r ) 2 r 3 d σ d r d λ = ∫ 0 1 ( 1 − σ ) ∫ 0 ∞ L ′ ( m σ ( r ) ) m f ( r ) 2 r 3 d r d σ . d_2(\rho,\bar{\rho})=\int_{0}^{1}\int_{0}^{\infty}\int_{0}^{\lambda}\frac{L'(m_\sigma(r))m_f(r)^2}{r^3}\,d\sigma\,dr\,d\lambda
=\int_{0}^{1}(1-\sigma)\int_{0}^{\infty}\frac{L'(m_\sigma(r))m_f(r)^2}{r^3}\,dr\,d\sigma. d 2 ( ρ , ρ ˉ ) = ∫ 0 1 ∫ 0 ∞ ∫ 0 λ r 3 L ′ ( m σ ( r )) m f ( r ) 2 d σ d r d λ = ∫ 0 1 ( 1 − σ ) ∫ 0 ∞ r 3 L ′ ( m σ ( r )) m f ( r ) 2 d r d σ .
Relabeling σ \sigma σ as λ \lambda λ establishes (4.9).
Modulated Entropy and Rotational Cancellation
We now introduce the modulated entropy associated with a rotating Euler-Riesz equilibrium.
Definition 4.4 (Rotating relative entropy). For an axisymmetric state ( ρ , M ) (\rho,\mathbf{M}) ( ρ , M ) satisfying the angular momentum compatibility condition (3.12), define
E L ( t ) : = 1 2 ∫ R 3 ( ∣ M r ∣ 2 + ∣ M z ∣ 2 ρ ) ( t , x ) d x + d 1 ( ρ ( t ) , ρ ˉ ) + d 2 ( ρ ( t ) , ρ ˉ ) . (48) \mathcal{E}_L(t):=\frac{1}{2}\int_{\mathbb{R}^3}\left(\frac{|\mathbf{M}^r|^2+|\mathbf{M}^z|^2}{\rho}\right)(t,x)\,dx+d_1(\rho(t),\bar{\rho})+d_2(\rho(t),\bar{\rho}). \tag*{(48)} E L ( t ) := 2 1 ∫ R 3 ( ρ ∣ M r ∣ 2 + ∣ M z ∣ 2 ) ( t , x ) d x + d 1 ( ρ ( t ) , ρ ˉ ) + d 2 ( ρ ( t ) , ρ ˉ ) . (48)
Under the hypotheses of Lemmas 4.1 and 4.3, including the axis condition (4.7), we have
E L ( t ) ≥ 0. \mathcal{E}_L(t)\ge0. E L ( t ) ≥ 0.
The following proposition establishes an exact cancellation of the centrifugal potential terms.
Proposition 4.5 (Rotational cancellation). Let ( ρ , M ) (\rho,\mathbf{M}) ( ρ , M ) be an axisymmetric finite-energy state satisfying (3.12). Assume that d 1 ( ρ , ρ ˉ ) d_1(\rho,\bar{\rho}) d 1 ( ρ , ρ ˉ ) and d 2 ( ρ , ρ ˉ ) d_2(\rho,\bar{\rho}) d 2 ( ρ , ρ ˉ ) are finite. Then all terms in the following identity are well defined and finite:
E L ( t ) = ∫ R 3 ( ∣ M ∣ 2 2 ρ + h ( ρ ) − h ( ρ ˉ ) + G ˉ ( ρ − ρ ˉ ) ) ( t , x ) d x − K L ( ρ ˉ ) . (49) \mathcal{E}_L(t)=\int_{\mathbb{R}^3}\left(\frac{|\mathbf{M}|^2}{2\rho}+h(\rho)-h(\bar{\rho})+\bar{G}(\rho-\bar{\rho})\right)(t,x)\,dx-\mathcal{K}_L(\bar{\rho}). \tag*{(49)} E L ( t ) = ∫ R 3 ( 2 ρ ∣ M ∣ 2 + h ( ρ ) − h ( ρ ˉ ) + G ˉ ( ρ − ρ ˉ ) ) ( t , x ) d x − K L ( ρ ˉ ) . (49) Proof. By angular momentum compatibility and (3.13),
1 2 ∫ R 3 ∣ M r ∣ 2 + ∣ M z ∣ 2 ρ d x + K L ( ρ ) = 1 2 ∫ R 3 ∣ M ∣ 2 ρ d x . \frac{1}{2}\int_{\mathbb{R}^3}\frac{|\mathbf{M}^r|^2+|\mathbf{M}^z|^2}{\rho}\,dx+\mathcal{K}_L(\rho)=\frac{1}{2}\int_{\mathbb{R}^3}\frac{|\mathbf{M}|^2}{\rho}\,dx. 2 1 ∫ R 3 ρ ∣ M r ∣ 2 + ∣ M z ∣ 2 d x + K L ( ρ ) = 2 1 ∫ R 3 ρ ∣ M ∣ 2 d x . On the other hand, adding (4.1) and (4.2) gives
d 1 ( ρ , ρ ˉ ) + d 2 ( ρ , ρ ˉ ) = ∫ R 3 ( h ( ρ ) − h ( ρ ˉ ) + ( V ˉ − Ψ ˉ ) ( ρ − ρ ˉ ) ) d x + K L ( ρ ) − K L ( ρ ˉ ) . d_1(\rho,\bar{\rho})+d_2(\rho,\bar{\rho})=\int_{\mathbb{R}^3}\left(h(\rho)-h(\bar{\rho})+(\bar{V}-\bar{\Psi})(\rho-\bar{\rho})\right)\,dx+\mathcal{K}_L(\rho)-\mathcal{K}_L(\bar{\rho}). d 1 ( ρ , ρ ˉ ) + d 2 ( ρ , ρ ˉ ) = ∫ R 3 ( h ( ρ ) − h ( ρ ˉ ) + ( V ˉ − Ψ ˉ ) ( ρ − ρ ˉ ) ) d x + K L ( ρ ) − K L ( ρ ˉ ) . Since V ˉ − Ψ ˉ = G ˉ \bar{V}-\bar{\Psi}=\bar{G} V ˉ − Ψ ˉ = G ˉ , the two linear terms involving Ψ ˉ \bar{\Psi} Ψ ˉ cancel exactly. Combining the preceding identities yields (4.18).
Weighted Estimates for the Riesz Potential
We first show that the gradient of the Riesz potential generated by an axisymmetric density perturbation has no azimuthal component. This observation is important because the rotational relative entropy controls only the radial and axial components of momentum.
Lemma 5.1 (Meridional structure of the Riesz force). Let α ∈ ( 0 , 2 ) \alpha\in(0,2) α ∈ ( 0 , 2 ) , and let f ∈ L s ( R 3 ) f\in L^s(\mathbb{R}^3) f ∈ L s ( R 3 ) be axisymmetric for some
1 < s < 3 2 − α . 1<s<\frac{3}{2-\alpha}. 1 < s < 2 − α 3 . Then F : = ∇ Φ α ∗ f F:=\nabla\Phi_\alpha*f F := ∇ Φ α ∗ f is well defined almost everywhere and satisfies
F = F r ( r , z ) e r + F z ( r , z ) e z a.e. in R 3 . (50) F=F^r(r,z)e_r+F^z(r,z)e_z\qquad\text{a.e. in }\mathbb{R}^3. \tag*{(50)} F = F r ( r , z ) e r + F z ( r , z ) e z a.e. in R 3 . (50) Consequently, for any axisymmetric momentum field M \mathbf{M} M ,
M ⋅ F = ( M r e r + M z e z ) ⋅ F a.e. (51) \mathbf{M}\cdot F=(M^r e_r+M^z e_z)\cdot F\qquad\text{a.e.} \tag*{(51)} M ⋅ F = ( M r e r + M z e z ) ⋅ F a.e. (51)
Proof. The Hardy-Littlewood-Sobolev inequality for the gradient Riesz potential gives
∥ ∇ Φ α ∗ f ∥ L q ( R 3 ) ≤ C α , p ∥ f ∥ L p ( R 3 ) with 1 q = 1 p − 2 − α 3 , (52) \lVert\nabla\Phi_\alpha* f\rVert_{L^q(\mathbb{R}^3)} \le C_{\alpha,p}\lVert f\rVert_{L^p(\mathbb{R}^3)} \qquad\text{with } \frac{1}{q}=\frac{1}{p}-\frac{2-\alpha}{3}, \tag*{(52)} ∥ ∇ Φ α ∗ f ∥ L q ( R 3 ) ≤ C α , p ∥ f ∥ L p ( R 3 ) with q 1 = p 1 − 3 2 − α , (52)
for 1 < p < 3 2 − α 1<p<\frac{3}{2-\alpha} 1 < p < 2 − α 3 .
We first assume that f ∈ C c ∞ ( R 3 ) f\in C^\infty_c(\mathbb{R}^3) f ∈ C c ∞ ( R 3 ) is axisymmetric. Since Φ α \Phi_\alpha Φ α is radial, for every rotation R R R about the z z z -axis,
( Φ α ∗ f ) ( R x ) = ( Φ α ∗ f ) ( x ) . (\Phi_\alpha*f)(Rx)=(\Phi_\alpha*f)(x). ( Φ α ∗ f ) ( R x ) = ( Φ α ∗ f ) ( x ) .
Differentiating gives
F ( R x ) = R F ( x ) . F(Rx)=RF(x). F ( R x ) = R F ( x ) .
Moreover, the potential is independent of the azimuthal angle, so
F ⋅ e θ = 0 F\cdot e_\theta=0 F ⋅ e θ = 0
away from the rotation axis.
For general axisymmetric f ∈ L s ( R 3 ) f\in L^s(\mathbb{R}^3) f ∈ L s ( R 3 ) , choose a radial mollifier η \eta η and radial cutoffs χ n ∈ C c ∞ ( R 3 ) \chi_n\in C^\infty_c(\mathbb{R}^3) χ n ∈ C c ∞ ( R 3 ) such that 0 ≤ χ n ≤ 1 0\le\chi_n\le1 0 ≤ χ n ≤ 1 and χ n → 1 \chi_n\to1 χ n → 1 pointwise. Set
f n : = χ n ( η 1 n ∗ f ) . f_n:=\chi_n\left(\eta_{\frac{1}{n}}*f\right). f n := χ n ( η n 1 ∗ f ) .
Then f n ∈ C c ∞ ( R 3 ) f_n\in C^\infty_c(\mathbb{R}^3) f n ∈ C c ∞ ( R 3 ) is axisymmetric and f n → f f_n\to f f n → f in L s ( R 3 ) L^s(\mathbb{R}^3) L s ( R 3 ) . Let
F n : = ∇ Φ α ∗ f n , 1 q = 1 s − 2 − α 3 . F_n:=\nabla\Phi_\alpha*f_n,\qquad\frac{1}{q}=\frac{1}{s}-\frac{2-\alpha}{3}. F n := ∇ Φ α ∗ f n , q 1 = s 1 − 3 2 − α .
By (5.3),
∥ F n − F ∥ L q ≤ C α , s ∥ f n − f ∥ L s ⟶ 0. \lVert F_n-F\rVert_{L^q}\le C_{\alpha,s}\lVert f_n-f\rVert_{L^s}\longrightarrow0. ∥ F n − F ∥ L q ≤ C α , s ∥ f n − f ∥ L s ⟶ 0.
Since F n ⋅ e θ = 0 F_n\cdot e_\theta=0 F n ⋅ e θ = 0 , we have
∥ F ⋅ e θ ∥ L q ≤ ∥ F − F n ∥ L q ⟶ 0. \lVert F\cdot e_\theta\rVert_{L^q}\le\lVert F-F_n\rVert_{L^q}\longrightarrow0. ∥ F ⋅ e θ ∥ L q ≤ ∥ F − F n ∥ L q ⟶ 0.
Thus, F F F has no azimuthal component.
Furthermore, for every fixed rotation R R R about the z z z -axis,
∥ F ( R ⋅ ) − R F ∥ L q ≤ ∥ F ( R ⋅ ) − F n ( R ⋅ ) ∥ L q + ∥ R ( F n − F ) ∥ L q = 2 ∥ F n − F ∥ L q ⟶ 0. \lVert F(R\cdot)-RF\rVert_{L^q}\le\lVert F(R\cdot)-F_n(R\cdot)\rVert_{L^q}+\lVert R(F_n-F)\rVert_{L^q}=2\lVert F_n-F\rVert_{L^q}\longrightarrow0. ∥ F ( R ⋅ ) − R F ∥ L q ≤ ∥ F ( R ⋅ ) − F n ( R ⋅ ) ∥ L q + ∥ R ( F n − F ) ∥ L q = 2 ∥ F n − F ∥ L q ⟶ 0.
Hence, F ( R x ) = R F ( x ) F(Rx)=RF(x) F ( R x ) = R F ( x ) almost everywhere. It follows that F r F^r F r and F z F^z F z depend only on ( r , z ) (r,z) ( r , z ) , proving (5.1). Finally, (5.2) follows from (3.11). □
We next establish the weighted Riesz potential estimates needed to control the nonlocal source term in the relative entropy inequality. Two regimes arise: the first requires only finite-energy control, while the second uses an additional high-integrability bound on the density and includes the Chandrasekhar pressure law for rotating white dwarfs.
Proposition 5.2 (Weighted Riesz potential estimates). Let ρ ˉ \bar{\rho} ρ ˉ be a rotating Euler-Riesz stellar density satisfying Definition 3.1, and let ρ ∈ L + 1 ( R 3 ) \rho\in L^1_+(\mathbb{R}^3) ρ ∈ L + 1 ( R 3 ) be axisymmetric with the same mass M M M . Assume that
d 1 : = d 1 ( ρ , ρ ˉ ) < ∞ , f : = ρ − ρ ˉ . d_1:=d_1(\rho,\bar{\rho})<\infty,\qquad f:=\rho-\bar{\rho}. d 1 := d 1 ( ρ , ρ ˉ ) < ∞ , f := ρ − ρ ˉ . Suppose that one of the following conditions holds:
Case 1 (finite-energy regime):
γ ≥ 9 7 − 2 α . (53) \gamma\ge\frac{9}{7-2\alpha}. \tag*{(53)} γ ≥ 7 − 2 α 9 . (53) Case 2 (high-integrability regime):
γ ≥ max { 1 , 6 7 − 2 α } , ρ ∈ L p ˉ ( R 3 ) for some p ˉ > 3 2 − α . (54) \gamma\ge\max\left\{1,\frac{6}{7-2\alpha}\right\},\qquad\rho\in L^{\bar p}(\mathbb{R}^3)\quad\text{for some }\bar p>\frac{3}{2-\alpha}. \tag*{(54)} γ ≥ max { 1 , 7 − 2 α 6 } , ρ ∈ L p ˉ ( R 3 ) for some p ˉ > 2 − α 3 . (54) Then, in Case 1,
∫ R 3 ρ ∣ ∇ Φ α ∗ f ∣ 2 d x ≤ C ( d 1 + d 1 χ ) , χ : = 7 − 2 α 3 ≥ 1. (55) \int_{\mathbb{R}^3}\rho\lvert\nabla\Phi_\alpha*f\rvert^2\,dx\le C(d_1+d_1^\chi),\qquad\chi:=\frac{7-2\alpha}{3}\ge1. \tag*{(55)} ∫ R 3 ρ ∣ ∇ Φ α ∗ f ∣ 2 d x ≤ C ( d 1 + d 1 χ ) , χ := 3 7 − 2 α ≥ 1. (55)
In Case 2,
∫ R 3 ρ ∣ ∇ Φ α ∗ f ∣ 2 d x ≤ C ( 1 + ∥ ρ ∥ L p ˉ 2 ) d 1 . (56) \int_{\mathbb{R}^{3}} \rho\lvert\nabla\Phi_{\alpha} * f\rvert^{2}\,dx \le C\left(1+\lVert\rho\rVert_{L^{\bar p}}^{2}\right)d_{1}. \tag*{(56)} ∫ R 3 ρ ∣ ∇ Φ α ∗ f ∣ 2 d x ≤ C ( 1 + ∥ ρ ∥ L p ˉ 2 ) d 1 . (56)
Proof. We may assume d 1 > 0 d_{1}>0 d 1 > 0 , since d 1 = 0 d_{1}=0 d 1 = 0 implies f = 0 f=0 f = 0 almost everywhere by Lemma 4.1.
Case 1. Set
p ∗ : = 9 7 − 2 α , q ∗ : = 9 1 + α . p_{*}:=\frac{9}{7-2\alpha}, \qquad q_{*}:=\frac{9}{1+\alpha}. p ∗ := 7 − 2 α 9 , q ∗ := 1 + α 9 .
Then
1 q ∗ = 1 p ∗ − 2 − α 3 , 1 p ∗ + 2 q ∗ = 1. \frac{1}{q_{*}}=\frac{1}{p_{*}}-\frac{2-\alpha}{3}, \qquad\frac{1}{p_{*}}+\frac{2}{q_{*}}=1. q ∗ 1 = p ∗ 1 − 3 2 − α , p ∗ 1 + q ∗ 2 = 1.
By (5.4), p ∗ ≤ γ p_{*}\le\gamma p ∗ ≤ γ . Choose p b p_{b} p b such that max { 1 , 6 7 − 2 α } < p b < p ∗ \max\left\{1,\frac{6}{7-2\alpha}\right\}<p_{b}<p_{*} max { 1 , 7 − 2 α 6 } < p b < p ∗ , and define
1 q b : = 1 p b − 2 − α 3 . \frac{1}{q_{b}}:=\frac{1}{p_{b}}-\frac{2-\alpha}{3}. q b 1 := p b 1 − 3 2 − α .
Then 1 < p b < p ∗ ≤ γ 1<p_{b}<p_{*}\le\gamma 1 < p b < p ∗ ≤ γ and q b > 2 q_{b}>2 q b > 2 .
Since ρ ˉ \bar\rho ρ ˉ is bounded and supported in the bounded set Γ \Gamma Γ , Hölder’s inequality and (5.3) yield
∫ R 3 ρ ˉ ∣ ∇ Φ α ∗ f ∣ 2 d x ≤ ∥ ρ ˉ ∥ L ∞ ∣ Γ ∣ 1 − 2 / q b ∥ ∇ Φ α ∗ f ∥ L q b 2 ≤ C ∥ f ∥ L p b 2 . \int_{\mathbb{R}^{3}}\bar\rho\lvert\nabla\Phi_{\alpha}*f\rvert^{2}\,dx\le\lVert\bar\rho\rVert_{L^{\infty}}\lvert\Gamma\rvert^{1-2/q_{b}}\lVert\nabla\Phi_{\alpha}*f\rVert_{L^{q_{b}}}^{2}\le C\lVert f\rVert_{L^{p_{b}}}^{2}. ∫ R 3 ρ ˉ ∣ ∇ Φ α ∗ f ∣ 2 d x ≤ ∥ ρ ˉ ∥ L ∞ ∣ Γ ∣ 1 − 2/ q b ∥ ∇ Φ α ∗ f ∥ L q b 2 ≤ C ∥ f ∥ L p b 2 .
Similarly, using 1 / p ∗ + 2 / q ∗ = 1 1/p_{*}+2/q_{*}=1 1/ p ∗ + 2/ q ∗ = 1 , we obtain
∫ R 3 ∣ f ∣ ∣ ∇ Φ α ∗ f ∣ 2 d x ≤ ∥ f ∥ L p ∗ ∥ ∇ Φ α ∗ f ∥ L q ∗ 2 ≤ C ∥ f ∥ L p ∗ 3 . \int_{\mathbb{R}^{3}}\lvert f\rvert\lvert\nabla\Phi_{\alpha}*f\rvert^{2}\,dx\le\lVert f\rVert_{L^{p_{*}}}\lVert\nabla\Phi_{\alpha}*f\rVert_{L^{q_{*}}}^{2}\le C\lVert f\rVert_{L^{p_{*}}}^{3}. ∫ R 3 ∣ f ∣ ∣ ∇ Φ α ∗ f ∣ 2 d x ≤ ∥ f ∥ L p ∗ ∥ ∇ Φ α ∗ f ∥ L q ∗ 2 ≤ C ∥ f ∥ L p ∗ 3 .
Since ρ ≤ ρ ˉ + ∣ f ∣ \rho\le\bar\rho+\lvert f\rvert ρ ≤ ρ ˉ + ∣ f ∣ ,
∫ R 3 ρ ∣ ∇ Φ α ∗ f ∣ 2 d x ≤ C ( ∥ f ∥ L p b 2 + ∥ f ∥ L p ∗ 3 ) . \int_{\mathbb{R}^{3}}\rho\lvert\nabla\Phi_{\alpha}*f\rvert^{2}\,dx\le C\left(\lVert f\rVert_{L^{p_{b}}}^{2}+\lVert f\rVert_{L^{p_{*}}}^{3}\right). ∫ R 3 ρ ∣ ∇ Φ α ∗ f ∣ 2 d x ≤ C ( ∥ f ∥ L p b 2 + ∥ f ∥ L p ∗ 3 ) .
Applying (4.5) with s = p b , p ∗ s=p_{b},p_{*} s = p b , p ∗ gives
∥ f ∥ L p b 2 ≤ C ( d 1 + d 1 2 / γ 1 + d 1 2 / p b ) , ∥ f ∥ L p ∗ 3 ≤ C ( d 1 2 / γ 1 + d 1 3 / γ 1 + d 1 3 / p ∗ ) . \lVert f\rVert_{L^{p_{b}}}^{2}\le C\left(d_{1}+d_{1}^{2/\gamma_{1}}+d_{1}^{2/p_{b}}\right), \qquad\lVert f\rVert_{L^{p_{*}}}^{3}\le C\left(d_{1}^{2/\gamma_{1}}+d_{1}^{3/\gamma_{1}}+d_{1}^{3/p_{*}}\right). ∥ f ∥ L p b 2 ≤ C ( d 1 + d 1 2/ γ 1 + d 1 2/ p b ) , ∥ f ∥ L p ∗ 3 ≤ C ( d 1 2/ γ 1 + d 1 3/ γ 1 + d 1 3/ p ∗ ) .
Choose p b p_{b} p b sufficiently close to p ∗ p_{*} p ∗ so that p b > 2 p ∗ 3 p_{b}>\frac{2p_{*}}{3} p b > 3 2 p ∗ . Since p ∗ ≤ γ ≤ γ 1 ≤ 2 p_{*}\le\gamma\le\gamma_{1}\le2 p ∗ ≤ γ ≤ γ 1 ≤ 2 , all exponents of d 1 d_{1} d 1 above belong to [ 1 , χ ] [1,\chi] [ 1 , χ ] , where χ = 3 p ∗ = 7 − 2 α 3 \chi=\frac{3}{p_{*}}=\frac{7-2\alpha}{3} χ = p ∗ 3 = 3 7 − 2 α . Using
d 1 β ≤ d 1 + d 1 χ for 1 ≤ β ≤ χ , d_{1}^{\beta}\le d_{1}+d_{1}^{\chi}\qquad\text{for }1\le\beta\le\chi, d 1 β ≤ d 1 + d 1 χ for 1 ≤ β ≤ χ ,
we obtain (5.6).
Case 2. Since ρ ˉ \bar\rho ρ ˉ is bounded and compactly supported,
f ∈ L 1 ( R 3 ) ∩ L p ˉ ( R 3 ) , ∥ f ∥ L 1 ≤ 2 M . f\in L^{1}(\mathbb{R}^{3})\cap L^{\bar p}(\mathbb{R}^{3}),\qquad\lVert f\rVert_{L^{1}}\le2M. f ∈ L 1 ( R 3 ) ∩ L p ˉ ( R 3 ) , ∥ f ∥ L 1 ≤ 2 M .
Let p ˉ ′ \bar p^{\prime} p ˉ ′ denote the conjugate exponent of p ˉ \bar p p ˉ . The condition p ˉ > 3 2 − α \bar p>\frac{3}{2-\alpha} p ˉ > 2 − α 3 implies
( α + 1 ) p ˉ ′ < 3. (\alpha+1)\bar p^{\prime}<3. ( α + 1 ) p ˉ ′ < 3.
Splitting the convolution into ∣ x − y ∣ < 1 \lvert x-y\rvert<1 ∣ x − y ∣ < 1 and its complement, Hölder’s inequality gives
∥ ∇ Φ α ∗ f ∥ L ∞ ≤ C α , p ˉ ( ∥ f ∥ L p ˉ + ∥ f ∥ L 1 ) ≤ B p ˉ , \lVert\nabla\Phi_{\alpha}*f\rVert_{L^{\infty}}\le C_{\alpha,\bar p}\left(\lVert f\rVert_{L^{\bar p}}+\lVert f\rVert_{L^{1}}\right)\le B_{\bar p}, ∥ ∇ Φ α ∗ f ∥ L ∞ ≤ C α , p ˉ ( ∥ f ∥ L p ˉ + ∥ f ∥ L 1 ) ≤ B p ˉ ,
where B p ˉ : = C α , p ˉ ( 2 M + ∥ ρ ˉ ∥ L p ˉ + ∥ ρ ∥ L p ˉ ) B_{\bar p}:=C_{\alpha,\bar p}\left(2M+\lVert\bar\rho\rVert_{L^{\bar p}}+\lVert\rho\rVert_{L^{\bar p}}\right) B p ˉ := C α , p ˉ ( 2 M + ∥ ρ ˉ ∥ L p ˉ + ∥ ρ ∥ L p ˉ ) .
Set
ℓ α : = max { 1 , 6 7 − 2 α } . \ell_{\alpha}:=\max\left\{1,\frac{6}{7-2\alpha}\right\}. ℓ α := max { 1 , 7 − 2 α 6 } .
By (5.5), we may choose
ℓ α ≤ p e ≤ γ , 1 < p e < 3 2 − α . \ell_{\alpha}\le p_{e}\le\gamma,\qquad1<p_{e}<\frac{3}{2-\alpha}. ℓ α ≤ p e ≤ γ , 1 < p e < 2 − α 3 .
sufficiently close to ℓ α \ell_\alpha ℓ α so that, with
1 q 0 : = 1 p e − 2 − α 3 , \frac{1}{q_0}:=\frac{1}{p_e}-\frac{2-\alpha}{3}, q 0 1 := p e 1 − 3 2 − α ,
we have
2 ≤ q 0 < 4. 2\le q_0<4. 2 ≤ q 0 < 4.
The Hardy–Littlewood–Sobolev inequality then gives
∥ ∇ Φ α ∗ f ∥ L q 0 ≤ C ∥ f ∥ L p e . \|\nabla\Phi_\alpha*f\|_{L^{q_0}}\le C\|f\|_{L^{p_e}}. ∥∇ Φ α ∗ f ∥ L q 0 ≤ C ∥ f ∥ L p e .
Suppose first that 0 < d 1 ≤ 1 0<d_1\le1 0 < d 1 ≤ 1 . By (37) ,
∥ ∇ Φ α ∗ f ∥ L q 0 ≤ C d 1 1 2 . \|\nabla\Phi_\alpha*f\|_{L^{q_0}}\le Cd_1^{\frac{1}{2}}. ∥∇ Φ α ∗ f ∥ L q 0 ≤ C d 1 2 1 .
Decompose
∣ f ∣ = g 2 + g 0 + g ∞ , |f|=g_2+g_0+g_\infty, ∣ f ∣ = g 2 + g 0 + g ∞ ,
where g 2 : = ∣ f ∣ 1 Γ ∩ { ∣ f ∣ ≤ 1 } g_2:=|f|\mathbf{1}_{\Gamma\cap\{|f|\le1\}} g 2 := ∣ f ∣ 1 Γ ∩ { ∣ f ∣ ≤ 1 } , g 0 : = ρ 1 Γ c ∩ { ρ ≤ 1 } g_0:=\rho\mathbf{1}_{\Gamma^c\cap\{\rho\le1\}} g 0 := ρ 1 Γ c ∩ { ρ ≤ 1 } , and g ∞ : = ∣ f ∣ 1 Γ ∩ { ∣ f ∣ > 1 } + ρ 1 Γ c ∩ { ρ > 1 } g_\infty:=|f|\mathbf{1}_{\Gamma\cap\{|f|>1\}}+\rho\mathbf{1}_{\Gamma^c\cap\{\rho>1\}} g ∞ := ∣ f ∣ 1 Γ ∩ { ∣ f ∣ > 1 } + ρ 1 Γ c ∩ { ρ > 1 } . Set
s 2 : = 2 , s 0 : = γ 1 , s ∞ : = min { γ 2 , 2 } . s_2:=2,\qquad s_0:=\gamma_1,\qquad s_\infty:=\min\{\gamma_2,2\}. s 2 := 2 , s 0 := γ 1 , s ∞ := min { γ 2 , 2 } .
Since s i ∈ ( 1 , 2 ] s_i\in(1,2] s i ∈ ( 1 , 2 ] , the coercivity estimate (34) gives
∥ g i ∥ L s i s i ≤ C d 1 for i ∈ { 2 , 0 , ∞ } . \|g_i\|_{L^{s_i}}^{s_i}\le Cd_1\qquad\text{for }i\in\{2,0,\infty\}. ∥ g i ∥ L s i s i ≤ C d 1 for i ∈ { 2 , 0 , ∞ } .
For each i i i , define
m i : = 2 s i s i − 1 , θ i : = q 0 m i . m_i:=\frac{2s_i}{s_i-1},\qquad\theta_i:=\frac{q_0}{m_i}. m i := s i − 1 2 s i , θ i := m i q 0 .
Since m i ≥ 4 > q 0 m_i\ge4>q_0 m i ≥ 4 > q 0 , interpolation between L q 0 L^{q_0} L q 0 and L ∞ L^\infty L ∞ yields
∥ ∇ Φ α ∗ f ∥ L m i ≤ ∥ ∇ Φ α ∗ f ∥ L q 0 θ i ∥ ∇ Φ α ∗ f ∥ L ∞ 1 − θ i ≤ C B p ˉ 1 − θ i d 1 θ i 2 . \|\nabla\Phi_\alpha*f\|_{L^{m_i}}
\le\|\nabla\Phi_\alpha*f\|_{L^{q_0}}^{\theta_i}\|\nabla\Phi_\alpha*f\|_{L^\infty}^{1-\theta_i}
\le CB_{\bar p}^{1-\theta_i}d_1^{\frac{\theta_i}{2}}. ∥∇ Φ α ∗ f ∥ L m i ≤ ∥∇ Φ α ∗ f ∥ L q 0 θ i ∥∇ Φ α ∗ f ∥ L ∞ 1 − θ i ≤ C B p ˉ 1 − θ i d 1 2 θ i .
By Hölder’s inequality,
∫ R 3 g i ∣ ∇ Φ α ∗ f ∣ 2 d x ≤ ∥ g i ∥ L s i ∥ ∇ Φ α ∗ f ∥ L m i 2 ≤ C B p ˉ 2 ( 1 − θ i ) d 1 1 s i + θ i . \int_{\mathbb{R}^3}g_i|\nabla\Phi_\alpha*f|^2\,dx
\le\|g_i\|_{L^{s_i}}\|\nabla\Phi_\alpha*f\|_{L^{m_i}}^2
\le CB_{\bar p}^{2(1-\theta_i)}d_1^{\frac{1}{s_i}+\theta_i}. ∫ R 3 g i ∣∇ Φ α ∗ f ∣ 2 d x ≤ ∥ g i ∥ L s i ∥∇ Φ α ∗ f ∥ L m i 2 ≤ C B p ˉ 2 ( 1 − θ i ) d 1 s i 1 + θ i .
Moreover,
1 s i + θ i − 1 = s i − 1 s i ( q 0 2 − 1 ) ≥ 0. \frac{1}{s_i}+\theta_i-1
=\frac{s_i-1}{s_i}\left(\frac{q_0}{2}-1\right)\ge0. s i 1 + θ i − 1 = s i s i − 1 ( 2 q 0 − 1 ) ≥ 0.
Since d 1 ≤ 1 d_1\le1 d 1 ≤ 1 ,
∫ R 3 g i ∣ ∇ Φ α ∗ f ∣ 2 d x ≤ C ( 1 + B p ˉ 2 ) d 1 . \int_{\mathbb{R}^3}g_i|\nabla\Phi_\alpha*f|^2\,dx
\le C(1+B_{\bar p}^2)d_1. ∫ R 3 g i ∣∇ Φ α ∗ f ∣ 2 d x ≤ C ( 1 + B p ˉ 2 ) d 1 .
For the equilibrium density,
∫ R 3 ρ ˉ ∣ ∇ Φ α ∗ f ∣ 2 d x ≤ ∥ ρ ˉ ∥ L ∞ ∣ Γ ∣ 1 − 2 q 0 ∥ ∇ Φ α ∗ f ∥ L q 0 2 ≤ C d 1 . \int_{\mathbb{R}^3}\bar\rho|\nabla\Phi_\alpha*f|^2\,dx
\le\|\bar\rho\|_{L^\infty}|\Gamma|^{1-\frac{2}{q_0}}\|\nabla\Phi_\alpha*f\|_{L^{q_0}}^2
\le Cd_1. ∫ R 3 ρ ˉ ∣∇ Φ α ∗ f ∣ 2 d x ≤ ∥ ρ ˉ ∥ L ∞ ∣Γ ∣ 1 − q 0 2 ∥∇ Φ α ∗ f ∥ L q 0 2 ≤ C d 1 .
Since ρ ≤ ρ ˉ + ∣ f ∣ \rho\le\bar\rho+|f| ρ ≤ ρ ˉ + ∣ f ∣ , summing these estimates gives
∫ R 3 ρ ∣ ∇ Φ α ∗ f ∣ 2 d x ≤ C ( 1 + B p ˉ 2 ) d 1 when 0 < d 1 ≤ 1. \int_{\mathbb{R}^3}\rho|\nabla\Phi_\alpha*f|^2\,dx
\le C(1+B_{\bar p}^2)d_1\qquad\text{when }0<d_1\le1. ∫ R 3 ρ ∣∇ Φ α ∗ f ∣ 2 d x ≤ C ( 1 + B p ˉ 2 ) d 1 when 0 < d 1 ≤ 1.
Finally, if d 1 > 1 d_1>1 d 1 > 1 , then
∫ R 3 ρ ∣ ∇ Φ α ∗ f ∣ 2 d x ≤ M ∥ ∇ Φ α ∗ f ∥ L ∞ 2 ≤ M B p ˉ 2 d 1 . \int_{\mathbb{R}^3}\rho|\nabla\Phi_\alpha*f|^2\,dx
\le M\|\nabla\Phi_\alpha*f\|_{L^\infty}^2
\le MB_{\bar p}^2d_1. ∫ R 3 ρ ∣∇ Φ α ∗ f ∣ 2 d x ≤ M ∥∇ Φ α ∗ f ∥ L ∞ 2 ≤ M B p ˉ 2 d 1 .
Combining the two ranges of d 1 d_1 d 1 and the definition of B p ˉ B_{\bar p} B p ˉ proves (56) . □
Relative Entropy Inequality and Finite-Time Stability
In this section, we derive the distributional relative entropy inequality and prove Theorem 6.2.
The Effective Entropy Inequality
We first derive the relative entropy inequality associated with the effective equilibrium potential G ˉ \bar{G} G ˉ . Define the effective entropy density and flux by
η ∗ ( ρ , M ) : = ∣ M ∣ 2 2 ρ + h ( ρ ) − h ( ρ ˉ ) + G ˉ ( ρ − ρ ˉ ) , q ∗ ( ρ , M ) : = q ( ρ , M ) + G ˉ M . \eta_*(\rho,\mathbf{M}) := \frac{|\mathbf{M}|^2}{2\rho} + h(\rho) - h(\bar{\rho}) + \bar{G}(\rho-\bar{\rho}), \qquad q_*(\rho,\mathbf{M}) := q(\rho,\mathbf{M})+\bar{G}\mathbf{M}. η ∗ ( ρ , M ) := 2 ρ ∣ M ∣ 2 + h ( ρ ) − h ( ρ ˉ ) + G ˉ ( ρ − ρ ˉ ) , q ∗ ( ρ , M ) := q ( ρ , M ) + G ˉ M .
The rotational cancellation established in Proposition 4.5 ensures that the centrifugal potential does not appear explicitly in these expressions. The remaining equilibrium potential is G ˉ = μ + W ˉ α \bar{G}=\mu+\bar{W}_\alpha G ˉ = μ + W ˉ α .
Lemma 6.1 (Relative entropy inequality). Let ( ρ , M ) (\rho,\mathbf{M}) ( ρ , M ) be an axisymmetric finite-energy entropy weak solution of (1.1) in the sense of Definition 2.4 . Assume that
M ⋅ ( ∇ Φ α ∗ ( ρ − ρ ˉ ) ) ∈ L l o c 1 ( ( 0 , T ) × R 3 ) . \mathbf{M}\cdot(\nabla\Phi_\alpha*(\rho-\bar{\rho}))\in L^1_{\mathrm{loc}}((0,T)\times\mathbb{R}^3). M ⋅ ( ∇ Φ α ∗ ( ρ − ρ ˉ )) ∈ L loc 1 (( 0 , T ) × R 3 ) . Then, for every nonnegative ψ ∈ W c 1 , ∞ ( [ 0 , T ) × R 3 ) \psi\in W_c^{1,\infty}([0,T)\times\mathbb{R}^3) ψ ∈ W c 1 , ∞ ([ 0 , T ) × R 3 ) ,
∫ 0 T ∫ R 3 ( η ∗ ( ρ , M ) ∂ t ψ + q ∗ ( ρ , M ) ⋅ ∇ ψ ) d x d t + ∫ R 3 η ∗ ( ρ 0 , M 0 ) ψ ( 0 , x ) d x ≥ ∫ 0 T ∫ R 3 ψ M ⋅ ( ∇ Φ α ∗ ( ρ − ρ ˉ ) ) d x d t . (57) \int_0^T\int_{\mathbb{R}^3}\left(\eta_*(\rho,\mathbf{M})\partial_t\psi+q_*(\rho,\mathbf{M})\cdot\nabla\psi\right)\,\mathrm{d}x\,\mathrm{d}t+\int_{\mathbb{R}^3}\eta_*(\rho_0,\mathbf{M}_0)\psi(0,x)\,\mathrm{d}x
\geq\int_0^T\int_{\mathbb{R}^3}\psi\,\mathbf{M}\cdot(\nabla\Phi_\alpha*(\rho-\bar{\rho}))\,\mathrm{d}x\,\mathrm{d}t. \tag*{(57)} ∫ 0 T ∫ R 3 ( η ∗ ( ρ , M ) ∂ t ψ + q ∗ ( ρ , M ) ⋅ ∇ ψ ) d x d t + ∫ R 3 η ∗ ( ρ 0 , M 0 ) ψ ( 0 , x ) d x ≥ ∫ 0 T ∫ R 3 ψ M ⋅ ( ∇ Φ α ∗ ( ρ − ρ ˉ )) d x d t . (57) Equivalently,
∂ t η ∗ ( ρ , M ) + div q ∗ ( ρ , M ) ≤ − M ⋅ ( ∇ Φ α ∗ ( ρ − ρ ˉ ) ) in D ′ ( ( 0 , T ) × R 3 ) . (58) \partial_t\eta_*(\rho,\mathbf{M})+\operatorname{div}q_*(\rho,\mathbf{M})\leq-\mathbf{M}\cdot(\nabla\Phi_\alpha*(\rho-\bar{\rho}))\qquad\text{in }D'((0,T)\times\mathbb{R}^3). \tag*{(58)} ∂ t η ∗ ( ρ , M ) + div q ∗ ( ρ , M ) ≤ − M ⋅ ( ∇ Φ α ∗ ( ρ − ρ ˉ )) in D ′ (( 0 , T ) × R 3 ) . (58) If ( ρ , M ) (\rho,\mathbf{M}) ( ρ , M ) also satisfies the angular momentum compatibility condition (3.12), then, whenever the terms are finite,
∫ R 3 η ∗ ( ρ , M ) ( t , x ) d x = E L ( t ) + K L ( ρ ˉ ) . (59) \int_{\mathbb{R}^3}\eta_*(\rho,\mathbf{M})(t,x)\,\mathrm{d}x=\mathcal{E}_L(t)+\mathcal{K}_L(\bar{\rho}). \tag*{(59)} ∫ R 3 η ∗ ( ρ , M ) ( t , x ) d x = E L ( t ) + K L ( ρ ˉ ) . (59)
Proof. By (30) , G ˉ ∈ W 1 , ∞ ( R 3 ) \bar{G}\in W^{1,\infty}(\mathbb{R}^3) G ˉ ∈ W 1 , ∞ ( R 3 ) . Thus G ˉ ψ \bar{G}\psi G ˉ ψ is an admissible test function in the weak continuity equation (15) . Since G ˉ \bar{G} G ˉ is independent of time, we obtain
∫ 0 T ∫ R 3 ( G ˉ ρ ∂ t ψ + G ˉ M ⋅ ∇ ψ ) d x d t + ∫ R 3 G ˉ ( x ) ρ 0 ( x ) ψ ( 0 , x ) d x = − ∫ 0 T ∫ R 3 ψ M ⋅ ∇ G ˉ d x d t . (60) \int_0^T\int_{\mathbb{R}^3}(\bar{G}\rho\partial_t\psi+\bar{G}\mathbf{M}\cdot\nabla\psi)\,\mathrm{d}x\,\mathrm{d}t+\int_{\mathbb{R}^3}\bar{G}(x)\rho_0(x)\psi(0,x)\,\mathrm{d}x=-\int_0^T\int_{\mathbb{R}^3}\psi\,\mathbf{M}\cdot\nabla\bar{G}\,\mathrm{d}x\,\mathrm{d}t. \tag*{(60)} ∫ 0 T ∫ R 3 ( G ˉ ρ ∂ t ψ + G ˉ M ⋅ ∇ ψ ) d x d t + ∫ R 3 G ˉ ( x ) ρ 0 ( x ) ψ ( 0 , x ) d x = − ∫ 0 T ∫ R 3 ψ M ⋅ ∇ G ˉ d x d t . (60)
Adding (60) to the mechanical entropy inequality (17) , and using ∇ G ˉ = ∇ W ˉ α \nabla\bar{G}=\nabla\bar{W}_\alpha ∇ G ˉ = ∇ W ˉ α , we find that the source term becomes
M ⋅ ( ∇ W α − ∇ W ˉ α ) = M ⋅ ( ∇ Φ α ∗ ( ρ − ρ ˉ ) ) . \mathbf{M}\cdot(\nabla W_\alpha-\nabla\bar{W}_\alpha)=\mathbf{M}\cdot(\nabla\Phi_\alpha*(\rho-\bar{\rho})). M ⋅ ( ∇ W α − ∇ W ˉ α ) = M ⋅ ( ∇ Φ α ∗ ( ρ − ρ ˉ )) .
Next, introduce the time-independent function
C ( x ) : = − h ( ρ ˉ ( x ) ) − G ˉ ( x ) ρ ˉ ( x ) . C(x):=-h(\bar{\rho}(x))-\bar{G}(x)\bar{\rho}(x). C ( x ) := − h ( ρ ˉ ( x )) − G ˉ ( x ) ρ ˉ ( x ) .
Since ρ ˉ \bar{\rho} ρ ˉ is bounded and compactly supported and G ˉ \bar{G} G ˉ is bounded, C ∈ L 1 ( R 3 ) C\in L^1(\mathbb{R}^3) C ∈ L 1 ( R 3 ) . Consequently,
∫ 0 T ∫ R 3 C ( x ) ∂ t ψ d x d t + ∫ R 3 C ( x ) ψ ( 0 , x ) d x = 0. \int_0^T\int_{\mathbb{R}^3}C(x)\partial_t\psi\,\mathrm{d}x\,\mathrm{d}t+\int_{\mathbb{R}^3}C(x)\psi(0,x)\,\mathrm{d}x=0. ∫ 0 T ∫ R 3 C ( x ) ∂ t ψ d x d t + ∫ R 3 C ( x ) ψ ( 0 , x ) d x = 0.
Adding this identity gives (57) . The distributional inequality (58) follows immediately. Finally, under angular momentum compatibility, (59) follows from the rotational cancellation identity (49) . □
To pass from the local entropy inequality to a global-in-space estimate, we impose the flux integrability condition
ρ ∣ u ∣ 3 + ρ ( h ′ ( ρ ) ) 3 2 ∈ L 1 ( ( 0 , T ) × R 3 ) for u : = M ρ . (61) \rho|\mathbf{u}|^3+\rho\bigl(h'(\rho)\bigr)^{\frac{3}{2}}\in L^1((0,T)\times\mathbb{R}^3)\qquad\text{for }\mathbf{u}:=\frac{\mathbf{M}}{\rho}. \tag*{(61)} ρ ∣ u ∣ 3 + ρ ( h ′ ( ρ ) ) 2 3 ∈ L 1 (( 0 , T ) × R 3 ) for u := ρ M . (61)
Here and throughout, u = 0 \mathbf{u}=0 u = 0 on { ρ = 0 } \{\rho=0\} { ρ = 0 } . This condition will ensure that the entropy flux vanishes in the limit when the spatial cutoff is removed.
Main Theorem
We now establish finite-time stability of compactly supported rotating Euler–Riesz equilibria within the class of axisymmetric finite-energy entropy weak solutions satisfying angular momentum compatibility.
Theorem 6.2 (Weighted L 2 L^2 L 2 -stability around rotating Euler–Riesz stars). Let α ∈ ( 0 , 2 ) \alpha\in(0,2) α ∈ ( 0 , 2 ) , let p p p satisfy Assumption 2.1, and let ( ρ ˉ , M ‾ ) (\bar{\rho},\overline{\mathbf{M}}) ( ρ ˉ , M ) represent a compactly supported rotating Euler–Riesz star of mass M M M in the sense of Definition 3.1. Let J J J be its prescribed angular momentum function, with L = J 2 L=J^2 L = J 2 satisfying (3.3). Let ( ρ , M ) (\rho,\mathbf{M}) ( ρ , M ) be an axisymmetric finite-energy entropy solution of (1.1) on [ 0 , T ] [0,T] [ 0 , T ] , in the sense of Definition 2.4. Assume further that the following conditions hold:
(i) The solution satisfies the angular momentum compatibility condition (3.12) for almost every t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) . The initial data ( ρ 0 , M 0 ) (\rho_0,\mathbf{M}_0) ( ρ 0 , M 0 ) satisfy the same compatibility condition and
∫ R 3 ρ 0 ( x ) d x = M . \int_{\mathbb{R}^3}\rho_0(x)\,dx=M. ∫ R 3 ρ 0 ( x ) d x = M . Moreover, the axis condition (4.7) holds for a.e. t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) , as well as for the initial density ρ 0 \rho_0 ρ 0 .
(ii) Either the finite-energy threshold (5.4) holds, or the high-integrability condition (5.5) holds with
K ‾ p , T : = ess sup 0 < t < T ∥ ρ ( t ) ∥ L p ˉ ( R 3 ) < ∞ . (62) \overline{K}_{p,T}:=\mathop{\operatorname{ess\,sup}}_{0<t<T}\|\rho(t)\|_{L^{\bar p}(\mathbb{R}^3)}<\infty. \tag*{(62)} K p , T := ess sup 0 < t < T ∥ ρ ( t ) ∥ L p ˉ ( R 3 ) < ∞. (62) (iii) The flux integrability condition (6.5) holds.
Then there exists a constant
C T = { C ( T , ρ ˉ , E L ( 0 ) ) in Case 1 , C ( T , ρ ˉ , K ‾ p , T ) in Case 2 , (63) C_T=
\begin{cases}
C(T,\bar{\rho},\mathcal{E}_L(0)) & \text{in Case 1},\\
C(T,\bar{\rho},\overline{K}_{p,T}) & \text{in Case 2},
\end{cases} \tag*{(63)} C T = { C ( T , ρ ˉ , E L ( 0 )) C ( T , ρ ˉ , K p , T ) in Case 1 , in Case 2 , (63) depending also on the fixed parameters of the problem, such that
E L ( t ) ≤ C T E L ( 0 ) for a.e. t ∈ ( 0 , T ) . (64) \mathcal{E}_L(t)\le C_T\mathcal{E}_L(0)\qquad\text{for a.e. }t\in(0,T). \tag*{(64)} E L ( t ) ≤ C T E L ( 0 ) for a.e. t ∈ ( 0 , T ) . (64) Moreover, define the weighted distance functional
D ( t ) : = d 2 ( ρ ( t ) , ρ ˉ ) + 1 2 ∫ R 3 ∣ M r ∣ 2 + ∣ M z ∣ 2 ρ d x + ∫ Γ ∩ { ∣ ρ − ρ ˉ ∣ ≤ 1 } ∣ ρ − ρ ˉ ∣ 2 d x + ∫ Γ ∩ { ∣ ρ − ρ ˉ ∣ > 1 } ∣ ρ − ρ ˉ ∣ γ 2 d x + ∫ Γ c ∩ { ρ ≤ 1 } ρ γ 1 d x + ∫ Γ c ∩ { ρ > 1 } ρ γ 2 d x + ∫ Γ c V ‾ ρ d x . (65) \begin{aligned}
\mathcal{D}(t):={}&d_2(\rho(t),\bar{\rho})+\frac{1}{2}\int_{\mathbb{R}^3}\frac{|\mathbf{M}^{r}|^2+|\mathbf{M}^{z}|^2}{\rho}\,dx+\int_{\Gamma\cap\{|\rho-\bar{\rho}|\le1\}}|\rho-\bar{\rho}|^2\,dx\\
&+\int_{\Gamma\cap\{|\rho-\bar{\rho}|>1\}}|\rho-\bar{\rho}|^{\gamma_2}\,dx+\int_{\Gamma^c\cap\{\rho\le1\}}\rho^{\gamma_1}\,dx+\int_{\Gamma^c\cap\{\rho>1\}}\rho^{\gamma_2}\,dx+\int_{\Gamma^c}\overline{V}\rho\,dx. \tag*{(65)}
\end{aligned} D ( t ) := d 2 ( ρ ( t ) , ρ ˉ ) + 2 1 ∫ R 3 ρ ∣ M r ∣ 2 + ∣ M z ∣ 2 d x + ∫ Γ ∩ { ∣ ρ − ρ ˉ ∣ ≤ 1 } ∣ ρ − ρ ˉ ∣ 2 d x + ∫ Γ ∩ { ∣ ρ − ρ ˉ ∣ > 1 } ∣ ρ − ρ ˉ ∣ γ 2 d x + ∫ Γ c ∩ { ρ ≤ 1 } ρ γ 1 d x + ∫ Γ c ∩ { ρ > 1 } ρ γ 2 d x + ∫ Γ c V ρ d x . (65) Then
D ( t ) ≤ C E L ( t ) ≤ C C T E L ( 0 ) for a.e. t ∈ ( 0 , T ) (66) \mathcal{D}(t)\le C\mathcal{E}_L(t)\le CC_T\mathcal{E}_L(0)\qquad\text{for a.e. }t\in(0,T) \tag*{(66)} D ( t ) ≤ C E L ( t ) ≤ C C T E L ( 0 ) for a.e. t ∈ ( 0 , T ) (66) for some constant C > 0 C>0 C > 0 independent of $T>0.
Proof. We divide the proof into six steps.
1. Finiteness and positivity of the modulated entropy: We first verify that E L ( t ) \mathcal{E}_L(t) E L ( t ) is well defined and nonnegative for almost every t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) . By angular momentum compatibility (3.12) and the finite kinetic energy assumption, the identity (3.13) gives
K L ( ρ ( t ) ) = 1 2 ∫ R 3 ∣ M θ ∣ 2 ρ d x < ∞ for a.e. t . \mathcal{K}_L(\rho(t))=\frac{1}{2}\int_{\mathbb{R}^3}\frac{|\mathbf{M}^{\theta}|^2}{\rho}\,dx<\infty\qquad\text{for a.e. }t. K L ( ρ ( t )) = 2 1 ∫ R 3 ρ ∣ M θ ∣ 2 d x < ∞ for a.e. t .
The finite internal energy, conservation of mass, and boundedness of G ‾ \overline{G} G ensure integrability of the terms involving h ( ρ ) h(\rho) h ( ρ ) and G ‾ ( ρ − ρ ˉ ) \overline{G}(\rho-\bar{\rho}) G ( ρ − ρ ˉ ) . Moreover, Lemma 3.2, together with the finiteness of K L ( ρ ) \mathcal{K}_L(\rho) K L ( ρ ) and K L ( ρ ˉ ) \mathcal{K}_L(\bar{\rho}) K L ( ρ ˉ ) , gives
∫ R 3 Ψ ‾ ∣ ρ − ρ ˉ ∣ d x < ∞ . \int_{\mathbb{R}^3}\overline{\Psi}|\rho-\bar{\rho}|\,dx<\infty. ∫ R 3 Ψ ∣ ρ − ρ ˉ ∣ d x < ∞.
Thus, d 1 ( ρ , ρ ˉ ) d_1(\rho,\bar{\rho}) d 1 ( ρ , ρ ˉ ) and d 2 ( ρ , ρ ˉ ) d_2(\rho,\bar{\rho}) d 2 ( ρ , ρ ˉ ) are finite.
By Lemma 4.1,
d 1 ( ρ , ρ ˉ ) ≥ 0 , d_1(\rho,\bar{\rho}) \ge0, d 1 ( ρ , ρ ˉ ) ≥ 0 ,
while Lemma 4.3, under the axis condition (4.7), yields
d 2 ( ρ , ρ ˉ ) ≥ 0. d_2(\rho,\bar{\rho}) \ge0. d 2 ( ρ , ρ ˉ ) ≥ 0.
Consequently, the definition (4.17) gives
E L ( t ) ≥ 0 for a.e. t ∈ ( 0 , T ) . (67) \mathcal{E}_{L}(t) \ge0 \qquad\text{for a.e. } t \in(0,T). \tag*{(67)} E L ( t ) ≥ 0 for a.e. t ∈ ( 0 , T ) . (67)
A priori bounds for the modulated entropy
Set
A ( t ) : = ∫ R 3 ( ∣ M ∣ 2 2 ρ + h ( ρ ) ) ( t , x ) d x . (68) A(t) := \int_{\mathbb{R}^{3}} \left(\frac{|\mathbf{M}|^{2}}{2\rho}+h(\rho)\right)(t,x)\,\mathrm{d}x. \tag*{(68)} A ( t ) := ∫ R 3 ( 2 ρ ∣ M ∣ 2 + h ( ρ ) ) ( t , x ) d x . (68)
By Definition 2.4,
A ∈ L ∞ ( 0 , T ) . A \in L^{\infty}(0,T). A ∈ L ∞ ( 0 , T ) .
We first note that the continuity equation implies conservation of total mass. Indeed, testing (2.14) with spatial cutoffs approaching one and using
∫ R 3 ∣ M ∣ d x ≤ ( ∫ R 3 ρ d x ) 1 / 2 ( ∫ R 3 ∣ M ∣ 2 ρ d x ) 1 / 2 , \int_{\mathbb{R}^{3}} |\mathbf{M}|\,\mathrm{d}x \le\left(\int_{\mathbb{R}^{3}} \rho\,\mathrm{d}x\right)^{1/2}\left(\int_{\mathbb{R}^{3}} \frac{|\mathbf{M}|^{2}}{\rho}\,\mathrm{d}x\right)^{1/2}, ∫ R 3 ∣ M ∣ d x ≤ ( ∫ R 3 ρ d x ) 1/2 ( ∫ R 3 ρ ∣ M ∣ 2 d x ) 1/2 ,
we may pass to the limit in the weak continuity equation. Then
∫ R 3 ρ ( t , x ) d x = M for a.e. t ∈ ( 0 , T ) . \int_{\mathbb{R}^{3}} \rho(t,x)\,\mathrm{d}x = M \qquad\text{for a.e. } t \in(0,T). ∫ R 3 ρ ( t , x ) d x = M for a.e. t ∈ ( 0 , T ) .
By the rotational cancellation identity (4.18),
E L ( t ) = A ( t ) − ∫ R 3 h ( ρ ˉ ) d x + ∫ R 3 G ˉ ( ρ − ρ ˉ ) d x − K L ( ρ ˉ ) . \mathcal{E}_{L}(t) = A(t) - \int_{\mathbb{R}^{3}} h(\bar{\rho})\,\mathrm{d}x + \int_{\mathbb{R}^{3}} \bar{G}(\rho-\bar{\rho})\,\mathrm{d}x - K_{L}(\bar{\rho}). E L ( t ) = A ( t ) − ∫ R 3 h ( ρ ˉ ) d x + ∫ R 3 G ˉ ( ρ − ρ ˉ ) d x − K L ( ρ ˉ ) .
Since G ˉ ∈ L ∞ ( R 3 ) \bar{G} \in L^{\infty}(\mathbb{R}^{3}) G ˉ ∈ L ∞ ( R 3 ) and both densities have mass M M M ,
∣ ∫ R 3 G ˉ ( ρ − ρ ˉ ) d x ∣ ≤ 2 M ∥ G ˉ ∥ L ∞ . \left|\int_{\mathbb{R}^{3}} \bar{G}(\rho-\bar{\rho})\,\mathrm{d}x\right| \le2M\|\bar{G}\|_{L^{\infty}}. ∫ R 3 G ˉ ( ρ − ρ ˉ ) d x ≤ 2 M ∥ G ˉ ∥ L ∞ .
Together with (6.11), this gives
0 ≤ E L ( t ) ≤ A ( t ) + ∫ R 3 h ( ρ ˉ ) d x + K L ( ρ ˉ ) + 2 M ∥ G ˉ ∥ L ∞ . (69) 0 \le\mathcal{E}_{L}(t) \le A(t) + \int_{\mathbb{R}^{3}} h(\bar{\rho})\,\mathrm{d}x + K_{L}(\bar{\rho}) + 2M\|\bar{G}\|_{L^{\infty}}. \tag*{(69)} 0 ≤ E L ( t ) ≤ A ( t ) + ∫ R 3 h ( ρ ˉ ) d x + K L ( ρ ˉ ) + 2 M ∥ G ˉ ∥ L ∞ . (69)
In particular,
E L ∈ L ∞ ( 0 , T ) . \mathcal{E}_{L} \in L^{\infty}(0,T). E L ∈ L ∞ ( 0 , T ) .
In Case 1, the pressure threshold implies
γ 2 ≥ γ ≥ 9 7 − 2 α > 3 + α 3 . \gamma_{2} \ge\gamma\ge\frac{9}{7-2\alpha} > \frac{3+\alpha}{3}. γ 2 ≥ γ ≥ 7 − 2 α 9 > 3 3 + α .
Thus Proposition 2.3 provides the fixed-mass coercivity estimate (2.9). Consequently,
H ( ρ ( t ) , M ( t ) ) = A ( t ) − P α ( ρ ( t ) ) ≥ ∫ R 3 ∣ M ∣ 2 2 ρ d x + ( 1 − ϑ M ) ∫ R 3 h ( ρ ) d x − C M ≥ c M A ( t ) − C M , (70) \begin{aligned}
\mathcal{H}(\rho(t),\mathbf{M}(t)) &= A(t)-\mathcal{P}_{\alpha}(\rho(t)) \\
&\ge\int_{\mathbb{R}^{3}} \frac{|\mathbf{M}|^{2}}{2\rho}\,\mathrm{d}x +(1-\vartheta_{M})\int_{\mathbb{R}^{3}} h(\rho)\,\mathrm{d}x-C_{M}\\
&\ge c_{M}A(t)-C_{M}, \tag*{(70)}
\end{aligned} H ( ρ ( t ) , M ( t )) = A ( t ) − P α ( ρ ( t )) ≥ ∫ R 3 2 ρ ∣ M ∣ 2 d x + ( 1 − ϑ M ) ∫ R 3 h ( ρ ) d x − C M ≥ c M A ( t ) − C M , (70)
where c M : = min { 1 , 1 − ϑ M } > 0 c_{M}:=\min\{1,1-\vartheta_{M}\}>0 c M := min { 1 , 1 − ϑ M } > 0 .
Since P α ( ρ 0 ) ≥ 0 \mathcal{P}_{\alpha}(\rho_{0}) \ge0 P α ( ρ 0 ) ≥ 0 , we have
H ( ρ 0 , M 0 ) ≤ A ( 0 ) . \mathcal{H}(\rho_{0},\mathbf{M}_{0}) \le A(0). H ( ρ 0 , M 0 ) ≤ A ( 0 ) .
The total energy inequality (2.8) therefore yields
A ( t ) ≤ c M − 1 ( A ( 0 ) + C M ) for a.e. t ∈ ( 0 , T ) . (71) A(t) \le c_{M}^{-1}(A(0)+C_{M}) \qquad\text{for a.e. } t \in(0,T). \tag*{(71)} A ( t ) ≤ c M − 1 ( A ( 0 ) + C M ) for a.e. t ∈ ( 0 , T ) . (71)
Applying (49) to the initial data, we obtain
A ( 0 ) = E L ( 0 ) + ∫ R 3 h ( ρ ˉ ) d x − ∫ R 3 G ˉ ( ρ 0 − ρ ˉ ) d x + K L ( ρ ˉ ) ≤ E L ( 0 ) + ∫ R 3 h ( ρ ˉ ) d x + K L ( ρ ˉ ) + 2 M ∥ G ˉ ∥ L ∞ . (72) \begin{aligned}
A(0)=\mathcal{E}_L(0)+\int_{\mathbb{R}^3}h(\bar{\rho})\,\mathrm{d}x-\int_{\mathbb{R}^3}\bar{G}(\rho_0-\bar{\rho})\,\mathrm{d}x+\mathcal{K}_L(\bar{\rho}) \\
&\leq\mathcal{E}_L(0)+\int_{\mathbb{R}^3}h(\bar{\rho})\,\mathrm{d}x+\mathcal{K}_L(\bar{\rho})+2M\lVert\bar{G}\rVert_{L^\infty}. \tag*{(72)}
\end{aligned} A ( 0 ) = E L ( 0 ) + ∫ R 3 h ( ρ ˉ ) d x − ∫ R 3 G ˉ ( ρ 0 − ρ ˉ ) d x + K L ( ρ ˉ ) ≤ E L ( 0 ) + ∫ R 3 h ( ρ ˉ ) d x + K L ( ρ ˉ ) + 2 M ∥ G ˉ ∥ L ∞ . (72)
Combining (69) , (71) , and (72) , we conclude that
ess sup 0 < t < T E L ( t ) ≤ B 0 : = C 0 ( 1 + E L ( 0 ) ) , (73) \operatorname*{ess\,sup}_{0<t<T}\mathcal{E}_L(t)\leq B_0:=C_0\left(1+\mathcal{E}_L(0)\right), \tag*{(73)} 0 < t < T ess sup E L ( t ) ≤ B 0 := C 0 ( 1 + E L ( 0 ) ) , (73)
where C 0 C_0 C 0 depends only on the fixed parameters, the prescribed mass, and the equilibrium. In Case 2, the fixed-mass coercivity estimate (10) is not required. The uniform boundedness of E L \mathcal{E}_L E L already follows from the finite-energy assumption and (69) .
Integrability of the nonlocal source term
Set
f : = ρ − ρ ˉ , F : = ∇ Φ α ∗ f . f:=\rho-\bar{\rho},\qquad F:=\nabla\Phi_\alpha*f. f := ρ − ρ ˉ , F := ∇ Φ α ∗ f .
By Lemma 5.1, axisymmetry implies that F F F has no azimuthal component. Therefore,
M ⋅ F = ( M r e r + M z e z ) ⋅ F a.e. (74) \mathbf{M}\cdot F=(M^r e_r+M^z e_z)\cdot F\qquad\text{a.e.} \tag*{(74)} M ⋅ F = ( M r e r + M z e z ) ⋅ F a.e. (74)
Using Cauchy–Schwarz and Young’s inequality, we obtain
∫ R 3 ∣ M ⋅ F ∣ d x ≤ 1 2 ∫ R 3 ∣ M r ∣ 2 + ∣ M z ∣ 2 ρ d x + 1 2 ∫ R 3 ρ ∣ F ∣ 2 d x . \int_{\mathbb{R}^3}\lvert\mathbf{M}\cdot F\rvert\,\mathrm{d}x\leq\frac{1}{2}\int_{\mathbb{R}^3}\frac{\lvert M^r\rvert^2+\lvert M^z\rvert^2}{\rho}\,\mathrm{d}x+\frac{1}{2}\int_{\mathbb{R}^3}\rho\lvert F\rvert^2\,\mathrm{d}x. ∫ R 3 ∣ M ⋅ F ∣ d x ≤ 2 1 ∫ R 3 ρ ∣ M r ∣ 2 + ∣ M z ∣ 2 d x + 2 1 ∫ R 3 ρ ∣ F ∣ 2 d x .
Since d 1 ≤ E L d_1\leq\mathcal{E}_L d 1 ≤ E L , Proposition 5.2 yields
∫ R 3 ∣ M ⋅ F ∣ d x ≤ { C ( E L ( t ) + E L ( t ) χ ) in Case 1 , C K E L ( t ) in Case 2 , (75) \int_{\mathbb{R}^3}\lvert\mathbf{M}\cdot F\rvert\,\mathrm{d}x\leq
\begin{cases}
C\left(\mathcal{E}_L(t)+\mathcal{E}_L(t)^\chi\right) & \text{in Case 1},\\
C_K\mathcal{E}_L(t) & \text{in Case 2},
\end{cases} \tag*{(75)} ∫ R 3 ∣ M ⋅ F ∣ d x ≤ { C ( E L ( t ) + E L ( t ) χ ) C K E L ( t ) in Case 1 , in Case 2 , (75)
where χ = 7 − 2 α 3 ≥ 1 \chi=\frac{7-2\alpha}{3}\geq1 χ = 3 7 − 2 α ≥ 1 and C K = C ( 1 + K ρ ˉ , T ) 2 C_K=C(1+K_{\bar{\rho},T})^2 C K = C ( 1 + K ρ ˉ , T ) 2 .
In Case 1, the right-hand side is uniformly bounded by (73) . In Case 2, the same conclusion follows from (69) and (62) . Hence
M ⋅ F ∈ L 1 ( ( 0 , T ) × R 3 ) . (76) \mathbf{M}\cdot F\in L^1\left((0,T)\times\mathbb{R}^3\right). \tag*{(76)} M ⋅ F ∈ L 1 ( ( 0 , T ) × R 3 ) . (76)
In particular, the integrability hypothesis of Lemma 6.1 is satisfied.
Removal of the spatial cutoff
Choose ζ ∈ C c ∞ ( R 3 ) \zeta\in C^\infty_c(\mathbb{R}^3) ζ ∈ C c ∞ ( R 3 ) such that
0 ≤ ζ ≤ 1 , ζ = 1 on B 1 , ζ = 0 on B 2 c , 0\leq\zeta\leq1,\qquad\zeta=1\text{ on }B_1,\qquad\zeta=0\text{ on }B_2^c, 0 ≤ ζ ≤ 1 , ζ = 1 on B 1 , ζ = 0 on B 2 c ,
and define
ζ R ( x ) : = ζ ( x R ) . \zeta_R(x):=\zeta\left(\frac{x}{R}\right). ζ R ( x ) := ζ ( R x ) .
Fix an increasing sequence R j → ∞ R_j\to\infty R j → ∞ .
Since G ˉ \bar{G} G ˉ is bounded,
∣ η ∗ ( ρ , M ) ∣ ≤ ∣ M ∣ 2 2 ρ + h ( ρ ) + C ρ + h ( ρ ˉ ) + C ρ ˉ . \lvert\eta_*(\rho,\mathbf{M})\rvert\leq\frac{\lvert\mathbf{M}\rvert^2}{2\rho}+h(\rho)+C\rho+h(\bar{\rho})+C\bar{\rho}. ∣ η ∗ ( ρ , M )∣ ≤ 2 ρ ∣ M ∣ 2 + h ( ρ ) + C ρ + h ( ρ ˉ ) + C ρ ˉ .
The finite-energy bounds and mass conservation therefore imply η ∗ ∈ L ∞ ( 0 , T ; L 1 ( R 3 ) ) \eta_*\in L^\infty(0,T;L^1(\mathbb{R}^3)) η ∗ ∈ L ∞ ( 0 , T ; L 1 ( R 3 )) . Consequently, for each j j j ,
I j ( τ ) : = ∫ R 3 ζ R j ( x ) η ∗ ( ρ , M ) ( τ , x ) d x I_j(\tau):=\int_{\mathbb{R}^3}\zeta_{R_j}(x)\eta_*(\rho,\mathbf{M})(\tau,x)\,\mathrm{d}x I j ( τ ) := ∫ R 3 ζ R j ( x ) η ∗ ( ρ , M ) ( τ , x ) d x
belongs to L 1 ( 0 , T ) L^1(0,T) L 1 ( 0 , T ) .
We fix t ∈ ( 0 , T ) t \in(0,T) t ∈ ( 0 , T ) that is a Lebesgue point of every I j I_j I j . Such times form a set of full measure, since the sequence is countable. For 0 < ε < T − t 0 < \varepsilon< T-t 0 < ε < T − t , define
ϑ ε ( τ ) : = { 1 for 0 ≤ τ ≤ t , 1 − τ − t ε for t < τ < t + ε , 0 for t + ε ≤ τ < T . \vartheta_\varepsilon(\tau):=
\begin{cases}
1 & \text{for } 0 \le\tau\le t,\\
1-\dfrac{\tau-t}{\varepsilon} & \text{for } t < \tau< t+\varepsilon,\\
0 & \text{for } t+\varepsilon\le\tau< T.
\end{cases} ϑ ε ( τ ) := ⎩ ⎨ ⎧ 1 1 − ε τ − t 0 for 0 ≤ τ ≤ t , for t < τ < t + ε , for t + ε ≤ τ < T .
For each fixed j j j , the nonnegative function
ψ ( τ , x ) : = ϑ ε ( τ ) ζ R j ( x ) \psi(\tau,x):=\vartheta_\varepsilon(\tau)\zeta_{R_j}(x) ψ ( τ , x ) := ϑ ε ( τ ) ζ R j ( x )
is admissible in (6.1). Since ∂ τ ϑ ε = − ε − 1 \partial_\tau\vartheta_\varepsilon=-\varepsilon^{-1} ∂ τ ϑ ε = − ε − 1 on ( t , t + ε ) (t,t+\varepsilon) ( t , t + ε ) , we obtain
1 ε ∫ t t + ε ∫ R 3 ζ R j η ∗ ( ρ , M ) d x d τ ≤ ∫ R 3 ζ R j η ∗ ( ρ 0 , M 0 ) d x + ∫ 0 t + ε ∫ R 3 ϑ ε q ∗ ⋅ ∇ ζ R j d x d τ − ∫ 0 t + ε ∫ R 3 ϑ ε ζ R j M ⋅ F d x d τ . \frac{1}{\varepsilon}\int_t^{t+\varepsilon}\int_{\mathbb{R}^3}\zeta_{R_j}\eta_*(\rho,\mathbf{M})\,\mathrm{d}x\,\mathrm{d}\tau
\leq\int_{\mathbb{R}^3}\zeta_{R_j}\eta_*(\rho_0,\mathbf{M}_0)\,\mathrm{d}x
+\int_0^{t+\varepsilon}\int_{\mathbb{R}^3}\vartheta_\varepsilon q_*\cdot\nabla\zeta_{R_j}\,\mathrm{d}x\,\mathrm{d}\tau
-\int_0^{t+\varepsilon}\int_{\mathbb{R}^3}\vartheta_\varepsilon\zeta_{R_j}\mathbf{M}\cdot\mathbf{F}\,\mathrm{d}x\,\mathrm{d}\tau. ε 1 ∫ t t + ε ∫ R 3 ζ R j η ∗ ( ρ , M ) d x d τ ≤ ∫ R 3 ζ R j η ∗ ( ρ 0 , M 0 ) d x + ∫ 0 t + ε ∫ R 3 ϑ ε q ∗ ⋅ ∇ ζ R j d x d τ − ∫ 0 t + ε ∫ R 3 ϑ ε ζ R j M ⋅ F d x d τ .
For fixed j j j , the Lebesgue differentiation theorem gives
1 ε ∫ t t + ε I j ( τ ) d τ ⟶ I j ( t ) as ε → 0. \frac{1}{\varepsilon}\int_t^{t+\varepsilon}I_j(\tau)\,\mathrm{d}\tau\longrightarrow I_j(t)
\qquad\text{as }\varepsilon\to0. ε 1 ∫ t t + ε I j ( τ ) d τ ⟶ I j ( t ) as ε → 0.
The remaining terms converge by dominated convergence theorem, using the local integrability of q ∗ q_* q ∗ and (6.20). We therefore obtain
∫ R 3 ζ R j η ∗ ( t ) d x ≤ ∫ R 3 ζ R j η ∗ ( 0 ) d x + ∫ 0 t ∫ R 3 q ∗ ⋅ ∇ ζ R j d x d τ − ∫ 0 t ∫ R 3 ζ R j M ⋅ F d x d τ . (77) \int_{\mathbb{R}^3}\zeta_{R_j}\eta_*(t)\,\mathrm{d}x
\leq\int_{\mathbb{R}^3}\zeta_{R_j}\eta_*(0)\,\mathrm{d}x
+\int_0^t\int_{\mathbb{R}^3}q_*\cdot\nabla\zeta_{R_j}\,\mathrm{d}x\,\mathrm{d}\tau
-\int_0^t\int_{\mathbb{R}^3}\zeta_{R_j}\mathbf{M}\cdot\mathbf{F}\,\mathrm{d}x\,\mathrm{d}\tau. \tag*{(77)} ∫ R 3 ζ R j η ∗ ( t ) d x ≤ ∫ R 3 ζ R j η ∗ ( 0 ) d x + ∫ 0 t ∫ R 3 q ∗ ⋅ ∇ ζ R j d x d τ − ∫ 0 t ∫ R 3 ζ R j M ⋅ F d x d τ . (77)
We now let j → ∞ j\to\infty j → ∞ . Recall that q ( ρ , M ) = M ( ∣ M ∣ 2 2 ρ 2 + h ′ ( ρ ) ) q(\rho,\mathbf{M})=\mathbf{M}\left(\dfrac{|\mathbf{M}|^2}{2\rho^2}+h'(\rho)\right) q ( ρ , M ) = M ( 2 ρ 2 ∣ M ∣ 2 + h ′ ( ρ ) ) . Since M = ρ u \mathbf{M}=\rho\mathbf{u} M = ρ u , Young’s inequality gives
∣ q ( ρ , M ) ∣ ≤ 1 2 ρ ∣ u ∣ 3 + ρ ∣ u ∣ h ′ ( ρ ) ≤ C ( ρ ∣ u ∣ 3 + ρ ( h ′ ( ρ ) ) 3 / 2 ) . |q(\rho,\mathbf{M})|\leq\frac{1}{2}\rho|\mathbf{u}|^3+\rho|\mathbf{u}|h'(\rho)\leq C\left(\rho|\mathbf{u}|^3+\rho(h'(\rho))^{3/2}\right). ∣ q ( ρ , M ) ∣ ≤ 2 1 ρ ∣ u ∣ 3 + ρ ∣ u ∣ h ′ ( ρ ) ≤ C ( ρ ∣ u ∣ 3 + ρ ( h ′ ( ρ ) ) 3/2 ) .
Moreover, boundedness of G ‾ \overline{G} G yields
∣ G ‾ M ∣ ≤ C ρ ∣ u ∣ ≤ C ( ρ + ρ ∣ u ∣ 3 ) . |\overline{G}\mathbf{M}|\leq C\rho|\mathbf{u}|\leq C(\rho+\rho|\mathbf{u}|^3). ∣ G M ∣ ≤ C ρ ∣ u ∣ ≤ C ( ρ + ρ ∣ u ∣ 3 ) .
Consequently,
∣ q ∗ ∣ ≤ C ( ρ ∣ u ∣ 3 + ρ ( h ′ ( ρ ) ) 3 / 2 + ρ ) . |q_*|\leq C\left(\rho|\mathbf{u}|^3+\rho(h'(\rho))^{3/2}+\rho\right). ∣ q ∗ ∣ ≤ C ( ρ ∣ u ∣ 3 + ρ ( h ′ ( ρ ) ) 3/2 + ρ ) .
By (6.5) and the mass conservation, q ∗ ∈ L 1 ( ( 0 , T ) × R 3 ) q_*\in L^1((0,T)\times\mathbb{R}^3) q ∗ ∈ L 1 (( 0 , T ) × R 3 ) . Since
∣ ∇ ζ R j ∣ ≤ C R j 1 { R j < ∣ x ∣ < 2 R j } , |\nabla\zeta_{R_j}|\leq\frac{C}{R_j}\mathbf{1}_{\{R_j<|x|<2R_j\}}, ∣∇ ζ R j ∣ ≤ R j C 1 { R j < ∣ x ∣ < 2 R j } ,
we have
∣ ∫ 0 t ∫ R 3 q ∗ ⋅ ∇ ζ R j d x d τ ∣ ≤ C R j ∫ 0 T ∫ R j < ∣ x ∣ < 2 R j ∣ q ∗ ∣ d x d τ ⟶ 0. \left|\int_0^t\int_{\mathbb{R}^3}q_*\cdot\nabla\zeta_{R_j}\,\mathrm{d}x\,\mathrm{d}\tau\right|
\leq\frac{C}{R_j}\int_0^T\int_{R_j<|x|<2R_j}|q_*|\,\mathrm{d}x\,\mathrm{d}\tau\longrightarrow0. ∫ 0 t ∫ R 3 q ∗ ⋅ ∇ ζ R j d x d τ ≤ R j C ∫ 0 T ∫ R j < ∣ x ∣ < 2 R j ∣ q ∗ ∣ d x d τ ⟶ 0.
Furthermore, since η ∗ ( t ) ∈ L 1 ( R 3 ) \eta_*(t)\in L^1(\mathbb{R}^3) η ∗ ( t ) ∈ L 1 ( R 3 ) for almost every t t t , dominated convergence theorem gives
∫ R 3 ζ R j η ∗ ( t ) d x ⟶ ∫ R 3 η ∗ ( t ) d x . \int_{\mathbb{R}^3}\zeta_{R_j}\eta_*(t)\,\mathrm{d}x\longrightarrow\int_{\mathbb{R}^3}\eta_*(t)\,\mathrm{d}x. ∫ R 3 ζ R j η ∗ ( t ) d x ⟶ ∫ R 3 η ∗ ( t ) d x .
The same argument applies to the initial entropy density. Finally, (6.20) yields
∫ 0 t ∫ R 3 ζ R j M ⋅ F d x d τ ⟶ ∫ 0 t ∫ R 3 M ⋅ F d x d τ . \int_0^t\int_{\mathbb{R}^3}\zeta_{R_j}\mathbf{M}\cdot\mathbf{F}\,\mathrm{d}x\,\mathrm{d}\tau
\longrightarrow
\int_0^t\int_{\mathbb{R}^3}\mathbf{M}\cdot\mathbf{F}\,\mathrm{d}x\,\mathrm{d}\tau. ∫ 0 t ∫ R 3 ζ R j M ⋅ F d x d τ ⟶ ∫ 0 t ∫ R 3 M ⋅ F d x d τ .
Passing to the limit in (6.21), using (6.3), and canceling the constant K L ( ρ ‾ ) K_L(\overline{\rho}) K L ( ρ ) , we obtain
E L ( t ) ≤ E L ( 0 ) − ∫ 0 t ∫ R 3 M ⋅ F d x d τ for a.e. t ∈ ( 0 , T ) . \mathcal{E}_L(t)\leq\mathcal{E}_L(0)-\int_0^t\int_{\mathbb{R}^3}\mathbf{M}\cdot\mathbf{F}\,\mathrm{d}x\,\mathrm{d}\tau
\qquad\text{for a.e. }t\in(0,T). E L ( t ) ≤ E L ( 0 ) − ∫ 0 t ∫ R 3 M ⋅ F d x d τ for a.e. t ∈ ( 0 , T ) .
Applying (6.19) now gives
E L ( t ) ≤ E L ( 0 ) + { C ∫ 0 t ( E L ( τ ) + E L ( τ ) χ ) d τ in Case 1 , C K ∫ 0 t E L ( τ ) d τ in Case 2 . \mathcal{E}_L(t) \le\mathcal{E}_L(0) + \begin{cases}
C\int_0^t \left(\mathcal{E}_L(\tau)+\mathcal{E}_L(\tau)^\chi\right)\,\mathrm{d}\tau& \text{in Case 1},\\
C_K\int_0^t \mathcal{E}_L(\tau)\,\mathrm{d}\tau& \text{in Case 2}.
\end{cases} E L ( t ) ≤ E L ( 0 ) + { C ∫ 0 t ( E L ( τ ) + E L ( τ ) χ ) d τ C K ∫ 0 t E L ( τ ) d τ in Case 1 , in Case 2 .
Grönwall estimate and finite-time stability
In Case 1, the uniform bound (6.17) and χ ≥ 1 \chi\ge1 χ ≥ 1 imply
E L ( τ ) χ ≤ B 0 χ − 1 E L ( τ ) for a.e. τ ∈ ( 0 , T ) . \mathcal{E}_L(\tau)^\chi\le B_0^{\chi-1}\mathcal{E}_L(\tau) \qquad\text{for a.e. }\tau\in(0,T). E L ( τ ) χ ≤ B 0 χ − 1 E L ( τ ) for a.e. τ ∈ ( 0 , T ) .
Thus the nonlinear source estimate reduces to a linear one.
In Case 2, the source estimate is already linear in E L \mathcal{E}_L E L .
Consequently, in both cases,
E L ( t ) ≤ E L ( 0 ) + { C ( 1 + B 0 χ − 1 ) ∫ 0 t E L ( τ ) d τ in Case 1 , C K ∫ 0 t E L ( τ ) d τ in Case 2 . (78) \mathcal{E}_L(t) \le\mathcal{E}_L(0) + \begin{cases}
C(1+B_0^{\chi-1})\int_0^t \mathcal{E}_L(\tau)\,\mathrm{d}\tau& \text{in Case 1},\\
C_K\int_0^t \mathcal{E}_L(\tau)\,\mathrm{d}\tau& \text{in Case 2}.
\end{cases} \tag*{(78)} E L ( t ) ≤ E L ( 0 ) + { C ( 1 + B 0 χ − 1 ) ∫ 0 t E L ( τ ) d τ C K ∫ 0 t E L ( τ ) d τ in Case 1 , in Case 2 . (78)
Since E L \mathcal{E}_L E L is nonnegative and belongs to L ∞ ( 0 , T ) L^\infty(0,T) L ∞ ( 0 , T ) , Grönwall’s inequality applies and yields
E L ( t ) ≤ C T E L ( 0 ) for a.e. t ∈ ( 0 , T ) , \mathcal{E}_L(t) \le C_T\mathcal{E}_L(0) \qquad\text{for a.e. }t \in(0,T), E L ( t ) ≤ C T E L ( 0 ) for a.e. t ∈ ( 0 , T ) ,
where C T C_T C T has the dependence stated in (6.7). This proves (6.8).
Control of the weighted distance
Finally, the coercivity estimate (4.3) gives
∫ Γ ∩ { ∣ f ∣ ≤ 1 } ∣ f ∣ 2 d x + ∫ Γ ∩ { ∣ f ∣ > 1 } ∣ f ∣ γ d x + ∫ Γ c ∩ { ρ ≤ 1 } ρ γ 1 d x + ∫ Γ c ∩ { ρ > 1 } ρ γ 2 d x + ∫ Γ c V ‾ ρ d x ≤ C d 1 ( ρ , ρ ˉ ) . \int_{\Gamma\cap\{|f|\le1\}} |f|^2\,\mathrm{d}x
+\int_{\Gamma\cap\{|f|>1\}} |f|^\gamma\,\mathrm{d}x
+\int_{\Gamma^c\cap\{\rho\le1\}} \rho^{\gamma_1}\,\mathrm{d}x
+\int_{\Gamma^c\cap\{\rho>1\}} \rho^{\gamma_2}\,\mathrm{d}x
+\int_{\Gamma^c} \overline{V}\rho\,\mathrm{d}x
\le C d_1(\rho,\bar{\rho}). ∫ Γ ∩ { ∣ f ∣ ≤ 1 } ∣ f ∣ 2 d x + ∫ Γ ∩ { ∣ f ∣ > 1 } ∣ f ∣ γ d x + ∫ Γ c ∩ { ρ ≤ 1 } ρ γ 1 d x + ∫ Γ c ∩ { ρ > 1 } ρ γ 2 d x + ∫ Γ c V ρ d x ≤ C d 1 ( ρ , ρ ˉ ) .
Together with the nonnegativity of d 2 d_2 d 2 and the definition (4.17), this implies that D ( t ) ≤ C E L ( t ) \mathcal{D}(t) \le C\mathcal{E}_L(t) D ( t ) ≤ C E L ( t ) .
Combining this with (6.8) establishes
D ( t ) ≤ C E L ( t ) ≤ C C T E L ( 0 ) \mathcal{D}(t) \le C\mathcal{E}_L(t) \le CC_T\mathcal{E}_L(0) D ( t ) ≤ C E L ( t ) ≤ C C T E L ( 0 )
for almost every t ∈ ( 0 , T ) t \in(0,T) t ∈ ( 0 , T ) , which proves (6.10) and completes the proof. □
Remark 6.3. An early application of the relative entropy method to L 2 L^2 L 2 -type stability for multidimensional compressible flows was developed by Chen–Chen [12 ] . They established stability results for rarefaction waves and vacuum states of the multidimensional Euler equations within a broad class of bounded entropy solutions. For earlier developments of the relative entropy method, we refer to Dafermos [22 ] and the references therein.
Relative Total Energy
We now relate the positive rotational relative entropy E L \mathcal{E}_L E L to the difference of the total energies of a perturbed state and the rotating equilibrium.
Define the relative total energy by
H ( ρ , M ∣ ρ ˉ , M ‾ ) : = H ( ρ , M ) − H ( ρ ˉ , M ‾ ) . \mathcal{H}(\rho,\mathbf{M}\mid\bar{\rho},\overline{\mathbf{M}}):=\mathcal{H}(\rho,\mathbf{M})-\mathcal{H}(\bar{\rho},\overline{\mathbf{M}}). H ( ρ , M ∣ ρ ˉ , M ) := H ( ρ , M ) − H ( ρ ˉ , M ) .
Since the equilibrium momentum is given by
M ‾ ( x ) = ρ ˉ ( x ) J ( m ρ ˉ ( r ( x ) ) ) r ( x ) e θ , \overline{\mathbf{M}}(x)=\bar{\rho}(x)\frac{J(m_{\bar{\rho}}(r(x)))}{r(x)}\mathbf{e}_\theta, M ( x ) = ρ ˉ ( x ) r ( x ) J ( m ρ ˉ ( r ( x ))) e θ ,
the angular momentum identity (3.13) yields
H ( ρ ˉ , M ‾ ) = K L ( ρ ˉ ) + ∫ R 3 h ( ρ ˉ ) d x + 1 2 ∫ R 3 ρ ˉ W ‾ α d x . \mathcal{H}(\bar{\rho},\overline{\mathbf{M}})
=\mathcal{K}_L(\bar{\rho})
+\int_{\mathbb{R}^3}h(\bar{\rho})\,\mathrm{d}x
+\frac{1}{2}\int_{\mathbb{R}^3}\bar{\rho}\overline{W}_\alpha\,\mathrm{d}x. H ( ρ ˉ , M ) = K L ( ρ ˉ ) + ∫ R 3 h ( ρ ˉ ) d x + 2 1 ∫ R 3 ρ ˉ W α d x .
The following proposition identifies the relative total energy as the sum of the positive modulated entropy and a nonpositive quadratic interaction term.
Proposition 7.1 (Relative energy identity). Let ( ρ , M ) (\rho,\mathbf{M}) ( ρ , M ) satisfy the angular momentum compatibility condition (3.12) and have the same total mass as ρ ˉ \bar{\rho} ρ ˉ . Assume that both total energies are finite and, with f : = ρ − ρ ˉ f:=\rho-\bar{\rho} f := ρ − ρ ˉ ,
f ∈ L 6 6 − α ( R 3 ) . (79) f\in L^{\frac{6}{6-\alpha}}(\mathbb{R}^{3}). \tag*{(79)} f ∈ L 6 − α 6 ( R 3 ) . (79) Suppose also that E L \mathcal{E}_{L} E L is well defined by (32) , (33) , and (48) ; in particular, K L ( ρ ) < ∞ \mathcal{K}_{L}(\rho)<\infty K L ( ρ ) < ∞ , Ψ ˉ ∣ ρ − ρ ˉ ∣ ∈ L 1 ( R 3 ) \bar{\Psi}|\rho-\bar{\rho}|\in L^{1}(\mathbb{R}^{3}) Ψ ˉ ∣ ρ − ρ ˉ ∣ ∈ L 1 ( R 3 ) .
Then
H ( ρ , M ∣ ρ ˉ , M ˉ ) = E L + Q α , (80) \mathcal{H}(\rho,\mathbf{M}\mid\bar{\rho},\bar{\mathbf{M}})=\mathcal{E}_{L}+Q_{\alpha}, \tag*{(80)} H ( ρ , M ∣ ρ ˉ , M ˉ ) = E L + Q α , (80) where
Q α : = 1 2 ∫ R 3 f ( Φ α ∗ f ) d x ≤ 0. (81) Q_{\alpha}:=\frac{1}{2}\int_{\mathbb{R}^{3}}f(\Phi_{\alpha}*f)\,\mathrm{d}x\leq0. \tag*{(81)} Q α := 2 1 ∫ R 3 f ( Φ α ∗ f ) d x ≤ 0. (81)
Proof. Since ρ \rho ρ and ρ ˉ \bar{\rho} ρ ˉ have the same total mass,
∫ R 3 μ ( ρ − ρ ˉ ) d x = 0. \int_{\mathbb{R}^{3}}\mu(\rho-\bar{\rho})\,\mathrm{d}x=0. ∫ R 3 μ ( ρ − ρ ˉ ) d x = 0.
Set f : = ρ − ρ ˉ f:=\rho-\bar{\rho} f := ρ − ρ ˉ . Using W ˉ α = Φ α ∗ ρ ˉ \bar{W}_{\alpha}=\Phi_{\alpha}*\bar{\rho} W ˉ α = Φ α ∗ ρ ˉ and the symmetry of the Riesz kernel, we obtain the quadratic expansion
1 2 ∫ R 3 ρ ( Φ α ∗ ρ ) d x − 1 2 ∫ R 3 ρ ˉ ( Φ α ∗ ρ ˉ ) d x = ∫ R 3 W ˉ α f d x + 1 2 ∫ R 3 f ( Φ α ∗ f ) d x . \frac{1}{2}\int_{\mathbb{R}^{3}}\rho(\Phi_{\alpha}*\rho)\,\mathrm{d}x-\frac{1}{2}\int_{\mathbb{R}^{3}}\bar{\rho}(\Phi_{\alpha}*\bar{\rho})\,\mathrm{d}x=\int_{\mathbb{R}^{3}}\bar{W}_{\alpha}f\,\mathrm{d}x+\frac{1}{2}\int_{\mathbb{R}^{3}}f(\Phi_{\alpha}*f)\,\mathrm{d}x. 2 1 ∫ R 3 ρ ( Φ α ∗ ρ ) d x − 2 1 ∫ R 3 ρ ˉ ( Φ α ∗ ρ ˉ ) d x = ∫ R 3 W ˉ α f d x + 2 1 ∫ R 3 f ( Φ α ∗ f ) d x .
All terms are finite under the stated assumptions, by using the Hardy–Littlewood–Sobolev inequality and the boundedness of W ˉ α \bar{W}_{\alpha} W ˉ α .
Since G ˉ = W ˉ α + μ \bar{G}=\bar{W}_{\alpha}+\mu G ˉ = W ˉ α + μ and the mass difference vanishes,
∫ R 3 G ˉ f d x = ∫ R 3 W ˉ α f d x . \int_{\mathbb{R}^{3}}\bar{G}f\,\mathrm{d}x=\int_{\mathbb{R}^{3}}\bar{W}_{\alpha}f\,\mathrm{d}x. ∫ R 3 G ˉ f d x = ∫ R 3 W ˉ α f d x .
Combining the preceding identities with the rotational cancellation formula (49) proves (80) . It remains to establish the sign of Q α Q_{\alpha} Q α . Set
p α : = 6 6 − α . p_{\alpha}:=\frac{6}{6-\alpha}. p α := 6 − α 6 .
Then 2 p α + α 3 = 2 \frac{2}{p_{\alpha}}+\frac{\alpha}{3}=2 p α 2 + 3 α = 2 . The Hardy–Littlewood–Sobolev inequality therefore gives
∣ ∬ R 3 × R 3 g ( x ) h ( y ) ∣ x − y ∣ α d x d y ∣ ≤ C α ∥ g ∥ L p α ∥ h ∥ L p α . (82) \left|\iint_{\mathbb{R}^{3}\times\mathbb{R}^{3}}\frac{g(x)h(y)}{|x-y|^{\alpha}}\,\mathrm{d}x\,\mathrm{d}y\right|\leq C_{\alpha}\lVert g\rVert_{L^{p_{\alpha}}}\lVert h\rVert_{L^{p_{\alpha}}}. \tag*{(82)} ∬ R 3 × R 3 ∣ x − y ∣ α g ( x ) h ( y ) d x d y ≤ C α ∥ g ∥ L p α ∥ h ∥ L p α . (82)
The Fourier transform of the Riesz kernel satisfies
∣ x ∣ − α ^ ( ξ ) = c α ∣ ξ ∣ α − 3 with c α > 0. \widehat{|\boldsymbol{x}|^{-\alpha}}(\xi)=c_{\alpha}|\xi|^{\alpha-3}\quad\text{with }c_{\alpha}>0. ∣ x ∣ − α ( ξ ) = c α ∣ ξ ∣ α − 3 with c α > 0.
For a real-valued g ∈ C c ∞ ( R 3 ) g\in C^{\infty}_{c}(\mathbb{R}^{3}) g ∈ C c ∞ ( R 3 ) , by Plancherel’s identity and the convolution theorem, we obtain
∬ R 3 × R 3 g ( x ) g ( y ) ∣ x − y ∣ α d x d y = c α ( 2 π ) 3 ∫ R 3 ∣ ξ ∣ α − 3 ∣ g ^ ( ξ ) ∣ 2 d ξ ≥ 0 , \iint_{\mathbb{R}^{3}\times\mathbb{R}^{3}}\frac{g(x)g(y)}{|x-y|^{\alpha}}\,\mathrm{d}x\,\mathrm{d}y=\frac{c_{\alpha}}{(2\pi)^{3}}\int_{\mathbb{R}^{3}}|\xi|^{\alpha-3}|\hat{g}(\xi)|^{2}\,\mathrm{d}\xi\geq0, ∬ R 3 × R 3 ∣ x − y ∣ α g ( x ) g ( y ) d x d y = ( 2 π ) 3 c α ∫ R 3 ∣ ξ ∣ α − 3 ∣ g ^ ( ξ ) ∣ 2 d ξ ≥ 0 ,
where the factor ( 2 π ) − 3 (2\pi)^{-3} ( 2 π ) − 3 corresponds to the Fourier convention.
Now let f n ∈ C c ∞ ( R 3 ) f_{n}\in C^{\infty}_{c}(\mathbb{R}^{3}) f n ∈ C c ∞ ( R 3 ) be real-valued and satisfy f n ⟶ f f_{n}\longrightarrow f f n ⟶ f in L p α ( R 3 ) L^{p_{\alpha}}(\mathbb{R}^{3}) L p α ( R 3 ) . By (82) , the associated quadratic forms converge. Then
∬ R 3 × R 3 f ( x ) f ( y ) ∣ x − y ∣ α d x d y ≥ 0. \iint_{\mathbb{R}^{3}\times\mathbb{R}^{3}}\frac{f(x)f(y)}{|x-y|^{\alpha}}\,\mathrm{d}x\,\mathrm{d}y\geq0. ∬ R 3 × R 3 ∣ x − y ∣ α f ( x ) f ( y ) d x d y ≥ 0.
Finally, recalling that Φ α ( x ) = − 1 α ∣ x ∣ − α \Phi_{\alpha}(x)=-\frac{1}{\alpha}|x|^{-\alpha} Φ α ( x ) = − α 1 ∣ x ∣ − α , we conclude that
Q α = − 1 2 α ∬ R 3 × R 3 f ( x ) f ( y ) ∣ x − y ∣ α d x d y ≤ 0. Q_{\alpha}=-\frac{1}{2\alpha}\iint_{\mathbb{R}^{3}\times\mathbb{R}^{3}}\frac{f(x)f(y)}{|x-y|^{\alpha}}\,\mathrm{d}x\,\mathrm{d}y\leq0. Q α = − 2 α 1 ∬ R 3 × R 3 ∣ x − y ∣ α f ( x ) f ( y ) d x d y ≤ 0.
This proves (81) and completes the proof.
We next verify that the integrability condition (79) is satisfied in both regimes of Theorem 6.2 . In Case 1, Lemma 4.2 applies because
6 6 − α < 9 7 − 2 α ≤ γ , 0 < α < 2. \frac{6}{6-\alpha}<\frac{9}{7-2\alpha}\leq\gamma,\qquad0<\alpha<2. 6 − α 6 < 7 − 2 α 9 ≤ γ , 0 < α < 2.
Consequently, f ∈ L 6 6 − α ( R 3 ) f\in L^{\frac{6}{6-\alpha}}(\mathbb{R}^{3}) f ∈ L 6 − α 6 ( R 3 ) .
In Case 2, the prescribed mass and high-integrability assumptions give
f ∈ L 1 ( R 3 ) ∩ L p ˉ ( R 3 ) . f\in L^{1}(\mathbb{R}^{3})\cap L^{\bar{p}}(\mathbb{R}^{3}). f ∈ L 1 ( R 3 ) ∩ L p ˉ ( R 3 ) .
Since
1 < 6 6 − α < 3 2 − α < p ˉ , 1<\frac{6}{6-\alpha}<\frac{3}{2-\alpha}<\bar{p}, 1 < 6 − α 6 < 2 − α 3 < p ˉ ,
interpolation yields f ∈ L 6 6 − α ( R 3 ) f\in L^{\frac{6}{6-\alpha}}(\mathbb{R}^{3}) f ∈ L 6 − α 6 ( R 3 ) . Therefore, Proposition 7.1 applies in both cases of the main theorem.
Corollary 7.2. (Relative total energy inequality). Under the assumptions of Theorem 6.2 , suppose additionally in Case 2 that ρ 0 ∈ L p ˉ ( R 3 ) \rho_{0}\in L^{\bar{p}}(\mathbb{R}^{3}) ρ 0 ∈ L p ˉ ( R 3 ) . Then
H ( ρ ( t ) , M ( t ) ∣ ρ ˉ , M ‾ ) ≤ H ( ρ 0 , M 0 ∣ ρ ˉ , M ‾ ) for a.e. t ∈ ( 0 , T ) . (83) \mathcal{H}(\rho(t),\mathbf{M}(t)\mid\bar{\rho},\overline{\mathbf{M}})\leq\mathcal{H}(\rho_{0},\mathbf{M}_{0}\mid\bar{\rho},\overline{\mathbf{M}})\qquad\text{for a.e. }t\in(0,T). \tag*{(83)} H ( ρ ( t ) , M ( t ) ∣ ρ ˉ , M ) ≤ H ( ρ 0 , M 0 ∣ ρ ˉ , M ) for a.e. t ∈ ( 0 , T ) . (83) In particular, both relative energies admit the decomposition (80) .
Proof. Since ( ρ ˉ , M ‾ ) (\bar{\rho},\overline{\mathbf{M}}) ( ρ ˉ , M ) is a stationary rotating equilibrium, its total energy H ( ρ ˉ , M ‾ ) \mathcal{H}(\bar{\rho},\overline{\mathbf{M}}) H ( ρ ˉ , M ) is independent of time.
Subtracting this constant from the total energy inequality (9) gives
H ( ρ ( t ) , M ( t ) ) − H ( ρ ˉ , M ‾ ) ≤ H ( ρ 0 , M 0 ) − H ( ρ ˉ , M ‾ ) for a.e. t ∈ ( 0 , T ) . \mathcal{H}(\rho(t),\mathbf{M}(t))-\mathcal{H}(\bar{\rho},\overline{\mathbf{M}})\leq\mathcal{H}(\rho_{0},\mathbf{M}_{0})-\mathcal{H}(\bar{\rho},\overline{\mathbf{M}})\qquad\text{for a.e. }t\in(0,T). H ( ρ ( t ) , M ( t )) − H ( ρ ˉ , M ) ≤ H ( ρ 0 , M 0 ) − H ( ρ ˉ , M ) for a.e. t ∈ ( 0 , T ) .
This proves (83) .
The integrability condition (79) follows for the time-dependent density from the preceding argument. For the initial density, it follows from Lemma 4.2 in Case 1 and from interpolation in Case 2, using the additional assumption ρ 0 ∈ L p ˉ ( R 3 ) \rho_{0}\in L^{\bar{p}}(\mathbb{R}^{3}) ρ 0 ∈ L p ˉ ( R 3 ) .
Proposition 7.1 therefore identifies the relative total energies with E L + Q α \mathcal{E}_{L}+Q_{\alpha} E L + Q α , as claimed.
Rotating White Dwarfs
We now apply the relative entropy stability theory to rotating white dwarfs governed by the Chandrasekhar equation of state. We specialize to the Newtonian gravitational interaction α = 1 \alpha=1 α = 1 and consider the pressure law
p w d ( ρ ) = A ∫ 0 B ρ 1 3 s 4 D + s 2 d s with A , B , D > 0. (84) p_{\mathrm{wd}}(\rho)=A\int_{0}^{B\rho^{\frac{1}{3}}}\frac{s^{4}}{\sqrt{D+s^{2}}}\,\mathrm{d}s\qquad\text{with }A,B,D>0. \tag*{(84)} p wd ( ρ ) = A ∫ 0 B ρ 3 1 D + s 2 s 4 d s with A , B , D > 0. (84)
Differentiating gives
p w d ′ ( ρ ) = A B 5 3 ρ 2 3 D + B 2 ρ 2 3 > 0 for ρ > 0. (85) p_{\mathrm{wd}}'(\rho)=\frac{AB^5}{3}\frac{\rho^{\frac{2}{3}}}{\sqrt{D+B^2\rho^{\frac{2}{3}}}}>0\qquad\text{for }\rho>0. \tag*{(85)} p wd ′ ( ρ ) = 3 A B 5 D + B 2 ρ 3 2 ρ 3 2 > 0 for ρ > 0. (85)
Let h w d h_{\mathrm{wd}} h wd denote the corresponding internal energy density defined by (2.1). The pressure exhibits different power-law behaviours at low and high densities. Indeed, direct asymptotic analysis of (8.1) yields
p w d ( ρ ) ∼ A B 5 5 D ρ 5 3 , h w d ( ρ ) ∼ 3 A B 5 10 D ρ 5 3 for ρ → 0 , (86) p_{\mathrm{wd}}(\rho)\sim\frac{AB^5}{5\sqrt{D}}\rho^{\frac{5}{3}},\qquad h_{\mathrm{wd}}(\rho)\sim\frac{3AB^5}{10\sqrt{D}}\rho^{\frac{5}{3}}\qquad\text{for }\rho\to0, \tag*{(86)} p wd ( ρ ) ∼ 5 D A B 5 ρ 3 5 , h wd ( ρ ) ∼ 10 D 3 A B 5 ρ 3 5 for ρ → 0 , (86)
p w d ( ρ ) ∼ A B 4 4 ρ 4 3 , h w d ( ρ ) ∼ 3 A B 4 4 ρ 4 3 for ρ → ∞ . (87) p_{\mathrm{wd}}(\rho)\sim\frac{AB^4}{4}\rho^{\frac{4}{3}},\qquad h_{\mathrm{wd}}(\rho)\sim\frac{3AB^4}{4}\rho^{\frac{4}{3}}\qquad\text{for }\rho\to\infty. \tag*{(87)} p wd ( ρ ) ∼ 4 A B 4 ρ 3 4 , h wd ( ρ ) ∼ 4 3 A B 4 ρ 3 4 for ρ → ∞. (87)
Together with (8.2), these asymptotics verify Assumption 2.1 with γ 1 = 5 3 \gamma_1=\frac{5}{3} γ 1 = 3 5 , γ 2 = 4 3 \gamma_2=\frac{4}{3} γ 2 = 3 4 , and γ = 4 3 \gamma=\frac{4}{3} γ = 3 4 . Moreover,
h w d ′ ′ ( ρ ) = p w d ′ ( ρ ) ρ = A B 5 3 ρ − 1 3 D + B 2 ρ 2 3 . h_{\mathrm{wd}}''(\rho)=\frac{p_{\mathrm{wd}}'(\rho)}{\rho}=\frac{AB^5}{3}\frac{\rho^{-\frac{1}{3}}}{\sqrt{D+B^2\rho^{\frac{2}{3}}}}. h wd ′′ ( ρ ) = ρ p wd ′ ( ρ ) = 3 A B 5 D + B 2 ρ 3 2 ρ − 3 1 .
In particular, for every R > 0 R>0 R > 0 ,
inf 0 < ρ ≤ R h w d ′ ′ ( ρ ) > 0 , \inf_{0<\rho\le R}h_{\mathrm{wd}}''(\rho)>0, 0 < ρ ≤ R inf h wd ′′ ( ρ ) > 0 ,
consistently with the uniform convexity property (2.6).
For α = 1 \alpha=1 α = 1 , the mass-critical exponent associated with the attractive interaction is
q c = 3 + α 3 = 4 3 , q_c=\frac{3+\alpha}{3}=\frac{4}{3}, q c = 3 3 + α = 3 4 ,
which coincides with the high-density adiabatic exponent of the Chandrasekhar pressure law.
The critical Hardy–Littlewood–Sobolev inequality takes the form
∬ R 3 × R 3 ϱ ( x ) ϱ ( y ) ∣ x − y ∣ d x d y ≤ C 1 ∗ M 2 3 ∫ R 3 ϱ 4 3 d x , \iint_{\mathbb{R}^3\times\mathbb{R}^3}\frac{\varrho(x)\varrho(y)}{|x-y|}\,\mathrm{d}x\,\mathrm{d}y\le C_1^*M^{\frac{2}{3}}\int_{\mathbb{R}^3}\varrho^{\frac{4}{3}}\,\mathrm{d}x, ∬ R 3 × R 3 ∣ x − y ∣ ϱ ( x ) ϱ ( y ) d x d y ≤ C 1 ∗ M 3 2 ∫ R 3 ϱ 3 4 d x ,
for every nonnegative density ϱ \varrho ϱ of mass M M M for which the right-hand side is finite.
By (8.4),
K ∞ : = lim inf ρ → ∞ h w d ( ρ ) ρ 4 3 = 3 A B 4 4 . K_\infty:=\liminf_{\rho\to\infty}\frac{h_{\mathrm{wd}}(\rho)}{\rho^{\frac{4}{3}}}=\frac{3AB^4}{4}. K ∞ := ρ → ∞ lim inf ρ 3 4 h wd ( ρ ) = 4 3 A B 4 .
Consequently, Proposition 2.3 shows that the fixed-mass coercivity condition (2.9) holds whenever
C 1 ∗ 2 M 2 3 < 3 A B 4 4 , \frac{C_1^*}{2}M^{\frac{2}{3}}<\frac{3AB^4}{4}, 2 C 1 ∗ M 3 2 < 4 3 A B 4 ,
or equivalently,
M < M H L S , M H L S : = ( 3 A B 4 2 C 1 ∗ ) 3 2 . M<M_{\mathrm{HLS}},\qquad M_{\mathrm{HLS}}:=\left(\frac{3AB^4}{2C_1^*}\right)^{\frac{3}{2}}. M < M HLS , M HLS := ( 2 C 1 ∗ 3 A B 4 ) 2 3 .
We remark that C 1 ∗ C_1^* C 1 ∗ in M H L S M_{\mathrm{HLS}} M HLS is the best interpolation constant in [5 Theorem 4] . This HLS critical-mass threshold is a sufficient condition for coercivity of the total energy. It is not, however, required for the relative entropy stability estimate established below.
Corollary 8.1 (Relative entropy stability of rotating white dwarfs). Let α = 1 \alpha=1 α = 1 and let p = p w d p=p_{\mathrm{wd}} p = p wd be the Chandrasekhar pressure law (8.1). Let ( ρ ˉ , M ˉ ) (\bar{\rho},\bar{M}) ( ρ ˉ , M ˉ ) be a rotating white dwarf equilibrium of mass M M M in the sense of Definition 3.1, with L = J ˉ 2 L=\bar{J}^{2} L = J ˉ 2 satisfying (3.3). Let ( ρ , M ) (\rho,M) ( ρ , M ) be an axisymmetric finite-energy entropy weak solution in the sense of Definition 2.4. Assume that the solution and its initial data satisfy conditions (i) and (iii) of Theorem 6.2. Suppose further that, for some p ˉ > 3 \bar{p}>3 p ˉ > 3 ,
ess sup 0 < t < T ∥ ρ ( t ) ∥ L p ˉ ( R 3 ) < ∞ . (88) \underset{0<t<T}{\operatorname{ess\,sup}}\|\rho(t)\|_{L^{\bar{p}}(\mathbb{R}^3)}<\infty. \tag*{(88)} 0 < t < T ess sup ∥ ρ ( t ) ∥ L p ˉ ( R 3 ) < ∞. (88)
Then there exists a constant C T > 0 C_T > 0 C T > 0 , depending on the fixed parameters, the equilibrium, T T T , and the bound in (88) , such that
E L ( t ) ≤ C T E L ( 0 ) for a.e. t ∈ ( 0 , T ) . (89) \mathcal{E}_L(t) \le C_T\mathcal{E}_L(0) \qquad\text{for a.e. } t \in(0,T). \tag*{(89)} E L ( t ) ≤ C T E L ( 0 ) for a.e. t ∈ ( 0 , T ) . (89)
In particular, no restriction M < M H L S M < M_{\mathrm{HLS}} M < M HLS is imposed.
Proof. As shown above, the Chandrasekhar pressure law satisfies Assumption 2.1 with γ = 4 3 \gamma= \frac{4}{3} γ = 3 4 . For α = 1 \alpha= 1 α = 1 , the threshold in Case 2 of Proposition 5.2 becomes
max { 1 , 6 7 − 2 α } = 6 5 < 4 3 = γ , \max\left\{1,\frac{6}{7-2\alpha}\right\} = \frac{6}{5} < \frac{4}{3} = \gamma, max { 1 , 7 − 2 α 6 } = 5 6 < 3 4 = γ ,
while 3 2 − α = 3 \frac{3}{2-\alpha} = 3 2 − α 3 = 3 . Thus, the pressure exponent satisfies the required threshold, and (88) provides the additional high-integrability assumption with p ˉ > 3 \bar{p} > 3 p ˉ > 3 . All hypotheses of Case 2 of Theorem 6.2 are therefore satisfied. Then applying Theorem 6.2 yields (89) . □
Acknowledgments. Gui-Qiang G. Chen was partially supported by EPSRC grants EP/V051121/1 and EP/V008854. Difan Yuan was partially supported by EPSRC grants EP/V051121/1 and the Fundamental Research Funds for the Central Universities No. 2233100021 and No. 2233300008. For the purpose of open access, the authors have applied a CC BY public copyright license to any Author Accepted Manuscript (AAM) version arising from this submission.
Conflict of Interest: The authors declare that they have no conflict of interest. The authors also declare that this manuscript has not been previously published, and will not be submitted elsewhere before your decision.
Data availability: Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.
References
Gui-Qiang G. Chen: Mathematical Institute, University of Oxford, Oxford OX2 6GG, UK; School of Mathematical Sciences, Fudan University, Shanghai 200433, China; AMSS, Chinese Academy of Sciences, Beijing 100190, China
Email address: gui-qiang.chen@maths.ox.ac.uk
Difan Yuan: School of Mathematical Sciences, Beijing Normal University and Laboratory of Mathematics and Complex Systems, Ministry of Education, Beijing 100875, China
Email address: yuandf@amss.ac.cn