Maximal Dissipation and Well-Posedness of the Euler System of Gas Dynamics Authors: Eduard Feireisl, Ansgar Jüngel, Mária Lukáčová-MedviĎová Published: 2026 ABSTRACT AbstractWe show that any dissipative (measure–valued) solution of the compressible Euler system that complies with Dafermos’ criterion of maximal dissipation is necessarily an admissible weak solution. In addition, we propose a simple, at most two-step, selection procedure to identify a unique semigroup solution in the class of dissipative solutions to the Euler system. Finally, we introduce a refined version of Dafermos’ criterion yielding a unique solution of the problem for any finite energy initial data. Introduction Mathematical models of perfect fluids in continuum mechanics exhibit the well known difficulties including formation of singularities and ill posedness with respect to the data. An iconic example is the barotropic Euler system of gas dynamics describing the time evolution of the density ϱ=ϱ(t,x) and the velocity u=u(t,x) of a compressible perfect (inviscid) fluid: Here p is the pressure depending solely on the fluid density, whereas the thermal effects are ignored. The fluid occupies a domain Ω⊂Rd, d=1,2,3, and the evolution is considered in the time interval t∈[0,∞). Relevant initial/boundary data must be prescribed to obtain, at least formally, a mathematically well-posed problem. Here, we consider the initial conditions together with the impermeability boundary condition More general boundary conditions are discussed in the monograph [4, Chapter 3,4]. We recall the well known facts concerning solvability of the Euler system: Problem (1.1)–(1.4) admits a local in time classical solution provided the initial data enjoy Sobolev regularity Wk,2, k>1+d2, the initial density ϱ0 is bounded below away from zero, and the initial velocity (momentum) m0 satisfies the relevant compatibility condition, see, e.g., [4] or [34] for a proper interpretation of the compatibility conditions. Problem (1.1)–(1.4) admits a local in time classical solution provided the initial data enjoy Sobolev regularity Wk,2, k>1+d2, the initial density ϱ0 is bounded below away from zero, and the initial velocity (momentum) m0 satisfies the relevant compatibility condition, see, e.g., [4] or [34] for a proper interpretation of the compatibility conditions. Smooth solutions develop shock type singularities in a finite time for a vast class of initial data, see e.g. Dafermos [15], Smoller [35]. Smooth solutions develop shock type singularities in a finite time for a vast class of initial data, see e.g. Dafermos [15], Smoller [35]. Singularities in the form of implosions may develop in the physically relevant 3D geometry, see Buckmaster et al. [9], Cao-Labora et al. [10], Merle et al. [32, 33]. Singularities in the form of implosions may develop in the physically relevant 3D geometry, see Buckmaster et al. [9], Cao-Labora et al. [10], Merle et al. [32, 33]. Weak Solutions To incorporate possible physically relevant singularities, the concept of weak distributional solutions was introduced, along with an admissibility condition where represents the energy of the system. The weak formulation of (1.1), (1.2), along with the admissibility condition (energy inequality) (1.5), is capable to identify a unique solution in many physically relevant situations, in particular the so–called Riemann problem in the 1-D geometry, see e.g. Bianchini, Bressan [5] or the monograph by Dafermos [15]. The same Riemann problem considered in the multi-D setting, however, gives rise to a large variety of different but still admissible weak solutions “constructed” by the method of convex integration, see e.g. Chiodaroli, De Lellis, and Kreml [12] among many others. Apart from the Riemann problem the ill-posedness of the barotropic Euler system was observed in the seminal paper De Lellis and Székehylidi [16]. More recently, ill posedness was shown even in the class of Hölder continuous admissible solutions, see Giri, Kwon [25]. Despite the abundance of weak solutions for certain “wild” data, a simple question of existence of an admissible weak solution for any finite energy initial data remains open, even in the incompressible case, cf. Wiedemann, Székelyhidi [36], and Wiedemann [37]. To fill this gap, in particular when identifying the limits of numerical approximations, a larger class of dissipative (measure–valued) solutions is needed, see e.g. [23]. Dissipative Solutions The origin of dissipative solutions goes back to DiPerna [18], see also DiPerna, Majda [17], and their concept of measure–valued solution to capture the qualitative properties of oscillatory approximations. Unfortunately, the class of Young measures and their various extensions to include possible concentrations (see e.g. Alibert, Bouchitté [2]) seems to be too large to give rise to a unique solution. In particular, the unphysical “numerical” oscillations may create different measure valued solutions depending on the approximation method. Instead, only certain physically observable quantities should be invariant and recovered independently of a specific approximation process. In [8], only the expected value (with respect to the associated Young measure) of the density ϱ, the momentum m=ϱu, and the scalar quantity E representing the total energy were proposed as a new concept of dissipative solution of the Euler system. Roughly speaking, dissipative solutions solve in the distributional sense an extended Euler system with a tensor R≥0 called Reynolds stress. In addition, there is a new variable E called total energy – a scalar quantity satisfying Moreover, the energy defect controls the amplitude of the Reynolds stress, more specifically, where the constant c(p)>0 is determined in terms of the structural properties of the equation of state p=p(ϱ). The integral on the right–hand side of (1.9) is computed in terms of the measure trace[R]. We call the quantity ∫ΩE(ϱ,m)dx mean energy. Admissible weak solutions (with a non–increasing total energy) are therefore characterized by the equality We refer to Sect. 2.1 for a precise definition of dissipative solutions. Remark 1.1 The terminology we use is reminiscent of the turbulence modelling. Note, however, that the meaning of the quantities like Reynolds stress or energy defect are rather different in the context of the (inviscid) Euler system and the models of viscous fluids used to describe turbulence. Comparison with Other Concepts of Generalized Solutions Recently, Eiter and Lasarzik [19] proposed a general approach applicable to systems of conservation laws. In the particular case of the barotropic Euler system, their concept of weak solution coincides with the dissipative solutions introduced in the previous section, see [19, Theorem 5.8]. Inspired by the definition of variational solutions to the incompressible Euler system by Lions [30], Lasarzik [29] introduced a similar concept of solutions based on the mere satisfaction of the relative energy inequality. A more sophisticated approach of similar “variational” character was proposed also by Brenier [6]. The resulting problem is quite elegant and its solution boils down to resolving a kind of convex variational problem. As the relative energy inequality is available also for the measure valued/dissipative solutions of the Euler system, cf. e.g. Gwiazda, Świerczewska-Gwiazda, and Wiedemann [26] or [22], a similar approach can also be developed also for the present problem. The class of possible solutions, however, is quite large and the desired solution of the Euler system is correctly identified only if it is smooth globally in time. As pointed out by Brenier [6], the resulting solutions may fail in attaining the prescribed initial data, and, in general, they may not be even (weakly) continuous with respect to time. In particular, the solution may not coincide with the smooth one on its life–span, see the example of Burgers equation discussed in [6]. Basic Properties of Dissipative Solutions The dissipative solutions enjoy the following properties: Global existence. Given the initial data ϱ0,m0,andE0≥∫ΩE(ϱ0,m0)dx, there exists a global–in–time dissipative solution of the Euler system. The functions ϱ(t,·), m(t,·) are weakly continuous in time and satisfy the initial conditions in the weak sense. Moreover, the total energy E can be identified with a non–increasing càglàd function of the time, with E(0)≡E(0-)=E0, see [8]. Global existence. Given the initial data ϱ0,m0,andE0≥∫ΩE(ϱ0,m0)dx, there exists a global–in–time dissipative solution of the Euler system. The functions ϱ(t,·), m(t,·) are weakly continuous in time and satisfy the initial conditions in the weak sense. Moreover, the total energy E can be identified with a non–increasing càglàd function of the time, with E(0)≡E(0-)=E0, see [8]. Compatibility. If 1.10E0=∫ΩE(ϱ0,m0)dx and ϱ, m are continuously differentiable on the time interval [0, T), ϱ>0, then R=0 and ϱ, m represent a classical solution on [0, T), see e.g. [22]. Compatibility. If 1.10E0=∫ΩE(ϱ0,m0)dx and ϱ, m are continuously differentiable on the time interval [0, T), ϱ>0, then R=0 and ϱ, m represent a classical solution on [0, T), see e.g. [22]. Weak–strong uniqueness. If (1.10) holds, and the Euler system admits a continuously differentiable (classical) solution in [0, T), then R=0 and any dissipative solution coincides with the classical solution in [0, T). Weak–strong uniqueness. If (1.10) holds, and the Euler system admits a continuously differentiable (classical) solution in [0, T), then R=0 and any dissipative solution coincides with the classical solution in [0, T). Asymptotic regularity. We define a partial ordering of the class of dissipative solutions emanating from the same initial data, 1.11(ϱ1,m1,E1)≺(ϱ2,m2,E2)⇔E1(t)≤E2(t)forallt>0. We say that a dissipative solution is admissible if it is minimal with respect to ≺. Admissible dissipative solutions exist for any finite energy initial data, see [8]. As shown in [20], the energy defect of admissible dissipative solutions vanishes in the long run, more specifically, 1.12E(t)-∫ΩE(ϱ,m)(t,·)dx→0⇒∫Ω¯dR(t,·)→0ast→∞. Asymptotic regularity. We define a partial ordering of the class of dissipative solutions emanating from the same initial data, 1.11(ϱ1,m1,E1)≺(ϱ2,m2,E2)⇔E1(t)≤E2(t)forallt>0. We say that a dissipative solution is admissible if it is minimal with respect to ≺. Admissible dissipative solutions exist for any finite energy initial data, see [8]. As shown in [20], the energy defect of admissible dissipative solutions vanishes in the long run, more specifically, 1.12E(t)-∫ΩE(ϱ,m)(t,·)dx→0⇒∫Ω¯dR(t,·)→0ast→∞. Fluids arising in real world applications are always viscous—they dissipate mechanical energy in some sense. The Euler system should be therefore understood as a zero viscosity limit. The same is true for numerical approximations containing “artificial” numerical viscosity as a stabilizing mechanism. On the one hand, a vanishing viscosity limit is a weak solution of the compressible Euler system only if the approximate solutions converge strongly (pointwise a.a.), see [21] or [22, Chapter 7, Theorem 7.5]. In particular, the weak convergence of approximate solutions observed for certain numerical schemes gives rise to truly dissipative solutions, meaning R≠0, see Fjordholm et al. [24], Lukáčová et al. [31]. On the other hand, as we show in the present paper, any dissipative solution satisfying Dafermos’ admissibility criterion of maximal dissipation is necessarily a weak solution. The incompatibility of weak convergence and the principle of maximal dissipation may be seen as a paradigm shift in our understanding of what is a correct limit of numerical approximations of models involving perfect fluids, cf. also Chiodaroli and Kreml [13]. Maximal Dissipation Principle, DiPerna’s Conjecture, Dafermos’ Admissibility Criterion The first group of results obtained in the present paper concerns regularity of dissipative solutions. It may seem that the dissipative solutions may deviate largely from the weak solutions due to a possibly large amplitude of the Reynolds stress R. However, as we show in Sect. 3.1, there exist dissipative solutions with arbitrarily small amplitude of R. Specifically, for any initial data and any δ>0, there exists a dissipative solution satisfying The proof is based on suitable partial ordering of the set of dissipative solutions and an application of Zorn’s lemma. Next, we recall DiPerna’s conjecture [18]: Any measure–valued solution minimal with respect to ≺ is in fact a weak solution of the Euler system. The validity of the above statement for the Euler system is still open. Motivated by the seminal work of Dafermos [14], we refine the relation ≺ introduced in (1.11) to for some 0≤T<∞, δ>0. In other words, the solutions (ϱ1,m1,E1), (ϱ2,m2,E2) coincide up to the time T, while (ϱ1,m1,E1) dissipates more energy than (ϱ2,m2,E2) on the interval (T,T+δ). Note carefully that the relation ≺loc is a strict partial order, meaning it is irreflexive, asymmetric, and transitive. We say that a dissipative solution is maximal dissipative if it is minimal with respect to the ordering ≺loc. Accordingly, maximal dissipative solutions comply with Dafermos’ admissibility criterion based on maximal dissipation. Our main result stated in Theorem 3.3 below asserts that any maximal dissipative solution is necessarily a weak solution of the Euler system with a non–increasing total energy. This can be seen as a rigorous verification of DiPerna’s conjecture (in the framework of Dafermos’ admissibility criterion) in the specific case of the barotropic Euler system. Selection Criteria The second main topic discussed in the present paper is the choice of a proper selection criterion applicable to the class of dissipative solutions. The first step in this direction was undertaken in [8] (cf. also [7]), where the approach of Cardona and Kapitanski [11] motivated by the original work by Krylov [28] was adapted to the Euler system. The selection procedure consists in a successive minimization of a family of cost functionals of the type where the family of bounded continuous functionals Fm separates points in a suitable phase space. As a result, a unique semigroup selection is identified as the asymptotic limit of the process. The selected solutions depend in the Borel measurable way on the initial data. The method is a bit awkward and suffers the obvious difficulties: Although the first functional may be clearly specified as 1.13F1,1=∫0∞exp(-t)β(E(t))dt, where β is a bounded increasing function, the subsequent choice of Fm as well as the exponents λn is entirely arbitrary. Obviously, different choices may give rise to different limits and there is no clear indication of possible preferences. Although the first functional may be clearly specified as 1.13F1,1=∫0∞exp(-t)β(E(t))dt, where β is a bounded increasing function, the subsequent choice of Fm as well as the exponents λn is entirely arbitrary. Obviously, different choices may give rise to different limits and there is no clear indication of possible preferences. The infinite selection process is unlikely to be efficiently numerically implemented/tested. The infinite selection process is unlikely to be efficiently numerically implemented/tested. We propose a different much simpler selection process consisting of only two steps. Similarly to [8], the first step is minimizing In comparison with (1.13), the function β is missing, in particular, the functional is not bounded. In order to select a measurable minimizer, a slight change of the solution space with respect to [8] is needed. Note that F1 can be obviously written in the form - the sum of the (weighted) time integral of the energy defect and the mean energy. There is a number of arguments discussed already in [8, Section 5.1] why F1 should be the first to minimize. In particular, the selected solutions are necessarily admissible (minimal) with respect to the relation ≺ introduced in (1.11). We conclude the selection process by the second step, namely, minimizing where is a suitable strictly convex function. We may call F “entropy” associated to the problem. In contrast to [8], the functional is neither bounded nor continuous with respect to the weak topologies pertinent to the solution space. Accordingly, the main new issue to be properly addressed is the Borel measurability of the selected minimizer with respect to the data. While measurability of the solution sets in [8] was reduced to measurability of the solution mapping in the Hausdorff topology on compact subsets of the phase space, we have to use the Wijsman topology on convex closed sets of suitable Lp-spaces. Fortunately, under certain conditions, convergence in the latter is equivalent to the so–called Mosco convergence of closed convex sets suitable to establish continuity of the minimizers. Once measurability is established, step 2 yields a minimizer of F2. As the cost functional is strictly convex, this concludes the selection process. Here, convexity of the set of all dissipative solutions emanating from the same initial data plays a crucial role. In sharp contrast with [8], ambiguity of the result is only due to the choice of the entropy F—a topic subject to future discussion. As shown in Theorem 4.2, the above selection process gives rise to a unique solution in the class of admissible dissipative solutions that enjoys the semigroup property. We call this solution a semigroup solution. Maximal Dissipation Principle Revisited Motivated by the selection procedure we introduce a concept of absolute energy minimizer emanating from the initial data (ϱ0,m0,E0), and satisfying for any other solution (ϱ,m,E) starting from the same initial data, where λ_=λ_(ϱ,m,E) depends on the solution (ϱ,m,E). We show that for each initial data there exists at most one absolute energy minimizer. Moreover, any absolute energy minimizer is minimal with respect to the order ≺loc; whence a weak solution of the Euler system with a non–increasing energy profile. Finally, we observe that any global minimizer with respect to the order ≺loc must be an absolute energy minimizer. This simple but quite important fact provides a local admissibility criterion applicable on any compact time interval [0, T] independent of the behaviour of solutions in the long run. The rest of this paper is organized as follow: in Sect. 2, we introduce the class of dissipative solutions to the barotropic Euler system and review their basic properties. In Sect. 3, we state and prove our main results concerning the existence of dissipative solutions with arbitrarily small Reynolds stress (Theorem 3.1 and Corollary 3.2). Moreover, we show that any maximal dissipative solution is necessarily a weak solution (Theorem 3.3). The selection process to identify a unique semigroup solution is discussed in Sect. 4. In Sect. 5, we introduce the class of absolute energy minimizers and discuss their basic properties. Dissipative Solutions to the Barotropic Euler System In this section, we introduce the class of dissipative solutions to the Euler system. For definiteness, we restrict ourselves to the isentropic state equation Accordingly, the corresponding pressure potential is given by The energy function is defined as Note that E:Rd+1→[0,∞] is a convex l.s.c. function, strictly convex on its domain. Finally, we suppose that Ω⊂Rd is a bounded Lipschitz domain. Definition of Dissipative Solutions The initial data (ϱ0,m0,E0) belong to the class As explained in [8], the total energy E, with its initial value E0, is formally added as a “new” state variable in the definition of dissipative solutions. Remark 2.2 The specific form of the constant r in (2.9) is pertinent to the isentropic equation of state. In the general case, we may consider that see [1]. The constant r in (2.9) is then determined solely in terms of a_, a¯, and the dimension d. As already pointed out the density—momentum component (ϱ,m) of a dissipative solution is nothing other than the expected value of a Young measure for “conventional” measure–valued solutions. The total energy E is the standard energy ∫ΩE(ϱ,m)dx augmented by the energy defect, where the latter corresponds to the sum of oscillation and concentration defects in the measure–valued formulation. Note that (2.8), (2.9) imply Rate of Energy Dissipation, Maximal Dissipative Solutions Let the data (ϱ0,m0,E0) be given. We denote as the set of all dissipative solutions on [0,∞) emanating from the data (ϱ0,m0,E0). To compare the rate of energy dissipation, we introduce two order relations already discussed in Sect. 1. The first one introduced by DiPerna [18] reads: For admissible dissipative solutions, the total energy E is uniquely determined by (ϱ,m). Specifically, the following holds: Proposition 2.4 Let (ϱi,mi,Ei) be two admissible solutions emanating from the data (ϱ0i,m0i,E0i), i=1,2. Suppose Then Proof Denote that It is easy to check that the solutions where are again dissipative solutions with the modified Reynolds stresses However, as both solutions are minimal with respect to ≺, we conclude that □ DiPerna’s conjecture stated in [18, Section 6, Part (b)] asserts that “any admissible dissipative solution is a weak solution”. To the best of our knowledge this conjecture is still not proven. Moreover, its formulation is not “deterministic” as it depends on the behaviour of solutions on the whole interval (0,∞). Motivated by Dafermos [14], we introduce a local version of (2.13), One of the main results of this paper—Theorem 3.3—is a rigorous proof of DiPerna’s conjecture in Dafermos’ framework. more Specifically, we show that any maximal dissipative solution is necessarily a weak solution of the Euler system. Properties of Dissipative Solutions We start by introducing the necessary function spaces framework. In accordance with (2.9), the instantaneous values of dissipative solutions (ϱ,m,E)(τ+,·) evaluated at a time τ≥0 belong to the convex set Next, we introduce the trajectory space T, in which solutions live. We consider two different topologies on the trajectory space, and 10 there exists T=T(δ)>0 such that 3.20≤DE(t)≤δforanyτ≤t<τ+T(δ). Indeed, as the energy is weakly l.s.c. and the total energy E non–increasing, we have 0≤lim supt→τ+DE(t)≤lim supt→τ+E(t+)-lim inft→τ∫ΩE(ϱ,m)(t,·)dx≤E(τ+)-∫ΩE(ϱ,m)(τ,·)dx=0. Initial time regularity. Suppose that 3.1E(τ+)=∫ΩE(ϱ,m)(τ)dx⇔DE(τ+)=0forsomeτ≥0. Then for any δ>0 there exists T=T(δ)>0 such that 3.20≤DE(t)≤δforanyτ≤t<τ+T(δ). Indeed, as the energy is weakly l.s.c. and the total energy E non–increasing, we have 0≤lim supt→τ+DE(t)≤lim supt→τ+E(t+)-lim inft→τ∫ΩE(ϱ,m)(t,·)dx≤E(τ+)-∫ΩE(ϱ,m)(τ,·)dx=0. Concatenation. Suppose (ϱ1,m1,E1)∈U[ϱ0,m0,E0] and (ϱ2,m2,E2)∈U[ϱ(T,·),m(T,·),ET2], ∫ΩE(ϱ,m)(T,·)dx≤ET2≤E1(T). Then (ϱ,m,E)(t,·)=(ϱ1,m1,E1)(t,·)for0≤t≤T,(ϱ2,m2,E2)(t-T,·)fort>T, belongs to the the solution set U[ϱ0,m0,E0]. Concatenation. Suppose (ϱ1,m1,E1)∈U[ϱ0,m0,E0] and (ϱ2,m2,E2)∈U[ϱ(T,·),m(T,·),ET2], ∫ΩE(ϱ,m)(T,·)dx≤ET2≤E1(T). Then (ϱ,m,E)(t,·)=(ϱ1,m1,E1)(t,·)for0≤t≤T,(ϱ2,m2,E2)(t-T,·)fort>T, belongs to the the solution set U[ϱ0,m0,E0]. Our main result stated below is based on an application of Zorn’s lemma equivalent to the axiom of choice. Proof For each global solution we define a “stopping” time We know from (3.2) that T(δ)>0. Moreover, Next, we introduce the order relation It is a routine matter to check that U[ϱ0,m0,E0] endowed with the relation ≺≺ is a partially ordered set eligible for the application of Zorn’s lemma. Strictly speaking, Zorn’s lemma is applied to the equivalence classes Let be an ordered chain. Without loss of generality, we may assume (ϱi,mi,Ei)≠(ϱj,mj,Ej) for i≠j. As the solution set is compact in Tweak, the sequence (ϱn,mn,En)n=1∞ contains a converging subsequence (not relabeled) In accordance with (3.5), Now, either T(δ)(ϱ~,m~,E~)=∞, meaning (ϱ~,m~,E~) is an upper bound for the chain, or T(δ)[ϱ~,m~,E~]=T~<∞. In the latter case, we consider a concatenation where Applying (3.1), (3.2) to (ϱ^,m^,E^) we can see (ϱ,m,E) is an upper bound for the chain. Thus in both cases, the ordered chain admits an upper bound in U[ϱ0,m0,E0]. By Zorn’s lemma, there exists a maximal solution (ϱ~,m~,E~)∈U[ϱ0,m0,E0] with respect to the relation ≺≺. We claim that which completes the proof. Indeed, if we could repeat the above argument concatenating the solution (ϱ~,m~,E~) with another solution starting from the initial data Obviously, the new solution (ϱ,m,E) satisfies and, by virtue of the initial regularity property, Consequently, (ϱ~,m~,E~) is not maximal which is a contradiction. □ Applying the same approach we can show the existence of a dissipative solution with the energy defect dominated by an arbitrary positive function, in particular, the defect may vanish for t→∞. Proof Apply the arguments of the proof of Theorem 3.1 to obtain a solution (ϱ,m,E), with the energy defect □ Maximal Dissipation Principle We recall the order relation introduced in (2.14), namely Similarly to the relation ≺≺ introduced in the proof of Theorem 3.1, the order relation ≺loc augmented formally by the identity relation represents a partial ordering of U[ϱ0,m0,E0]. Recall that a dissipative solution is maximal (dissipative) if it is minimal with respect to ≺loc. Regularity of Maximal Dissipative Solutions Our next goal is to show that any maximal dissipative solution in the sense of Definition 2.5 is necessarily an admissible (with non–increasing total energy) weak solution. The result can be seen as a rigorous verification of DiPerna’s conjecture [18, Section 6] in the context of the barotropic Euler system. Proof Suppose for some T≥0. As E is non–increasing, there exists δ>0 such that Next, by concatenation, we may construct a new solution of the same problem such that where Obviously, while, by virtue of (3.6), for any t∈(T,T+δ). We conclude that (ϱ,m,E) is not minimal. □ As we shall see in Sect. 4, the existence of admissible dissipative solutions, meaning solutions minimal with respect to the order ≺ is guaranteed for any finite energy data. Unfortunately, however, the existence of maximal dissipative solutions, meaning minimal with respect to the order ≺loc remains an open problem. Semigroup (Semiflow) Selection Before discussing any selection process, we define two operations on the trajectory space Tweak: Time shift. For any T≥0 and any (ϱ,m,E)∈Tweak, we define ST[ϱ,m,E](t,·)=(ϱ(t+T),m(t+T),E(t+T)),t≥0. Time shift. For any T≥0 and any (ϱ,m,E)∈Tweak, we define ST[ϱ,m,E](t,·)=(ϱ(t+T),m(t+T),E(t+T)),t≥0. Concatenation. For (ϱi,mi,Ei)∈Tweak,i=1,2,T>0, we define that (ϱ1,m1,E1)∪T(ϱ2,m2,E2)=(ϱ,m,E),(ϱ,m,E)=(ϱ1,m1,E1)(t,·)for0≤t≤T,(ϱ2,m2,E2)(t-T,·)fort>T Note that (ϱ1,m1,E1)∪T(ϱ2,m2,E2) belongs to Tweak, provided that (ϱ2(0,·),m2(0,·))=(ϱ1(T,·),m1(T,·)). Concatenation. For (ϱi,mi,Ei)∈Tweak,i=1,2,T>0, we define that (ϱ1,m1,E1)∪T(ϱ2,m2,E2)=(ϱ,m,E),(ϱ,m,E)=(ϱ1,m1,E1)(t,·)for0≤t≤T,(ϱ2,m2,E2)(t-T,·)fort>T Note that (ϱ1,m1,E1)∪T(ϱ2,m2,E2) belongs to Tweak, provided that (ϱ2(0,·),m2(0,·))=(ϱ1(T,·),m1(T,·)). Properties of the Solution Sets We recall the basic properties of the solution sets U established in Sect. 2.2. [A1] For any data (ϱ0,m0,E0)∈D, the solution set U[ϱ0,m0,E0]⊂Tweak∩Tstrong is a non–empty convex set, compact in Tweak and closed bounded in Tstrong. [A1] For any data (ϱ0,m0,E0)∈D, the solution set U[ϱ0,m0,E0]⊂Tweak∩Tstrong is a non–empty convex set, compact in Tweak and closed bounded in Tstrong. [A2] If (ϱ,m,E)∈U[ϱ0,m0,E0], than the time shift satisfies ST[ϱ,m,E]∈U[ϱ(T,·),m(T,·),E(T)] for any T≥0. [A2] If (ϱ,m,E)∈U[ϱ0,m0,E0], than the time shift satisfies ST[ϱ,m,E]∈U[ϱ(T,·),m(T,·),E(T)] for any T≥0. [A3] If (ϱ1,m1,E1)∈U[ϱ0,m0,E0] and (ϱ2,m2,E2)∈UF[ϱ1(T,·),m1(T,·),E02],where∫ΩE(ϱ1,m1)(T,·)dx≤E02≤E1(T), then (ϱ1,m1,E1)∪T(ϱ2,m2,E2)∈U[ϱ0,m0,E0]. [A3] If (ϱ1,m1,E1)∈U[ϱ0,m0,E0] and (ϱ2,m2,E2)∈UF[ϱ1(T,·),m1(T,·),E02],where∫ΩE(ϱ1,m1)(T,·)dx≤E02≤E1(T), then (ϱ1,m1,E1)∪T(ϱ2,m2,E2)∈U[ϱ0,m0,E0]. [A4] The set valued mappings (ϱ0,m0,E0)∈D↦U[ϱ0,m0,E0]∈comp[Tweak] and (ϱ0,m0,E0)∈D↦U[ϱ0,m0,E0]∈closed[Tstrong] are Borel measurable with respect to the Hausdorff and the Wijsman topology, respectively. [A4] The set valued mappings (ϱ0,m0,E0)∈D↦U[ϱ0,m0,E0]∈comp[Tweak] and (ϱ0,m0,E0)∈D↦U[ϱ0,m0,E0]∈closed[Tstrong] are Borel measurable with respect to the Hausdorff and the Wijsman topology, respectively. Selection Process We consider a functional of the form where is a convex l.s.c. functional. For each solution set let be the set of minimizers of F on U[ϱ0,m0,E0]. It is easy to see that the restricted solution set UF[ϱ0,m0,E0] is again non–empty and convex. Indeed, suppose infU[ϱ0,m0,E0]=I<∞, since otherwise the claim is trivial. As the solution set U[ϱ0,m0,E0] is compact in Tweak, there is a minimizing sequence (ϱn,mn,En)→(ϱ,m,E)∈U[ϱ0,m0,E0] in Tweak. By Fatou’s lemma, whence the limit (ϱ,m,E) is a minimizer. The convexity of the set of minimizers follows from the convexity of F. Next, we show that the properties [A2], [A3] are satisfied. As a matter of fact, the arguments are almost identical to those of [8, Section 5]. As the present setting is slightly more general, we reproduce the proof for reader’s convenience. Lemma 4.1 Shift property. (ϱ,m,E)∈UF[ϱ0,m0,E0]⇒ST(ϱ,m,E)∈UF[ϱ(T,·),m(T,·),E(T)] Shift property. (ϱ,m,E)∈UF[ϱ0,m0,E0]⇒ST(ϱ,m,E)∈UF[ϱ(T,·),m(T,·),E(T)] Concatenation property.(ϱ1,m1,E1)∈UF[ϱ0,m0,E0] and (ϱ2,m2,E2)∈UF[ϱ1(T,·),m1(T,·),E02],where∫ΩE(ϱ1,m1)(T,·)dx≤E02≤E1(T),⇒(ϱ1,m1,E1)∪T(ϱ2,m2,E2)∈UF[ϱ0,m0,E0]. Concatenation property.(ϱ1,m1,E1)∈UF[ϱ0,m0,E0] and (ϱ2,m2,E2)∈UF[ϱ1(T,·),m1(T,·),E02],where∫ΩE(ϱ1,m1)(T,·)dx≤E02≤E1(T),⇒(ϱ1,m1,E1)∪T(ϱ2,m2,E2)∈UF[ϱ0,m0,E0]. Proof The proof is basically the same as in Cardona, Kapintanski [11] or [8]. We present it here for the reader’s convenience. Shift property. First, by the continuation property, Now Concatenation property. Compute □ First Selection Criterion The first selection criterion consists in replacing the set by the set of minimizers of the functional considering the topology of the space Tweak. Note that F1(ϱ,m,E) is a bounded (continuous) linear form on both Tweak and Tstrong. Thus we set The resulting solution sets are again non–empty, convex and compact in Tweak. Moreover, as shown in [8] the mapping is continuous in comp[Tweak] endowed with the Hausdorff topology. It is worth noting that we minimize where is the energy defect, and the “real” energy of the fluid flow. Moreover, as observed in [8, Section 5], all selected solutions are minimal with respect to the relation ≺, meaning admissible. Second Selection Criterion The next step is to minimize on UF1[ϱ0,m0,E0] for a fixed 10. Seeing that the weak and strong topology generate the same family of Borel sets, we may also conclude the mapping is strongly measurable. Let us summarize the results obtained. Absolute Energy Minimizers Our final goal is to introduce a refined maximal dissipation criterion to recover a unique weak solution. Consider a solution (ϱ1,m1,E1)∈U[ϱ0,m0,E0] that is not maximal. Consequently, there exists (ϱ2,m2,E2)∈U[ϱ0,m0,E0] such that meaning Next, we compute provided λ>0 is large enough. This motivates the following definition. It follows from the previous discussion that any absolute energy minimizer is minimal with respect to the order ≺loc; whence an admissible weak solution of the Euler system. Next, we report the following variant of Lerch theorem, see [7, Lemma 5.3]. Lemma 5.2 (Lerch lemma) Suppose there exists λ_>0 such that Then It follows from Lemma 5.2 that there exists at most one absolute energy minimizer. Indeed it is easy to see that the set of all absolute minimizers is convex. Suppose that are two absolute minimizers (ϱ_1,m_1,E_) and (ϱ_2,m_2,E_), where The convex combination is another absolute energy minimizers; whence a weak solution satisfying As E is strictly convex, the relations (5.3), (5.4) are compatible only if Let us summarize the previous discussion. Finally, it follows from inequality (5.1) that any absolute minimizer with respect to order ≺loc is an absolute energy minimizer. Recall that (ϱ_,m_,E_)∈U[ϱ0,m0,E0] is an absolute minimizer with respect to order ≺loc if It is worth noting that the property of being absolute minimizer with respect to ≺loc is a local property that can be defined independently of the actual length of the time interval on which the problem is considered. Next, we show that the family of absolute energy minimizers, provided they exist for any initial data, enjoy the semigroup properties. Proposition 5.4 The following properties hold: Suppose (ϱ_,m_,E_)∈U[ϱ0,m0,E0] is an absolute energy minimizer. Then its time shift ST(ϱ_,m_,E_) is an absolute energy minimizer in U[(ϱ_,m_,E_)(T,·)]. Suppose (ϱ_,m_,E_)∈U[ϱ0,m0,E0] is an absolute energy minimizer. Then its time shift ST(ϱ_,m_,E_) is an absolute energy minimizer in U[(ϱ_,m_,E_)(T,·)]. Let (ϱ_1,m_1,E_1) be an absolute energy minimizer in U[ϱ0,m0,E0], and let (ϱ_2,m_2,E_2) be an absolute energy minimizer in U[(ϱ_1,m_1,E_1)(T,·)]. Then their concatenation (ϱ_1,m_1,E_1)∪T(ϱ_2,m_2,E_2) is an absolute energy minimizer in U[ϱ0,m0,E0], meaning (ϱ_1,m_1,E_1)=(ϱ_1,m_1,E_1)∪T(ϱ_2,m_2,E_2). Let (ϱ_1,m_1,E_1) be an absolute energy minimizer in U[ϱ0,m0,E0], and let (ϱ_2,m_2,E_2) be an absolute energy minimizer in U[(ϱ_1,m_1,E_1)(T,·)]. Then their concatenation (ϱ_1,m_1,E_1)∪T(ϱ_2,m_2,E_2) is an absolute energy minimizer in U[ϱ0,m0,E0], meaning (ϱ_1,m_1,E_1)=(ϱ_1,m_1,E_1)∪T(ϱ_2,m_2,E_2). Proof Shift property. Let (ϱ,m,E)∈U[(ϱ_,m_,E_)(T,·)] be an arbitrary solution. We have for a suitable sequence λ≥λ_ associated to (ϱ_,m_,E_)∪T(ϱ,m,E). Concatenation property. A straightforward manipulation yields for all λ≥λ_ related to STE_1. As a matter of fact, it follows from the previous step and uniqueness of absolute minimizers that We conclude that □ Conclusion We have identified five different classes of dissipative solutions emanating from given data (ϱ0,m0,E0) (see Fig. 1): Admissible dissipative solutions do exist for any finite energy initial data and admit a semigroup selection resulting from a two step process. Maximal dissipative solutions are admissible weak solution of the Euler system. They comply with Dafermos’ principle of maximal dissipation. Their existence for general initial data, however, remains an open problem. The absolute energy minimizer, provided it exists, is a unique weak solution in U[ϱ0,m0,E0]. Relations between the various solution concepts. The black dot denotes a solution set with a unique element. The gray color indicates solution sets with at least one element, while the existence of elements of the white sets is open. We prove that any maximal dissipative solution is in fact an admissible weak solution Funding Open access funding provided by TU Wien (TUW). Data availability Data sharing is not applicable to this article as no data sets were generated during the current study. Declarations Conflict of interest The authors have no relevant financial or non-financial interests to disclose. Footnotes Eduard Feireisl: The authors gratefully acknowledge Research in Teams Fellowship “Entropy methods for evolutionary systems: Analysis and Numerics” of the Erwin Schrödinger International Institute for Mathematics and Physics, Vienna. The work of E.F. was partially supported by the Czech Sciences Foundation (GAČR), Grant Agreement 24–11034 S. The Institute of Mathematics of the Academy of Sciences of the Czech Republic is supported by RVO:67985840. E.F. is a member of the Nečas Center for Mathematical Modelling. Ansgar Jüngel: The work of A.J. was funded in part by the Austrian Science Fund (FWF), grant 10.55776/P33010, 10.55776/F65, and 10.55776/PAT2687825. This work has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme, ERC Advanced Grant NEUROMORPH, no. 101018153. For open access purposes, the authors have applied a CC BY public copyright license to any author-accepted manuscript version arising from this submission. Mária Lukáčová-Medviďová: The work of M.L.-M. was supported by the Gutenberg Research College and by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – project number 233630050 – TRR 146 and project number 525800857 – SPP 2410 “Hyperbolic Balance Laws: Complexity, Scales and Randomness”. She is also grateful to the Mainz Institute of Multiscale Modelling for supporting her research. Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Change history 6/27/2026 The original online version of this article was revised Change history 7/2/2026 A Correction to this paper has been published: 10.1007/s00205-026-02222-3 References Associated Data Data Availability Statement Data sharing is not applicable to this article as no data sets were generated during the current study.