Overview
Relativistic hydrodynamics describes the long-wavelength evolution of conserved energy, momentum and charge. Its applications in nuclear physics and astrophysics require a consistent treatment of dissipation. However, constitutive relations truncated at first order in gradients need not define stable and causal evolution equations when treated as an exact system [1 ] . This matters in numerical simulations, which also probe wavelengths outside the hydrodynamic regime: an instability there can prevent the extraction of long-wavelength observables.
Müller-Israel-Stewart (MIS) theories address this problem by promoting dissipative stresses and currents to additional variables with relaxation equations [2 , 3 ] . Another approach is the density frame formulation [4 ] , in which conserved densities define the hydrodynamic variables on a chosen time foliation. Its linearized theory is stable, and boost symmetry is recovered to the retained order in the derivative expansion. This diffusive formulation provides an alternative organization of relativistic hydrodynamics, without requiring the truncated equations to be an exactly causal hyperbolic system.
The Bemfica-Disconzi-Noronha-Kovtun (BDNK) formulation instead uses hydrodynamic frame freedom to obtain causal and stable evolution while retaining covariant constitutive relations in temperature, chemical potential and velocity [5 , 6 , 7 , 8 , 9 ] . Time derivatives in the first-order constitutive relations produce second-order conservation equations and additional nonhydrodynamic modes. Suitable frame choices give causal, linearly stable systems, with local well-posedness established for several constitutive classes.
Numerical studies provide encouraging tests of this formulation. Stable, convergent evolutions of smooth plane-symmetric conformal flows have been compared with MIS [10 ] , and conservative finite-volume methods have been developed for multidimensional flows [11 ] . Nonconformal ideal-gas calculations include heat flow and shock profiles [12 ] , while holographically initialized evolutions test hydrodynamics against microscopic relaxation [13 ] . Spherically symmetric neutron-star calculations in the Cowling approximation also exhibit stable evolution within a restricted parameter range [14 ] . These developments motivate further tests of BDNK as an effective classical description.
The main obstacle to BDNK hydrodynamics is finding linearly stable and causal frames for a given equation of state and physical transport coefficients.1 General conditions for causality and stability are available, including charged constructions [8 , 9 ] , necessary and sufficient conditions for generic charged frames, including non-thermodynamic choices [24 ] ,2 and a strongly hyperbolic class with a first-order corrected charge current, constraints on the equation of state, and large frame parameters [27 ] . General criteria characterize admissible coefficients, but using them still requires solving complicated inequalities for a given equation of state and set of physical transport coefficients.
In this work we provide a concrete strategy for finding such frames for any equation of state (EoS) and physical transport coefficients. We take the point of view that, within BDNK theory, the EoS and physical transport coefficients are the properties that characterise the fluid microscopically and can be obtained from e.g. kinetic theory or holography. The remaining BDNK coefficients can be freely chosen [19 ] . With this in mind, we brute-force the reduction of the Routh-Hurwitz stability criteria in order to find analytically tractable solutions and brute-force the factorisation of the longitudinal (sound) polynomial to simplify the analysis and find concrete families of frames that are general enough to accommodate all EoS and physical transport coefficients.3
In § 1.1 we provide the constitutive relations and basic thermodynamic properties of the fluids considered here. In § 1.2 we explain the construction method, allowing both thermodynamic and non-thermodynamic frames. In sections 1.3 and 1.4 we provide six families of frames and their properties for both neutral and charged fluids while in § 1.7 we conclude with a summary and open questions. In Appendix § 3 we give various derivations, checks, and proofs. Appendix § 4 compares these constructions with existing neutral and ideal-current charged frames.
Constitutive relations and physical data
To begin with, we consider a parity-even fluid with one conserved U ( 1 ) U(1) U ( 1 ) current and no external electromagnetic field following mostly the conventions of [8 ] . We introduce the timelike fluid velocity u μ u^\mu u μ normalised such that u 2 = − 1 u^2=-1 u 2 = − 1 , and define the operators Δ μ ν = g μ ν + u μ u ν \Delta^{\mu\nu}=g^{\mu\nu}+u^\mu u^\nu Δ μν = g μν + u μ u ν , D = u μ ∇ μ D=u^\mu\nabla_\mu D = u μ ∇ μ , Θ = ∇ μ u μ \Theta=\nabla_\mu u^\mu Θ = ∇ μ u μ , a μ = D u μ a^\mu=Du^\mu a μ = D u μ , ∇ ⊥ μ = Δ μ ν ∇ ν \nabla_\perp^\mu=\Delta^{\mu\nu}\nabla_\nu ∇ ⊥ μ = Δ μν ∇ ν , and α = μ / T \alpha=\mu/T α = μ / T as well as the shear tensor
σ μ ν = 2 Δ μ ρ Δ ν λ ∇ ( ρ u λ ) − 2 3 Δ μ ν Θ . (1) \sigma^{\mu\nu}=2\Delta^{\mu\rho}\Delta^{\nu\lambda}\nabla_{(\rho}u_{\lambda)}-\frac{2}{3}\Delta^{\mu\nu}\Theta. \tag*{(1)} σ μν = 2 Δ μ ρ Δ ν λ ∇ ( ρ u λ ) − 3 2 Δ μν Θ. (1)
The conservation equations determining the evolution of the fluid variables are ∇ μ T μ ν = 0 \nabla_\mu T^{\mu\nu}=0 ∇ μ T μν = 0 and ∇ μ J μ = 0 \nabla_\mu J^\mu=0 ∇ μ J μ = 0 , where the most general stress tensor and charge current under the above assumptions are, respectively
T μ ν = ( ϵ + E ) u μ u ν + ( p + P ) Δ μ ν + 2 u ( μ Q ν ) − η σ μ ν , (2) T^{\mu\nu}=(\epsilon+\mathcal{E})u^\mu u^\nu+(p+\mathcal{P})\Delta^{\mu\nu}+2u^{(\mu}Q^{\nu)}-\eta\sigma^{\mu\nu}, \tag*{(2)} T μν = ( ϵ + E ) u μ u ν + ( p + P ) Δ μν + 2 u ( μ Q ν ) − η σ μν , (2)
J μ = ( n + N ) u μ + J μ , (3) J^\mu=(n+\mathcal{N})u^\mu+\mathcal{J}^\mu, \tag*{(3)} J μ = ( n + N ) u μ + J μ , (3)
\begin{pmatrix}
\mathcal{E}\\
\mathcal{P}\\
\mathcal{N}
\end{pmatrix}
=
\begin{pmatrix}
\epsilon_1 & \epsilon_2 & \epsilon_3\\
\pi_1 & \pi_2 & \pi_3\\
\nu_1 & \nu_2 & \nu_3
\end{pmatrix}
\begin{pmatrix}
D\ln T\\
\Theta\\
D\alpha
\end{pmatrix}, \tag*{(4)}\end{pmatrix}
Q μ = θ 1 a μ + θ 2 ∇ ⊥ μ ln T + θ 3 ∇ ⊥ μ α , (5) Q^\mu=\theta_1a^\mu+\theta_2\nabla_\perp^\mu\ln T+\theta_3\nabla_\perp^\mu\alpha, \tag*{(5)} Q μ = θ 1 a μ + θ 2 ∇ ⊥ μ ln T + θ 3 ∇ ⊥ μ α , (5)
J μ = γ 1 a μ + γ 2 ∇ ⊥ μ ln T + γ 3 ∇ ⊥ μ α . (6) \mathcal{J}^\mu=\gamma_1a^\mu+\gamma_2\nabla_\perp^\mu\ln T+\gamma_3\nabla_\perp^\mu\alpha. \tag*{(6)} J μ = γ 1 a μ + γ 2 ∇ ⊥ μ ln T + γ 3 ∇ ⊥ μ α . (6)
All transport coefficients are functions of the temperature T T T and chemical potential μ \mu μ . In the case of neutral fluids they are functions of T T T only and six non-trivial coefficients, ϵ 1 , 2 \epsilon_{1,2} ϵ 1 , 2 , π 1 , 2 \pi_{1,2} π 1 , 2 , θ 1 , 2 \theta_{1,2} θ 1 , 2 , remain while the others vanish. In contrast to the thermodynamic-frame choices used in Refs. [5 , 6 , 7 , 8 , 9 ] , we do not impose θ 1 = θ 2 \theta_1=\theta_2 θ 1 = θ 2 or, for charged fluids, γ 1 = γ 2 \gamma_1=\gamma_2 γ 1 = γ 2 . We allow these vector coefficients to differ, subject to the entropy constraint stated below. This enlarges the frame choices without changing the equilibrium equation of state.
We are only interested in thermodynamically consistent equations of state (EoS) defined via the pressure p ( T , μ ) p(T,\mu) p ( T , μ ) satisfying ϵ = − p + T p T + μ p μ \epsilon=-p+Tp_T+\mu p_\mu ϵ = − p + T p T + μ p μ where we introduced the charge density n = p μ = ∂ p / ∂ μ n=p_\mu=\partial p/\partial\mu n = p μ = ∂ p / ∂ μ , the entropy density s = p T = ∂ p / ∂ T s=p_T=\partial p/\partial T s = p T = ∂ p / ∂ T and the energy density ϵ \epsilon ϵ . To specify the constraints that some of these functions must obey, it is useful to define
r = n w , A = ∂ p ∂ ϵ ∣ n , T B = ∂ p ∂ n ∣ ϵ , x = c s 2 = A + r T B , (7) r=\frac{n}{w},\qquad A=\left.\frac{\partial p}{\partial\epsilon}\right|_n,\qquad TB=\left.\frac{\partial p}{\partial n}\right|_\epsilon,\qquad x=c_s^2=A+rTB, \tag*{(7)} r = w n , A = ∂ ϵ ∂ p n , T B = ∂ n ∂ p ϵ , x = c s 2 = A + r T B , (7)
where w = ϵ + p w=\epsilon+p w = ϵ + p is the enthalpy and x x x is the square of the sound speed. We also introduce
E = ∂ T ϵ ∣ α , F = ∂ α ϵ ∣ T , N = ∂ T n ∣ α , H = ∂ α n ∣ T , E=\left.\partial_T\epsilon\right|_\alpha,\qquad F=\left.\partial_\alpha\epsilon\right|_T,\qquad N=\left.\partial_T n\right|_\alpha,\qquad H=\left.\partial_\alpha n\right|_T, E = ∂ T ϵ ∣ α , F = ∂ α ϵ ∣ T , N = ∂ T n ∣ α , H = ∂ α n ∣ T ,
Δ J = E H − F N = T ( ϵ T n μ − ϵ μ n T ) , q T = N − r E , q α = H − r F , (8) \Delta_J=EH-FN=T(\epsilon_Tn_\mu-\epsilon_\mu n_T),\qquad q_T=N-rE,\qquad q_\alpha=H-rF, \tag*{(8)} Δ J = E H − F N = T ( ϵ T n μ − ϵ μ n T ) , q T = N − r E , q α = H − r F , (8)
where the subscript T T T (or μ \mu μ ) denotes a thermodynamic derivative with respect to T T T (or μ \mu μ ). Given these functions, we work on a regular and causal thermodynamic domain T > 0 T>0 T > 0 and w > 0 w>0 w > 0 with the constraints
T > 0 , w > 0 , Δ J > 0 , 0 < x < 1 , T>0,\qquad w>0,\qquad\Delta_J>0,\qquad0<x<1, T > 0 , w > 0 , Δ J > 0 , 0 < x < 1 ,
n μ ≥ 0 , E ≥ 0 , A + n μ w ( T B ) 2 ≥ r T B , n μ ≥ n 2 w x . (9) n_\mu\geq0,\qquad E\geq0,\qquad A+\frac{n_\mu}{w}(TB)^2\geq rTB,\qquad n_\mu\geq\frac{n^2}{wx}. \tag*{(9)} n μ ≥ 0 , E ≥ 0 , A + w n μ ( T B ) 2 ≥ r T B , n μ ≥ w x n 2 . (9)
The strict inequality in Δ j \Delta_j Δ j and x x x avoid singular susceptibilities and degenerate or luminal sound speeds. Five of these conditions follow from equilibrium fluctuation positivity [8 ] and their derivation is recalled in Appendix § 3 .2.1. In addition, the second law of thermodynamics enforces [8 ]
η , ζ , D ≥ 0 , γ 2 − γ 1 = r ( θ 2 − θ 1 ) , (10) \eta,\zeta,\mathcal{D} \ge0,\qquad\gamma_2-\gamma_1=r(\theta_2-\theta_1), \tag*{(10)} η , ζ , D ≥ 0 , γ 2 − γ 1 = r ( θ 2 − θ 1 ) , (10)
as we re-derive in Section A.2.2, where η \eta η is the shear viscosity, ζ \zeta ζ is the bulk viscosity and D \mathcal{D} D the conductivity defined as
ζ = − π 2 − ϵ 1 A 2 − T ν 3 B 2 + A ( ϵ 2 + π 1 ) + B ( π 3 + T ν 2 ) − A B ( ϵ 3 + T ν 1 ) , D T = r ( θ 3 − r T θ 1 ) + r T γ 1 − γ 3 . (11) \begin{aligned}
\zeta&= -\pi_2-\epsilon_1 A^2-T\nu_3 B^2+A(\epsilon_2+\pi_1)+B(\pi_3+T\nu_2)-AB(\epsilon_3+T\nu_1),\\
\mathcal{D}T &= r(\theta_3-rT\theta_1)+rT\gamma_1-\gamma_3. \tag*{(11)}
\end{aligned} ζ D T = − π 2 − ϵ 1 A 2 − T ν 3 B 2 + A ( ϵ 2 + π 1 ) + B ( π 3 + T ν 2 ) − A B ( ϵ 3 + T ν 1 ) , = r ( θ 3 − r T θ 1 ) + r T γ 1 − γ 3 . (11)
We consider the EoS p ( T , μ ) p(T,\mu) p ( T , μ ) and the transport coefficients η , ζ , D \eta,\zeta,\mathcal{D} η , ζ , D as the physical data characterising the fluid, and derivable for instance via kinetic theory, holography, or any kind of microscopic model. Our goal is to find families of frames, i.e. choices of the remaining coefficients, in terms of this physical data, satisfying (9) –(10) but otherwise generic, that render the BDNK theory linearly causal and stable following the strategy described in the next subsection.
Frame construction strategy
We begin by linearising the conservation equations ∇ μ T μ ν = 0 \nabla_\mu T^{\mu\nu}=0 ∇ μ T μν = 0 and ∇ μ J μ = 0 \nabla_\mu J^\mu=0 ∇ μ J μ = 0 in 3 + 1 3+1 3 + 1 dimensions around a uniform equilibrium state T → T + δ T T\to T+\delta T T → T + δ T , μ → μ + δ μ \mu\to\mu+\delta\mu μ → μ + δ μ and u μ → u μ + δ u μ u^\mu\to u^\mu+\delta u^\mu u μ → u μ + δ u μ in flat space g μ ν = η μ ν g_{\mu\nu}=\eta_{\mu\nu} g μν = η μν using plane-wave perturbations of the form e − i ω t + i k ⋅ x e^{-i\omega t+i\mathbf{k}\cdot\mathbf{x}} e − iω t + i k ⋅ x where ω \omega ω is the frequency and k \mathbf{k} k the wavevector. The frequencies are given by the roots of a dispersion polynomial P ( ς , k ; v ) P(\varsigma,\mathbf{k};\mathbf{v}) P ( ς , k ; v ) . Up to a nonzero normalization, rotational symmetry in the fluid rest frame and boost invariance [6 , 8 ] gives
P ( ς , k ; v ) ∝ P T ( ς , k ; v ) 2 P L ( ς , k ; v ) = 0 , ς = − i ω . (12) P(\varsigma,\mathbf{k};\mathbf{v})\propto P_T(\varsigma,\mathbf{k};\mathbf{v})^2P_L(\varsigma,\mathbf{k};\mathbf{v})=0,\qquad\varsigma=-i\omega. \tag*{(12)} P ( ς , k ; v ) ∝ P T ( ς , k ; v ) 2 P L ( ς , k ; v ) = 0 , ς = − iω . (12)
Here v \mathbf{v} v is the equilibrium spatial fluid velocity which can be generated by a boost from the rest state, ς \varsigma ς is the growth rate, P T P_T P T is the polynomial factor for each of the two transverse velocity polarizations, and P L P_L P L describes the longitudinal velocity that couples to temperature and, for charged fluids, to the chemical potential. Our goal is to find dispersion relations ω a ( k ) \omega_a(\mathbf{k}) ω a ( k ) that are linearly stable and causal as we now explain.
What the roots must satisfy
For every root ω a \omega_a ω a , every real wavevector, and every uniform velocity ∣ v ∣ < 1 |\mathbf{v}|<1 ∣ v ∣ < 1 , linear spectral stability requires
Im ω a ( k ; v ) = Re ς a ( k ; v ) ≤ 0. (13) \operatorname{Im}\omega_a(\mathbf{k};\mathbf{v})=\operatorname{Re}\varsigma_a(\mathbf{k};\mathbf{v})\le0. \tag*{(13)} Im ω a ( k ; v ) = Re ς a ( k ; v ) ≤ 0. (13)
This bound must be satisfied for all k \mathbf{k} k including k = 0 \mathbf{k}=0 k = 0 . In order to turn stability into algebraically tractable inequalities, we first work at rest v = 0 \mathbf{v}=0 v = 0 and define q = ∣ k ∣ 2 q=|\mathbf{k}|^2 q = ∣ k ∣ 2 . For any real coefficient factor p m ( ς , q ) = ∑ j = 0 m a j ( q ) ς j p_m(\varsigma,q)=\sum_{j=0}^{m}a_j(q)\varsigma^j p m ( ς , q ) = ∑ j = 0 m a j ( q ) ς j , normalized with a m ( q ) > 0 a_m(q)>0 a m ( q ) > 0 , the Routh-Hurwitz criterion says that all roots have Re ς < 0 \operatorname{Re}\varsigma<0 Re ς < 0 if and only if all its Hurwitz determinants are positive:
Δ j ( q ) = det [ a m + i − 2 ℓ ( q ) ] i , ℓ = 1 j > 0 , j = 1 , … , m . (14) \Delta_j(q)=\det\left[a_{m+i-2\ell}(q)\right]_{i,\ell=1}^{j}>0,\qquad j=1,\ldots,m. \tag*{(14)} Δ j ( q ) = det [ a m + i − 2 ℓ ( q ) ] i , ℓ = 1 j > 0 , j = 1 , … , m . (14)
Coefficients outside 0 ≤ j ≤ m 0 \le j \le m 0 ≤ j ≤ m are set to zero. In particular, Δ 1 = a m − 1 \Delta_1 = a_{m-1} Δ 1 = a m − 1 and Δ 2 = a m − 1 a m − 2 − a m a m − 3 \Delta_2 = a_{m-1}a_{m-2} - a_ma_{m-3} Δ 2 = a m − 1 a m − 2 − a m a m − 3 . We impose their positivity for every q > 0 q > 0 q > 0 while zeros at q = 0 q = 0 q = 0 are factored out before testing the remaining roots. At nonzero boosts v ≠ 0 \mathbf{v} \ne0 v = 0 the coefficients generally become complex, so this real coefficient test is supplemented by the direct boosted proofs in Appendix § 3 .3. Causality in turn additionally requires real, finite characteristic speeds not exceeding the speed of light. These are the large k \mathbf{k} k slopes of the dispersion relations. In particular for k = k k ^ \mathbf{k} = k\hat{\mathbf{k}} k = k k ^ ,
lim k → ∞ ω a ( k k ^ ; v ) k = c a ( k ^ ; v ) ∈ R , ∣ c a ∣ ≤ 1. (15) \lim_{k\to\infty} \frac{\omega_a(k\hat{\mathbf{k}};\mathbf{v})}{k} = c_a(\hat{\mathbf{k}};\mathbf{v}) \in\mathbb{R}, \qquad|c_a| \le1. \tag*{(15)} k → ∞ lim k ω a ( k k ^ ; v ) = c a ( k ^ ; v ) ∈ R , ∣ c a ∣ ≤ 1. (15)
The speed bound is therefore lim k → ∞ ∣ ω a ∣ / k ≤ 1 \lim_{k\to\infty} |\omega_a|/k \le1 lim k → ∞ ∣ ω a ∣/ k ≤ 1 . We apply it to the principal part of the full local conservation system. A bound on an isolated dispersion branch is not, by itself, a causality proof [20 , 21 ] . The polynomial criterion also requires that the degree in frequency match the total derivative degree, so that no additional modes arise from spatial derivatives after a boost [20 ] . Below we check this for the full neutral and charged determinants, and Appendix § 3 .3 verifies their characteristic speeds for every propagation angle.
Transverse sector: the telegrapher equation
The rest frame transverse polynomial is the same for neutral and charged fluids, and given by a simple quadratic polynomial of the form
P T ( ς , q ) = θ 1 ς 2 + w ς + η q . (16) P_T(\varsigma, q) = \theta_1\varsigma^2 + w\varsigma+ \eta q. \tag*{(16)} P T ( ς , q ) = θ 1 ς 2 + w ς + η q . (16)
Indeed, a transverse perturbation has δ T = δ α = 0 \delta T = \delta\alpha= 0 δ T = δ α = 0 and k ⋅ δ u = 0 \mathbf{k} \cdot\delta\mathbf{u} = 0 k ⋅ δ u = 0 , leaving only the acceleration coefficient θ 1 \theta_1 θ 1 , enthalpy w w w , and shear viscosity η \eta η in the linear analysis. Dividing the polynomial by w w w gives the Fourier polynomial of the telegrapher equation for a transverse velocity amplitude V ⊥ V_\perp V ⊥ :
τ T ∂ t 2 V ⊥ + ∂ t V ⊥ − ν T ∇ 2 V ⊥ = 0 , τ T = θ 1 w , ν T = η w . (17) \tau_T \partial_t^2 V_\perp+ \partial_t V_\perp- \nu_T \nabla^2 V_\perp= 0, \qquad\tau_T = \frac{\theta_1}{w}, \quad\nu_T = \frac{\eta}{w}. \tag*{(17)} τ T ∂ t 2 V ⊥ + ∂ t V ⊥ − ν T ∇ 2 V ⊥ = 0 , τ T = w θ 1 , ν T = w η . (17)
This is the diffusion equation with an additional second time derivative term. The telegrapher form, which we exploit further in the charged case, makes it straightforward to derive stable and causal bounds on the BDNK coefficients as we now explain. For τ T , ν T > 0 \tau_T, \nu_T > 0 τ T , ν T > 0 and at small q q q , it has one diffusive root ς = − ν T q + O ( q 2 ) \varsigma= -\nu_T q + O(q^2) ς = − ν T q + O ( q 2 ) and one relaxation root tending to − 1 / τ T -1/\tau_T − 1/ τ T as q → 0 q \to0 q → 0 . The quadratic Routh-Hurwitz test is simply positivity of its coefficients while its squared characteristic speed is y T = ν T / τ T = η / θ 1 y_T = \nu_T/\tau_T = \eta/\theta_1 y T = ν T / τ T = η / θ 1 . Consequently the stability and causality requirements (14) and (15) reduce to
θ 1 ≥ η > 0 , θ 1 = α T η , α T ≥ 1. (18) \theta_1 \ge\eta> 0, \qquad\theta_1 = \alpha_T\eta, \quad\alpha_T \ge1. \tag*{(18)} θ 1 ≥ η > 0 , θ 1 = α T η , α T ≥ 1. (18)
The choice α T > 1 \alpha_T > 1 α T > 1 makes this shear channel strictly subluminal. The direct boosted calculation shows that the same bound controls the shear channel at all momenta and boosts, including α T = 1 \alpha_T = 1 α T = 1 (see Appendix § 3 .3.2). We note that it does not require a thermodynamic frame equality between θ 1 \theta_1 θ 1 and θ 2 \theta_2 θ 2 . The coefficient θ 1 \theta_1 θ 1 also enters the longitudinal sector, so its bound must be solved together with the conditions on the longitudinal (sound) channel.
Neutral longitudinal sector: reducing the quartic
In the case of neutral fluids, the longitudinal polynomial is a function of the six frame coefficients ε 1 , 2 \varepsilon_{1,2} ε 1 , 2 , π 1 , 2 \pi_{1,2} π 1 , 2 and θ 1 , 2 \theta_{1,2} θ 1 , 2 . A choice of these coefficients will also serve as a seed for the charged construction that follows later. We denote its bulk viscosity invariant by ζ ∗ \zeta_\ast ζ ∗ , i.e. for neutral fluids ζ ∗ = ζ \zeta_\ast=\zeta ζ ∗ = ζ . We also introduce Γ = 4 η / 3 + ζ ∗ \Gamma=4\eta/3+\zeta_\ast Γ = 4 η /3 + ζ ∗ , which is the combination controlling sound attenuation. The second law (10) gives Γ ≥ 0 \Gamma\geq0 Γ ≥ 0 , with Γ > 0 \Gamma>0 Γ > 0 in the positive-shear regime considered here. In this case, the longitudinal polynomial P L = P N P_L=P_N P L = P N is given by the quartic
P N = A L ς 4 + B L ς 3 + ( w 2 + C L q ) ς 2 + D L q ς + x w 2 q + F L q 2 (19) P_N=A_L\varsigma^4+B_L\varsigma^3+(w^2+C_Lq)\varsigma^2+D_Lq\varsigma+xw^2q+F_Lq^2 \tag*{(19)} P N = A L ς 4 + B L ς 3 + ( w 2 + C L q ) ς 2 + D L q ς + x w 2 q + F L q 2 (19)
with coefficients in the form
A L = x ε 1 θ 1 , B L = w ( x ε 1 + θ 1 ) , C L = x { ε 1 ( 4 η / 3 − π 2 ) + ε 2 ( θ 2 + π 1 ) + θ 1 π 1 } , D L = x B L + w Γ , F L = x θ 2 ( π 2 − 4 η / 3 ) , ζ ∗ = − π 2 − x 2 ε 1 + x ( ε 2 + π 1 ) . (20) \begin{aligned}
A_L&=x\varepsilon_1\theta_1,\qquad B_L=w(x\varepsilon_1+\theta_1),\qquad C_L=x\{\varepsilon_1(4\eta/3-\pi_2)+\varepsilon_2(\theta_2+\pi_1)+\theta_1\pi_1\},\\
D_L&=xB_L+w\Gamma,\qquad F_L=x\theta_2(\pi_2-4\eta/3),\\
\zeta_\ast&=-\pi_2-x^2\varepsilon_1+x(\varepsilon_2+\pi_1). \tag*{(20)}
\end{aligned} A L D L ζ ∗ = x ε 1 θ 1 , B L = w ( x ε 1 + θ 1 ) , C L = x { ε 1 ( 4 η /3 − π 2 ) + ε 2 ( θ 2 + π 1 ) + θ 1 π 1 } , = x B L + w Γ , F L = x θ 2 ( π 2 − 4 η /3 ) , = − π 2 − x 2 ε 1 + x ( ε 2 + π 1 ) . (20)
The last relation is the neutral specialization of (11) . Before applying Routh-Hurwitz, we check the zero-momentum modes:
P N ( ς , 0 ) = ς 2 ( x ε 1 ς + w ) ( θ 1 ς + w ) . (21) P_N(\varsigma,0)=\varsigma^2(x\varepsilon_1\varsigma+w)(\theta_1\varsigma+w). \tag*{(21)} P N ( ς , 0 ) = ς 2 ( x ε 1 ς + w ) ( θ 1 ς + w ) . (21)
For nonzero ε 1 , θ 1 \varepsilon_1,\theta_1 ε 1 , θ 1 , this gives two gapless modes and the gaps − w / ( x ε 1 ) -w/(x\varepsilon_1) − w / ( x ε 1 ) and − w / θ 1 -w/\theta_1 − w / θ 1 . Since w , x > 0 w,x>0 w , x > 0 , damping of both gaps requires ε 1 , θ 1 > 0 \varepsilon_1,\theta_1>0 ε 1 , θ 1 > 0 . We work in this nondegenerate regime and introduce the dimensionless combinations
R = D L B L = x + Γ x ε 1 + θ 1 , Σ = C L A L , Φ = F L A L , L = R ( Σ − R ) − Φ (22) R=\frac{D_L}{B_L}=x+\frac{\Gamma}{x\varepsilon_1+\theta_1},\qquad\Sigma=\frac{C_L}{A_L},\qquad\Phi=\frac{F_L}{A_L},\qquad\mathcal{L}=R(\Sigma-R)-\Phi \tag*{(22)} R = B L D L = x + x ε 1 + θ 1 Γ , Σ = A L C L , Φ = A L F L , L = R ( Σ − R ) − Φ (22)
in terms of which the Hurwitz determinants defined in (14) gives
Δ 1 = B L , Δ 4 = ( x w 2 q + F L q 2 ) Δ 3 , Δ 2 = B L w 2 + A L B L ( Σ − R ) q , Δ 3 = B L w 3 Γ q + A L B L 2 L q 2 . (23) \begin{aligned}
\Delta_1&=B_L,\qquad\Delta_4=(xw^2q+F_Lq^2)\Delta_3,\\
\Delta_2&=B_Lw^2+A_LB_L(\Sigma-R)q,\\
\Delta_3&=B_Lw^3\Gamma q+A_LB_L^2Lq^2. \tag*{(23)}
\end{aligned} Δ 1 Δ 2 Δ 3 = B L , Δ 4 = ( x w 2 q + F L q 2 ) Δ 3 , = B L w 2 + A L B L ( Σ − R ) q , = B L w 3 Γ q + A L B L 2 L q 2 . (23)
The requirement at every momentum has thus become a sign problem for the coefficients of these low-degree polynomials in q q q . For causality, substituting ς = − i c k \varsigma=-ick ς = − i c k , q = k 2 q=k^2 q = k 2 into the terms of total derivative order four gives y 2 − Σ y + Φ = 0 y^2-\Sigma y+\Phi=0 y 2 − Σ y + Φ = 0 , where y = c 2 y=c^2 y = c 2 . The squared longitudinal and transverse speeds are
y ± = Σ ± Σ 2 − 4 Φ 2 , y T = η θ 1 . (24) y_\pm=\frac{\Sigma\pm\sqrt{\Sigma^2-4\Phi}}{2},\qquad y_T=\frac{\eta}{\theta_1}. \tag*{(24)} y ± = 2 Σ ± Σ 2 − 4Φ , y T = θ 1 η . (24)
A sufficient set of rest frame design inequalities is therefore
Φ ≥ 0 , L ≥ 0 , 0 ≤ y − ≤ y + ≤ 1 , θ 1 ≥ η . (25) \Phi\geq0,\qquad\mathcal{L}\geq0,\qquad0\leq y_-\leq y_+\leq1,\qquad\theta_1\geq\eta. \tag*{(25)} Φ ≥ 0 , L ≥ 0 , 0 ≤ y − ≤ y + ≤ 1 , θ 1 ≥ η . (25)
Indeed, L ≥ 0 \mathcal{L}\geq0 L ≥ 0 places R R R between y − y_- y − and y + y_+ y + , so Σ ≥ R \Sigma\geq R Σ ≥ R and all four Hurwitz determinants are strictly positive at every q > 0 q>0 q > 0 . The remaining task is to choose actual frame coefficients satisfying these inequalities and entropy matching ζ ∗ = ζ \zeta_\ast=\zeta ζ ∗ = ζ . In summary, the Routh-Hurwitz inequalities and causality condition for a quartic polynomial are still analytically tractable - a realisation that we will be exploiting in the charged case.
Charged longitudinal sector: extracting a telegrapher factor
For charged fluids, P L = P c h P_L = P_{\mathrm{ch}} P L = P ch comes from three coupled longitudinal equations and is generically a sextic:
P c h ( ς , q ) = ∑ j = 0 6 a j ( q ) ς j . (26) P_{\mathrm{ch}}(\varsigma,q)=\sum_{j=0}^{6}a_j(q)\varsigma^j. \tag*{(26)} P ch ( ς , q ) = j = 0 ∑ 6 a j ( q ) ς j . (26)
Directly imposing its six Routh-Hurwitz inequalities for all q > 0 q>0 q > 0 is considerably harder to solve explicitly for arbitrary physical inputs. We instead seek frame coefficients for which the original determinant factorizes exactly as
P c h ( ς , q ) = N 0 F q ( ς , q ) P N ( ς , q ) , F q ( ς , q ) = τ q ς 2 + ς + D q q . (27) P_{\mathrm{ch}}(\varsigma,q)=\mathcal{N}_0F_q(\varsigma,q)P_N(\varsigma,q), \qquad F_q(\varsigma,q)=\tau_q\varsigma^2+\varsigma+D_qq. \tag*{(27)} P ch ( ς , q ) = N 0 F q ( ς , q ) P N ( ς , q ) , F q ( ς , q ) = τ q ς 2 + ς + D q q . (27)
Here N 0 ≠ 0 \mathcal{N}_0\ne0 N 0 = 0 is an overall normalisation that depends on thermodynamic data, D q D_q D q is a charge diffusivity, and τ q \tau_q τ q is its relaxation time to be defined precisely below. The factor F q F_q F q has precisely the telegrapher form introduced in the shear sector in (16) . With D q > 0 D_q>0 D q > 0 , choosing τ q ≥ D q \tau_q\ge D_q τ q ≥ D q gives stable rest roots and a squared characteristic speed y q = D q / τ q ≤ 1 y_q=D_q/\tau_q\le1 y q = D q / τ q ≤ 1 . The other factor has the neutral quartic form, evaluated at the charged values w w w , x x x , ζ ∗ \zeta_* ζ ∗ , where ζ ∗ = ζ + D T 2 B 2 / x \zeta_*=\zeta+DT^2B^2/x ζ ∗ = ζ + D T 2 B 2 / x as defined in (29) . The charged bulk viscosity remains the prescribed ζ \zeta ζ in (11) . Thus the search reduces to realizing this factorization while preserving the physical transport coefficients. At rest and q = 0 q=0 q = 0 , P c h P_{\mathrm{ch}} P ch gives three gapless and three gapped modes.4
Five step frame construction strategy
Below we detail the strategy used to find families of BDNK frames for any given EoS and transport coefficients.
Solve the entropy relations. The first step is to solve the entropy relations (11) for some of the BDNK coefficients in terms of the physical coefficients D \mathcal{D} D , ζ \zeta ζ and the remaining BDNK coefficients. In the neutral case, for instance, we solve the last relation in (20) , for example as π 1 = x ϵ 1 + ( ζ ∗ + π 2 ) / x − ϵ 2 \pi_1=x\epsilon_1+(\zeta_*+\pi_2)/x-\epsilon_2 π 1 = x ϵ 1 + ( ζ ∗ + π 2 ) / x − ϵ 2 . The remaining coefficients will be fixed in the later steps.
Choose branches that make the reduced quartic inequalities solvable. Next, we compute R R R , Σ \Sigma Σ , Φ \Phi Φ , L \mathcal{L} L using the coefficients in the previous step and impose (25) . To simplify the inequalities, we set π 2 = 4 η / 3 \pi_2=4\eta/3 π 2 = 4 η /3 in order to obtain Φ = 0 \Phi=0 Φ = 0 , which reduces the sound bounds to R ≤ Σ ≤ 1 R\le\Sigma\le1 R ≤ Σ ≤ 1 . Alternatively taking ϵ 2 = 0 \epsilon_2=0 ϵ 2 = 0 extracts the exact factor θ 1 ς + w \theta_1\varsigma+w θ 1 ς + w from the polynomial, leaving a simpler cubic polynomial to analyse. Keeping ϵ 2 ≠ 0 \epsilon_2\ne0 ϵ 2 = 0 and choosing Σ = 1 \Sigma=1 Σ = 1 leaves room for families that admit conformal limits. Other simplifying alternatives are possible and we discuss them in § 1.4 .
Optional: impose a finite conformal limit. Certain EoS and transport coefficients may admit a conformal point. At that point, conformal symmetry requires finite coefficients satisfying [6 , 8 ]
ε i = 3 π i , π 1 = 3 π 2 , ν 1 = 3 ν 2 , θ 1 = θ 2 , γ 1 = γ 2 . (28) \varepsilon_i = 3\pi_i,\quad\pi_1 = 3\pi_2,\quad\nu_1 = 3\nu_2,\quad\theta_1 = \theta_2,\quad\gamma_1 = \gamma_2. \tag*{(28)} ε i = 3 π i , π 1 = 3 π 2 , ν 1 = 3 ν 2 , θ 1 = θ 2 , γ 1 = γ 2 . (28)
In the neutral case, these constraints are equivalently expressed by requiring the seed to approach
( ε 1 , ε 2 , π 1 , π 2 , θ 1 , θ 2 ) ⟶ ( 3 Π , Π , Π , Π / 3 , θ , θ ) (\varepsilon_1,\varepsilon_2,\pi_1,\pi_2,\theta_1,\theta_2) \longrightarrow(3\Pi,\Pi,\Pi,\Pi/3,\theta,\theta) ( ε 1 , ε 2 , π 1 , π 2 , θ 1 , θ 2 ) ⟶ ( 3Π , Π , Π , Π/3 , θ , θ )
as x → 1 / 3 x \to1/3 x → 1/3 and ζ ∗ → 0 \zeta_* \to0 ζ ∗ → 0 , with finite Π \Pi Π , θ \theta θ . This means that non-thermodynamic frame choices are allowed away from the conformal point, but θ 2 − θ 1 \theta_2-\theta_1 θ 2 − θ 1 must vanish at the conformal point.
Uplift each neutral seed to the charged factorization. Each admissible neutral seed can be uplifted to a charged family by choosing the additional coefficients so that, after linearising about uniform equilibrium and Fourier transforming in its rest frame, δ [ ∇ μ J μ + r u ν ∇ μ T μ ν ] = F q ( ς , q ) h = 0 \delta[\nabla_\mu J^\mu+ r u_\nu\nabla_\mu T^{\mu\nu}] = F_q(\varsigma,q)h = 0 δ [ ∇ μ J μ + r u ν ∇ μ T μν ] = F q ( ς , q ) h = 0 . Here h = δ n − r δ ε = q T δ T + q α δ α h = \delta n - r\delta\varepsilon= q_T\delta T + q_\alpha\delta\alpha h = δ n − r δ ε = q T δ T + q α δ α , and r = n / w r = n/w r = n / w is evaluated at equilibrium. The coefficient map in § 1.5 realizes (27) , with
Ξ = q α − r T 2 q T = T Δ J x w > 0 , D q = D T Ξ , τ q = α q D q , α q ≥ 1 , \Xi= q_\alpha- rT^2q_T = \frac{T\Delta_J x}{w} > 0,\qquad D_q = \frac{DT}{\Xi},\qquad\tau_q = \alpha_qD_q,\qquad\alpha_q \ge1, Ξ = q α − r T 2 q T = w T Δ J x > 0 , D q = Ξ D T , τ q = α q D q , α q ≥ 1 ,
ζ ∗ = ζ + D T 2 B 2 x . (29) \zeta_* = \zeta+ \frac{DT^2B^2}{x}. \tag*{(29)} ζ ∗ = ζ + x D T 2 B 2 . (29)
Here Ξ \Xi Ξ is a charge susceptibility and α q \alpha_q α q is a free dimensionless relaxation parameter. Entropy matching (11) requires evaluating the neutral seed at the charged values w w w , x x x , ζ ∗ \zeta_* ζ ∗ . The physical charged bulk viscosity remains the prescribed ζ \zeta ζ .
Verify the original determinant, boosts and boundaries. We substitute the proposed coefficients into the constitutive relations to verify (11) with the prescribed transport coefficients and compute the determinant of the resulting linearized conservation equations to verify the factorization of the polynomial exactly. We then use the boosted polynomial at real momentum, including its actual k = 0 k = 0 k = 0 gaps to make sure that the boosted modes are stable. In practice, in Appendix § 3 .3 we show that the reduced quartic inequalities and telegrapher bounds already ensure stability and causality in every inertial frame. Hence no additional coefficient inequalities arise from boosting, and the result applies to every family of frames found by following the previous steps above. When frame coefficients or physical transport coefficients vanish, the polynomials can change degree. We then check the reduced equations directly, as summarized in footnotes 7, 11 and 12.
The steps outlined above, together with brute-force polynomial factorisation, form the basis of the AI research protocol. We will now apply this strategy to neutral and charged fluids.
Neutral construction and charged telegrapher factorization
We apply the strategy of § 1.2 to construct the neutral families F 1 F_1 F 1 –F 3 F_3 F 3 , F 4 F_4 F 4 , F 5 F_5 F 5 and F 6 F_6 F 6 , and then uplift them to the charged case. F 4 F_4 F 4 , F 5 F_5 F 5 and F 6 F_6 F 6 admit finite conformal limits with the matching conditions stated below.
Neutral branches without conformal limits
We begin by solving the reduced conditions (25) by choosing the remaining frame coefficients after entropy matching (11) . Setting π 2 = 4 η / 3 \pi_2 = 4\eta/3 π 2 = 4 η /3 makes Φ = 0 \Phi= 0 Φ = 0 and reduces the sound constraints to the interval R ≤ Σ ≤ 1 R \le\Sigma\le1 R ≤ Σ ≤ 1 . Setting ϵ 2 = 0 \epsilon_2 = 0 ϵ 2 = 0 yields, for any θ 2 \theta_2 θ 2 , the exact factorization
P N = ( θ 1 ς + w ) [ x ϵ 1 ς 3 + w ς 2 + x π 1 q ς + x w q ] . (30) P_N = (\theta_1\varsigma+ w)\left[x\epsilon_1\varsigma^3 + w\varsigma^2 + x\pi_1q\varsigma+ xwq\right]. \tag*{(30)} P N = ( θ 1 ς + w ) [ x ϵ 1 ς 3 + w ς 2 + x π 1 q ς + x w q ] . (30)
Equation (30) is an exact simplification of the neutral sound polynomial: it separates the momentum-independent relaxation root ς = − w / θ 1 \varsigma= -w/\theta_1 ς = − w / θ 1 (when θ 1 > 0 \theta_1 > 0 θ 1 > 0 ) from a cubic polynomial describing two sound modes and a second relaxation mode. Using entropy matching (11) , the cubic part of the polynomial becomes
( w + x ϵ 1 ς ) ( ς 2 + x q ) + Γ ς q . (w + x\epsilon_1\varsigma)(\varsigma^2 + xq) + \Gamma\varsigma q. ( w + x ϵ 1 ς ) ( ς 2 + x q ) + Γ ς q .
The term Γ ς q \Gamma\varsigma q Γ ς q couples sound to relaxation.5 Its cubic Routh-Hurwitz inequality reduces to positivity of Γ \Gamma Γ which is already guaranteed by the second law (10) while causality fixes ϵ 1 ≥ Γ / [ x ( 1 − x ) ] \epsilon_1 \ge\Gamma/[x(1-x)] ϵ 1 ≥ Γ/ [ x ( 1 − x )] . Together with θ 1 ≥ η \theta_1 \ge\eta θ 1 ≥ η from (18) , this defines the families F 1 F_1 F 1 and F 2 F_2 F 2 by choosing θ 2 = 0 \theta_2 = 0 θ 2 = 0 (non-thermodynamic) and θ 2 = θ 1 \theta_2 = \theta_1 θ 2 = θ 1 (thermodynamic), respectively. The family F 1 F_1 F 1 allows equality in both causality bounds, ϵ 1 = Γ / [ x ( 1 − x ) ] \epsilon_1 = \Gamma/[x(1-x)] ϵ 1 = Γ/ [ x ( 1 − x )] and θ 1 = η \theta_1 = \eta θ 1 = η .
Alternatively, we can retain ϵ 2 ≠ 0 \epsilon_2 \ne0 ϵ 2 = 0 and impose Σ = 1 \Sigma= 1 Σ = 1 , thereby determining
θ 2 = ( ϵ 1 − π 1 ) θ 1 − ϵ 2 π 1 ϵ 2 , x ϵ 1 + θ 1 ≥ Γ 1 − x . (31) \theta_2 = \frac{(\epsilon_1 - \pi_1)\theta_1 - \epsilon_2\pi_1}{\epsilon_2}, \qquad x\epsilon_1 + \theta_1 \ge\frac{\Gamma}{1-x}. \tag*{(31)} θ 2 = ϵ 2 ( ϵ 1 − π 1 ) θ 1 − ϵ 2 π 1 , x ϵ 1 + θ 1 ≥ 1 − x Γ . (31)
This defines a general luminal branch: ϵ 1 \epsilon_1 ϵ 1 , ϵ 2 ≠ 0 \epsilon_2 \ne0 ϵ 2 = 0 and θ 1 \theta_1 θ 1 remain free, subject to ϵ 1 > 0 \epsilon_1 > 0 ϵ 1 > 0 , θ 1 ≥ η \theta_1 \ge\eta θ 1 ≥ η and the inequality in (31) . Entropy matching fixes π 1 = x ϵ 1 + Γ / x − ϵ 2 \pi_1 = x\epsilon_1 + \Gamma/x - \epsilon_2 π 1 = x ϵ 1 + Γ/ x − ϵ 2 , and (31) fixes θ 2 \theta_2 θ 2 . The particular choices ϵ 2 = − 2 η \epsilon_2 = -2\eta ϵ 2 = − 2 η and θ 1 = η \theta_1 = \eta θ 1 = η give rise to the family that we refer to as F 3 F_3 F 3 and its remaining bound on ϵ 1 \epsilon_1 ϵ 1 below. In summary, the three prescriptions just described are
F 1 , F 2 : ( ϵ 1 , ϵ 2 , π 1 , π 2 , θ 1 ) = ( e , 0 , x e + Γ / x , 4 η / 3 , α T η ) , e = λ Γ x ( 1 − x ) , θ 2 = 0 ( F 1 ) , θ 2 = θ 1 ( F 2 ) , λ , α T ≥ 1 , F 3 : ϵ 2 = − 2 η , π 1 = x ϵ 1 + Γ / x + 2 η , π 2 = 4 η / 3 , θ 1 = η , θ 2 = − ( ϵ 1 + π 1 ) / 2 , ϵ 1 ≥ Γ − η ( 1 − x ) x ( 1 − x ) . (32) \begin{aligned}
F_1, F_2:\quad&(\epsilon_1,\epsilon_2,\pi_1,\pi_2,\theta_1) = (e,0,xe+\Gamma/x,4\eta/3,\alpha_T\eta), \qquad e = \frac{\lambda\Gamma}{x(1-x)},\\
&\theta_2 = 0\ (F_1), \qquad\theta_2 = \theta_1\ (F_2), \qquad\lambda,\alpha_T \ge1,\\
F_3:\quad&\epsilon_2 = -2\eta,\qquad\pi_1 = x\epsilon_1+\Gamma/x+2\eta,\qquad\pi_2 = 4\eta/3,\qquad\theta_1 = \eta,\\
&\theta_2 = -(\epsilon_1+\pi_1)/2,\qquad\epsilon_1 \ge\frac{\Gamma-\eta(1-x)}{x(1-x)}. \tag*{(32)}
\end{aligned} F 1 , F 2 : F 3 : ( ϵ 1 , ϵ 2 , π 1 , π 2 , θ 1 ) = ( e , 0 , x e + Γ/ x , 4 η /3 , α T η ) , e = x ( 1 − x ) λ Γ , θ 2 = 0 ( F 1 ) , θ 2 = θ 1 ( F 2 ) , λ , α T ≥ 1 , ϵ 2 = − 2 η , π 1 = x ϵ 1 + Γ/ x + 2 η , π 2 = 4 η /3 , θ 1 = η , θ 2 = − ( ϵ 1 + π 1 ) /2 , ϵ 1 ≥ x ( 1 − x ) Γ − η ( 1 − x ) . (32)
These bounds ensure ε 1 , θ 1 > 0 \varepsilon_1,\theta_1>0 ε 1 , θ 1 > 0 , so the gapped modes in (21) are damped. For F 3 F_3 F 3 , positivity follows from Γ − η ( 1 − x ) = ζ ∗ + η ( x + 1 / 3 ) > 0 \Gamma-\eta(1-x)=\zeta_*+\eta(x+1/3)>0 Γ − η ( 1 − x ) = ζ ∗ + η ( x + 1/3 ) > 0 . Here λ \lambda λ increases the scalar relaxation coefficient and α T \alpha_T α T increases the shear relaxation time relative to their causal lower bounds and can be chosen freely. At λ = α T = 1 \lambda=\alpha_T=1 λ = α T = 1 , the family F 1 F_1 F 1 has π 1 = ε 1 \pi_1=\varepsilon_1 π 1 = ε 1 . Substituting its coefficients (32) into the characteristic-speed formulas (24) gives y − = 0 y_-=0 y − = 0 and y + = y T = 1 y_+=y_T=1 y + = y T = 1 : one longitudinal characteristic speed vanishes, while the other longitudinal speed and the shear speed equal the speed of light. For positive limiting shear viscosity, these families do not admit conformal limits, as shown in Section A.1.4. We now return to the general branch to construct families with conformal limits.
Neutral families with finite conformal limits
We now construct the neutral families F 4 F_4 F 4 , F 5 F_5 F 5 and F 6 F_6 F 6 by solving the reduced inequalities and then imposing the finite conformal limits required by (28) . To construct F 4 F_4 F 4 , we return to the general branch (31) with π 2 = 4 η / 3 \pi_2=4\eta/3 π 2 = 4 η /3 and Σ = 1 \Sigma=1 Σ = 1 , before imposing F 3 F_3 F 3 ’s choices ε 2 = − 2 η \varepsilon_2=-2\eta ε 2 = − 2 η and θ 1 = η \theta_1=\eta θ 1 = η . A well defined conformal limit requires ε 2 → 3 π 2 = 4 η c > 0 \varepsilon_2\to3\pi_2=4\eta_c>0 ε 2 → 3 π 2 = 4 η c > 0 when the limiting shear viscosity η c \eta_c η c is positive. We therefore choose ε 2 \varepsilon_2 ε 2 and π 1 \pi_1 π 1 both positive and parametrise their freedom in the following way. Entropy matching (11) fixes only the sum ε 2 + π 1 = x ε 1 + Γ / x \varepsilon_2+\pi_1=x\varepsilon_1+\Gamma/x ε 2 + π 1 = x ε 1 + Γ/ x so we can parametrise the remaining freedom by ε 1 = L Γ \varepsilon_1=L\Gamma ε 1 = L Γ , θ 1 = K Γ \theta_1=K\Gamma θ 1 = K Γ for some L L L , and a fraction f f f of that sum assigned to π 1 \pi_1 π 1 . The luminal condition (31) then fixes θ 2 \theta_2 θ 2 , giving
ε 1 = L Γ , ε 2 = ( 1 − f ) S 0 Γ , π 1 = f S 0 Γ , S 0 = x L + 1 / x , \varepsilon_1=L\Gamma,\qquad\varepsilon_2=(1-f)S_0\Gamma,\qquad\pi_1=fS_0\Gamma,\qquad S_0=xL+1/x, ε 1 = L Γ , ε 2 = ( 1 − f ) S 0 Γ , π 1 = f S 0 Γ , S 0 = xL + 1/ x ,
π 2 = 4 η / 3 , θ 1 = K Γ , θ 2 = Γ [ K ( L − f S 0 ) ( 1 − f ) S 0 − f S 0 ] . (33) \pi_2=4\eta/3,\qquad\theta_1=K\Gamma,\qquad\theta_2=\Gamma\left[\frac{K(L-fS_0)}{(1-f)S_0}-fS_0\right]. \tag*{(33)} π 2 = 4 η /3 , θ 1 = K Γ , θ 2 = Γ [ ( 1 − f ) S 0 K ( L − f S 0 ) − f S 0 ] . (33)
The choices L ≥ 2 / [ x ( 1 − x ) ] L\ge2/[x(1-x)] L ≥ 2/ [ x ( 1 − x )] , K ≥ 3 / 4 K\ge3/4 K ≥ 3/4 and 0 < f < 1 0<f<1 0 < f < 1 solve the remaining bounds as they give R ≤ ( 1 + x ) / 2 < 1 R\le(1+x)/2<1 R ≤ ( 1 + x ) /2 < 1 , θ 1 ≥ η \theta_1\ge\eta θ 1 ≥ η and a nonzero denominator in θ 2 \theta_2 θ 2 . Thus Φ = 0 \Phi=0 Φ = 0 , Σ = 1 \Sigma=1 Σ = 1 and L = R ( 1 − R ) > 0 \mathcal{L}=R(1-R)>0 L = R ( 1 − R ) > 0 satisfying (25) . These bounds establish stability and causality but not yet a conformal limit. In order to do so we assume that η c > 0 \eta_c>0 η c > 0 denotes the limiting conformal form of the shear viscosity. Since π 2 → 4 η c / 3 \pi_2\to4\eta_c/3 π 2 → 4 η c /3 , Step 3 in § 1.3.3 fixes Π = 4 η c \Pi=4\eta_c Π = 4 η c . The limits ε 1 → 3 Π \varepsilon_1\to3\Pi ε 1 → 3Π and ε 2 , π 1 → Π \varepsilon_2,\pi_1\to\Pi ε 2 , π 1 → Π therefore require L → 9 L\to9 L → 9 and f → 1 / 2 f\to1/2 f → 1/2 . Using these scalar limits and θ 1 → θ \theta_1\to\theta θ 1 → θ , the luminal condition (31) gives θ 2 → 2 θ − 4 η c \theta_2\to2\theta-4\eta_c θ 2 → 2 θ − 4 η c . Requiring θ 2 → θ \theta_2\to\theta θ 2 → θ fixes θ = 4 η c \theta=4\eta_c θ = 4 η c , hence K → 3 K\to3 K → 3 . These implement the conformal restrictions leading to a family F 4 F_4 F 4 with a well defined conformal limit and is in general a non-thermodynamic frame.6
In turn, to construct the family F 5 F_5 F 5 with a conformal limit, we first pick the F 4 F_4 F 4 scalar parameters to be L = 2 / [ x ( 1 − x ) ] L=2/[x(1-x)] L = 2/ [ x ( 1 − x )] and f = 1 / 2 f=1/2 f = 1/2 , which satisfy the preceding bounds and conformal restrictions and imply m = ε 2 = π 1 m=\varepsilon_2=\pi_1 m = ε 2 = π 1 . Contrary to the family F 4 F_4 F 4 , we now impose a thermodynamic frame by choosing θ 1 = θ 2 = θ 0 \theta_1=\theta_2=\theta_0 θ 1 = θ 2 = θ 0 and hence (31) gives ( ε 1 − 2 m ) θ 0 = m 2 (\varepsilon_1-2m)\theta_0=m^2 ( ε 1 − 2 m ) θ 0 = m 2 . This determines the parameters
ε 1 = 2 Γ x ( 1 − x ) , m = Γ ( 1 + x ) 2 x ( 1 − x ) , π 2 = 4 3 η , θ 0 = Γ ( 1 + x ) 2 4 x ( 1 − x ) 2 . (34) \varepsilon_1 = \frac{2\Gamma}{x(1-x)}, \qquad m = \frac{\Gamma(1+x)}{2x(1-x)}, \qquad\pi_2 = \frac{4}{3}\eta, \qquad\theta_0 = \frac{\Gamma(1+x)^2}{4x(1-x)^2}. \tag*{(34)} ε 1 = x ( 1 − x ) 2Γ , m = 2 x ( 1 − x ) Γ ( 1 + x ) , π 2 = 3 4 η , θ 0 = 4 x ( 1 − x ) 2 Γ ( 1 + x ) 2 . (34)
Keeping these parameters as they are, we can deform away from the luminal branch Φ = 0 \Phi= 0 Φ = 0 , Σ = 1 \Sigma= 1 Σ = 1 by increasing slightly the coefficients θ 1 , θ 2 \theta_1,\theta_2 θ 1 , θ 2 such that θ 1 = θ 2 = λ θ 0 \theta_1 = \theta_2 = \lambda\theta_0 θ 1 = θ 2 = λ θ 0 , with λ > 1 \lambda> 1 λ > 1 . In this case the condition Φ = 0 \Phi= 0 Φ = 0 remains unchanged, while
Σ = ρ 0 = 1 + x 2 + 1 − x 2 λ < 1. \Sigma= \rho_0 = \frac{1+x}{2} + \frac{1-x}{2\lambda} < 1. Σ = ρ 0 = 2 1 + x + 2 λ 1 − x < 1.
For these choices one finds that R < ρ 0 R < \rho_0 R < ρ 0 , and λ θ 0 > η \lambda\theta_0 > \eta λ θ 0 > η also makes the shear speed (24) strictly subluminal. It is possible to slightly generalise this family to non-thermodynamic frames by keeping θ 1 = λ θ 0 \theta_1 = \lambda\theta_0 θ 1 = λ θ 0 and choosing θ 2 = θ 1 + χ ( λ − 1 ) m \theta_2 = \theta_1 + \chi(\lambda- 1)m θ 2 = θ 1 + χ ( λ − 1 ) m , where
λ > 1 , χ = β ( 3 x − 1 ) 2 1 + ( 3 x − 1 ) 2 , 0 ≤ β ≤ 1. (35) \lambda> 1, \qquad\chi= \beta\frac{(3x-1)^2}{1+(3x-1)^2}, \qquad0 \le\beta\le1. \tag*{(35)} λ > 1 , χ = β 1 + ( 3 x − 1 ) 2 ( 3 x − 1 ) 2 , 0 ≤ β ≤ 1. (35)
Here the parameter β \beta β parametrises the freedom in choosing χ \chi χ . This changes the upper squared sound speed y + y_+ y + (24) by a fraction χ < 1 \chi< 1 χ < 1 of the available margin to the light cone:
y − = 0 , y + = ρ = ρ 0 + χ ( 1 − ρ 0 ) = 1 − ( 1 − x ) ( λ − 1 ) ( 1 − χ ) 2 λ < 1. y_- = 0, \qquad y_+ = \rho= \rho_0 + \chi(1-\rho_0) = 1 - \frac{(1-x)(\lambda-1)(1-\chi)}{2\lambda} < 1. y − = 0 , y + = ρ = ρ 0 + χ ( 1 − ρ 0 ) = 1 − 2 λ ( 1 − x ) ( λ − 1 ) ( 1 − χ ) < 1.
Hence R < ρ 0 ≤ ρ < 1 R < \rho_0 \le\rho< 1 R < ρ 0 ≤ ρ < 1 preserves the reduced stability and causality inequalities. The explicit positive margins are proved in Section A.1.5. Since χ → 0 \chi\to0 χ → 0 at conformality, finite parameters with λ → λ c > 1 \lambda\to\lambda_c > 1 λ → λ c > 1 give the Step 3 limits in Section 1.2.5 Π = 4 η c \Pi= 4\eta_c Π = 4 η c and θ = 4 λ c η c \theta= 4\lambda_c\eta_c θ = 4 λ c η c . Thus F 5 F_5 F 5 achieves strict subluminality without losing its conformal limit while its lower sound speed y − y_- y − remains zero.
The family F 6 F_6 F 6 that we now describe allows for non-zero lower sound speed. In order to construct it, we choose π 2 > 4 η / 3 \pi_2 > 4\eta/3 π 2 > 4 η /3 and θ 2 > 0 \theta_2 > 0 θ 2 > 0 , which makes Φ > 0 \Phi> 0 Φ > 0 in (20) . We introduce a parameter Λ \Lambda Λ to parametrise this freedom, while the function χ \chi χ in (35) controls the difference between the θ 1 \theta_1 θ 1 and θ 2 \theta_2 θ 2 . Choosing 1 < Λ ≤ 2 1 < \Lambda\le2 1 < Λ ≤ 2 , 0 ≤ β ≤ 1 0 \le\beta\le1 0 ≤ β ≤ 1 , this leads to
ε 1 = 2 Λ Γ x ( 1 − x ) , ε 2 = π 1 = Λ Γ ( 1 + x ) 2 x ( 1 − x ) , π 2 = 4 3 η + ( Λ − 1 ) Γ , θ 1 = Λ Γ ( 1 + x ) 2 2 x ( 1 − x ) 2 , θ 2 = θ 1 [ 1 + 4 χ ( 1 − Σ 0 ) 1 + x ] , Σ 0 = 3 + x 4 − 2 ( Λ − 1 ) x ( 1 − x ) 2 Λ ( 1 + x ) 2 . (36) \begin{aligned}
\varepsilon_1 &= \frac{2\Lambda\Gamma}{x(1-x)}, & \varepsilon_2 = \pi_1 &= \frac{\Lambda\Gamma(1+x)}{2x(1-x)},\\
\pi_2 &= \frac{4}{3}\eta+ (\Lambda-1)\Gamma, & \theta_1 &= \frac{\Lambda\Gamma(1+x)^2}{2x(1-x)^2},\\
\theta_2 &= \theta_1\left[1+\frac{4\chi(1-\Sigma_0)}{1+x}\right], & \Sigma_0 &= \frac{3+x}{4} - \frac{2(\Lambda-1)x(1-x)^2}{\Lambda(1+x)^2}. \tag*{(36)}
\end{aligned} ε 1 π 2 θ 2 = x ( 1 − x ) 2ΛΓ , = 3 4 η + ( Λ − 1 ) Γ , = θ 1 [ 1 + 1 + x 4 χ ( 1 − Σ 0 ) ] , ε 2 = π 1 θ 1 Σ 0 = 2 x ( 1 − x ) ΛΓ ( 1 + x ) , = 2 x ( 1 − x ) 2 ΛΓ ( 1 + x ) 2 , = 4 3 + x − Λ ( 1 + x ) 2 2 ( Λ − 1 ) x ( 1 − x ) 2 . (36)
The squared sound speed combinations defined in (22) are then
Σ = Σ 0 + χ ( 1 − Σ 0 ) , Φ = ( Λ − 1 ) x ( 1 − x ) 2 Λ + 2 ( Λ − 1 ) x ( 1 − x ) Λ ( 1 + x ) χ ( 1 − Σ 0 ) , R = x + 2 x ( 1 − x ) 2 Λ ( 1 + 6 x − 3 x 2 ) . (37) \begin{aligned}
\Sigma&= \Sigma_0+\chi(1-\Sigma_0), & \Phi&= \frac{(\Lambda-1)x(1-x)}{2\Lambda} + \frac{2(\Lambda-1)x(1-x)}{\Lambda(1+x)}\chi(1-\Sigma_0),\\
R &= x+\frac{2x(1-x)^2}{\Lambda(1+6x-3x^2)}. \tag*{(37)}
\end{aligned} Σ R = Σ 0 + χ ( 1 − Σ 0 ) , = x + Λ ( 1 + 6 x − 3 x 2 ) 2 x ( 1 − x ) 2 . Φ = 2Λ ( Λ − 1 ) x ( 1 − x ) + Λ ( 1 + x ) 2 ( Λ − 1 ) x ( 1 − x ) χ ( 1 − Σ 0 ) , (37)
and satisfy Φ > 0 \Phi> 0 Φ > 0 , Σ < 1 \Sigma< 1 Σ < 1 and L > 0 \mathcal{L} > 0 L > 0 given in (25) . The resulting neutral speed bounds are
0 < y − < R < y + < 1 , 0 < y T < 1 2 Λ < 1. (38) 0 < y_{-} < R < y_{+} < 1,\qquad0 < y_{T} < \frac{1}{2\Lambda} < 1. \tag*{(38)} 0 < y − < R < y + < 1 , 0 < y T < 2Λ 1 < 1. (38)
The choice β = 0 \beta= 0 β = 0 yields a thermodynamic frame while β > 0 \beta> 0 β > 0 permits non-thermodynamic frames away from x = 1 / 3 x = 1/3 x = 1/3 . Bounded parameters with finite conformal limits yield finite coefficients. In particular, Λ → Λ c \Lambda\to\Lambda_{c} Λ → Λ c and χ → 0 \chi\to0 χ → 0 give the Step 3 limits (see § 1.3.3 ) Π = 4 Λ c η c \Pi= 4\Lambda_{c}\eta_{c} Π = 4 Λ c η c and θ = 8 Λ c η c \theta= 8\Lambda_{c}\eta_{c} θ = 8 Λ c η c . For η c > 0 \eta_{c} > 0 η c > 0 and 1 < Λ c ≤ 2 1 < \Lambda_{c} \le2 1 < Λ c ≤ 2 , both longitudinal speeds remain nonzero and strictly subluminal at conformality. The full derivation is given in Section A.1.5.
Charged uplift: an additional exact telegrapher factor
Having constructed the neutral families F 1 F_{1} F 1 –F 3 F_{3} F 3 , F 4 F_{4} F 4 , F 5 F_{5} F 5 and F 6 F_{6} F 6 , we now provide the coefficient map between neutral and charged frames implementing Step 4 in § 1.3.3 . Assuming D > 0 \mathcal{D} > 0 D > 0 , we take D q \mathcal{D}_{q} D q , τ q \tau_{q} τ q , ζ ∗ \zeta_{*} ζ ∗ from (29) and evaluate the chosen neutral families at the charged values of w w w , x x x , η \eta η , with Γ = 4 η / 3 + ζ ∗ \Gamma= 4\eta/3 + \zeta_{*} Γ = 4 η /3 + ζ ∗ . In addition we retain the values of the six stress coefficients ϵ 1 , 2 \epsilon_{1,2} ϵ 1 , 2 , π 1 , 2 \pi_{1,2} π 1 , 2 , θ 1 , 2 \theta_{1,2} θ 1 , 2 as given by each family and set the charge coefficients according to
ϵ 3 = r T ϵ 1 , π 3 = r T π 1 , θ 3 = r T θ 2 , (39) \epsilon_{3} = rT\epsilon_{1},\qquad\pi_{3} = rT\pi_{1},\qquad\theta_{3} = rT\theta_{2}, \tag*{(39)} ϵ 3 = r T ϵ 1 , π 3 = r T π 1 , θ 3 = r T θ 2 , (39)
( ν 1 , ν 2 , ν 3 ) = ( r ϵ 1 + T τ q q T , r ϵ 2 + T D q q T , r 2 T ϵ 1 + τ q q α ) , ( γ 1 , γ 2 , γ 3 ) = ( r θ 1 − T D q q T , r θ 2 − T D q q T , r 2 T θ 2 − D q q α ) . (40) \begin{aligned}
(\nu_{1},\nu_{2},\nu_{3}) &= (r\epsilon_{1} + T\tau_{q}q_{T},\ r\epsilon_{2} + T\mathcal{D}_{q}q_{T},\ r^{2}T\epsilon_{1} + \tau_{q}q_{\alpha}),\\
(\gamma_{1},\gamma_{2},\gamma_{3}) &= (r\theta_{1} - T\mathcal{D}_{q}q_{T},\ r\theta_{2} - T\mathcal{D}_{q}q_{T},\ r^{2}T\theta_{2} - \mathcal{D}_{q}q_{\alpha}). \tag*{(40)}
\end{aligned} ( ν 1 , ν 2 , ν 3 ) ( γ 1 , γ 2 , γ 3 ) = ( r ϵ 1 + T τ q q T , r ϵ 2 + T D q q T , r 2 T ϵ 1 + τ q q α ) , = ( r θ 1 − T D q q T , r θ 2 − T D q q T , r 2 T θ 2 − D q q α ) . (40)
These coefficients preserve the entropy identities η \eta η , ζ \zeta ζ , D \mathcal{D} D in (11) , factorise the polynomial into a telegrapher part times the quartic and fix the normalization in (27) such that
det M c h a r g e d = − Δ J w F q ( ς , q ) P N ( ς , q ) . (41) \det M_{\mathrm{charged}} = -\frac{\Delta_{J}}{w}F_{q}(\varsigma,q)P_{N}(\varsigma,q). \tag*{(41)} det M charged = − w Δ J F q ( ς , q ) P N ( ς , q ) . (41)
We verify this factorization explicitly in Appendix § 3 .2.2 while stability at arbitrary boosts is verified in Appendix § 3 .3.3. With this map, choosing α q > 1 \alpha_{q} > 1 α q > 1 makes the extra squared speed y q = 1 / α q y_{q} = 1/\alpha_{q} y q = 1/ α q strictly subluminal as described in § 1.2 . For a regular conformal EoS limit with ζ → 0 \zeta\to0 ζ → 0 , one has B B B , q T → 0 q_{T} \to0 q T → 0 and ζ ∗ → 0 \zeta_{*} \to0 ζ ∗ → 0 leading to finite charged coefficients satisfying the conformal requirements (28) .
We note that in the case of the F 6 F_{6} F 6 family, choosing α q > 1 \alpha_{q} > 1 α q > 1 , the bounds (38) , together with 0 < y q = 1 / α q < 1 0 < y_{q} = 1/\alpha_{q} < 1 0 < y q = 1/ α q < 1 , make every comoving characteristic speed nonzero and strictly subluminal when η , D > 0 \eta,\mathcal{D} > 0 η , D > 0 . As a useful example, at the conformal point, picking Λ = α q = 2 \Lambda= \alpha_{q} = 2 Λ = α q = 2 gives
( ϵ 1 , ϵ 2 , π 1 , π 2 , θ 1 , θ 2 ) = η ( 24 , 8 , 8 , 8 / 3 , 16 , 16 ) , ( y − , y + , y T , y q ) = ( 1 / 12 , 2 / 3 , 1 / 16 , 1 / 2 ) . (42) (\epsilon_{1},\epsilon_{2},\pi_{1},\pi_{2},\theta_{1},\theta_{2}) = \eta(24,8,8,8/3,16,16),\qquad(y_{-},y_{+},y_{T},y_{q}) = (1/12,2/3,1/16,1/2). \tag*{(42)} ( ϵ 1 , ϵ 2 , π 1 , π 2 , θ 1 , θ 2 ) = η ( 24 , 8 , 8 , 8/3 , 16 , 16 ) , ( y − , y + , y T , y q ) = ( 1/12 , 2/3 , 1/16 , 1/2 ) . (42)
as well as q T = 0 q_{T} = 0 q T = 0 and Ξ = q α \Xi= q_{\alpha} Ξ = q α so (29) gives D q q α = D T \mathcal{D}_{q}q_{\alpha} = \mathcal{D}T D q q α = D T and τ q q α = 2 D T \tau_{q}q_{\alpha} = 2\mathcal{D}T τ q q α = 2 D T . The current coefficients in (40) therefore become
( ν 1 , ν 2 , ν 3 ) = ( 24 r η , 8 r η , 24 r 2 T η + 2 D T ) , (\nu_{1},\nu_{2},\nu_{3}) = (24r\eta,8r\eta,24r^{2}T\eta+ 2\mathcal{D}T), ( ν 1 , ν 2 , ν 3 ) = ( 24 r η , 8 r η , 24 r 2 T η + 2 D T ) ,
( γ 1 , γ 2 , γ 3 ) = ( 16 r η , 16 r η , 16 r 2 T η − D T ) . (43) (\gamma_{1},\gamma_{2},\gamma_{3}) = (16r\eta,16r\eta,16r^{2}T\eta- \mathcal{D}T). \tag*{(43)} ( γ 1 , γ 2 , γ 3 ) = ( 16 r η , 16 r η , 16 r 2 T η − D T ) . (43)
Here r = η / w r = \eta/w r = η / w and the physical conductivity D \mathcal{D} D remains a free parameter, in particular, ν 1 = 3 ν 2 \nu_{1} = 3\nu_{2} ν 1 = 3 ν 2 and γ 1 = γ 2 \gamma_{1} = \gamma_{2} γ 1 = γ 2 satisfy the conformal current relations in (28) . We overview all families in the next subsection.
Catalogue of families of frames and remaining freedom
Table 1 collects the six families. Each has a neutral version and a charged uplift; the latter uses Γ \Gamma Γ from (20) and adds the charge speed y q y_q y q . Families F 1 F_1 F 1 –F 3 F_3 F 3 do not have a finite conformal limit with positive limiting shear viscosity. Families F 4 F_4 F 4 –F 6 F_6 F 6 do, subject to the parameter limits specified in § 1.4.2 . Some families overlap at limiting parameter values. As λ ↓ 1 \lambda\downarrow1 λ ↓ 1 , F 5 F_5 F 5 approaches the thermodynamic member of F 4 F_4 F 4 with
Table 1. Squared comoving characteristic speeds, with η > 0 \eta>0 η > 0 and, for charged uplifts, D > 0 D>0 D > 0 . Here ρ = 1 − ( 1 − x ) ( λ − 1 ) ( 1 − χ ) / ( 2 λ ) < 1 \rho=1-(1-x)(\lambda-1)(1-\chi)/(2\lambda)<1 ρ = 1 − ( 1 − x ) ( λ − 1 ) ( 1 − χ ) / ( 2 λ ) < 1 in F 5 F_{5} F 5 while F 6 F_{6} F 6 uses (37) . Each charged uplift adds y q = 1 / α q y_{q}=1/\alpha_{q} y q = 1/ α q , with finite α q ≥ 1 \alpha_{q}\ge1 α q ≥ 1 (strictly > 1 >1 > 1 for F 5 F_{5} F 5 and F 6 F_{6} F 6 ). The shear speed y T y_{T} y T occurs in two transverse polarizations. The column with the conformal limit assumes a regular endpoint with η c > 0 \eta_{c}>0 η c > 0 . In F 5 F_{5} F 5 and F 6 F_{6} F 6 , χ \chi χ is defined by (35) and vanishes at conformality even for β > 0 \beta>0 β > 0 . In F 4 F_{4} F 4 , the thermodynamic subfamily is selected by θ 1 = θ 2 \theta_{1}=\theta_{2} θ 1 = θ 2 in (33) , which involves a relation among L L L , K K K , f f f . For vanishing transport coefficients, the reduced modes are described in footnotes 7, 11 and 12.
L = 2 x ( 1 − x ) , f = 1 2 , K = ( 1 + x ) 2 4 x ( 1 − x ) 2 , (44) L = \frac{2}{x(1-x)}, \qquad f = \frac{1}{2}, \qquad K = \frac{(1+x)^2}{4x(1-x)^2}, \tag*{(44)} L = x ( 1 − x ) 2 , f = 2 1 , K = 4 x ( 1 − x ) 2 ( 1 + x ) 2 , (44)
for which θ 1 = θ 2 = θ 0 \theta_1=\theta_2=\theta_0 θ 1 = θ 2 = θ 0 and y + = 1 y_+=1 y + = 1 . As Λ ↓ 1 \Lambda\downarrow1 Λ ↓ 1 , F 6 F_6 F 6 approaches the λ = 2 \lambda=2 λ = 2 member of F 5 F_5 F 5 with the same β \beta β , while y − y_- y − reaches zero. These relations are proved in Section A.1.5.
The catalogue is not exhaustive. Even after fixing the charged telegrapher factor, some coefficient changes preserve the spectrum and the physical transport. For arbitrary regular functions f E , f P , f Q f_E,f_P,f_Q f E , f P , f Q of the allowed thermodynamic and transport data, let δ \delta δ denote the additive changes
( δ ε 1 / T , δ ε 3 ) = f E ( q T , q α ) , ( δ π 1 / T , δ π 3 ) = f P ( q T , q α ) , ( δ θ 2 / T , δ θ 3 ) = f Q ( q T , q α ) , (45) \begin{aligned}
(\delta\varepsilon_1/T,\delta\varepsilon_3) &= f_E(q_T,q_\alpha), \qquad
(\delta\pi_1/T,\delta\pi_3) = f_P(q_T,q_\alpha), \\
(\delta\theta_2/T,\delta\theta_3) &= f_Q(q_T,q_\alpha), \tag*{(45)}
\end{aligned} ( δ ε 1 / T , δ ε 3 ) ( δ θ 2 / T , δ θ 3 ) = f E ( q T , q α ) , ( δ π 1 / T , δ π 3 ) = f P ( q T , q α ) , = f Q ( q T , q α ) , (45)
with δ ν 1 , 3 = r δ ε 1 , 3 \delta\nu_{1,3}=r\delta\varepsilon_{1,3} δ ν 1 , 3 = r δ ε 1 , 3 , δ γ 2 , 3 = r δ θ 2 , 3 \delta\gamma_{2,3}=r\delta\theta_{2,3} δ γ 2 , 3 = r δ θ 2 , 3 and all other coefficient shifts zero. These changes preserve the full determinant, the shear polynomial and the entropy invariants. Finite conformal matching requires, in particular, f E − 3 f P → 0 f_E-3f_P\to0 f E − 3 f P → 0 . For a fixed neutral representative and fixed charge factor, these are the general changes within the charge-row construction used here, as proved in Appendix § 3 .2.4. This is a completeness statement for that row identity; it does not classify all possible ways of obtaining a telegrapher factor in a charged determinant.
Discussion and numerical use
Table 1 provides explicit coefficient prescriptions that can be used in numerical implementations of neutral and charged BDNK hydrodynamics. Given an equation of state and prescribed η \eta η , ζ \zeta ζ , D D D , one evaluates the thermodynamic quantities, checks (1.8)–(1.9), and selects frame parameters satisfying the bounds of the chosen family. The neutral coefficients and the common charged uplift are then given in Sections 1.3.1–1.3.3. The construction avoids solving the full stability inequalities anew for each equation of state.
The families allow different choices of characteristic structure while keeping physical transport fixed. With positive shear viscosity and conductivity, F 6 F_{6} F 6 has strictly positive, strictly subluminal squared comoving characteristic speeds, including at its finite conformal endpoint. It is therefore a natural starting point. Families F 1 F_{1} F 1 -F 5 F_{5} F 5 instead allow a zero sound characteristic speed, and some also allow luminal propagation. It would be interesting to test these families in numerical implementations. In particular, various numerical studies in multiple contexts could provide useful tests of these BDNK families [10 , 11 , 12 , 13 , 14 , 32 , 33 , 34 , 35 ] .
Nonlinear causality and local existence and uniqueness in Gevrey classes have been established for several neutral constitutive classes [5 , 7 , 28 ] . Applying these results requires matching the constitutive relations as well as checking the coefficient, regularity and initial-data hypotheses. For the thermodynamic neutral choices θ 1 = θ 2 \theta_{1}=\theta_{2} θ 1 = θ 2 , the coefficients used in Refs. [7 , 29 ] are related to ours by
( χ 1 , χ 2 , χ 3 , χ 4 ) = ( x ε 1 , ε 2 , x π 1 , π 2 ) , λ B D N = θ 1 = θ 2 . (\chi_{1},\chi_{2},\chi_{3},\chi_{4})=(x\varepsilon_{1},\varepsilon_{2},x\pi_{1},\pi_{2}),\qquad\lambda_{\mathrm{BDN}}=\theta_{1}=\theta_{2}. ( χ 1 , χ 2 , χ 3 , χ 4 ) = ( x ε 1 , ε 2 , x π 1 , π 2 ) , λ BDN = θ 1 = θ 2 .
Under this identification, our causal bounds imply the algebraic conditions of Ref. [7 ] . Consequently, its results on local existence, uniqueness and nonlinear causality apply to the thermodynamic neutral members of our families, provided the constitutive functions and initial data satisfy the regularity and positivity assumptions stated there.
Under the same coefficient identification, the neutral F 6 F_6 F 6 subfamily with β = 0 \beta= 0 β = 0 , constant 1 < Λ ≤ 2 1 < \Lambda\le2 1 < Λ ≤ 2 and η > 0 \eta> 0 η > 0 satisfies the algebraic conditions of Ref. [29 ] . Its result on local well-posedness in Sobolev spaces therefore applies in Minkowski spacetime, provided the EoS and constitutive functions satisfy the analyticity and positivity assumptions stated there, and the initial data satisfy the required Sobolev regularity and compatibility conditions.
For charged fluids, Ref. [9 ] establishes nonlinear causality and local well-posedness in a constitutive class with J μ = n u μ J^\mu= n u^\mu J μ = n u μ . Our current contains additional derivative terms, so those results do not apply directly. Nonlinear causality conditions for generic charged frames, including non-thermodynamic frames, were derived in Ref. [24 ] from the characteristics of the full nonlinear equations. All six charged frame families constructed here satisfy these conditions for η > 0 \eta> 0 η > 0 and D > 0 \mathcal{D} > 0 D > 0 , within their stated thermodynamic and parameter domains. Strong hyperbolicity has also been established for a restricted class of charged constitutive relations in Ref. [27 ] . Those results do not apply directly to the general families constructed here. Their result does apply to the exactly conformal family F 6 F_6 F 6 with Λ = α q = 2 \Lambda= \alpha_q = 2 Λ = α q = 2 , provided the prescribed conductivity satisfies the additional restriction D = 4 η Δ J / ( 3 w 2 ) \mathcal{D} = 4\eta\Delta_J/(3w^2) D = 4 η Δ J / ( 3 w 2 ) . Establishing well-posedness for the full families constructed here remains a further step beyond their linear stability and characteristic-speed bounds.
A natural extension of the method developed here is magnetohydrodynamics (MHD) for which a BDNK approach has been developed in various works [30 , 31 , 36 , 37 ] . These developments motivate seeking coefficient families that keep anisotropic viscosities, resistivities and other physical transport prescribed. The magnetic field introduces direction dependent Alfvén and magnetosonic sectors and a magnetic constraint, so the neutral-to-charged map we established cannot simply be reused but the overall strategy of telegrapher factorisation of the longitudinal polynomial may work. The same strategy may also be useful for chiral fluids [24 ] and for relativistic superfluids [38 ] . It would be interesting to apply the AI research protocol developed through this work to these more complicated cases.
Acknowledgements
J. Armas is partly funded by the Dutch Institute for Emergent Phenomena (DIEP) cluster at the University of Amsterdam via the DIEP programme Foundations and Applications of Emergence (FAEME) and the national NWA-ORC 2023 consortium EAAS. The author used OpenAI Codex Astra 6 Pro to assist with manuscript revision, literature cross-checks, and review of mathematical arguments and analysis code. The author performed human checks on the AI-assisted material and takes responsibility for the final content and conclusions.
Derivations, checks and proofs
This appendix is organized into three parts. Section Section A.1 derives the neutral entropy production, rest-frame stability and causality conditions, and frame families. Section Section A.2 develops the charged uplift, verifies entropy matching and determinant factorization, and describes the remaining frame freedom. Section A.3 proves stability and causality for sound, shear and charge modes in arbitrary inertial frames. Throughout, we use the constitutive notation and thermodynamic data of § 1.1 , the Fourier convention in (12) , and the spectral and characteristic criteria (13) and (15) . All coefficients in the linear calculations are evaluated at ( T 0 , μ 0 ) (T_0,\mu_0) ( T 0 , μ 0 ) , with unsubscripted state quantities denoting those equilibrium values.
Neutral frames
For a neutral fluid, d p = s d T dp=s\,dT d p = s d T , d ϵ = T d s d\epsilon=T\,ds d ϵ = T d s and w = T s w=Ts w = T s imply
x = p T ϵ T = w T ϵ T ∈ ( 0 , 1 ) . (46) x=\frac{p_T}{\epsilon_T}=\frac{w}{T\epsilon_T}\in(0,1). \tag*{(46)} x = ϵ T p T = T ϵ T w ∈ ( 0 , 1 ) . (46)
We require T , w > 0 T,w>0 T , w > 0 and the viscosity bounds in (10) , and set ζ ∗ = ζ \zeta_*=\zeta ζ ∗ = ζ in the neutral calculations. The conformal coefficients Π , θ \Pi,\theta Π , θ are those introduced in Step 3 of § 1.3.3 .
Neutral entropy production
Here we re-derive neutral entropy production in a general frame in order to highlight the physical transport coefficients and their constraints. We write T ( 1 ) μ ν T^{\mu\nu}_{(1)} T ( 1 ) μν for the first-derivative correction in (2) . The canonical neutral entropy current is
T S μ = p u μ − u ν T μ ν , S μ = s u μ − u ν T ( 1 ) μ ν T . (47) TS^\mu=pu^\mu-u_\nu T^{\mu\nu},\qquad S^\mu=su^\mu-\frac{u_\nu T^{\mu\nu}_{(1)}}{T}. \tag*{(47)} T S μ = p u μ − u ν T μν , S μ = s u μ − T u ν T ( 1 ) μν . (47)
The energy equation gives D ϵ + w Θ = u ν ∇ μ T ( 1 ) μ ν D\epsilon+w\Theta=u_\nu\nabla_\mu T^{\mu\nu}_{(1)} D ϵ + w Θ = u ν ∇ μ T ( 1 ) μν . Using d ϵ = T d s d\epsilon=T\,ds d ϵ = T d s and w = T s w=Ts w = T s , one therefore obtains
∇ μ ( s u μ ) = u ν T ∇ μ T ( 1 ) μ ν , (48) \nabla_\mu(su^\mu)=\frac{u_\nu}{T}\nabla_\mu T^{\mu\nu}_{(1)}, \tag*{(48)} ∇ μ ( s u μ ) = T u ν ∇ μ T ( 1 ) μν , (48)
∇ μ S μ = − T ( 1 ) μ ν ∇ μ ( u ν T ) . (49) \nabla_\mu S^\mu=-T^{\mu\nu}_{(1)}\nabla_\mu\left(\frac{u_\nu}{T}\right). \tag*{(49)} ∇ μ S μ = − T ( 1 ) μν ∇ μ ( T u ν ) . (49)
Using the operators of § 1.1 , direct contraction of (2) in (A.4) yields
T ∇ μ S μ = − ϵ 1 ( D ln T ) 2 − ( ϵ 2 + π 1 ) ( D ln T ) Θ − π 2 Θ 2 − θ 1 a μ a μ − ( θ 1 + θ 2 ) a μ ∇ ⊥ μ ln T − θ 2 ( ∇ ⊥ ln T ) 2 + η 2 σ μ ν σ μ ν + O ( ∂ 3 ) . (50) \begin{aligned}
T\nabla_\mu S^\mu={}&-\epsilon_1(D\ln T)^2-(\epsilon_2+\pi_1)(D\ln T)\Theta-\pi_2\Theta^2\\
&-\theta_1a_\mu a^\mu-(\theta_1+\theta_2)a_\mu\nabla_\perp^\mu\ln T-\theta_2(\nabla_\perp\ln T)^2\\
&+\frac{\eta}{2}\sigma_{\mu\nu}\sigma^{\mu\nu}+O(\partial^3). \tag*{(50)}
\end{aligned} T ∇ μ S μ = − ϵ 1 ( D ln T ) 2 − ( ϵ 2 + π 1 ) ( D ln T ) Θ − π 2 Θ 2 − θ 1 a μ a μ − ( θ 1 + θ 2 ) a μ ∇ ⊥ μ ln T − θ 2 ( ∇ ⊥ ln T ) 2 + 2 η σ μν σ μν + O ( ∂ 3 ) . (50)
The ideal energy and momentum equations imply
D ln T = − x Θ + O ( ∂ 2 ) , a μ + ∇ ⊥ μ ln T = O ( ∂ 2 ) . (51) D\ln T=-x\Theta+O(\partial^2),\qquad a^\mu+\nabla_\perp^\mu\ln T=O(\partial^2). \tag*{(51)} D ln T = − x Θ + O ( ∂ 2 ) , a μ + ∇ ⊥ μ ln T = O ( ∂ 2 ) . (51)
The vector contribution is − Q μ ( a μ + ∇ ⊥ μ ln T ) -Q_\mu(a^\mu+\nabla_\perp^\mu\ln T) − Q μ ( a μ + ∇ ⊥ μ ln T ) and is consequently O ( ∂ 3 ) O(\partial^3) O ( ∂ 3 ) even when θ 1 ≠ θ 2 \theta_1\ne\theta_2 θ 1 = θ 2 . Thus
T ∇ μ S μ = ζ ∗ Θ 2 + η 2 σ μ ν σ μ ν + O ( ∂ 3 ) , (52) T\nabla_\mu S^\mu=\zeta_*\Theta^2+\frac{\eta}{2}\sigma_{\mu\nu}\sigma^{\mu\nu}+O(\partial^3), \tag*{(52)} T ∇ μ S μ = ζ ∗ Θ 2 + 2 η σ μν σ μν + O ( ∂ 3 ) , (52)
where ζ ∗ \zeta_* ζ ∗ is the neutral bulk viscosity frame-invariant in (20) . This derives that entropy relation from the canonical current. The squares in (A.7) are nonnegative, since the shear tensor is transverse. Matching the prescribed ζ \zeta ζ by ζ ∗ = ζ ≥ 0 \zeta_*=\zeta\ge0 ζ ∗ = ζ ≥ 0 , together with η ≥ 0 \eta\ge0 η ≥ 0 , proves the required second law at this order for on-shell configurations.
Neutral rest-frame sound matrix and polynomial
To obtain the matrix of linearised perturbations and sound polynomial, we linearize around the fluid at rest using longitudinal perturbations ( δ T / T 0 , δ u ∥ ) (\delta T/T_0,\delta u_{\parallel}) ( δ T / T 0 , δ u ∥ ) . With q = k 2 q=k^2 q = k 2 , the conservation matrix is
M ( ς , k ) = ( ϵ 1 ς 2 + w χ ς − θ 2 k 2 i k [ w + ( ϵ 2 + θ 1 ) ς ] i k [ w + ( θ 2 + π 1 ) ς ] θ 1 ς 2 + w ς + ( 4 3 η − π 2 ) k 2 ) . (53) M(\varsigma,k)=\begin{pmatrix}
\epsilon_1\varsigma^2+\frac{w}{\chi}\varsigma-\theta_2 k^2 & ik\left[w+(\epsilon_2+\theta_1)\varsigma\right] \\
ik\left[w+(\theta_2+\pi_1)\varsigma\right] & \theta_1\varsigma^2+w\varsigma+\left(\frac{4}{3}\eta-\pi_2\right)k^2
\end{pmatrix}. \tag*{(53)} M ( ς , k ) = ( ϵ 1 ς 2 + χ w ς − θ 2 k 2 ik [ w + ( θ 2 + π 1 ) ς ] ik [ w + ( ϵ 2 + θ 1 ) ς ] θ 1 ς 2 + w ς + ( 3 4 η − π 2 ) k 2 ) . (53)
Expanding P N ( ς , q ) = χ det M ( ς , k ) P_N(\varsigma,q)=\chi\det M(\varsigma,k) P N ( ς , q ) = χ det M ( ς , k ) and using the neutral bulk invariant gives exactly (19) with coefficients (20) . The positive normalization leaves the roots unchanged. In particular, setting π 2 = 4 η / 3 \pi_2=4\eta/3 π 2 = 4 η /3 makes F L F_L F L in (20) vanish, but leaves the physical sound attenuation unchanged. Expanding in small-k k k gives the usual sound mode
ω ± = ± χ k − i Γ 2 w k 2 + O ( k 3 ) . (54) \omega_{\pm}=\pm\sqrt{\chi}\,k-i\frac{\Gamma}{2w}k^2+O(k^3). \tag*{(54)} ω ± = ± χ k − i 2 w Γ k 2 + O ( k 3 ) . (54)
The rest frame gaps are those in Section 1.2. Other gaps must instead be extracted after boosting, as in Section A.3.1.
Neutral rest-frame stability and causality conditions
We now derive the reduced sound conditions used in Section 1.2. We first consider η > 0 \eta>0 η > 0 , so entropy matching gives Γ = 4 η / 3 + ζ > 0 \Gamma=4\eta/3+\zeta>0 Γ = 4 η /3 + ζ > 0 . The conditions ϵ 1 , θ 1 > 0 \epsilon_1,\theta_1>0 ϵ 1 , θ 1 > 0 follow from damping of the two rest-frame gaps in (21) .7 The shear sector additionally requires (18) , as derived in Section 1.2.2 and proved for arbitrary boosts in Section A.3.2. We therefore focus below on the longitudinal polynomial. The family constructions follow in Sections A.1.4 and A.1.5.
Reducing the Routh-Hurwitz inequalities for every momentum
To reduce the inequalities, we use R , Σ , Φ , L R,\Sigma,\Phi,\mathcal{L} R , Σ , Φ , L from (22) and the Hurwitz determinants (23) . Their highest-order terms follow from the identity
A L B L 2 L = D L ( B L C L − A L D L ) − B L 2 F L . (55) A_L B_L^2\mathcal{L}=D_L(B_L C_L-A_L D_L)-B_L^2F_L. \tag*{(55)} A L B L 2 L = D L ( B L C L − A L D L ) − B L 2 F L . (55)
For example, expanding D L q { B L ( w 2 + C L q ) − A L D L q } − B L 2 ( χ w 2 q + F L q 2 ) D_Lq\{B_L(w^2+C_Lq)-A_LD_Lq\}-B_L^2(\chi w^2q+F_Lq^2) D L q { B L ( w 2 + C L q ) − A L D L q } − B L 2 ( χ w 2 q + F L q 2 ) and using D L − χ B L = w Γ D_L-\chi B_L=w\Gamma D L − χ B L = w Γ gives Δ 3 \Delta_3 Δ 3 in (23) . With a 0 ( q ) = χ w 2 q + F L q 2 a_0(q)=\chi w^2q+F_Lq^2 a 0 ( q ) = χ w 2 q + F L q 2 , the Routh array is
ς 4 A L w 2 + C L q a 0 ( q ) ς 3 B L D L q 0 ς 2 Δ 2 / B L a 0 ( q ) 0 ς 1 Δ 3 / Δ 2 0 0 ς 0 a 0 ( q ) 0 0 (56) \begin{array}{c|cccc}
\varsigma^4 & A_L & w^2+C_Lq & a_0(q) \\
\varsigma^3 & B_L & D_Lq & 0 \\
\varsigma^2 & \Delta_2/B_L & a_0(q) & 0 \\
\varsigma^1 & \Delta_3/\Delta_2 & 0 & 0 \\
\varsigma^0 & a_0(q) & 0 & 0 \tag*{(56)}
\end{array} ς 4 ς 3 ς 2 ς 1 ς 0 A L B L Δ 2 / B L Δ 3 / Δ 2 a 0 ( q ) w 2 + C L q D L q a 0 ( q ) 0 0 a 0 ( q ) 0 0 0 0 (56)
For each q > 0 q > 0 q > 0 , strict stability requires the first column to be positive. Under the stated positivity assumptions, imposing this for every q > 0 q > 0 q > 0 gives
F L ≥ 0 , B L C L − A L D L ≥ 0 , L ≥ 0. (57) F_L \ge0,\qquad B_L C_L - A_L D_L \ge0,\qquad\mathcal{L} \ge0. \tag*{(57)} F L ≥ 0 , B L C L − A L D L ≥ 0 , L ≥ 0. (57)
For example, a negative coefficient of the highest power of q q q in a 0 a_0 a 0 , Δ 2 \Delta_2 Δ 2 , or Δ 3 \Delta_3 Δ 3 eventually violates the corresponding condition. Conversely, the positive lower-power terms ensure strict positivity at every finite q > 0 q > 0 q > 0 even if a displayed inequality saturates. Because D L > 0 D_L > 0 D L > 0 , the first and third conditions imply the second. Thus the rest-frame sound stability test reduces to Φ , L ≥ 0 \Phi,\mathcal{L} \ge0 Φ , L ≥ 0 in (25) .
Adding the sound causality conditions
The sound characteristic polynomial is A L y 2 − C L y + F L A_L y^2-C_L y+F_L A L y 2 − C L y + F L . Its two roots are real and lie in [ 0 , 1 ] [0,1] [ 0 , 1 ] if and only if
F L ≥ 0 , C L ≥ 0 , C L 2 ≥ 4 A L F L , C L ≤ 2 A L , A L − C L + F L ≥ 0. (58) F_L \ge0,\qquad C_L \ge0,\qquad C_L^2 \ge4A_LF_L,\qquad C_L \le2A_L,\qquad A_L-C_L+F_L \ge0. \tag*{(58)} F L ≥ 0 , C L ≥ 0 , C L 2 ≥ 4 A L F L , C L ≤ 2 A L , A L − C L + F L ≥ 0. (58)
Indeed, the first three conditions give real nonnegative roots. The last two give y − + y + ≤ 2 y_-+y_+ \le2 y − + y + ≤ 2 and ( 1 − y − ) ( 1 − y + ) ≥ 0 (1-y_-)(1-y_+) \ge0 ( 1 − y − ) ( 1 − y + ) ≥ 0 , which force both roots to be at most one. This argument includes the boundary roots 0 and 1 without replacing a strict stability criterion by a non-strict one. Moreover,
A L R 2 − C L R + F L = − A L L ≤ 0. (59) A_L R^2-C_L R+F_L=-A_L\mathcal{L}\le0. \tag*{(59)} A L R 2 − C L R + F L = − A L L ≤ 0. (59)
With R > x > 0 R>x>0 R > x > 0 and F L ≥ 0 F_L \ge0 F L ≥ 0 , this already implies reality, nonnegativity, and y − ≤ R ≤ y + y_- \le R \le y_+ y − ≤ R ≤ y + . Hence after solving the Routh-Hurwitz conditions, the remaining sound speed restrictions can be imposed as C L ≤ 2 A L C_L \le2A_L C L ≤ 2 A L and A L − C L + F L ≥ 0 A_L-C_L+F_L \ge0 A L − C L + F L ≥ 0 .
On the branch π 2 = 4 η / 3 \pi_2=4\eta/3 π 2 = 4 η /3 , one has Φ = 0 \Phi=0 Φ = 0 and the sound conditions reduce to x < R ≤ Σ ≤ 1 x<R\le\Sigma\le1 x < R ≤ Σ ≤ 1 . Setting Σ = 1 \Sigma=1 Σ = 1 fixes θ 2 \theta_2 θ 2 through (31) and gives
P N , p r = A L ξ 2 ( ξ 2 + q ) , P_{N,\mathrm{pr}}=A_L\xi^2(\xi^2+q), P N , pr = A L ξ 2 ( ξ 2 + q ) ,
P N = P N , p r + B L ξ ( ξ 2 + R q ) + w 2 ( ξ 2 + x q ) . (60) P_N=P_{N,\mathrm{pr}}+B_L\xi(\xi^2+Rq)+w^2(\xi^2+xq). \tag*{(60)} P N = P N , pr + B L ξ ( ξ 2 + R q ) + w 2 ( ξ 2 + x q ) . (60)
Only the principal part factorizes in this way. The remaining terms retain sound attenuation. Indeed, division by ξ 2 + x q \xi^2+xq ξ 2 + x q leaves a nonzero ξ q \xi q ξ q coefficient D L − x B L = w Γ D_L-xB_L=w\Gamma D L − x B L = w Γ from (20) , ruling out an undamped ideal sound factor when Γ > 0 \Gamma>0 Γ > 0 .
Neutral families F 1 F_1 F 1 –F 3 F_3 F 3
We now verify the prescriptions (32) using the reduced Φ = 0 \Phi=0 Φ = 0 conditions from § 1.4.1 : x < R ≤ Σ ≤ 1 x<R\le\Sigma\le1 x < R ≤ Σ ≤ 1 and θ 1 ≥ η \theta_1\ge\eta θ 1 ≥ η .
Family F 1 F_1 F 1
The choices π 2 = 4 η / 3 \pi_2=4\eta/3 π 2 = 4 η /3 and ϵ 2 = 0 \epsilon_2=0 ϵ 2 = 0 make the second column of (A.8) divisible by w + θ 1 ξ w+\theta_1\xi w + θ 1 ξ . Taking the determinant proves (30) for arbitrary θ 2 \theta_2 θ 2 while entropy matching (11) gives x π 1 − x 2 ϵ 1 = Γ x\pi_1-x^2\epsilon_1=\Gamma x π 1 − x 2 ϵ 1 = Γ . The cubic factor in (30) has only one nontrivial Routh-Hurwitz inequality:
w ( x π 1 q ) − ( x ϵ 1 ) ( x w q ) = w Γ q > 0. (61) w(x\pi_1q)-(x\epsilon_1)(xwq)=w\Gamma q>0. \tag*{(61)} w ( x π 1 q ) − ( x ϵ 1 ) ( x w q ) = w Γ q > 0. (61)
The sound and shear speed bounds give the parameter range in (32) , with θ 2 = 0 \theta_2=0 θ 2 = 0 selecting F 1 F_1 F 1 . The cubic calculation proves rest-frame stability after extracting the damped factor w + θ 1 ς w+\theta_1\varsigma w + θ 1 ς . To apply the common boosted proof in Section A.3.1, we also express the result in terms of the full quartic (30) . It has Φ = 0 \Phi=0 Φ = 0 , and
Σ − R = Γ θ 1 x ε 1 ( x ε 1 + θ 1 ) > 0. (62) \Sigma-R=\frac{\Gamma\theta_1}{x\varepsilon_1(x\varepsilon_1+\theta_1)}>0. \tag*{(62)} Σ − R = x ε 1 ( x ε 1 + θ 1 ) Γ θ 1 > 0. (62)
Hence L = R ( Σ − R ) > 0 \mathcal{L}=R(\Sigma-R)>0 L = R ( Σ − R ) > 0 . Together with the sound and shear bounds above, this verifies (1.24) for the full quartic.
Family F 2 F_2 F 2
Instead setting θ 2 = θ 1 \theta_2=\theta_1 θ 2 = θ 1 gives the family F 2 F_2 F 2 in (32) . Because ε 2 = 0 \varepsilon_2=0 ε 2 = 0 and π 2 = 4 η / 3 \pi_2=4\eta/3 π 2 = 4 η /3 , neither C L C_L C L nor F L F_L F L depends on θ 2 \theta_2 θ 2 . The entropy invariant, full sound determinant and shear polynomial are therefore unchanged. The F 1 F_1 F 1 proof given above applies, including its boosted polynomial and saturated bounds.
Family F 3 F_3 F 3
For this family, we substitute θ 1 = η \theta_1=\eta θ 1 = η and ε 2 = − 2 η \varepsilon_2=-2\eta ε 2 = − 2 η into (31) . Entropy matching (11) and Σ = 1 \Sigma=1 Σ = 1 then give the F 3 F_3 F 3 coefficients in (32) . The remaining condition R ≤ 1 R\leq1 R ≤ 1 is
x ( 1 − x ) ε 1 ≥ Γ − η ( 1 − x ) = ζ + η ( x + 1 / 3 ) > 0. (63) x(1-x)\varepsilon_1\geq\Gamma-\eta(1-x)=\zeta+\eta(x+1/3)>0. \tag*{(63)} x ( 1 − x ) ε 1 ≥ Γ − η ( 1 − x ) = ζ + η ( x + 1/3 ) > 0. (63)
Substitution into the original determinant gives
ε 2 ( θ 2 + π 1 ) + θ 1 π 1 = η ε 1 , (64) \varepsilon_2(\theta_2+\pi_1)+\theta_1\pi_1=\eta\varepsilon_1, \tag*{(64)} ε 2 ( θ 2 + π 1 ) + θ 1 π 1 = η ε 1 , (64)
so Σ = 1 \Sigma=1 Σ = 1 . The inequality R ≤ 1 R\leq1 R ≤ 1 is precisely (A.18). At equality L = 0 \mathcal{L}=0 L = 0 , but Δ 3 = B L w 3 Γ q > 0 \Delta_3=B_Lw^3\Gamma q>0 Δ 3 = B L w 3 Γ q > 0 for every finite q > 0 q>0 q > 0 . The boosted proof in Section A.3.1 includes this endpoint.8 The nonzero sound and shear speeds become equal to the speed of light.
The reduced inequalities do not impose a well defined conformal limit. In fact, the coefficients just derived have the exact residuals
ε 2 − 3 π 2 = { − 4 η , families F 1 and F 2 , − 6 η , F 3 . (65) \varepsilon_2-3\pi_2=
\begin{cases}
-4\eta, & \text{families }F_1\text{ and }F_2,\\
-6\eta, & F_3.
\end{cases} \tag*{(65)} ε 2 − 3 π 2 = { − 4 η , − 6 η , families F 1 and F 2 , F 3 . (65)
Therefore no choice of their free parameters can impose all conformal coefficient relations (28) at the conformal limit with η c > 0 \eta_c>0 η c > 0 . The family F 2 F_2 F 2 obeys θ 1 = θ 2 \theta_1=\theta_2 θ 1 = θ 2 , but that vector relation does not repair its scalar mismatch. The charged uplift retains the seed coefficients ε 2 \varepsilon_2 ε 2 and π 2 \pi_2 π 2 , so the same residual excludes the charged families F 1 F_1 F 1 –F 3 F_3 F 3 with a conformal limit.
Neutral families F 4 F_4 F 4 –F 6 F_6 F 6 and conformal limits
We now verify the three constructions with finite conformal limits in § 1.4.2 , then discuss the overlaps between different families.9
General luminal family F 4 F_4 F 4
For the coefficients and parameter bounds in (33) , ε 2 = ( 1 − f ) S 0 Γ > 0 \varepsilon_2=(1-f)S_0\Gamma>0 ε 2 = ( 1 − f ) S 0 Γ > 0 when Γ > 0 \Gamma>0 Γ > 0 , so θ 2 \theta_2 θ 2 is regular. Substitution into (20) gives Σ = 1 \Sigma=1 Σ = 1 and Φ = 0 \Phi=0 Φ = 0 . Moreover,
x ε 1 + θ 1 ≥ 2 Γ 1 − x + θ 1 , x < R ≤ 1 + x 2 < 1. (66) x\varepsilon_1+\theta_1\ge\frac{2\Gamma}{1-x}+\theta_1,\qquad x<R\le\frac{1+x}{2}<1. \tag*{(66)} x ε 1 + θ 1 ≥ 1 − x 2Γ + θ 1 , x < R ≤ 2 1 + x < 1. (66)
Hence L = R ( 1 − R ) > 0 \mathcal{L}=R(1-R)>0 L = R ( 1 − R ) > 0 . The bound K ≥ 3 / 4 K\ge3/4 K ≥ 3/4 gives θ 1 ≥ η \theta_1\ge\eta θ 1 ≥ η because Γ ≥ 4 η / 3 \Gamma\ge4\eta/3 Γ ≥ 4 η /3 . This verifies (25) directly. A well defined conformal limit (28) further requires the parameter limits derived in § 1.4.2 .
Family F 5 F_5 F 5 : strict subluminality
We check the strict limits for the deformation of the F 5 F_5 F 5 family constructed in § 1.4.2 . Substituting (34) and the F 5 F_5 F 5 vector coefficients into (20) gives Σ = ρ \Sigma=\rho Σ = ρ and Φ = 0 \Phi=0 Φ = 0 , with ρ 0 , ρ \rho_0,\rho ρ 0 , ρ defined in § 1.4.2 . For Γ > 0 \Gamma>0 Γ > 0 , the distance to the causal bound is
ρ − ρ 0 = m ( θ 2 − θ 1 ) ε 1 θ 1 = χ ( λ − 1 ) ( 1 − x ) 2 λ , 1 − ρ = ( λ − 1 ) ( 1 − x ) ( 1 − χ ) 2 λ > 0 , 1 − χ = 1 + ( 1 − β ) ( 3 x − 1 ) 2 1 + ( 3 x − 1 ) 2 > 0. (67) \begin{aligned}
\rho-\rho_0&=\frac{m(\theta_2-\theta_1)}{\varepsilon_1\theta_1}=\frac{\chi(\lambda-1)(1-x)}{2\lambda},\\
1-\rho&=\frac{(\lambda-1)(1-x)(1-\chi)}{2\lambda}>0,\\
1-\chi&=\frac{1+(1-\beta)(3x-1)^2}{1+(3x-1)^2}>0. \tag*{(67)}
\end{aligned} ρ − ρ 0 1 − ρ 1 − χ = ε 1 θ 1 m ( θ 2 − θ 1 ) = 2 λ χ ( λ − 1 ) ( 1 − x ) , = 2 λ ( λ − 1 ) ( 1 − x ) ( 1 − χ ) > 0 , = 1 + ( 3 x − 1 ) 2 1 + ( 1 − β ) ( 3 x − 1 ) 2 > 0. (67)
To prove the remaining stability inequality, we introduce the positive denominator
d 5 = 8 x ( 1 − x ) + λ ( 1 + x ) 2 > 0 , R = D L B L = x + 4 x ( 1 − x ) 2 d 5 . (68) d_5=8x(1-x)+\lambda(1+x)^2>0,\qquad R=\frac{D_L}{B_L}=x+\frac{4x(1-x)^2}{d_5}. \tag*{(68)} d 5 = 8 x ( 1 − x ) + λ ( 1 + x ) 2 > 0 , R = B L D L = x + d 5 4 x ( 1 − x ) 2 . (68)
Direct subtraction gives the strictly positive expression
ρ 0 − R = ( 1 − x ) [ 8 x ( 1 − x ) + λ ( λ + 1 ) ( 1 + x ) 2 ] 2 λ d 5 > 0. (69) \rho_0-R=\frac{(1-x)\left[8x(1-x)+\lambda(\lambda+1)(1+x)^2\right]}{2\lambda d_5}>0. \tag*{(69)} ρ 0 − R = 2 λ d 5 ( 1 − x ) [ 8 x ( 1 − x ) + λ ( λ + 1 ) ( 1 + x ) 2 ] > 0. (69)
Hence
x < R < ρ 0 ≤ ρ < 1 , L = R ( ρ − R ) > 0. (70) x<R<\rho_0\le\rho<1,\qquad\mathcal{L}=R(\rho-R)>0. \tag*{(70)} x < R < ρ 0 ≤ ρ < 1 , L = R ( ρ − R ) > 0. (70)
The shear speed also has a strict bound. Since Γ ≥ 4 η / 3 \Gamma\ge4\eta/3 Γ ≥ 4 η /3 ,
0 ≤ y T = η θ 1 ≤ 3 x ( 1 − x ) 2 λ ( 1 + x ) 2 < 1 λ < 1. (71) 0\le y_T=\frac{\eta}{\theta_1}\le\frac{3x(1-x)^2}{\lambda(1+x)^2}<\frac{1}{\lambda}<1. \tag*{(71)} 0 ≤ y T = θ 1 η ≤ λ ( 1 + x ) 2 3 x ( 1 − x ) 2 < λ 1 < 1. (71)
The strict middle inequality follows from ( 1 + x ) 2 − 3 x ( 1 − x ) 2 = ( 1 − x ) + x 2 ( 7 − 3 x ) > 0 (1+x)^2-3x(1-x)^2=(1-x)+x^2(7-3x)>0 ( 1 + x ) 2 − 3 x ( 1 − x ) 2 = ( 1 − x ) + x 2 ( 7 − 3 x ) > 0 . The sound and shear characteristic speeds therefore have magnitude strictly less than one while one of the sound branches has vanishing characteristic speed. Entropy matching (11) follows from − π 2 − x 2 ϵ 1 + 2 x ϵ 2 = Γ − 4 η / 3 = ζ -\pi_2-x^2\epsilon_1+2x\epsilon_2=\Gamma-4\eta/3=\zeta − π 2 − x 2 ϵ 1 + 2 x ϵ 2 = Γ − 4 η /3 = ζ . We have therefore verified (25) for this family. The conformal limits of this family are given in § 1.4.2 . The remaining residual is θ 2 − θ 1 = χ ( λ − 1 ) m \theta_2-\theta_1=\chi(\lambda-1)m θ 2 − θ 1 = χ ( λ − 1 ) m . For η c > 0 \eta_c>0 η c > 0 and finite λ c > 1 \lambda_c>1 λ c > 1 , it vanishes only if χ → 0 \chi\to0 χ → 0 , as ensured by (35) . At λ c = 1 \lambda_c=1 λ c = 1 the upper sound speed becomes luminal.
Family F 6 F_6 F 6 : nonzero characteristic speeds
Here we prove the bounds (38) for the neutral coefficients (36) , with η > 0 \eta>0 η > 0 and ζ ≥ 0 \zeta\geq0 ζ ≥ 0 . The coefficients ϵ 1 , ϵ 2 , π 1 \epsilon_1,\epsilon_2,\pi_1 ϵ 1 , ϵ 2 , π 1 in (36) are obtained by multiplying their values in (34) by Λ \Lambda Λ , with ϵ 2 = π 1 \epsilon_2=\pi_1 ϵ 2 = π 1 throughout. Consequently, − x 2 ϵ 1 + 2 x ϵ 2 = Λ Γ -x^2\epsilon_1+2x\epsilon_2=\Lambda\Gamma − x 2 ϵ 1 + 2 x ϵ 2 = ΛΓ . Entropy matching (11) fixes π 2 \pi_2 π 2 through − π 2 + Λ Γ = ζ -\pi_2+\Lambda\Gamma=\zeta − π 2 + ΛΓ = ζ , preserving the prescribed bulk viscosity. It also gives F L = x θ 2 ( Λ − 1 ) Γ > 0 F_L=x\theta_2(\Lambda-1)\Gamma>0 F L = x θ 2 ( Λ − 1 ) Γ > 0 . We now check that the vector coefficients leave sufficient margins for both stability and causality. We write Φ 0 = ( Λ − 1 ) x ( 1 − x ) / ( 2 Λ ) \Phi_0=(\Lambda-1)x(1-x)/(2\Lambda) Φ 0 = ( Λ − 1 ) x ( 1 − x ) / ( 2Λ ) for the value at χ = 0 \chi=0 χ = 0 , with Σ 0 \Sigma_0 Σ 0 as defined in (36) . Substitution into the original neutral matrix (A.8) gives the invariants (37) . For the remaining inequality we set
M = 1 + 6 x − 3 x 2 > 0 , j = 2 ( Λ − 1 ) x ( 1 − x ) Λ ( 1 + x ) . (72) M=1+6x-3x^2>0,\qquad j=\frac{2(\Lambda-1)x(1-x)}{\Lambda(1+x)}. \tag*{(72)} M = 1 + 6 x − 3 x 2 > 0 , j = Λ ( 1 + x ) 2 ( Λ − 1 ) x ( 1 − x ) . (72)
Thus Φ = Φ 0 + j χ ( 1 − Σ 0 ) \Phi=\Phi_0+j\chi(1-\Sigma_0) Φ = Φ 0 + j χ ( 1 − Σ 0 ) . We define
Q = ( Λ − 1 ) 2 ( 3 x − 1 ) 2 M 2 + 2 ( Λ − 1 ) ( 1 + x ) M [ 5 ( 1 − x ) + x 2 ( 19 − 3 x ) ] + ( 3 − x ) ( 1 + x ) 3 ( 3 + 10 x − x 2 ) , \begin{aligned}
Q={}&(\Lambda-1)^2(3x-1)^2M^2\\
&+2(\Lambda-1)(1+x)M[5(1-x)+x^2(19-3x)]\\
&+(3-x)(1+x)^3(3+10x-x^2),
\end{aligned} Q = ( Λ − 1 ) 2 ( 3 x − 1 ) 2 M 2 + 2 ( Λ − 1 ) ( 1 + x ) M [ 5 ( 1 − x ) + x 2 ( 19 − 3 x )] + ( 3 − x ) ( 1 + x ) 3 ( 3 + 10 x − x 2 ) ,
L 0 = x ( 1 − x ) Q 4 Λ 2 ( 1 + x ) 2 M 2 > 0. (73) \mathcal{L}_0=\frac{x(1-x)Q}{4\Lambda^2(1+x)^2M^2}>0. \tag*{(73)} L 0 = 4 Λ 2 ( 1 + x ) 2 M 2 x ( 1 − x ) Q > 0. (73)
Every bracket is positive on 0 < x < 1 0<x<1 0 < x < 1 , and the last term in Q Q Q is strictly positive. An algebra calculation gives
R ( Σ 0 − R ) − Φ 0 = L 0 , L = L 0 + χ ( 1 − Σ 0 ) ( R − j ) > 0. (74) R(\Sigma_0-R)-\Phi_0=\mathcal{L}_0,\qquad\mathcal{L}=\mathcal{L}_0+\chi(1-\Sigma_0)(R-j)>0. \tag*{(74)} R ( Σ 0 − R ) − Φ 0 = L 0 , L = L 0 + χ ( 1 − Σ 0 ) ( R − j ) > 0. (74)
The final sign follows from the explicit bounds
1 − Σ 0 = 1 − x 4 + 2 ( Λ − 1 ) x ( 1 − x ) 2 Λ ( 1 + x ) 2 > 0 , (75) 1-\Sigma_0=\frac{1-x}{4}+\frac{2(\Lambda-1)x(1-x)^2}{\Lambda(1+x)^2}>0, \tag*{(75)} 1 − Σ 0 = 4 1 − x + Λ ( 1 + x ) 2 2 ( Λ − 1 ) x ( 1 − x ) 2 > 0 , (75)
x − j = x [ 2 − Λ + ( 3 Λ − 2 ) x ] Λ ( 1 + x ) > 0 , R > x , x-j=\frac{x[2-\Lambda+(3\Lambda-2)x]}{\Lambda(1+x)}>0,\qquad R>x, x − j = Λ ( 1 + x ) x [ 2 − Λ + ( 3Λ − 2 ) x ] > 0 , R > x ,
1 − Σ = ( 1 − χ ) ( 1 − Σ 0 ) > 0 , Φ > 0. 1-\Sigma=(1-\chi)(1-\Sigma_0)>0,\qquad\Phi>0. 1 − Σ = ( 1 − χ ) ( 1 − Σ 0 ) > 0 , Φ > 0.
The upper restriction Λ ≤ 2 \Lambda\leq2 Λ ≤ 2 is a sufficient bound ensuring x − j > 0 x-j>0 x − j > 0 for all x x x , including the non-thermodynamic case.
Now define f ( y ) = y 2 − Σ y + Φ f(y)=y^2-\Sigma y+\Phi f ( y ) = y 2 − Σ y + Φ . Then f ( 0 ) = Φ > 0 f(0)=\Phi>0 f ( 0 ) = Φ > 0 , f ( R ) = − L < 0 f(R)=-L<0 f ( R ) = − L < 0 , and R > 0 R>0 R > 0 . Hence f f f has two distinct positive roots y − < R < y + y_-<R<y_+ y − < R < y + . Their sum is Σ < 1 \Sigma<1 Σ < 1 , which proves the sound bounds in (38) . In particular Σ − R > 0 \Sigma-R>0 Σ − R > 0 . The shear squared speed obeys
y T = η θ 1 ≤ 3 x ( 1 − x ) 2 2 Λ ( 1 + x ) 2 < 1 2 Λ < 1. (76) y_{\mathrm{T}}=\frac{\eta}{\theta_1}\leq\frac{3x(1-x)^2}{2\Lambda(1+x)^2}<\frac{1}{2\Lambda}<1. \tag*{(76)} y T = θ 1 η ≤ 2Λ ( 1 + x ) 2 3 x ( 1 − x ) 2 < 2Λ 1 < 1. (76)
The strict upper bound uses the same positive polynomial as (A.26). With η > 0 \eta>0 η > 0 , all sound and shear characteristic speeds are nonzero. These inequalities verify (25) .
At a regular conformal endpoint, multiplying these three coefficients by a finite Λ c \Lambda_{\mathrm{c}} Λ c gives Π = 4 Λ c η c \Pi=4\Lambda_{\mathrm{c}}\eta_{\mathrm{c}} Π = 4 Λ c η c . The value of π 2 \pi_2 π 2 fixed by entropy matching also tends to Π / 3 \Pi/3 Π/3 , giving the conformal scalar relations in § 1.4.2 . Conformality does not select Λ = 2 \Lambda=2 Λ = 2 : the range 1 < Λ ≤ 2 1<\Lambda\leq2 1 < Λ ≤ 2 comes from the stability bounds. The vector residual has endpoint θ 2 , c − θ 1 , c = 4 η c ( 2 Λ c − 1 ) χ c \theta_{2,\mathrm{c}}-\theta_{1,\mathrm{c}}=4\eta_{\mathrm{c}}(2\Lambda_{\mathrm{c}}-1)\chi_{\mathrm{c}} θ 2 , c − θ 1 , c = 4 η c ( 2 Λ c − 1 ) χ c . For η c > 0 \eta_{\mathrm{c}}>0 η c > 0 and Λ c > 1 \Lambda_{\mathrm{c}}>1 Λ c > 1 , conformal symmetry therefore requires χ c = 0 \chi_{\mathrm{c}}=0 χ c = 0 . A limit Λ c = 1 \Lambda_{\mathrm{c}}=1 Λ c = 1 reaches F 5 F_5 F 5 with λ = 2 \lambda=2 λ = 2 and loses the nonzero lower sound speed.
Overlaps between the families F 4 F_4 F 4 , F 5 F_5 F 5 and F 6 F_6 F 6
Substituting the parameters (44) into the family F 4 F_4 F 4 prescription (33) gives the values of ϵ 1 , ϵ 2 = π 1 = m \epsilon_1,\epsilon_2=\pi_1=m ϵ 1 , ϵ 2 = π 1 = m and π 2 \pi_2 π 2 in (34) , together with θ 1 = θ 2 = θ 0 \theta_1=\theta_2=\theta_0 θ 1 = θ 2 = θ 0 . They lie in the F 4 F_4 F 4 domain: the scalar lower bound is saturated, f = 1 / 2 f=1/2 f = 1/2 , and K > 3 / 4 K>3/4 K > 3/4 follows from the polynomial positivity used in (A.26). Setting λ = 1 \lambda=1 λ = 1 in F 5 F_5 F 5 gives this member of F 4 F_4 F 4 , independently of β \beta β , proving the first overlap stated in § 1.6 . For F 6 F_6 F 6 , setting Λ = 1 \Lambda=1 Λ = 1 gives the same scalar coefficients (34) and
π 2 = 4 η 3 , Σ 0 = 3 + x 4 , θ 1 = 2 θ 0 , θ 2 = 2 θ 0 + χ m . \pi_2=\frac{4\eta}{3},\qquad\Sigma_0=\frac{3+x}{4},\qquad\theta_1=2\theta_0,\qquad\theta_2=2\theta_0+\chi m. π 2 = 3 4 η , Σ 0 = 4 3 + x , θ 1 = 2 θ 0 , θ 2 = 2 θ 0 + χ m .
which is exactly F 5 F_5 F 5 at λ = 2 \lambda=2 λ = 2 and the same β \beta β . At Γ > 0 \Gamma>0 Γ > 0 , matching F 5 F_5 F 5 's value π 2 = 4 η / 3 \pi_2=4\eta/3 π 2 = 4 η /3 to F 6 F_6 F 6 forces Λ = 1 \Lambda=1 Λ = 1 . Matching θ 1 \theta_1 θ 1 then forces λ = 2 \lambda=2 λ = 2 . Thus this overlap with F 5 F_5 F 5 occurs only at Λ = 1 \Lambda=1 Λ = 1 in F 6 F_6 F 6 , and only at λ = 2 \lambda=2 λ = 2 .
Charged frames
We now construct charged frames from the neutral ones. We first recall the charged thermodynamic conditions, then derive the coefficient map and check its entropy matching, factorization and remaining freedom.
Thermodynamic domain and susceptibilities
We use the full domain (9) . The charged uplift evaluates the neutral seed at the charged values in ζ ∗ \zeta_* ζ ∗ in (29) . We now derive the fluctuation-positivity inequalities in (9) from equilibrium thermodynamics. Let δ p = p T T p μ μ − p T μ 2 \delta_p=p_{TT}p_{\mu\mu}-p_{T\mu}^2 δ p = p T T p μμ − p T μ 2 denote the pressure-Hessian determinant, with entropy density s = p T s=p_T s = p T as in § 1.1 . Positivity of equilibrium energy-charge fluctuations makes the pressure Hessian positive semidefinite, equivalently n μ ≥ 0 n_\mu\geq0 n μ ≥ 0 , E ≥ 0 E\geq0 E ≥ 0 , Δ J ≥ 0 \Delta_J\geq0 Δ J ≥ 0 , as in [8 ] . On our regular domain it is positive definite and direct differentiation gives
n μ = p μ > 0 , Δ J = T 2 δ p > 0 , n_\mu=p_\mu>0,\qquad\Delta_J=T^2\delta_p>0, n μ = p μ > 0 , Δ J = T 2 δ p > 0 ,
E = T ( p T T + 2 α p T μ + α 2 p μ μ ) > 0 , w x = p μ μ s 2 − 2 p T μ s n + p T T n 2 δ p . E=T\left(p_{TT}+2\alpha p_{T\mu}+\alpha^2p_{\mu\mu}\right)>0,\qquad wx=\frac{p_{\mu\mu}s^2-2p_{T\mu}sn+p_{TT}n^2}{\delta_p}. E = T ( p T T + 2 α p T μ + α 2 p μμ ) > 0 , w x = δ p p μμ s 2 − 2 p T μ s n + p T T n 2 .
Consequently
w x n μ − n 2 = ( p μ μ s − p T μ n ) 2 δ p ≥ 0 , wxn_{\mu}-n^2=\frac{(p_{\mu\mu}s-p_{T\mu}n)^2}{\delta_p}\ge0, w x n μ − n 2 = δ p ( p μμ s − p T μ n ) 2 ≥ 0 ,
which proves the last inequality in (9) . Using x = A + r T B x=A+rTB x = A + r T B from (7) , the other inequality follows from
A + n μ w ( T B ) 2 − r T B = n μ w ( T B − n n μ ) 2 + x − n 2 w n μ ≥ 0. A+\frac{n_\mu}{w}(TB)^2-rTB=\frac{n_\mu}{w}\left(TB-\frac{n}{n_\mu}\right)^2+x-\frac{n^2}{wn_\mu}\ge0. A + w n μ ( T B ) 2 − r T B = w n μ ( T B − n μ n ) 2 + x − w n μ n 2 ≥ 0.
The condition Δ J > 0 \Delta_J>0 Δ J > 0 selects regular susceptibilities, and 0 < x < 1 0<x<1 0 < x < 1 restricts the ideal sound cone. We do not consider the case of singular susceptibilities nor vanishing/light speeds x = 0 , 1 x=0,1 x = 0 , 1 .
Charged entropy production and the coefficient uplift
We now derive the map (39) –(40) for an arbitrary neutral seed satisfying the preceding stability and causality requirements, and prove (41) directly using the original charged matrix. We use the thermodynamic domain (9) , transport constraints (10) and susceptibilities (8) .
Charged entropy production and matching identities
We begin by re-deriving the entropy constraints in an arbitrary frame stated in Section 1. For the canonical current S μ = ( p u μ − u ν T μ ν − μ J μ ) / T S^\mu=(p u^\mu-u_\nu T^{\mu\nu}-\mu J^\mu)/T S μ = ( p u μ − u ν T μν − μ J μ ) / T , conservation and thermodynamics give
∇ μ S μ = − T ( 1 ) μ ν ∇ μ ( u ν / T ) − J ( 1 ) μ ∇ μ α , (77) \nabla_\mu S^\mu=-T^{\mu\nu}_{(1)}\nabla_\mu(u_\nu/T)-J^\mu_{(1)}\nabla_\mu\alpha, \tag*{(77)} ∇ μ S μ = − T ( 1 ) μν ∇ μ ( u ν / T ) − J ( 1 ) μ ∇ μ α , (77)
where J ( 1 ) μ J^\mu_{(1)} J ( 1 ) μ is the first-derivative correction in (3) . Using the scalar and vector corrections of (4) –(1.5), the contraction is
T ∇ μ S μ = − E D ln T − P Θ − T N D α − Q μ ( a μ + ∇ ⊥ μ ln T ) − T J μ ∇ ⊥ μ α + η 2 σ μ ν σ μ ν + O ( ∂ 3 ) . (78) T\nabla_\mu S^\mu=-\mathcal{E}D\ln T-\mathcal{P}\Theta-TN D\alpha
-Q_\mu(a^\mu+\nabla_\perp^\mu\ln T)-TJ_\mu\nabla_\perp^\mu\alpha+\frac{\eta}{2}\sigma_{\mu\nu}\sigma^{\mu\nu}+O(\partial^3). \tag*{(78)} T ∇ μ S μ = − E D ln T − P Θ − T N D α − Q μ ( a μ + ∇ ⊥ μ ln T ) − T J μ ∇ ⊥ μ α + 2 η σ μν σ μν + O ( ∂ 3 ) . (78)
To reduce the entropy production using the ideal conservation equations, we first express A A A and B B B in terms of the susceptibilities (8) . The pressure identities ( ∂ T p ) α = w / T (\partial_Tp)_\alpha=w/T ( ∂ T p ) α = w / T and ( ∂ α p ) T = T n (\partial_\alpha p)_T= Tn ( ∂ α p ) T = T n , together with the Maxwell relation F = T 2 N F=T^2N F = T 2 N , give
A = w q α T Δ J , B = − w q T Δ J . (79) A=\frac{wq_\alpha}{T\Delta_J},\qquad B=-\frac{wq_T}{\Delta_J}. \tag*{(79)} A = T Δ J w q α , B = − Δ J w q T . (79)
These follow by expressing the derivatives in (7) as Jacobians with respect to ( T , α ) (T,\alpha) ( T , α ) , whose transformation to ( ϵ , n ) (\epsilon,n) ( ϵ , n ) has determinant Δ J \Delta_J Δ J . The ideal scalar equations E D T + F D α = − w Θ E D T+F D\alpha=-w\Theta E D T + F D α = − w Θ and N D T + H D α = − n Θ N D T+H D\alpha=-n\Theta N D T + H D α = − n Θ give D ln T = − A Θ D\ln T=-A\Theta D ln T = − A Θ and D α = − B Θ D\alpha=-B\Theta D α = − B Θ on inversion. Ideal momentum conservation gives a μ = − ∇ ⊥ μ ln T − r T ∇ ⊥ μ α a^\mu=-\nabla_\perp^\mu\ln T-rT\nabla_\perp^\mu\alpha a μ = − ∇ ⊥ μ ln T − r T ∇ ⊥ μ α .
Substituting the constitutive relations gives the following quadratic expression, before imposing any frame-coefficient relation to cancel a contribution:
∇ μ S μ = η 2 T σ μ ν σ μ ν − 1 T [ ε 1 ( D ln T ) 2 + ( ε 2 + π 1 ) ( D ln T ) Θ + π 2 Θ 2 + ( ε 3 + T ν 1 ) ( D ln T ) ( D α ) + ( π 3 + T ν 2 ) Θ D α + T ν 3 ( D α ) 2 + θ 1 a μ a μ + ( θ 1 + θ 2 ) a μ ∇ ⊥ μ ln T + θ 2 ( ∇ ⊥ ln T ) 2 + ( θ 3 + T γ 1 ) a μ ∇ ⊥ μ α + ( θ 3 + T γ 2 ) ∇ ⊥ μ ln T ∇ ⊥ μ α + T γ 3 ( ∇ ⊥ α ) 2 ] + O ( ∂ 3 ) = η 2 T σ μ ν σ μ ν + ζ T Θ 2 + D T ( ∇ ⊥ α ) 2 + [ r ( θ 2 − θ 1 ) − ( γ 2 − γ 1 ) ] ∇ ⊥ μ ln T ∇ ⊥ μ α + O ( ∂ 3 ) . (80) \begin{aligned}
\nabla_\mu S^\mu= \frac{\eta}{2T}\sigma_{\mu\nu}\sigma^{\mu\nu} - \frac{1}{T}\left[\varepsilon_1(D\ln T)^2 + (\varepsilon_2+\pi_1)(D\ln T)\Theta+ \pi_2\Theta^2 \\
+(\varepsilon_3+T\nu_1)(D\ln T)(D\alpha) + (\pi_3+T\nu_2)\Theta D\alpha+ T\nu_3(D\alpha)^2 \\
+\theta_1 a_\mu a^\mu+ (\theta_1+\theta_2)a_\mu\nabla_\perp^\mu\ln T + \theta_2(\nabla_\perp\ln T)^2 \\
+(\theta_3+T\gamma_1)a_\mu\nabla_\perp^\mu\alpha+ (\theta_3+T\gamma_2)\nabla_{\perp\mu}\ln T\nabla_\perp^\mu\alpha\\
+T\gamma_3(\nabla_\perp\alpha)^2\right] + O(\partial^3) \\
&= \frac{\eta}{2T}\sigma_{\mu\nu}\sigma^{\mu\nu}+\frac{\zeta}{T}\Theta^2+DT(\nabla_\perp\alpha)^2 \\
+\left[r(\theta_2-\theta_1)-(\gamma_2-\gamma_1)\right]\nabla_{\perp\mu}\ln T\nabla_\perp^\mu\alpha+O(\partial^3). \tag*{(80)}
\end{aligned} ∇ μ S μ = 2 T η σ μν σ μν − T 1 [ ε 1 ( D ln T ) 2 + ( ε 2 + π 1 ) ( D ln T ) Θ + π 2 Θ 2 + ( ε 3 + T ν 1 ) ( D ln T ) ( D α ) + ( π 3 + T ν 2 ) Θ D α + T ν 3 ( D α ) 2 + θ 1 a μ a μ + ( θ 1 + θ 2 ) a μ ∇ ⊥ μ ln T + θ 2 ( ∇ ⊥ ln T ) 2 + ( θ 3 + T γ 1 ) a μ ∇ ⊥ μ α + ( θ 3 + T γ 2 ) ∇ ⊥ μ ln T ∇ ⊥ μ α + T γ 3 ( ∇ ⊥ α ) 2 ] + O ( ∂ 3 ) + [ r ( θ 2 − θ 1 ) − ( γ 2 − γ 1 ) ] ∇ ⊥ μ ln T ∇ ⊥ μ α + O ( ∂ 3 ) . = 2 T η σ μν σ μν + T ζ Θ 2 + D T ( ∇ ⊥ α ) 2 (80)
Only the last equality in (A.35) uses these ideal equation relations, together with the definitions in (11) . The mixed term has not been set to zero. There is no remaining ( ∇ ⊥ ln T ) 2 (\nabla_\perp\ln T)^2 ( ∇ ⊥ ln T ) 2 term, so positivity for arbitrary independent ∇ ⊥ μ ln T \nabla_\perp\mu\ln T ∇ ⊥ μ ln T and ∇ ⊥ μ α \nabla_\perp^\mu\alpha ∇ ⊥ μ α requires the vector relation in (10) . Imposing that equation makes the displayed quadratic production nonnegative. This condition permits unequal θ 1 , θ 2 \theta_1,\theta_2 θ 1 , θ 2 and unequal γ 1 , γ 2 \gamma_1,\gamma_2 γ 1 , γ 2 . It does not impose a thermodynamic frame.
Derivation of the coefficient map and entropy matching
We start with a set of neutral coefficients or one the six neutral families we found but evaluated at the charged values in (29) . We replace its temperature gradient by the pressure gradient using
d p w = d T T + r T d α . (81) \frac{dp}{w}=\frac{dT}{T}+rT\,d\alpha. \tag*{(81)} w d p = T d T + r T d α . (81)
This replacement yields the third stress coefficients in (39) . We then choose the current coefficients so that, after linearising about uniform equilibrium and Fourier transforming in its rest frame, the fluctuation h = δ n − r δ ϵ h=\delta n-r\delta\epsilon h = δ n − r δ ϵ obeys the independent telegrapher equation F q ( ς , q ) h = ( τ q ς 2 + ς + D q q ) h = 0 F_q(\varsigma,q)h=(\tau_q\varsigma^2+\varsigma+D_{qq})h=0 F q ( ς , q ) h = ( τ q ς 2 + ς + D q q ) h = 0 , as described in Step 4 of § 1.3.3 , with F q F_q F q defined in (27) . We obtain this equation by subtracting r r r times the linearised energy-conservation equation from the charge-conservation equation, with r = n / w r=n/w r = n / w evaluated at equilibrium. Imposing this condition together with entropy matching (11) gives the current coefficients (40) . We verify the corresponding matrix identity (A.41) in the following subsection.
To check entropy matching (11) , we first combine (A.34) with x = A + r T B x=A+rTB x = A + r T B to obtain x = w ( q α − r T 2 q T ) / ( T Δ J ) = w Ξ / ( T Δ J ) x=w(q_\alpha-rT^2q_T)/(T\Delta_J)=w\Xi/(T\Delta_J) x = w ( q α − r T 2 q T ) / ( T Δ J ) = w Ξ/ ( T Δ J ) . This proves the susceptibility identity in (29) , with Ξ > 0 \Xi>0 Ξ > 0 on the domain (9) , and gives D q = D w / ( Δ J x ) D_q=D_w/(\Delta_Jx) D q = D w / ( Δ J x ) . The vector matching relations in (11) then follow directly from the map:
γ 2 − γ 1 = r ( θ 2 − θ 1 ) , (82) \gamma_2-\gamma_1=r(\theta_2-\theta_1), \tag*{(82)} γ 2 − γ 1 = r ( θ 2 − θ 1 ) , (82)
r ( θ 3 − r T θ 1 ) + r T γ 1 − γ 3 = D q Ξ = D T . (83) r(\theta_3-rT\theta_1)+rT\gamma_1-\gamma_3=D_q\Xi=DT. \tag*{(83)} r ( θ 3 − r T θ 1 ) + r T γ 1 − γ 3 = D q Ξ = D T . (83)
Substituting the map into the charged bulk invariant, using q α B + T q T A = 0 q_{\alpha}B+Tq_{T}A=0 q α B + T q T A = 0 from (A.34) and A + r T B = x A+r_{T}B=x A + r T B = x , gives
ζ = − π 2 − x 2 ϵ 1 + x ( ϵ 2 + π 1 ) − D q T 2 Δ J B 2 w = ζ ∗ − D T 2 B 2 x . (84) \zeta=-\pi_{2}-x^{2}\epsilon_{1}+x(\epsilon_{2}+\pi_{1})-\frac{D_{q}T^{2}\Delta_{J}B^{2}}{w}=\zeta_{*}-\frac{DT^{2}B^{2}}{x}. \tag*{(84)} ζ = − π 2 − x 2 ϵ 1 + x ( ϵ 2 + π 1 ) − w D q T 2 Δ J B 2 = ζ ∗ − x D T 2 B 2 . (84)
This derives the charge values in (29) from (11) . These identities cancel the mixed term in (A.35) and leave the three nonnegative terms with the prescribed η , ζ , D \eta,\zeta,D η , ζ , D .
Exact factorization of the original charged matrix
The purpose of this subsection is to verify the map introduced above directly from the linearised matrix. With equation order energy, longitudinal momentum, charge and unknowns ( δ T , δ α , δ u ∥ ) (\delta T,\delta\alpha,\delta u_{\parallel}) ( δ T , δ α , δ u ∥ ) , the original sound matrix is
M c h a r g e d ( ς , k ) = ( E ς + ϵ 1 ς 2 / T − θ 2 q / T F ς + ϵ 3 ς 2 − θ 3 q i k [ w + ( ϵ 2 + θ 1 ) ς ] i k [ w / T + ( θ 2 + π 1 ) ς / T ] i k [ T n + ( θ 3 + π 3 ) ς ] θ 1 ς 2 + w ς + ( 4 3 η − π 2 ) q N ς + ν 1 ς 2 / T − γ 2 q / T H ς + ν 3 ς 2 − γ 3 q i k [ n + ( ν 2 + γ 1 ) ς ] ) . (85) M_{\mathrm{charged}}(\varsigma,k)=
\begin{pmatrix}
E\varsigma+\epsilon_{1}\varsigma^{2}/T-\theta_{2}q/T & F\varsigma+\epsilon_{3}\varsigma^{2}-\theta_{3}q & ik[w+(\epsilon_{2}+\theta_{1})\varsigma]\\
ik[w/T+(\theta_{2}+\pi_{1})\varsigma/T] & ik[Tn+(\theta_{3}+\pi_{3})\varsigma] & \theta_{1}\varsigma^{2}+w\varsigma+(\frac{4}{3}\eta-\pi_{2})q\\
N\varsigma+\nu_{1}\varsigma^{2}/T-\gamma_{2}q/T & H\varsigma+\nu_{3}\varsigma^{2}-\gamma_{3}q & ik[n+(\nu_{2}+\gamma_{1})\varsigma]
\end{pmatrix}. \tag*{(85)} M charged ( ς , k ) = E ς + ϵ 1 ς 2 / T − θ 2 q / T ik [ w / T + ( θ 2 + π 1 ) ς / T ] N ς + ν 1 ς 2 / T − γ 2 q / T F ς + ϵ 3 ς 2 − θ 3 q ik [ T n + ( θ 3 + π 3 ) ς ] H ς + ν 3 ς 2 − γ 3 q ik [ w + ( ϵ 2 + θ 1 ) ς ] θ 1 ς 2 + w ς + ( 3 4 η − π 2 ) q ik [ n + ( ν 2 + γ 1 ) ς ] . (85)
Here E , F , N , H E,F,N,H E , F , N , H are the derivatives at fixed α \alpha α or T T T defined in (8) . Substituting ω = i ς \omega=i\varsigma ω = i ς into the Fourier space conservation equations gives exactly this matrix, including the minus signs on its spatial-gradient entries.
We define R ϵ \mathcal{R}_{\epsilon} R ϵ and R n \mathcal{R}_{n} R n as the first and third rows of M c h a r g e d M_{\mathrm{charged}} M charged in (A.40), corresponding respectively to the linearised energy- and charge-conservation equations. Equations (39) –(40) give the exact row identity
R n − r R ϵ = F q ( ς , q ) ( q T , q α , 0 ) . (86) \mathcal{R}_{n}-r\mathcal{R}_{\epsilon}=F_{q}(\varsigma,q)(q_{T},q_{\alpha},0). \tag*{(86)} R n − r R ϵ = F q ( ς , q ) ( q T , q α , 0 ) . (86)
Here F q F_{q} F q is defined in (27) , and R ϵ \mathcal{R}_{\epsilon} R ϵ is the row of ∂ μ T μ 0 = 0 \partial_{\mu}T^{\mu0}=0 ∂ μ T μ 0 = 0 . Its covariant energy projection is u ν ∇ μ T μ ν = − ∂ μ T μ 0 u_{\nu}\nabla_{\mu}T^{\mu\nu}=-\partial_{\mu}T^{\mu0} u ν ∇ μ T μν = − ∂ μ T μ 0 in the rest frame. Thus the row difference is the Fourier transform of δ [ ∇ μ J μ + r u ν ∇ μ T μ ν ] \delta[\nabla_{\mu}J^{\mu}+ru_{\nu}\nabla_{\mu}T^{\mu\nu}] δ [ ∇ μ J μ + r u ν ∇ μ T μν ] at uniform equilibrium. It is a linear identity. We use ( δ ϵ , h ) (\delta\epsilon,h) ( δ ϵ , h ) , with h h h from Step 4 of Section 1.2.5 and δ ϵ = E δ T + F δ α \delta\epsilon=E\delta T+F\delta\alpha δ ϵ = E δ T + F δ α from (8) . The Jacobian of this change is E q α − F q T = Δ J Eq_{\alpha}-Fq_{T}=\Delta_{J} E q α − F q T = Δ J . On the h = 0 h=0 h = 0 subspace,
δ T T + r T δ α = x w δ ϵ . (87) \frac{\delta T}{T}+r_{T}\delta\alpha=\frac{x}{w}\delta\epsilon. \tag*{(87)} T δ T + r T δ α = w x δ ϵ . (87)
Consequently the remaining energy-momentum block is
\begin{pmatrix}
\varsigma+\frac{x\epsilon_{1}}{w}\varsigma^{2}-\frac{x\theta_{2}}{w}q & ik[w+(\epsilon_{2}+\theta_{1})\varsigma]\\
ik\left[x+\frac{x(\theta_{2}+\pi_{1})}{w}\varsigma\right] & \theta_{1}\varsigma^{2}+w\varsigma+\left(\frac{4}{3}\eta-\pi_{2}\right)q
\end{pmatrix}. \tag*{(88)}\end{pmatrix}
Its determinant is P N / w P_{N}/w P N / w , where P N P_{N} P N is exactly the neutral polynomial (19) with the indicated coefficient choices. Expanding the charge row, including the row/column sign and Jacobian, therefore proves (41) on the original determinant. The shear polynomial remains (16) .
Charged families and inherited properties
We now check the conformal limits and family overlaps of the charged uplifts defined in § 1.5 . For the conformal limits of F 4 F_4 F 4 –F 6 F_6 F 6 , the additional relations (28) are controlled by the differences
ϵ 3 − 3 π 3 = r T ( ϵ 1 − 3 π 1 ) , ν 1 − 3 ν 2 = r ( ϵ 1 − 3 ϵ 2 ) + T ( τ q − 3 D q ) q T , γ 1 − γ 2 = r ( θ 1 − θ 2 ) . (89) \begin{aligned}
\epsilon_3-3\pi_3 &= rT(\epsilon_1-3\pi_1),\\
\nu_1-3\nu_2 &= r(\epsilon_1-3\epsilon_2)+T(\tau_q-3D_q)q_T,\\
\gamma_1-\gamma_2 &= r(\theta_1-\theta_2). \tag*{(89)}
\end{aligned} ϵ 3 − 3 π 3 ν 1 − 3 ν 2 γ 1 − γ 2 = r T ( ϵ 1 − 3 π 1 ) , = r ( ϵ 1 − 3 ϵ 2 ) + T ( τ q − 3 D q ) q T , = r ( θ 1 − θ 2 ) . (89)
At a regular conformal endpoint, we take the conductivity D \mathcal{D} D and the parameter α q \alpha_q α q to have finite limits. The coefficients D q , τ q D_q,\tau_q D q , τ q in (29) are then finite, while B , q T → 0 B,q_T\to0 B , q T → 0 and ζ ∗ − ζ → 0 \zeta_*-\zeta\to0 ζ ∗ − ζ → 0 . The uplift leaves the six neutral stress coefficients unchanged. If these have finite limits satisfying (28) , the differences in (A.44) also vanish, and all additional coefficients in (39) –(40) remain finite. Thus the charged families F 4 F_4 F 4 –F 6 F_6 F 6 have finite conformal limits under the parameter conditions given in § 1.4.2 .
To evaluate the characteristic speeds at these endpoints, we substitute the neutral limits into the quartic and retain the charge factor (27) . For F 5 F_5 F 5 , y − = 0 y_-=0 y − = 0 . At the conformal endpoint with η , D > 0 \eta,\mathcal{D}>0 η , D > 0 , the nonzero squared characteristic speeds are
ρ c = 2 λ + 1 3 λ < 1 , y T , c = 1 4 λ < 1 , y q , c = 1 α q < 1. (90) \rho_c=\frac{2\lambda+1}{3\lambda}<1,\qquad
y_{T,c}=\frac{1}{4\lambda}<1,\qquad
y_{q,c}=\frac{1}{\alpha_q}<1. \tag*{(90)} ρ c = 3 λ 2 λ + 1 < 1 , y T , c = 4 λ 1 < 1 , y q , c = α q 1 < 1. (90)
For example, λ = α q = 2 \lambda=\alpha_q=2 λ = α q = 2 gives the nonzero conformal squared speeds ( 5 / 6 , 1 / 8 , 1 / 2 ) (5/6,1/8,1/2) ( 5/6 , 1/8 , 1/2 ) . The choice β = 0 \beta=0 β = 0 gives thermodynamic frames, while β = 1 \beta=1 β = 1 gives non-thermodynamic frames for x ≠ 1 / 3 x\ne1/3 x = 1/3 . Both approach the same conformal coefficients.
For F 6 F_6 F 6 , the charged principal branches consist of two sound pairs, one charge pair and two shear pairs. At the conformal endpoint, with positive limiting η , D \eta,\mathcal{D} η , D , their squared comoving characteristic speeds are nonzero and strictly subluminal:
y ± = 4 Λ + 1 ± 24 Λ + 1 12 Λ , y T = 1 8 Λ , y q = 1 α q . (91) y_{\pm}=\frac{4\Lambda+1\pm\sqrt{24\Lambda+1}}{12\Lambda},\qquad
y_T=\frac{1}{8\Lambda},\qquad
y_q=\frac{1}{\alpha_q}. \tag*{(91)} y ± = 12Λ 4Λ + 1 ± 24Λ + 1 , y T = 8Λ 1 , y q = α q 1 . (91)
The choice Λ = α q = 2 \Lambda=\alpha_q=2 Λ = α q = 2 gives the result (42) .
The neutral family overlaps proved in Section A.1.5 also hold after applying the common charged uplift at the same physical inputs and the same α q > 1 \alpha_q>1 α q > 1 .
Additional frame freedom preserving the dispersion relations
The charged uplift is not unique. We now derive the additional coefficient freedom in (45) that preserves its dispersion relations and physical transport coefficients. We retain the same equation F q h = 0 F_qh=0 F q h = 0 for the charge fluctuation h = δ n − r δ ϵ h=\delta n-r\delta\epsilon h = δ n − r δ ϵ , implemented by (A.41). Comparing coefficients in the original matrix gives
ν 1 − r ϵ 1 = T τ q q T , ν 3 − r ϵ 3 = τ q q α , γ 2 − r θ 2 = − T D q q T , γ 3 − r θ 3 = − D q q α , ν 2 + γ 1 = r ( ϵ 2 + θ 1 ) . (92) \begin{aligned}
\nu_1-r\epsilon_1 &= T\tau_q q_T, & \nu_3-r\epsilon_3 &= \tau_q q_\alpha,\\
\gamma_2-r\theta_2 &= -TD_q q_T, & \gamma_3-r\theta_3 &= -D_q q_\alpha,\\
\nu_2+\gamma_1 &= r(\epsilon_2+\theta_1). \tag*{(92)}
\end{aligned} ν 1 − r ϵ 1 γ 2 − r θ 2 ν 2 + γ 1 = T τ q q T , = − T D q q T , = r ( ϵ 2 + θ 1 ) . ν 3 − r ϵ 3 γ 3 − r θ 3 = τ q q α , = − D q q α , (92)
Together with the vector entropy relation in (10) , these conditions determine the current coefficients. Four retain the form in (40) , while ν 3 = r ϵ 3 + τ q q α \nu_3 = r\epsilon_3 + \tau_q q_\alpha ν 3 = r ϵ 3 + τ q q α and γ 3 = r θ 3 − D q q α \gamma_3 = r\theta_3 - D_q q_\alpha γ 3 = r θ 3 − D q q α now allow ϵ 3 , θ 3 \epsilon_3,\theta_3 ϵ 3 , θ 3 to differ from (39) . Conductivity matching gives D T = D q Ξ D T = D_q \Xi D T = D q Ξ , as in (29) . We keep τ q \tau_q τ q independent to include zero-diffusion cases.10 For D q > 0 D_q > 0 D q > 0 , we use the stable and causal range in (29) .
The existing frame freedom can be parametrised by three functions. Each pair ( h 1 , h 3 ) = ( ϵ 1 , ϵ 3 ) (h_1,h_3) = (\epsilon_1,\epsilon_3) ( h 1 , h 3 ) = ( ϵ 1 , ϵ 3 ) , ( π 1 , π 3 ) (\pi_1,\pi_3) ( π 1 , π 3 ) or ( θ 2 , θ 3 ) (\theta_2,\theta_3) ( θ 2 , θ 3 ) can be separated into a representative obeying (39) and an additional term. We define
h ˉ = q α h 1 − T q T h 3 Ξ , f h = h 3 − r T h 1 Ξ . (93) \bar{h} = \frac{q_\alpha h_1 - Tq_T h_3}{\Xi}, \qquad f_h = \frac{h_3-rT h_1}{\Xi}. \tag*{(93)} h ˉ = Ξ q α h 1 − T q T h 3 , f h = Ξ h 3 − r T h 1 . (93)
Then
( h 1 , h 3 ) = ( h ˉ , r T h ˉ ) + f h ( T q T , q α ) . (94) (h_1,h_3) = (\bar{h},rT\bar{h}) + f_h(Tq_T,q_\alpha). \tag*{(94)} ( h 1 , h 3 ) = ( h ˉ , r T h ˉ ) + f h ( T q T , q α ) . (94)
The two directions ( 1 , r T ) (1,rT) ( 1 , r T ) and ( T q T , q α ) (Tq_T,q_\alpha) ( T q T , q α ) have determinant Ξ > 0 \Xi> 0 Ξ > 0 , so this decomposition is unique and remains regular when B B B or q T q_T q T vanishes. We denote the representative coefficients by ϵ ˉ 1 , π ˉ 1 , θ ˉ 2 \bar{\epsilon}_1,\bar{\pi}_1,\bar{\theta}_2 ϵ ˉ 1 , π ˉ 1 , θ ˉ 2 and the three functions by f E , f P , f Q f_E,f_P,f_Q f E , f P , f Q . The other representative coefficients are unchanged: ϵ ˉ 2 = ϵ 2 \bar{\epsilon}_2 = \epsilon_2 ϵ ˉ 2 = ϵ 2 , π ˉ 2 = π 2 \bar{\pi}_2 = \pi_2 π ˉ 2 = π 2 and θ ˉ 1 = θ 1 \bar{\theta}_1 = \theta_1 θ ˉ 1 = θ 1 . Together with (A.47) and the vector entropy relation, this decomposition accounts for every choice of frame in the class specified above. The additional terms are precisely the shifts (45) , including their associated current shifts.
Preservation of dispersion relations and physical transport
The preceding decomposition identifies the remaining coefficient freedom. Here we show that varying f E , f P , f Q f_E,f_P,f_Q f E , f P , f Q at fixed representative coefficients, thermodynamic state and charge factor F q F_q F q preserves the dispersion relations and the prescribed η , ζ , D \eta,\zeta,D η , ζ , D . The shifts therefore give additional frame choices with the same frequency roots and characteristic speeds, while changing the individual BDNK coefficients. To see why the dispersion relations are unchanged, we use the variables ( δ ϵ , δ u ∥ , h ) (\delta\epsilon,\delta u_\parallel,h) ( δ ϵ , δ u ∥ , h ) and the matrix reduction of Section A.2.2. Its last equation is F q h = 0 F_q h = 0 F q h = 0 , so the last row is ( 0 , 0 , F q ) (0,0,F_q) ( 0 , 0 , F q ) . The shifts (45) add
( f E ς 2 − f Q q ) h , i k ς ( f P + f Q ) h , (95) (f_E\varsigma^2-f_Qq)h,\qquad ik\varsigma(f_P+f_Q)h, \tag*{(95)} ( f E ς 2 − f Q q ) h , ik ς ( f P + f Q ) h , (95)
to the energy and longitudinal momentum equations, respectively. Both additions multiply h h h , so they change only the last column of these two equations. The last row and the energy-momentum block acting on ( δ ϵ , δ u ∥ ) (\delta\epsilon,\delta u_\parallel) ( δ ϵ , δ u ∥ ) are unchanged. Expanding the determinant along the last row gives F q F_q F q times the determinant of this unchanged block. Restoring the normalization of the original charged matrix therefore gives
P n e w ( ς , q ) = P o l d ( ς , q ) = − Δ J w F q ( ς , q ) P ˉ N ( ς , q ) . (96) P_{\mathrm{new}}(\varsigma,q)=P_{\mathrm{old}}(\varsigma,q)=-\frac{\Delta_J}{w}F_q(\varsigma,q)\bar{P}_N(\varsigma,q). \tag*{(96)} P new ( ς , q ) = P old ( ς , q ) = − w Δ J F q ( ς , q ) P ˉ N ( ς , q ) . (96)
where P ˉ N \bar{P}_N P ˉ N is (19) evaluated at the representative coefficients. This proves that the longitudinal roots, damping rates and gaps are unchanged. The shear polynomial is unchanged because θ 1 \theta_1 θ 1 and η \eta η are fixed. The characteristic speeds obtained from these polynomials are consequently unchanged as well. The identity also holds for complex ς , q \varsigma,q ς , q , giving the boosted statement in footnote 13.
Preservation of physical transport follows from the entropy matching identities (11) . Using T A q T + B q α = 0 T A q_T+B q_\alpha=0 T A q T + B q α = 0 from (A.34) and A + r T B = x A+rT B=x A + r T B = x , the change in the bulk viscosity is
δ ζ = ( T A q T + B q α ) ( f P − x f E ) = 0. (97) \delta\zeta=(T A q_T+B q_\alpha)(f_P-x f_E)=0. \tag*{(97)} δ ζ = ( T A q T + B q α ) ( f P − x f E ) = 0. (97)
The conductivity changes by ( r δ θ 3 − δ γ 3 ) / T = 0 (r\delta\theta_3-\delta\gamma_3)/T=0 ( r δ θ 3 − δ γ 3 ) / T = 0 , and the vector entropy relation changes by δ γ 2 − r δ θ 2 = 0 \delta\gamma_2-r\delta\theta_2=0 δ γ 2 − r δ θ 2 = 0 . Together with δ η = 0 \delta\eta=0 δ η = 0 , these identities prove preservation of all physical transport coefficients and entropy matching.
We now determine which shifts preserve the finite conformal limit of a frame that already satisfies the conformal relations. At a regular conformal endpoint, q T → 0 q_T\to0 q T → 0 and q α → T Δ J / ( 3 w ) > 0 q_\alpha\to T\Delta J/(3w)>0 q α → T Δ J / ( 3 w ) > 0 . If f E , f P , f Q f_E,f_P,f_Q f E , f P , f Q have finite limits, the changes in π 1 − 3 π 2 \pi_1-3\pi_2 π 1 − 3 π 2 , ν 1 − 3 ν 2 \nu_1-3\nu_2 ν 1 − 3 ν 2 , θ 1 − θ 2 \theta_1-\theta_2 θ 1 − θ 2 and γ 1 − γ 2 \gamma_1-\gamma_2 γ 1 − γ 2 vanish because they contain q T q_T q T . The remaining condition is f E − 3 f P → 0 f_E-3f_P\to0 f E − 3 f P → 0 , as stated in § 1.6 , since δ ( ϵ 3 − 3 π 3 ) = q α ( f E − 3 f P ) \delta(\epsilon_3-3\pi_3)=q_\alpha(f_E-3f_P) δ ( ϵ 3 − 3 π 3 ) = q α ( f E − 3 f P ) . Thus f Q f_Q f Q may have any finite limit, while the limiting values of f E f_E f E and f P f_P f P must satisfy this relation. If we also want to retain the original frame’s limiting coefficients themselves, we take all three functions to vanish at conformality. For example, we can choose
f a = c a Γ Ξ ( 3 x − 1 ) 2 1 + ( 3 x − 1 ) 2 , a ∈ { E , P , Q } , c a ∈ R , (98) f_a=c_a\frac{\Gamma}{\Xi}\frac{(3x-1)^2}{1+(3x-1)^2},\qquad a\in\{E,P,Q\},\qquad c_a\in\mathbb{R}, \tag*{(98)} f a = c a Ξ Γ 1 + ( 3 x − 1 ) 2 ( 3 x − 1 ) 2 , a ∈ { E , P , Q } , c a ∈ R , (98)
with finite dimensionless constants c a c_a c a and Γ \Gamma Γ evaluated using the bulk viscosity in (29) . On the regular domain, these functions are finite and vanish as x → 1 / 3 x\to1/3 x → 1/3 , including at Γ = 0 \Gamma=0 Γ = 0 .
Generalisations to arbitrary boosts
The connection between causality and stability in different inertial frames is established in Refs. [9 , 22 , 23 ] . Here we verify stability and causality directly for our families by analysing the sound, shear and charge polynomials in an arbitrarily boosted frame. We check stability for all real wavevectors, including k = 0 k=0 k = 0 , and causal propagation in every spatial direction. The proof also covers cases in which characteristic speeds equal the speed of light.
Boosted sound quartic: gaps, stability and causality
As explained in § 1.2 , boosting a damped mode with a real wavevector in the fluid rest frame generally makes its transformed wavevector complex. To test stability in a moving frame, we instead prescribe a real wavevector k k k in that frame and solve for its complex growth rate ς \varsigma ς . Throughout this section, k = ∣ k ∣ k=|\boldsymbol{k}| k = ∣ k ∣ , and primes denote quantities in the fluid rest frame. We first check the gaps at k = 0 k=0 k = 0 . For k > 0 k>0 k > 0 , we then show that a root cannot acquire a positive real part as the boost varies: it can neither cross the imaginary axis nor enter from infinity. Finally, we check the characteristic speeds for every propagation angle. The common shear and charge proof follows in Section A.3.2, and the full charged system is treated in Section A.3.3.
The characteristic roots are y ± y_{\pm} y ± in (24) , so
A L y 2 − C L y + F L = A L ( y − y − ) ( y − y + ) . (99) A_L y^2-C_L y+F_L=A_L(y-y_-)(y-y_+). \tag*{(99)} A L y 2 − C L y + F L = A L ( y − y − ) ( y − y + ) . (99)
Proposition A.1. Suppose w > 0 w>0 w > 0 , 0 < x < 1 0<x<1 0 < x < 1 , Γ > 0 \Gamma>0 Γ > 0 , A L > 0 A_L>0 A L > 0 , B L > 0 B_L>0 B L > 0 , and the sound inequalities in (25) hold. These are its first three conditions, involving Φ \Phi Φ , L \mathcal{L} L and y ± y_{\pm} y ± . Then every sound root at a real wavevector with k > 0 k>0 k > 0 satisfies Re ς < 0 \operatorname{Re}\varsigma<0 Re ς < 0 for every ∣ v ∣ < 1 |v|<1 ∣ v ∣ < 1 and every propagation angle. At k = 0 k=0 k = 0 there are two roots at ς = 0 \varsigma=0 ς = 0 and two gapped modes with negative real parts. All sound characteristic velocities are real and have magnitude at most one.
Rest-frame stability follows from the earlier Routh–Hurwitz analysis in Section A.1.3 and the gaps in Eq. (21) . The assumptions A L , B L > 0 A_L,B_L>0 A L , B L > 0 imply ε 1 , θ 1 > 0 \varepsilon_1,\theta_1>0 ε 1 , θ 1 > 0 by Eq. (20) . We now establish stability under arbitrary boosts, first checking the k = 0 k=0 k = 0 gaps and then treating k > 0 k>0 k > 0 .
Boosted k = 0 k=0 k = 0 gaps
We align the boost with the z z z axis. In the convention ς = − i ω \varsigma=-i\omega ς = − iω , the Lorentz transformation gives
ς ′ = γ ( ς + i v k z ) , q ′ = k ⊥ 2 + γ 2 ( k z − i v ς ) 2 , γ = ( 1 − v 2 ) − 1 / 2 . (100) \varsigma'=\gamma(\varsigma+ivk_z),\qquad q'=k_\perp^2+\gamma^2(k_z-iv\varsigma)^2,\qquad\gamma=(1-v^2)^{-1/2}. \tag*{(100)} ς ′ = γ ( ς + i v k z ) , q ′ = k ⊥ 2 + γ 2 ( k z − i v ς ) 2 , γ = ( 1 − v 2 ) − 1/2 . (100)
The boosted polynomial is P N , b o o s t ( ς , k z , k ⊥ ) = P N ( ς ′ , q ′ ) P_{N,\mathrm{boost}}(\varsigma,k_z,k_\perp)=P_N(\varsigma',q') P N , boost ( ς , k z , k ⊥ ) = P N ( ς ′ , q ′ ) . Here k ⊥ k_\perp k ⊥ is the magnitude of the wavevector perpendicular to the boost. At k = 0 k=0 k = 0 , we have ς ′ = γ ς \varsigma'=\gamma\varsigma ς ′ = γ ς and q ′ = − γ 2 v 2 ς 2 q'=-\gamma^2v^2\varsigma^2 q ′ = − γ 2 v 2 ς 2 . Substituting these into (19) and removing the factor γ 2 ς 2 \gamma^2\varsigma^2 γ 2 ς 2 associated with the two gapless modes gives
P N , b o o s t ( ς , 0 , 0 ) γ 2 ς 2 = γ 2 ( A L − C L v 2 + F L v 4 ) ς 2 + γ ( B L − D L v 2 ) ς + w 2 ( 1 − x v 2 ) . (101) \frac{P_{N,\mathrm{boost}}(\varsigma,0,0)}{\gamma^2\varsigma^2}=\gamma^2(A_L-C_Lv^2+F_Lv^4)\varsigma^2+\gamma(B_L-D_Lv^2)\varsigma+w^2(1-xv^2). \tag*{(101)} γ 2 ς 2 P N , boost ( ς , 0 , 0 ) = γ 2 ( A L − C L v 2 + F L v 4 ) ς 2 + γ ( B L − D L v 2 ) ς + w 2 ( 1 − x v 2 ) . (101)
The leading coefficient is positive because
A L − C L v 2 + F L v 4 = A L ( 1 − y − v 2 ) ( 1 − y + v 2 ) > 0 ( ∣ v ∣ < 1 ) . (102) A_L-C_Lv^2+F_Lv^4=A_L(1-y_-v^2)(1-y_+v^2)>0\qquad(|v|<1). \tag*{(102)} A L − C L v 2 + F L v 4 = A L ( 1 − y − v 2 ) ( 1 − y + v 2 ) > 0 ( ∣ v ∣ < 1 ) . (102)
Equation (A.14) gives y − ≤ R ≤ y + ≤ 1 y_-\le R\le y_+\le1 y − ≤ R ≤ y + ≤ 1 , so the linear coefficient is positive as well: B L − D L v 2 = B L ( 1 − R v 2 ) > 0 B_L-D_Lv^2=B_L(1-Rv^2)>0 B L − D L v 2 = B L ( 1 − R v 2 ) > 0 . Finally, 0 < x < 1 0<x<1 0 < x < 1 makes the constant term in (A.56) positive. All three coefficients are therefore positive, proving that both gaps have negative real parts, including at luminal saturation y + = 1 y_+=1 y + = 1 .
Stability at every nonzero wavevector
We fix real k z , k ⊥ k_z,k_\perp k z , k ⊥ with k 2 = k z 2 + k ⊥ 2 > 0 k^2=k_z^2+k_\perp^2>0 k 2 = k z 2 + k ⊥ 2 > 0 . Either component may vanish, so this includes propagation parallel and perpendicular to the boost. Suppose that a root were on the imaginary axis of the growth-rate plane, ς = i σ \varsigma=i\sigma ς = iσ with σ ∈ R \sigma\in\mathbb{R} σ ∈ R . Equation (A.55) then gives
ς ′ = i Ω , Ω = γ ( σ + v k z ) , K 2 = q ′ = k ⊥ 2 + γ 2 ( k z + v σ ) 2 ≥ 0. (103) \varsigma'=i\Omega,\qquad\Omega=\gamma(\sigma+vk_z),\qquad K^2=q'=k_\perp^2+\gamma^2(k_z+v\sigma)^2\ge0. \tag*{(103)} ς ′ = i Ω , Ω = γ ( σ + v k z ) , K 2 = q ′ = k ⊥ 2 + γ 2 ( k z + v σ ) 2 ≥ 0. (103)
Thus Ω \Omega Ω and K K K are real. Substitution into (19) gives
Im P N ( i Ω , K 2 ) = Ω ( − B L Ω 2 + D L K 2 ) , (104) \operatorname{Im} P_N(i\Omega,K^2)=\Omega(-B_L\Omega^2+D_LK^2), \tag*{(104)} Im P N ( i Ω , K 2 ) = Ω ( − B L Ω 2 + D L K 2 ) , (104)
Re P N ( i Ω , K 2 ) = A L Ω 4 − ( w 2 + C L K 2 ) Ω 2 + χ w 2 K 2 + F L K 4 . (105) \operatorname{Re} P_N(i\Omega,K^2)=A_L\Omega^4-(w^2+C_LK^2)\Omega^2+\chi w^2K^2+F_LK^4. \tag*{(105)} Re P N ( i Ω , K 2 ) = A L Ω 4 − ( w 2 + C L K 2 ) Ω 2 + χ w 2 K 2 + F L K 4 . (105)
For a root, both parts must vanish. The imaginary part allows only Ω = 0 \Omega=0 Ω = 0 or Ω 2 = R K 2 \Omega^2=RK^2 Ω 2 = R K 2 , since R = D L / B L R=D_L/B_L R = D L / B L . The possibility K = Ω = 0 K=\Omega=0 K = Ω = 0 is excluded: invertibility of the boost would give σ = k z = k ⊥ = 0 \sigma=k_z=k_\perp=0 σ = k z = k ⊥ = 0 , contrary to k > 0 k>0 k > 0 . If Ω = 0 \Omega=0 Ω = 0 , we therefore have K > 0 K>0 K > 0 and Re P N = χ w 2 K 2 + F L K 4 > 0 \operatorname{Re}P_N=\chi w^2K^2+F_LK^4>0 Re P N = χ w 2 K 2 + F L K 4 > 0 . If Ω 2 = R K 2 \Omega^2=RK^2 Ω 2 = R K 2 , then (A.14) instead gives
Re P N = w 2 ( χ − R ) K 2 − A L L K 4 < 0. (106) \operatorname{Re}P_N=w^2(\chi-R)K^2-A_L\mathcal{L}K^4<0. \tag*{(106)} Re P N = w 2 ( χ − R ) K 2 − A L L K 4 < 0. (106)
Here R > χ R>\chi R > χ and L ≥ 0 \mathcal{L}\geq0 L ≥ 0 ensure the strict inequality. Neither case allows a root on the imaginary axis.
It remains to exclude roots entering the right half-plane from infinity. For the fixed wavevector, we vary v v v from zero to any chosen v ∗ v_* v ∗ with ∣ v ∗ ∣ < 1 |v_*|<1 ∣ v ∗ ∣ < 1 . Writing the boosted quartic as ∑ j = 0 4 a j ( v ) ς j \sum_{j=0}^{4}a_j(v)\varsigma^j ∑ j = 0 4 a j ( v ) ς j , its leading coefficient is
a 4 ( v ) = γ 4 A L ( 1 − y − v 2 ) ( 1 − y + v 2 ) > 0. a_4(v)=\gamma^4A_L(1-y_-v^2)(1-y_+v^2)>0. a 4 ( v ) = γ 4 A L ( 1 − y − v 2 ) ( 1 − y + v 2 ) > 0.
It has a positive lower bound on the closed interval between 0 and v ∗ v_* v ∗ . The other coefficients are continuous and bounded there. The polynomial root bound
∣ ς ∣ ≤ 1 + max 0 ≤ j < 4 ∣ a j ( v ) / a 4 ( v ) ∣ (107) |\varsigma|\leq1+\max_{0\leq j<4}\left|a_j(v)/a_4(v)\right| \tag*{(107)} ∣ ς ∣ ≤ 1 + 0 ≤ j < 4 max ∣ a j ( v ) / a 4 ( v ) ∣ (107)
therefore bounds all roots uniformly along this interval. Since roots can neither cross the imaginary axis nor escape to infinity, all four remain in the left half-plane as v v v varies from 0 to v ∗ v_* v ∗ . The wavevector and final boost were arbitrary. Together with the k = 0 k=0 k = 0 gap calculation, this proves stability for every real wavevector and every finite boost.
Causality for all propagation angles
For either squared characteristic speed y = y ± ∈ [ 0 , 1 ] y=y_\pm\in[0,1] y = y ± ∈ [ 0 , 1 ] , the corresponding factor of the principal polynomial (A.54) gives ω ′ 2 = y ∣ k ′ ∣ 2 \omega^{\prime2}=y|k'|^2 ω ′2 = y ∣ k ′ ∣ 2 in the fluid rest frame. To express this in the moving frame, we use the Lorentz transformations underlying (A.55),
ω ′ = γ ( ω − v k z ) , k z ′ = γ ( k z − v ω ) , k ⊥ ′ = k ⊥ . \omega'=\gamma(\omega-vk_z),\qquad k'_z=\gamma(k_z-v\omega),\qquad k'_\perp=k_\perp. ω ′ = γ ( ω − v k z ) , k z ′ = γ ( k z − v ω ) , k ⊥ ′ = k ⊥ .
Writing k z = k cos ϕ k_z=k\cos\phi k z = k cos ϕ , k ⊥ = k sin ϕ k_\perp=k\sin\phi k ⊥ = k sin ϕ and ω = c k \omega=ck ω = c k , with k > 0 k>0 k > 0 , and dividing the rest-frame equation by γ 2 k 2 \gamma^2k^2 γ 2 k 2 gives
( c − v cos ϕ ) 2 = y [ ( cos ϕ − v c ) 2 + ( 1 − v 2 ) sin 2 ϕ ] . (c-v\cos\phi)^2=y\left[(\cos\phi-vc)^2+(1-v^2)\sin^2\phi\right]. ( c − v cos ϕ ) 2 = y [ ( cos ϕ − v c ) 2 + ( 1 − v 2 ) sin 2 ϕ ] .
Expanding both sides gives the quadratic equation for the characteristic velocity c c c in the moving frame,
( 1 − y v 2 ) c 2 − 2 v ( 1 − y ) cos ϕ c + v 2 cos 2 ϕ − y ( 1 − v 2 + v 2 cos 2 ϕ ) = 0. (108) (1-yv^2)c^2-2v(1-y)\cos\phi\,c+v^2\cos^2\phi-y(1-v^2+v^2\cos^2\phi)=0. \tag*{(108)} ( 1 − y v 2 ) c 2 − 2 v ( 1 − y ) cos ϕ c + v 2 cos 2 ϕ − y ( 1 − v 2 + v 2 cos 2 ϕ ) = 0. (108)
Its roots are
c ± ( y , ϕ ) = v ( 1 − y ) cos ϕ ± y ( 1 − v 2 ) [ 1 − v 2 + v 2 ( 1 − y ) sin 2 ϕ ] 1 − y v 2 (109) c_{\pm}(y,\phi)=\frac{v(1-y)\cos\phi\pm\sqrt{y(1-v^2)\left[1-v^2+v^2(1-y)\sin^2\phi\right]}}{1-yv^2} \tag*{(109)} c ± ( y , ϕ ) = 1 − y v 2 v ( 1 − y ) cos ϕ ± y ( 1 − v 2 ) [ 1 − v 2 + v 2 ( 1 − y ) sin 2 ϕ ] (109)
The discriminant is nonnegative and 1 − y v 2 > 0 1-yv^2>0 1 − y v 2 > 0 . Moreover the characteristic covector obeys
∣ k ∣ 2 − ω 2 = ∣ k ′ ∣ 2 − ω ′ 2 = ( 1 − y ) ∣ k ′ ∣ 2 ≥ 0 , \lvert k\rvert^2-\omega^2=\lvert k'\rvert^2-\omega'^2=(1-y)\lvert k'\rvert^2\ge0, ∣ k ∣ 2 − ω 2 = ∣ k ′ ∣ 2 − ω ′2 = ( 1 − y ) ∣ k ′ ∣ 2 ≥ 0 ,
so ∣ c ± ∣ ≤ 1 \lvert c_{\pm}\rvert\le1 ∣ c ± ∣ ≤ 1 . At y = 0 y=0 y = 0 the root is the advective speed v cos ϕ v\cos\phi v cos ϕ . At y = 1 y=1 y = 1 the roots are exactly ± 1 \pm1 ± 1 . Applying this calculation to both y ± y_{\pm} y ± completes the causality proof in Proposition A.1 .
Boosted shear and charge factors
Both the shear and charge factors have the form
Q ( ς , q ) = a ς 2 + b ς + d q , a ≥ d > 0 , b > 0 , Q(\varsigma,q)=a\varsigma^2+b\varsigma+dq,\qquad a\ge d>0,\qquad b>0, Q ( ς , q ) = a ς 2 + b ς + d q , a ≥ d > 0 , b > 0 ,
where a a a , b b b , d d d are constant coefficients at the equilibrium state. At rest, positivity gives two damped roots for q > 0 q>0 q > 0 . After (A.55), the leading coefficient is γ 2 ( a − d v 2 ) > 0 \gamma^2(a-dv^2)>0 γ 2 ( a − d v 2 ) > 0 , and the k = 0 k=0 k = 0 roots are 0 0 0 and − b / [ γ ( a − d v 2 ) ] < 0 -b/[\gamma(a-dv^2)]<0 − b / [ γ ( a − d v 2 )] < 0 . For k > 0 k>0 k > 0 , an imaginary-axis root would satisfy
Im Q ( i Ω , K 2 ) = b Ω = 0 , Re Q ( 0 , K 2 ) = d K 2 > 0 , \operatorname{Im}Q(i\Omega,K^2)=b\Omega=0,\qquad\operatorname{Re}Q(0,K^2)=dK^2>0, Im Q ( i Ω , K 2 ) = b Ω = 0 , Re Q ( 0 , K 2 ) = d K 2 > 0 ,
where Ω \Omega Ω , K K K are defined in (A.58). The second inequality follows from K > 0 K>0 K > 0 when Ω = 0 \Omega=0 Ω = 0 and k > 0 k>0 k > 0 , and excludes such a root. The leading coefficient is bounded away from zero on each closed finite boost interval, so the root-bound argument of Section A.3.1 excludes roots entering from infinity. Both roots therefore remain damped at every nonzero real wavevector. Causality follows from (A.64) with y = d / a ≤ 1 y=d/a\le1 y = d / a ≤ 1 .
For the shear sector, ( a , b , d ) = ( θ 1 , w , η ) (a,b,d)=(\theta_1,w,\eta) ( a , b , d ) = ( θ 1 , w , η ) , so the required bound is precisely (18) . Each transverse polarization has the gap
ς T , g a p = − w γ ( θ 1 − η v 2 ) < 0. (110) \varsigma_{\mathrm{T,gap}}=-\frac{w}{\gamma(\theta_1-\eta v^2)}<0. \tag*{(110)} ς T , gap = − γ ( θ 1 − η v 2 ) w < 0. (110)
No relation between θ 1 \theta_1 θ 1 and θ 2 \theta_2 θ 2 is used, and the proof includes the luminal endpoint θ 1 = η \theta_1=\eta θ 1 = η .11 For the charge sector, we take D > 0 \mathcal{D}>0 D > 0 and τ q = α q D q ≥ D q > 0 \tau_q=\alpha_qD_q\ge D_q>0 τ q = α q D q ≥ D q > 0 . The same proof applies with ( a , b , d ) = ( τ q , 1 , D q ) (a,b,d)=(\tau_q,1,D_q) ( a , b , d ) = ( τ q , 1 , D q ) , establishing stability at all real wavevectors and the squared comoving characteristic speed y q = D q / τ q ≤ 1 y_q=D_q/\tau_q\le1 y q = D q / τ q ≤ 1 .
Stability and causality of the full boosted charged system
We now apply the charged uplift to a neutral seed satisfying Proposition A.1 , with Γ > 0 \Gamma>0 Γ > 0 and θ 1 ≥ η > 0 \theta_1\ge\eta>0 θ 1 ≥ η > 0 . We first take D > 0 \mathcal{D}>0 D > 0 , so D q > 0 D_q>0 D q > 0 , and choose τ q = α q D q \tau_q=\alpha_qD_q τ q = α q D q with α q ≥ 1 \alpha_q\ge1 α q ≥ 1 .12
factorization (41) holds for complex ς \varsigma ς , q q q and therefore remains valid after the simultaneous substitution (A.55). The sound factor is covered by Section A.3.1, and both the shear and charge factors by Section A.3.2. Together these results establish stability at nonzero real wavevectors. We now extract the combined k = 0 k=0 k = 0 gaps and collect the causality results for the full system.
At k = 0 k=0 k = 0 , we extract the gaps directly from the boosted determinant. For the charge factor, (A.55) gives F q ( γ ς , − γ 2 v 2 ς 2 ) = γ ς ( 1 + ℓ q ς ) F_q(\gamma\varsigma,-\gamma^2v^2\varsigma^2)=\gamma\varsigma(1+\ell_q\varsigma) F q ( γ ς , − γ 2 v 2 ς 2 ) = γ ς ( 1 + ℓ q ς ) . The sound factor is γ 2 ς 2 \gamma^2\varsigma^2 γ 2 ς 2 times the quadratic in (A.56). The coefficients in these two factors are
ℓ q = γ ( τ q − D q v 2 ) > 0 , a g = γ 2 ( A L − C L v 2 + F L v 4 ) > 0 , \ell_q=\gamma(\tau_q-D_qv^2)>0,\qquad a_g=\gamma^2(A_L-C_Lv^2+F_Lv^4)>0, ℓ q = γ ( τ q − D q v 2 ) > 0 , a g = γ 2 ( A L − C L v 2 + F L v 4 ) > 0 ,
b g = γ ( B L − D L v 2 ) > 0 , c g = w 2 ( 1 − χ v 2 ) > 0. (111) b_g=\gamma(B_L-D_Lv^2)>0,\qquad c_g=w^2(1-\chi v^2)>0. \tag*{(111)} b g = γ ( B L − D L v 2 ) > 0 , c g = w 2 ( 1 − χ v 2 ) > 0. (111)
Their positivity follows from τ q ≥ D q > 0 \tau_q\ge D_q>0 τ q ≥ D q > 0 and the sound-gap calculation above. Multiplying the two factors gives
P c h , b o o s t ( ς , 0 ) γ 3 ς 3 = − Δ J w ( 1 + ℓ q ς ) ( a g ς 2 + b g ς + c g ) . (112) \frac{P_{\mathrm{ch,boost}}(\varsigma,0)}{\gamma^3\varsigma^3}
=-\frac{\Delta_J}{w}(1+\ell_q\varsigma)(a_g\varsigma^2+b_g\varsigma+c_g). \tag*{(112)} γ 3 ς 3 P ch , boost ( ς , 0 ) = − w Δ J ( 1 + ℓ q ς ) ( a g ς 2 + b g ς + c g ) . (112)
The linear factor gives the negative gap − 1 / ℓ q -1/\ell_q − 1/ ℓ q . The quadratic has positive coefficients a g , b g , c g a_g,b_g,c_g a g , b g , c g , so both of its roots have negative real parts. This proves stability of all three longitudinal gaps directly from the original boosted determinant. The extracted factor ς 3 \varsigma^3 ς 3 accounts for the three zero-frequency longitudinal modes. The two shear gaps are given by (A.65).
For causality, the squared speeds in the fluid rest frame are y − , y + , y T , y q y_-,y_+,y_T,y_q y − , y + , y T , y q , as listed in Table 1 . Applying (A.64) to each gives real characteristic velocities with ∣ c ∣ ≤ 1 |c|\le1 ∣ c ∣ ≤ 1 in every finite boosted frame. When y < 1 y<1 y < 1 the inequality is strict, including the advective branch c = v cos ϕ c=v\cos\phi c = v cos ϕ at y = 0 y=0 y = 0 . At y = 1 y=1 y = 1 the velocities are ± 1 \pm1 ± 1 .13
Comments on existing frames
The constitutive relations of Ref. [9 ] provide another route to stable and causal frames with prescribed physical transport. In this appendix, we show this directly for neutral fluids and for charged fluids at nonzero density, and comment on the obstructions for zero density states.
Constitutive class and neutral construction
In our notation, the charged class of Ref. [9 ] takes the form
E = τ E ( D ϵ + w Θ ) , P = τ P ( D ϵ + w Θ ) − ζ Θ , \mathcal{E}=\tau_{\mathcal{E}}(D\epsilon+w\Theta),\qquad
\mathcal{P}=\tau_{\mathcal{P}}(D\epsilon+w\Theta)-\zeta\Theta, E = τ E ( D ϵ + w Θ ) , P = τ P ( D ϵ + w Θ ) − ζ Θ ,
Q μ = τ Q ( w a μ + ∇ ⊥ μ p ) + D T w n ∇ ⊥ μ α , J μ = n u μ . (113) Q^\mu=\tau_Q(wa^\mu+\nabla_\perp^\mu p)+\frac{DTw}{n}\nabla_\perp^\mu\alpha,\qquad
J^\mu=nu^\mu. \tag*{(113)} Q μ = τ Q ( w a μ + ∇ ⊥ μ p ) + n D T w ∇ ⊥ μ α , J μ = n u μ . (113)
The three relaxation parameters τ ε , τ P , τ Q \tau_\varepsilon,\tau_P,\tau_Q τ ε , τ P , τ Q parametrise the frame choice. The coefficients η \eta η , ζ \zeta ζ and D D D remain physical inputs. Using T ε T = w / x T\varepsilon_T=w/x T ε T = w / x , the neutral coefficients become
ε 1 = w τ ε x , ε 2 = w τ ε , π 1 = w τ P x , \varepsilon_1=\frac{w\tau_\varepsilon}{x},\qquad\varepsilon_2=w\tau_\varepsilon,\qquad\pi_1=\frac{w\tau_P}{x}, ε 1 = x w τ ε , ε 2 = w τ ε , π 1 = x w τ P ,
π 2 = w τ P − ζ , θ 1 = θ 2 = w τ Q . (114) \pi_2=w\tau_P-\zeta,\qquad\theta_1=\theta_2=w\tau_Q. \tag*{(114)} π 2 = w τ P − ζ , θ 1 = θ 2 = w τ Q . (114)
For both the neutral and charged calculations below, we write d = ( 4 η / 3 + ζ ) / w d=(4\eta/3+\zeta)/w d = ( 4 η /3 + ζ ) / w , using the physical bulk viscosity. In the neutral case this is Γ / w \Gamma/w Γ/ w , with Γ \Gamma Γ as in (20) , and a sufficient prescription in order to attain causal and stable fluids for generic EoS and prescribed transport is
τ ε = τ Q = τ , τ P = x τ , τ = d [ 1 x + 1 + x ( 1 − x ) 2 ] . (115) \tau_\varepsilon=\tau_Q=\tau,\qquad\tau_P=x\tau,\qquad\tau=d\left[\frac{1}{x}+\frac{1+x}{(1-x)^2}\right]. \tag*{(115)} τ ε = τ Q = τ , τ P = xτ , τ = d [ x 1 + ( 1 − x ) 2 1 + x ] . (115)
For d > 0 d>0 d > 0 , all relaxation parameters are positive. Substitution into the original matrix (A.8) gives the normalized polynomial P ^ N = P N / w 2 = x det M / w 2 \widehat{P}_N=P_N/w^2=x\det M/w^2 P N = P N / w 2 = x det M / w 2 :
P ^ N ( ς , q ) = τ 2 ς 4 + 2 τ ς 3 + [ 1 + ( 2 x τ 2 + d τ ) q ] ς 2 + ( 2 x τ + d ) q ς + x q + ( x 2 τ 2 − x d τ ) q 2 . (116) \widehat{P}_N(\varsigma,q)=\tau^2\varsigma^4+2\tau\varsigma^3+\left[1+(2x\tau^2+d\tau)q\right]\varsigma^2
+(2x\tau+d)q\varsigma+xq+(x^2\tau^2-xd\tau)q^2. \tag*{(116)} P N ( ς , q ) = τ 2 ς 4 + 2 τ ς 3 + [ 1 + ( 2 x τ 2 + d τ ) q ] ς 2 + ( 2 xτ + d ) q ς + x q + ( x 2 τ 2 − x d τ ) q 2 . (116)
Since τ > d / x \tau>d/x τ > d / x , its coefficients are positive for q > 0 q>0 q > 0 . The Hurwitz determinants of this normalized polynomial are
Δ 1 = 2 τ , Δ 2 = 2 τ + τ 2 ( 2 x τ + d ) q , \Delta_1=2\tau,\qquad\Delta_2=2\tau+\tau^2(2x\tau+d)q, Δ 1 = 2 τ , Δ 2 = 2 τ + τ 2 ( 2 xτ + d ) q ,
Δ 3 = τ d q [ 2 + τ ( 8 x τ + d ) q ] , \Delta_3=\tau dq[2+\tau(8x\tau+d)q], Δ 3 = τ d q [ 2 + τ ( 8 xτ + d ) q ] ,
Δ 4 = [ x q + ( x 2 τ 2 − x d τ ) q 2 ] Δ 3 . (117) \Delta_4=[xq+(x^2\tau^2-xd\tau)q^2]\Delta_3. \tag*{(117)} Δ 4 = [ x q + ( x 2 τ 2 − x d τ ) q 2 ] Δ 3 . (117)
They are strictly positive for every q > 0 q>0 q > 0 . At q = 0 q=0 q = 0 , P ^ N = ς 2 ( τ ς + 1 ) 2 \widehat{P}_N=\varsigma^2(\tau\varsigma+1)^2 P N = ς 2 ( τ ς + 1 ) 2 . The squared longitudinal characteristic speeds are
y ± = x + δ τ 2 ± 1 2 δ τ 2 + 8 x δ τ , δ τ = d τ . (118) y_\pm=x+\frac{\delta_\tau}{2}\pm\frac{1}{2}\sqrt{\delta_\tau^2+8x\delta_\tau},\qquad\delta_\tau=\frac{d}{\tau}. \tag*{(118)} y ± = x + 2 δ τ ± 2 1 δ τ 2 + 8 x δ τ , δ τ = τ d . (118)
The prescription gives 0 < δ τ < x 0<\delta_\tau<x 0 < δ τ < x and δ τ < ( 1 − x ) 2 / ( 1 + x ) \delta_\tau<(1-x)^2/(1+x) δ τ < ( 1 − x ) 2 / ( 1 + x ) . The speed polynomial has positive product x ( x − δ τ ) x(x-\delta_\tau) x ( x − δ τ ) , positive sum 2 x + δ τ < 2 2x+\delta_\tau<2 2 x + δ τ < 2 , and positive value ( 1 − x ) 2 − δ τ ( 1 + x ) (1-x)^2-\delta_\tau(1+x) ( 1 − x ) 2 − δ τ ( 1 + x ) at squared speed one. Its distinct real roots therefore obey 0 < y − < y + < 1 0<y_-<y_+<1 0 < y − < y + < 1 . The shear polynomial, divided by w w w , is τ ς 2 + ς + ( η / w ) q \tau\varsigma^2+\varsigma+(\eta/w)q τ ς 2 + ς + ( η / w ) q . Its squared speed η / ( w τ ) < 1 \eta/(w\tau)<1 η / ( w τ ) < 1 follows from τ > d / x ≥ 4 η / ( 3 w x ) \tau>d/x\ge4\eta/(3wx) τ > d / x ≥ 4 η / ( 3 w x ) when η > 0 \eta>0 η > 0 . At η = 0 \eta=0 η = 0 it factors as ς ( τ ς + 1 ) \varsigma(\tau\varsigma+1) ς ( τ ς + 1 ) , retaining an advective root. We discuss the case of arbitrary boosts in Section B.3.
If d = 0 d=0 d = 0 , both viscosities vanish and (B.3) sets all frame coefficients to zero. The reduced system has ideal sound ς 2 + x q = 0 \varsigma^2+xq=0 ς 2 + x q = 0 and advected transverse perturbations. At a regular conformal limit x → 1 / 3 x\to1/3 x → 1/3 , ζ → 0 \zeta\to0 ζ → 0 , the prescription instead gives the finite values τ → 8 η / w \tau\to8\eta/w τ → 8 η / w , ε 1 → 24 η \varepsilon_1\to24\eta ε 1 → 24 η , ε 2 , π 1 → 8 η \varepsilon_2,\pi_1\to8\eta ε 2 , π 1 → 8 η , π 2 → 8 η / 3 \pi_2\to8\eta/3 π 2 → 8 η /3 and θ 1 , θ 2 → 8 η \theta_1,\theta_2\to8\eta θ 1 , θ 2 → 8 η , which obey (28) .
The class of frames introduced in (B.2) requires ε 2 = x ε 1 \varepsilon_2=x\varepsilon_1 ε 2 = x ε 1 . Entropy matching (11) then gives π 2 = x π 1 − ζ \pi_2=x\pi_1-\zeta π 2 = x π 1 − ζ . The subset of F 4 F_4 F 4 specified in footnote 6 satisfies these relations. On the domain η > 0 \eta>0 η > 0 , F 1 F_1 F 1 -F 3 F_3 F 3 do not, while F 5 F_5 F 5 and F 6 F_6 F 6 satisfy them at x = 1 / 3 x=1/3 x = 1/3 only.
Charged construction at nonzero density
In the charged case and for n ≠ 0 n \ne0 n = 0 , the coefficients of (B.1) in our basis are
( ϵ 1 , ϵ 2 , ϵ 3 ) = τ ϵ ( T E , w , F ) , ( π 1 , π 2 , π 3 ) = ( τ P T E , τ P w − ζ , τ P F ) , θ 1 = θ 2 = τ Q w , θ 3 = τ Q r w T + D T r , ν i = γ i = 0. (119) \begin{aligned}
(\epsilon_1,\epsilon_2,\epsilon_3) &= \tau_{\epsilon}(TE,w,F),\\
(\pi_1,\pi_2,\pi_3) &= (\tau_PTE,\tau_Pw-\zeta,\tau_PF),\\
\theta_1 &= \theta_2=\tau_Qw,\qquad\theta_3=\tau_QrwT+\frac{\mathcal{D}T}{r},\qquad\nu_i=\gamma_i=0. \tag*{(119)}
\end{aligned} ( ϵ 1 , ϵ 2 , ϵ 3 ) ( π 1 , π 2 , π 3 ) θ 1 = τ ϵ ( T E , w , F ) , = ( τ P T E , τ P w − ζ , τ P F ) , = θ 2 = τ Q w , θ 3 = τ Q r w T + r D T , ν i = γ i = 0. (119)
Equations (11) give precisely the prescribed ζ \zeta ζ and D \mathcal{D} D . All identities in (10) hold. With d d d as above, we introduce
g I = D T B r w , h I = D T E Δ J , K I = x ( h I − D q ) = ( x − A ) g I = D T 2 B 2 w ≥ 0. (120) \begin{aligned}
g_I &= \frac{\mathcal{D}TB}{rw},\qquad h_I=\frac{\mathcal{D}TE}{\Delta_J},\\
K_I &= x(h_I-D_q)=(x-A)g_I=\frac{\mathcal{D}T^2B^2}{w}\ge0. \tag*{(120)}
\end{aligned} g I K I = r w D T B , h I = Δ J D T E , = x ( h I − D q ) = ( x − A ) g I = w D T 2 B 2 ≥ 0. (120)
Here x D q = D w / Δ J xD_q=\mathcal{D}w/\Delta_J x D q = D w / Δ J follows from (29) . The remaining identities use E = w / ( T x ) + T Δ J B 2 / ( w x ) E=w/(Tx)+T\Delta_JB^2/(wx) E = w / ( T x ) + T Δ J B 2 / ( w x ) and (7) . In particular h I , D q ≥ 0 h_I,D_q\ge0 h I , D q ≥ 0 , whereas A A A and g I g_I g I can have either sign. For charged fluids, the Γ \Gamma Γ used in our uplift satisfies Γ / w = d + K I / x \Gamma/w=d+K_I/x Γ/ w = d + K I / x , whereas d d d contains only the physical viscosities. For the original matrix (A.40), direct expansion gives det M c h a r g e d = − Δ J w P I \det M_{\mathrm{charged}}=-\Delta_JwP_I det M charged = − Δ J w P I , with
P I ( ς , q ) = τ ϵ τ Q ς 5 + ( τ ϵ + τ Q ) ς 4 + ( 1 + H q ) ς 3 + D q ς 2 + ( x q + C q 2 ) ς + x D q q 2 . (121) P_I(\varsigma,q)=\tau_{\epsilon}\tau_Q\varsigma^5+(\tau_{\epsilon}+\tau_Q)\varsigma^4+(1+Hq)\varsigma^3+D_q\varsigma^2+(xq+Cq^2)\varsigma+xD_qq^2. \tag*{(121)} P I ( ς , q ) = τ ϵ τ Q ς 5 + ( τ ϵ + τ Q ) ς 4 + ( 1 + H q ) ς 3 + D q ς 2 + ( x q + C q 2 ) ς + x D q q 2 . (121)
H = τ ϵ ( x τ Q + d + g I ) + τ P τ Q , D = x ( τ ϵ + τ Q ) + d + h I , C = τ P ( x τ Q + g I ) − d ( τ Q A + g I − h I ) . (122) \begin{aligned}
H &= \tau_{\epsilon}(x\tau_Q+d+g_I)+\tau_P\tau_Q,\\
D &= x(\tau_{\epsilon}+\tau_Q)+d+h_I,\\
C &= \tau_P(x\tau_Q+g_I)-d(\tau_QA+g_I-h_I). \tag*{(122)}
\end{aligned} H D C = τ ϵ ( x τ Q + d + g I ) + τ P τ Q , = x ( τ ϵ + τ Q ) + d + h I , = τ P ( x τ Q + g I ) − d ( τ Q A + g I − h I ) . (122)
The shear polynomial is τ Q ς 2 + ς + ( η / w ) q \tau_Q\varsigma^2+\varsigma+(\eta/w)q τ Q ς 2 + ς + ( η / w ) q . We can pick any positive time scale ℓ \ell ℓ and set
a ⋆ = 1 + ( A / x ) 2 , y ⋆ = x − x ( 1 − x ) g I ℓ ℓ 2 + g I 2 , τ Q = τ ⋆ , τ ϵ = a ⋆ τ ⋆ , τ P = a ⋆ y ⋆ τ ⋆ . (123) \begin{aligned}
a_{\star} &= 1+(A/x)^2,\qquad y_{\star}=x-x(1-x)\frac{g_I\ell}{\ell^2+g_I^2},\\
\tau_Q &= \tau_{\star},\qquad\tau_{\epsilon}=a_{\star}\tau_{\star},\qquad\tau_P=a_{\star}y_{\star}\tau_{\star}. \tag*{(123)}
\end{aligned} a ⋆ τ Q = 1 + ( A / x ) 2 , y ⋆ = x − x ( 1 − x ) ℓ 2 + g I 2 g I ℓ , = τ ⋆ , τ ϵ = a ⋆ τ ⋆ , τ P = a ⋆ y ⋆ τ ⋆ . (123)
Our aim is to choose a finite, sufficiently large τ ⋆ > 0 \tau_{\star}>0 τ ⋆ > 0 , without changing the physical inputs. Since ∣ g I ℓ / ( ℓ 2 + g I 2 ) ∣ ≤ 1 / 2 \lvert g_I\ell/(\ell^2+g_I^2)\rvert\le1/2 ∣ g I ℓ / ( ℓ 2 + g I 2 )∣ ≤ 1/2 , we have 0 < y ⋆ < 1 0<y_{\star}<1 0 < y ⋆ < 1 and
g I ( x − y ⋆ ) = x ( 1 − x ) ℓ g I 2 ℓ 2 + g I 2 ≥ 0 , S ⋆ = a ⋆ x + A > 0 , M ⋆ = y ⋆ + a ⋆ x + A ( 1 + 1 / a ⋆ ) > 0. (124) \begin{aligned}
g_I(x-y_{\star}) &= \frac{x(1-x)\ell g_I^2}{\ell^2+g_I^2}\ge0,\\
S_{\star} &= a_{\star}x+A>0,\qquad M_{\star}=y_{\star}+a_{\star}x+A(1+1/a_{\star})>0. \tag*{(124)}
\end{aligned} g I ( x − y ⋆ ) S ⋆ = ℓ 2 + g I 2 x ( 1 − x ) ℓ g I 2 ≥ 0 , = a ⋆ x + A > 0 , M ⋆ = y ⋆ + a ⋆ x + A ( 1 + 1/ a ⋆ ) > 0. (124)
Indeed, S ⋆ = x [ ( A / x + 1 / 2 ) 2 + 3 / 4 ] S_{\star}=x[(A/x+1/2)^2+3/4] S ⋆ = x [( A / x + 1/2 ) 2 + 3/4 ] . For A ≥ 0 A\ge0 A ≥ 0 , positivity of M ⋆ M_{\star} M ⋆ is immediate. For A < 0 A<0 A < 0 , write z A = − A / x > 0 z_A=-A/x>0 z A = − A / x > 0 . Then a ⋆ = 1 + z A 2 ≥ 2 z A a_{\star}=1+z_A^2\ge2z_A a ⋆ = 1 + z A 2 ≥ 2 z A gives M ⋆ ≥ y ⋆ + x ( a ⋆ − 1 ) / 2 > 0 M_{\star}\ge y_{\star}+x(a_{\star}-1)/2>0 M ⋆ ≥ y ⋆ + x ( a ⋆ − 1 ) /2 > 0 . Consequently
Z ⋆ = d x M ⋆ + g I ( x − y ⋆ ) S ⋆ > 0 , d x + K I > 0 , if d > 0 or g I ≠ 0. (125) Z_{\star}=dxM_{\star}+g_I(x-y_{\star})S_{\star}>0,\qquad dx+K_I>0,\qquad\text{if }d>0\text{ or }g_I\ne0. \tag*{(125)} Z ⋆ = d x M ⋆ + g I ( x − y ⋆ ) S ⋆ > 0 , d x + K I > 0 , if d > 0 or g I = 0. (125)
First assume d > 0 d > 0 d > 0 or g I ≠ 0 g_I \ne0 g I = 0 . For each of the first four Hurwitz determinants of (B.9), the table lists every nonzero coefficient of a power of q q q , its degree in τ ⋆ \tau_\star τ ⋆ , and its leading coefficient in which the relation x h I = x D q + ( x − A ) g I x h_I = x D_q + (x - A)g_I x h I = x D q + ( x − A ) g I has been used.
Table 2.
All entries in the last column are positive. There are only finitely many coefficient polynomials, so one finite τ ⋆ \tau_\star τ ⋆ makes all of them positive simultaneously, uniformly for every q > 0 q > 0 q > 0 at the given state. If D > 0 D > 0 D > 0 , then D q > 0 D_q > 0 D q > 0 and Δ 5 = x D q q 2 Δ 4 > 0 \Delta_5 = xD_q q^2\Delta_4 > 0 Δ 5 = x D q q 2 Δ 4 > 0 , proving strict rest-frame stability. If D = 0 D = 0 D = 0 , then g I = h I = D q = K I = 0 g_I = h_I = D_q = K_I = 0 g I = h I = D q = K I = 0 and P I = ς P 4 P_I = \varsigma P_4 P I = ς P 4 . The first four Hurwitz determinants are those of P 4 P_4 P 4 and remain positive for d > 0 d > 0 d > 0 , while the exact advective root is retained. At q = 0 q = 0 q = 0 , P I = ς 3 ( τ ϵ ς + 1 ) ( τ Q ς + 1 ) P_I = \varsigma^3(\tau_\epsilon\varsigma+ 1)(\tau_Q\varsigma+ 1) P I = ς 3 ( τ ϵ ς + 1 ) ( τ Q ς + 1 ) .
To choose a sufficiently large τ ⋆ \tau_\star τ ⋆ , we first fix the physical inputs and choose any reference time ℓ > 0 \ell> 0 ℓ > 0 . We write τ ⋆ = ℓ z \tau_\star= \ell z τ ⋆ = ℓ z , so that only the dimensionless parameter z z z remains to be chosen. Each required inequality takes the form p ⋆ ( z ) > 0 p_\star(z) > 0 p ⋆ ( z ) > 0 , where p ⋆ ( z ) = c m z m + ⋯ + c 0 p_\star(z) = c_m z^m + \cdots+ c_0 p ⋆ ( z ) = c m z m + ⋯ + c 0 has a positive leading coefficient c m c_m c m . For large z z z , the positive highest-power term dominates the remaining terms. An explicit sufficient bound is therefore
z > 1 + 1 c m ∑ j = 0 m − 1 ∣ c j ∣ . z > 1 + \frac{1}{c_m}\sum_{j=0}^{m-1}|c_j|. z > 1 + c m 1 j = 0 ∑ m − 1 ∣ c j ∣.
For example, labelling these finitely many polynomials by i i i and writing p i ( z ) = ∑ j = 0 m i c i j z j p_i(z) = \sum_{j=0}^{m_i} c_{ij}z^j p i ( z ) = ∑ j = 0 m i c ij z j , where m i m_i m i is the degree at the chosen state and c i m i > 0 c_{im_i} > 0 c i m i > 0 , an explicit sufficient choice is
τ ⋆ = ℓ [ 2 + max i ∑ j = 0 m i − 1 ∣ c i j ∣ c i m i ] . \tau_\star= \ell\left[2 + \max_i \sum_{j=0}^{m_i-1}\frac{|c_{ij}|}{c_{im_i}}\right]. τ ⋆ = ℓ [ 2 + i max j = 0 ∑ m i − 1 c i m i ∣ c ij ∣ ] .
The maximum includes the coefficient inequalities in the table and the characteristic and gap inequalities below. This choice makes z = τ ⋆ / ℓ z = \tau_\star/\ell z = τ ⋆ / ℓ strictly larger than every required bound, so all these conditions hold for the prescribed inputs. The reference time ℓ \ell ℓ remains fixed. The resulting τ ⋆ \tau_\star τ ⋆ is finite at the chosen state, although it need not be the smallest admissible value. Writing y = c 2 y = c^2 y = c 2 as in (24) , the principal sound speeds squared are the roots of
a ⋆ τ ⋆ 2 y 2 − H y + C = 0 , a_\star\tau_\star^2 y^2 - Hy + C = 0, a ⋆ τ ⋆ 2 y 2 − H y + C = 0 ,
H = a ⋆ ( x + y ⋆ ) τ ⋆ 2 + a ⋆ ( d + g I ) τ ⋆ , C = a ⋆ x y ⋆ τ ⋆ 2 + ( a ⋆ y ⋆ g I − A d ) τ ⋆ − d ( g I − h I ) . (126) H = a_\star(x + y_\star)\tau_\star^2 + a_\star(d + g_I)\tau_\star,\qquad C = a_\star x y_\star\tau_\star^2 + (a_\star y_\star g_I - Ad)\tau_\star- d(g_I - h_I). \tag*{(126)} H = a ⋆ ( x + y ⋆ ) τ ⋆ 2 + a ⋆ ( d + g I ) τ ⋆ , C = a ⋆ x y ⋆ τ ⋆ 2 + ( a ⋆ y ⋆ g I − A d ) τ ⋆ − d ( g I − h I ) . (126)
Their discriminant is
H 2 − 4 a ⋆ τ ⋆ 2 C = a ⋆ 2 ( x − y ⋆ ) 2 τ ⋆ 4 + 2 a ⋆ { d [ a ⋆ ( x + y ⋆ ) + 2 A ] + a ⋆ g I ( x − y ⋆ ) } τ ⋆ 3 + { a ⋆ 2 ( d + g I ) 2 + 4 a ⋆ d ( g I − h I ) } τ ⋆ 2 . (127) H^2-4a_{\star}\tau_{\star}^2C=a_{\star}^2(x-y_{\star})^2\tau_{\star}^4+2a_{\star}\{d[a_{\star}(x+y_{\star})+2A]+a_{\star}g_I(x-y_{\star})\}\tau_{\star}^3+\{a_{\star}^2(d+g_I)^2+4a_{\star}d(g_I-h_I)\}\tau_{\star}^2. \tag*{(127)} H 2 − 4 a ⋆ τ ⋆ 2 C = a ⋆ 2 ( x − y ⋆ ) 2 τ ⋆ 4 + 2 a ⋆ { d [ a ⋆ ( x + y ⋆ ) + 2 A ] + a ⋆ g I ( x − y ⋆ )} τ ⋆ 3 + { a ⋆ 2 ( d + g I ) 2 + 4 a ⋆ d ( g I − h I )} τ ⋆ 2 . (127)
The cubic coefficient is positive: a ⋆ ( x + y ⋆ ) + 2 A > 0 a_{\star}(x+y_{\star})+2A>0 a ⋆ ( x + y ⋆ ) + 2 A > 0 follows, for negative A A A , from a ⋆ ≥ 2 ∣ A ∣ / x a_{\star}\ge2\lvert A\rvert/x a ⋆ ≥ 2 ∣ A ∣ / x , and g I ( x − y ⋆ ) > 0 g_I(x-y_{\star})>0 g I ( x − y ⋆ ) > 0 if g I ≠ 0 g_I\ne0 g I = 0 . The discriminant is therefore positive for sufficiently large τ ⋆ \tau_{\star} τ ⋆ , also when x = y ⋆ x=y_{\star} x = y ⋆ and its quartic coefficient vanishes. For large τ ⋆ \tau_{\star} τ ⋆ , the inequalities
C > 0 , H > 0 , 2 a ⋆ τ ⋆ 2 − H > 0 , a ⋆ τ ⋆ 2 − H + C > 0 (128) C>0,\qquad H>0,\qquad2a_{\star}\tau_{\star}^2-H>0,\qquad a_{\star}\tau_{\star}^2-H+C>0 \tag*{(128)} C > 0 , H > 0 , 2 a ⋆ τ ⋆ 2 − H > 0 , a ⋆ τ ⋆ 2 − H + C > 0 (128)
follow from their positive leading coefficients. The last is a ⋆ ( 1 − x ) ( 1 − y ⋆ ) a_{\star}(1-x)(1-y_{\star}) a ⋆ ( 1 − x ) ( 1 − y ⋆ ) . As in the neutral case, these give two distinct squared speeds 0 < y − < y + < 1 0<y_-<y_+<1 0 < y − < y + < 1 , together with the advective characteristic. Taking τ ⋆ > η / w \tau_{\star}>\eta/w τ ⋆ > η / w also makes the shear cone subluminal.
Cases with vanishing transport
The case d = g I = 0 d=g_I=0 d = g I = 0 is not covered by the strict Hurwitz argument above. It implies η = ζ = 0 \eta=\zeta=0 η = ζ = 0 and D q = h I D_q=h_I D q = h I . Direct factorisation of the original quintic gives
P I = ( ς 2 + x q ) [ a ⋆ τ ⋆ 2 ς 3 + ( a ⋆ + 1 ) τ ⋆ ς 2 + ( 1 + a ⋆ y ⋆ τ q 2 ) ς + h I q ] . (129) P_I=(\varsigma^2+xq)\left[a_{\star}\tau_{\star}^2\varsigma^3+(a_{\star}+1)\tau_{\star}\varsigma^2+(1+a_{\star}y_{\star}\tau_q^2)\varsigma+h_Iq\right]. \tag*{(129)} P I = ( ς 2 + x q ) [ a ⋆ τ ⋆ 2 ς 3 + ( a ⋆ + 1 ) τ ⋆ ς 2 + ( 1 + a ⋆ y ⋆ τ q 2 ) ς + h I q ] . (129)
Here the prescription has y ⋆ = x y_{\star}=x y ⋆ = x . The first factor is an undamped causal sound wave. If h I > 0 h_I>0 h I > 0 , the cubic has positive coefficients for q > 0 q>0 q > 0 and its remaining Hurwitz condition is
( a ⋆ + 1 ) τ ⋆ + a ⋆ τ ⋆ 2 [ ( a ⋆ + 1 ) y ⋆ τ ⋆ − h I ] q > 0 , (130) (a_{\star}+1)\tau_{\star}+a_{\star}\tau_{\star}^2[(a_{\star}+1)y_{\star}\tau_{\star}-h_I]q>0, \tag*{(130)} ( a ⋆ + 1 ) τ ⋆ + a ⋆ τ ⋆ 2 [( a ⋆ + 1 ) y ⋆ τ ⋆ − h I ] q > 0 , (130)
which holds whenever τ ⋆ > h I / [ ( a ⋆ + 1 ) y ⋆ ] \tau_{\star}>h_I/[(a_{\star}+1)y_{\star}] τ ⋆ > h I / [( a ⋆ + 1 ) y ⋆ ] . If h I = 0 h_I=0 h I = 0 , the cubic instead factors as ς [ a ⋆ τ ⋆ 2 ς 2 + ( a ⋆ + 1 ) τ ⋆ ς + 1 + a ⋆ y ⋆ τ q 2 ] \varsigma[a_{\star}\tau_{\star}^2\varsigma^2+(a_{\star}+1)\tau_{\star}\varsigma+1+a_{\star}y_{\star}\tau_q^2] ς [ a ⋆ τ ⋆ 2 ς 2 + ( a ⋆ + 1 ) τ ⋆ ς + 1 + a ⋆ y ⋆ τ q 2 ] . The quadratic is strictly stable and the advective root remains. The principal squared speeds are x x x , y ⋆ y_{\star} y ⋆ and zero, with coincident sound speeds allowed in this case. At η = 0 \eta=0 η = 0 , each shear factor is ς ( τ ⋆ ς + 1 ) \varsigma(\tau_{\star}\varsigma+1) ς ( τ ⋆ ς + 1 ) . These exact reduced factors establish spectral stability and causal characteristics.
Generalisations to arbitrary boosts
We use the Lorentz substitution (A.55), keeping the wavevector real in the moving frame to check stability and causality at arbitrary boost. For the neutral polynomial, direct evaluation at k = 0 k=0 k = 0 gives
P ^ N , b o o s t γ 2 ς 2 = 1 − x v 2 + γ [ 2 τ − ( 2 x τ + d ) v 2 ] ς + γ 2 [ τ 2 − ( 2 x τ 2 + d τ ) v 2 + ( x 2 τ 2 − x d τ ) v 4 ] ς 2 . (131) \frac{\widehat{P}_{N,\mathrm{boost}}}{\gamma^2\varsigma^2}=1-xv^2+\gamma[2\tau-(2x\tau+d)v^2]\varsigma+\gamma^2[\tau^2-(2x\tau^2+d\tau)v^2+(x^2\tau^2-xd\tau)v^4]\varsigma^2. \tag*{(131)} γ 2 ς 2 P N , boost = 1 − x v 2 + γ [ 2 τ − ( 2 xτ + d ) v 2 ] ς + γ 2 [ τ 2 − ( 2 x τ 2 + d τ ) v 2 + ( x 2 τ 2 − x d τ ) v 4 ] ς 2 . (131)
The quadratic coefficient is γ 2 τ 2 ( 1 − y + v 2 ) ( 1 − y − v 2 ) > 0 \gamma^2\tau^2(1-y_+v^2)(1-y_-v^2)>0 γ 2 τ 2 ( 1 − y + v 2 ) ( 1 − y − v 2 ) > 0 . The linear bracket is bounded below by
2 τ ( 1 − x ) − d = d [ 2 ( 1 − x ) x + 1 + 3 x 1 − x ] > 0 ( d > 0 ) . (132) 2\tau(1-x)-d=d\left[\frac{2(1-x)}{x}+\frac{1+3x}{1-x}\right]>0\qquad(d>0). \tag*{(132)} 2 τ ( 1 − x ) − d = d [ x 2 ( 1 − x ) + 1 − x 1 + 3 x ] > 0 ( d > 0 ) . (132)
Thus both nonzero gaps are damped for every ∣ v ∣ < 1 |v| < 1 ∣ v ∣ < 1 . For the charged quintic the corresponding direct calculation is
P I , b o o s t γ 3 s 3 = 1 − x v 2 + γ [ ( a ⋆ + 1 ) τ ⋆ − D v 2 + x D q v 4 ] ς + γ 2 [ a ⋆ τ ⋆ 2 − H v 2 + C v 4 ] ς 2 . (133) \frac{P_{I,\mathrm{boost}}}{\gamma^{3}s^{3}} = 1 - xv^{2} + \gamma\left[(a_{\star}+1)\tau_{\star} - \mathcal{D}v^{2} + xD_{q}v^{4}\right]\varsigma+ \gamma^{2}\left[a_{\star}\tau_{\star}^{2} - Hv^{2} + Cv^{4}\right]\varsigma^{2}. \tag*{(133)} γ 3 s 3 P I , boost = 1 − x v 2 + γ [ ( a ⋆ + 1 ) τ ⋆ − D v 2 + x D q v 4 ] ς + γ 2 [ a ⋆ τ ⋆ 2 − H v 2 + C v 4 ] ς 2 . (133)
The constant and quadratic coefficients are positive, the latter because its bracket is a ⋆ τ ⋆ 2 ( 1 − y + v 2 ) ( 1 − y − v 2 ) a_{\star}\tau_{\star}^{2}(1-y_{+}v^{2})(1-y_{-}v^{2}) a ⋆ τ ⋆ 2 ( 1 − y + v 2 ) ( 1 − y − v 2 ) . The linear bracket is at least ( a ⋆ + 1 ) τ ⋆ ( 1 − x ) − ( d + h I ) > 0 (a_{\star}+1)\tau_{\star}(1-x)-(d+h_{I})>0 ( a ⋆ + 1 ) τ ⋆ ( 1 − x ) − ( d + h I ) > 0 after increasing τ ⋆ \tau_{\star} τ ⋆ if necessary. The shear factor at k = 0 k=0 k = 0 similarly has a negative nonzero gap, since τ Q − η v 2 / w > 0 \tau_{Q}-\eta v^{2}/w>0 τ Q − η v 2 / w > 0 .
For a fixed nonzero real wavevector, vary the boost continuously from zero to any prescribed ∣ v ∣ < 1 |v|<1 ∣ v ∣ < 1 . The leading coefficients of the boosted neutral quartic and charged quintic are, respectively,
γ 4 τ 2 ( 1 − y + v 2 ) ( 1 − y − v 2 ) , γ 5 a ⋆ τ ⋆ 2 ( 1 − y + v 2 ) ( 1 − y − v 2 ) . (134) \gamma^{4}\tau^{2}(1-y_{+}v^{2})(1-y_{-}v^{2}), \qquad\gamma^{5}a_{\star}\tau_{\star}^{2}(1-y_{+}v^{2})(1-y_{-}v^{2}). \tag*{(134)} γ 4 τ 2 ( 1 − y + v 2 ) ( 1 − y − v 2 ) , γ 5 a ⋆ τ ⋆ 2 ( 1 − y + v 2 ) ( 1 − y − v 2 ) . (134)
Here y ± y_{\pm} y ± denote the squared speeds of the corresponding neutral or charged polynomial. These coefficients are nonzero on the compact boost path, so no root enters from infinity. A crossing of the imaginary ς \varsigma ς axis would give a real frequency–momentum covector, whose Lorentz transform is also real. Strict rest-frame stability excludes such a root at nonzero rest-frame momentum. At zero rest-frame momentum the only imaginary root is zero, hence the entire covector is zero, which cannot correspond to a nonzero wavevector in the moving frame. This excludes every crossing, as in Section A.3.1.
Marginal factors are separated before this argument is applied. At D = 0 \mathcal{D}=0 D = 0 , remove the exact advective factor of P I P_{I} P I . The remaining quartic has nonzero boosted leading coefficient γ 4 a ⋆ τ ⋆ 2 ( 1 − y + v 2 ) ( 1 − y − v 2 ) \gamma^{4}a_{\star}\tau_{\star}^{2}(1-y_{+}v^{2})(1-y_{-}v^{2}) γ 4 a ⋆ τ ⋆ 2 ( 1 − y + v 2 ) ( 1 − y − v 2 ) . At d = g I = 0 d=g_{I}=0 d = g I = 0 , remove the causal wave factor in (B.17). The strictly stable cubic has boosted leading coefficient γ 3 a ⋆ τ ⋆ 2 ( 1 − y ⋆ v 2 ) > 0 \gamma^{3}a_{\star}\tau_{\star}^{2}(1-y_{\star}v^{2})>0 γ 3 a ⋆ τ ⋆ 2 ( 1 − y ⋆ v 2 ) > 0 . If h I = 0 h_{I}=0 h I = 0 , also remove its advective factor, leaving a strictly stable quadratic with leading coefficient γ 2 a ⋆ τ ⋆ 2 ( 1 − y ⋆ v 2 ) > 0 \gamma^{2}a_{\star}\tau_{\star}^{2}(1-y_{\star}v^{2})>0 γ 2 a ⋆ τ ⋆ 2 ( 1 − y ⋆ v 2 ) > 0 . The advective roots and the causal wave roots remain on the imaginary axis under boosts. The shear factors obey the same argument when η > 0 \eta>0 η > 0 , with leading coefficient γ 2 ( τ Q − η v 2 / w ) > 0 \gamma^{2}(\tau_{Q}-\eta v^{2}/w)>0 γ 2 ( τ Q − η v 2 / w ) > 0 . When η = 0 \eta=0 η = 0 , their advective and decaying factors are explicit. Together with the zero-momentum gaps, this proves stability for all real wavevectors and finite boosts. All characteristic cones used here are causal, so their boosted characteristic velocities are also real and bounded by one.
Zero density case
The charged construction above requires n ≠ 0 n\ne0 n = 0 . The associated restriction to baryon-rich matter was already noted in Ref. [9 ] . The following identity makes its conductivity obstruction at zero density explicit, without choosing relaxation parameters. Since r i = 0 r_{i}=0 r i = 0 for the ideal current, the physical conductivity identity (11) reads
D T = r ( θ 3 − r T θ 1 ) . (135) \mathcal{D}T = r(\theta_{3}-rT\theta_{1}). \tag*{(135)} D T = r ( θ 3 − r T θ 1 ) . (135)
At r = 0 r=0 r = 0 , finite θ 1 , θ 3 \theta_{1},\theta_{3} θ 1 , θ 3 therefore force D = 0 \mathcal{D}=0 D = 0 . The 1 / n 1/n 1/ n term in (B.1) expresses the same obstruction. A divergent τ Q \tau_{Q} τ Q cannot remove it while keeping θ 1 = τ Q ω \theta_{1}=\tau_{Q}\omega θ 1 = τ Q ω finite. Thus this class of frames does not cover the entire charged thermodynamic domain consider here, which includes zero density with prescribed positive conductivity.
For every fixed regular state with n ≠ 0 n \ne0 n = 0 , the preceding construction accommodates arbitrary finite prescribed η \eta η , ζ \zeta ζ , D ≥ 0 D \ge0 D ≥ 0 , without imposing A > 0 A > 0 A > 0 or a relation among the physical transports. On compact regular state patches with ∣ n ∣ |n| ∣ n ∣ and d d d bounded away from zero, the positive Hurwitz leading coefficients admit uniform lower bounds. The positive cubic coefficient of (B.15) controls the discriminant uniformly even where its quartic coefficient vanishes. All other coefficients are bounded, so one sufficiently large constant τ ⋆ \tau_\star τ ⋆ works throughout such a patch. Smooth constitutive data then give smooth frame coefficients. Positive shear viscosity suffices for the stated condition on d d d , while bulk viscosity may vanish.
At a regular charged conformal limit with n ≠ 0 n \ne0 n = 0 and d > 0 d > 0 d > 0 , B → 0 B \to0 B → 0 , A , x → 1 / 3 A, x \to1/3 A , x → 1/3 give a ⋆ → 2 a_\star\to2 a ⋆ → 2 , y ⋆ → 1 / 3 y_\star\to1/3 y ⋆ → 1/3 and τ P → τ ϵ / 3 \tau_P \to\tau_\epsilon/3 τ P → τ ϵ /3 . The relaxation parameters can remain finite on such a compact patch. Our constructions discussed in § 1 supply additional explicit frame choices, including non-thermodynamic members and regular charged prescriptions at n = 0 n = 0 n = 0 .
References