A variational derivation of the thermodynamics of a moist atmosphere with irreversible processes Franc¸ois Gay-Balmaz1 7 17 January 2017 1 0 2 Abstract n a Irreversible processes play a major role in the description and prediction of atmo- J spheric dynamics. In this paper, we present a variational derivation of the evolution 4 equations for a moist atmosphere subject to the irreversible processes of molecular 1 viscosity, heat conduction, diffusion, and phase transition. This derivation is based on the general variational formalism for nonequilibrium thermodynamics established ] inGay-Balmaz and Yoshimura[2017a,b],whichextendsHamilton’sprincipletoincor- h porates irreversible processes. It is valid for any state equation and thus also covers p thecase of the atmosphere of other planets. In this approach, the second law of ther- - h modynamics is understood as a nonlinear constraint formulated with the help of new t variables, called thermodynamic displacements, whose time derivative coincides with a m the thermodynamic force of the irreversible process. The formulation is written both in the Lagrangian and Eulerian descriptions and can be directly adapted to oceanic [ dynamics. 1 v 1 1 Introduction 2 9 3 The partial differential equations governing the thermodynamic of the atmosphere are of 0 obviousimportanceforweatherandclimateprediction. These equationsarewell-knownfor . their extreme complexity,bothfromthe theoreticalandthe computationalside, whichis in 1 0 part due to the many physical processes that they involve, such as phase transition, cloud 7 formation, precipitation, and radiation. 1 : Inabsenceofirreversibleprocesses,theequationsofatmosphericdynamiccanbederived v byapplyingHamilton’svariationalprincipletothe Lagrangianfunctionofthe fluid. Thisis i X in agreement with a fundamental fact from classical reversible mechanics, namely that the r motionofthemechanicalsystemisgovernedbytheEuler-Lagrangeequationswhich,inturn, a describethecriticalpointsoftheactionfunctionalofthisLagrangianamongallpossibletra- jectories with prescribed values at the temporal extremities. Hamilton’s principle for fluid mechanicsintheLagrangiandescriptionhasbeendiscussedatleastsincetheworksofHerivel [1955], for an incompressible fluid and Serrin [1959] and Eckart [1960] for compressible flows, see also Truesdell and Toupin [1960] for further references. While in the Lagrangian description this principle is a straightforward extension of the Hamilton principle of parti- cles mechanics, in the Eulerian description the variational principle is much more involved and several approaches have been developed, see Lin [1963], Seliger and Whitham [1968], Bretherton[1970]. WerefertoSalmon[1983]andSalmon[1988]forfurtherdevelopmentsin 1LMD/IPSL, CNRS, Ecole normale sup´erieure, PSL Research University, Ecole polytechnique, Uni- versit´e Paris-Saclay, Sorbonne Universit´es, UPMC Univ Paris 06, 24 Rue Lhomond 75005 Paris, France, [email protected] 1 1 Introduction 2 thecontextofgeophysicalfluids,seealsoMu¨ller[1995]. InHolm, Marsden and Ratiu[2002], the variational principle in Eulerian description is obtained via the Euler-Poincar´e reduc- tion theory for several geophysical fluid models, by exploiting the relabelling symmetries. Physically,the existence ofa variationalHamilton’s principle manifests itself in specific La- grangianconservationlaws,themostcelebratedbeingtheconservationofpotentialvorticity which alone allows one to understand many processes taking place in the atmosphere, the ocean, or in laboratory experiments (e.g. Hoskins, McIntyre, and Robertson [1985]). Con- servation of potential vorticity is related to Kelvin’s circulation theorem which, thanks to variational principle, can be interpreted in terms of the general Noether theorem, linking conservation laws to symmetries. More recently, the variational formulation of geophysi- cal fluids has been crucially used in Desbrun, Gawlik, Gay-Balmaz, and Zeitlin [2014] and Bauer and Gay-Balmaz [2016] to derive structure preserving numerical schemes for rotat- ing stratified fluids in the Boussinesq and pseudo-incompressible approximations,following Pavlov et al. [2010], Gawlik et al. [2011]. Irreversibleprocessessuchasphasetransition,cloudformation,precipitation,andradia- tion,playamajorroleinthedescriptionandpredictionofatmosphericdynamicsandithas been a longstanding questionwhether ornot the variationalformulationsmentionedabove canbe extendedto include allthese irreversibleprocesses. Inthis paper, weshallpositively answerthisquestionbypresentingavariationalderivationforthethermodynamicofamoist atmosphere that includes the irreversible processes of molecular viscosity, heat conduction, diffusion, and phase transition. We shall use the variational formulation of nonequilibrium thermodynamics recently developed in Gay-Balmaz and Yoshimura [2017a,b]. The main aspect of this approach is the introduction, for each of the irreversible process, of a new variable, called the thermodynamic displacement, whose time derivative coincides with the thermodynamicforceoftheprocess. Thisnewvariableallowsustoformulatethesecondlaw ofthermodynamicsasanonlinearconstrainttobeusedinthevariationalformulation,both fortheLagrangianandEuleriandescriptions. Thisconstraintinvolvesthephenomenological expressionof eachof the thermodynamic fluxes in terms of the thermodynamic forces. It is hence called the phenomenologicalconstraint. From the point of view of atmospheric mod- elling,thisconstraintencodesthevariousparametrizationsofsubgridscaleeffectsingeneral circulation models of the atmosphere. In absence of irreversible processes, the constraint disappears and our variational formulation consistently recovers the classical variational principles in Lagrangianand Eulerian descriptions. The atmosphereof the Earthis composed of dry air, watersubstance in any of its three phases, and atmospheric aerosols. For practical use in meteorology, dry air, whose main components are Nitrogen and Oxygen, can be regarded as an ideal gas. Unlike the compo- nents of dry air, the proportion of the gas phase of water,i.e., water vapor, is very variable and plays a major role in the thermodynamics of the atmosphere because of its ability to condense under atmospheric conditions. The condensed phases of water consist of cloud particles(waterdropletoricecrystal)thatfollowthemotionofdryairandofhydrometeors (such as rainand snow)that are falling throughthe air. The atmospheric aerosolsaresolid and liquid particles in suspension (other than that of water substance) whose study is very important for atmospheric chemistry, cloud and precipitation physics, and for atmospheric radiation and optics. It is not significant for atmospheric thermodynamics, and shall not be considered here. In the first part of the paper (Section 2), we restrict our approach to the case of dry air. In this case, the only irreversible processes are viscosity and heat conduction. This situation allows us to introduce the main ideas of our approach in a sim- plified context. In the second part of the paper (Section 3) we consider the case of a moist atmospherebyincludingwatersubstanceinitsdifferentphases,namely,watervapor,liquid water,andsolidwater. The irreversibleprocessesconsideredareviscosity,heatconduction, 2 Variational derivation of the thermodynamic of a dry atmosphere 3 vapordiffusion,andphasechanges. Wealsocomputetheimpactoftheirreversibleprocesses on the potential vorticity equation and on Kelvin’s circulation theorem, by staying in the general Lagrangian framework, which provides a useful unified treatment of potential vor- ticity and circulation theorems for various approximations of the equations of atmospheric dynamics. Finally, in Section 4, we quickly mention that our approach directly applies to oceanic dynamics with the irreversible process of molecular viscosity, heat conduction, and salt diffusion included. 2 Variational derivation of the thermodynamic of a dry atmosphere In this section we consider the dynamics of a dry atmosphere subject to the irreversible processes of molecular viscosity and of heat exchange by molecular conduction. The equation of state of dry air is the ideal gas law R T d v =v(p,T)= , (2.1) p where v is the specific volume, p is the pressure, T is the temperature, R =R∗/m is the d d gas constant for dry air written in terms of the universal gas constant R∗ and the mean molecular weight of dry air m . The expression of all other thermodynamic variables for d dry air in terms of p and T can be derived by using the equation of state (2.1) and the fact that the specific heat C at constant pressure can be assumed to be constant in the p atmosphere, which is the hypothesis for a perfect gas. For example, it is deduced that the specific internal energy and the specific entropy are u=C T, η =C lnT −R lnp+η , v p d 0 where C is the specific heat at constant volume, η is a constant, and we assume that the v 0 internal energy at T =0 K is zero. Our variational formulation is more naturally expressed in terms of density variables rather than specific variables, hence we will write the Lagrangian in terms of the mass density ρ = 1 and entropy density s = η. In addition, the variational derivation has a v v simpler form in Lagrangian description, this is why we shall first write it in Lagrangian variables and then later deduce from it the Eulerian form. Letusdenotebyx=ϕ(t,X)theLagrangiantrajectoryofdryairparticles. Thevariable X refers to the label of the particle and x is its current location. The Lagrangian and Eulerian wind velocities ϕ˙ = dϕ and v are related as dt d ϕ(t,X)=v(t,ϕ(t,X)). dt We shall denote by S(t,X) the entropy density in Lagrangian representation, related to s(t,x) as S(t,X) = s(t,ϕ(t,X))|∇ϕ(t,X)|, where |∇ϕ| denotes the Jacobian of ϕ. The mass density in Lagrangian representation is ̺ (X) = ρ(t,ϕ(t,X))|∇ϕ(t,X)|. Note that 0 because of mass conservation, the mass density ̺ in Lagrangian representation is time 0 independent. This is in contrastwith the time dependence of S, due to the presence of the irreversible processes. The Lagrangian of the dry atmosphere consists of the sum of the kinetic energy and of thecontributionofEarthrotation,towhichissubstractedthegravitationalpotentialΦand the internal energy u. In terms of the Lagrangianvariables ϕ, ϕ˙, and S, it reads 1 S |∇ϕ| L(ϕ,ϕ˙,S)= ̺ |ϕ˙|2+R(ϕ)·ϕ˙ −Φ(ϕ)−u , dX=: LdX. (2.2) 0 2 ̺ ̺ ZD (cid:20) (cid:18) 0 0 (cid:19)(cid:21) ZD 2.1 Variational formulation in Lagrangian description 4 Here the vector field R is the vector potential for the Coriolis parameter, i.e., curlR=2Ω, whereΩ is the angularvelocityofthe Earth. Onthe righthandside of(2.2)we definedthe Lagrangiandensity L as the integrand of the LagrangianL. 2.1 Variational formulation in Lagrangian description In absence of irreversible processes and in absence of heat or matter exchange with the exterior, the entropy is conserved. This means that in Lagrangian variables the entropy is time independent, S(t,X) = S (X). In this case, the equations of motion follow from the 0 Hamilton principle: t2 δ LdXdt=0, (2.3) Zt1 ZD wherethecriticalconditionistakenwithrespecttoarbitraryvariationsδϕoftheLagrangian trajectory ϕ, with δϕ(t )=δϕ(t )=0. 1 2 WeshallnowextendtheHamiltonprinciple(2.3)inordertoincorporatetheirreversible processes of viscosity and heat conduction. Following Gay-Balmaz and Yoshimura [2017b], we consider the variational condition t2 δ L+(S−Σ)Γ˙ dXdt=0, (2.4) Zt1 ZD (cid:0) (cid:1) subject to the phenomenological constraint ∂L Σ˙ =−Pfr :∇ϕ˙ +J ·∇Γ˙ (2.5) S ∂S and with respect to variations δϕ, δS, δΣ, δΓ subject to the variational constraint ∂L δΣ=−Pfr :∇δϕ+J ·∇δΓ (2.6) S ∂S and with δϕ(t )=0, δΓ(t )=0, i=1,2. i i This variationalformulationintroduces two new functions, Σ and Γ, whose time deriva- tive will be ultimately identified with the entropygenerationrate density and the tempera- ture, respectively. The tensor Pfr is the Piola-Kirchhofffriction stress tensor and J is the S entropyfluxdensityinLagrangianrepresentation. Thelinkwiththecorrespondingfamiliar Eulerian objects will be explained below. The notation “:” indicates the contraction of tensors with respect to two indices. The δ-notation in (2.4) indicates that we compute the variation of the functional with respect to all the fields, namely ϕ,S,Σ,Γ. The constraint (2.5) is referred to as a phe- nomenological constraint, since the expression of the thermodynamic fluxes Pfr and J are S obtained through phenomenological laws, also called parameterizations of the irreversible processes. It is a nonlinear constraint on the time derivatives of the field variables. We note that the right hand side of this constraint is of the generic form J Λ˙ , for the α α α thermodynamic fluxes J (here Pfr and J ) acting on the rate of thermodynamic displace- α S ments Λ˙ (here ϕ˙ and Γ˙). The constraint (2.6) is referred to as a variaPtional constraint, α since it is a constraint on the variations to be considered in (2.4). One passes from (2.5) to (2.6) by formally replacing each occurrence of a time derivative by a δ-variation, i.e., J Λ˙ J δΛ . This is reminiscent of what happens in the Lagrange-d’Alembert α α α α α α principle in nonholonomic mechanics, see Remark 2.1 below. P P 2.2 Variational formulation in Eulerian description 5 Taking the variations in (2.4), using the variational constraint (2.6) and collecting the terms proportional to δϕ, δΓ, and δS, we get the three conditions d ∂L ∂L ∂L−1 ∂L +DIV +Γ˙ Pfr − =0 dt∂ϕ˙ ∂∇ϕ ∂S ∂ϕ (cid:18) (cid:19) ∂L−1 ∂L S˙ =DIV Γ˙ J +Σ˙, Γ˙ =− , S ∂S ∂S (cid:18) (cid:19) where DIV denotes the divergence with respect to the labels X. From the last equation, we have Γ˙ = −∂L = T, the temperature in material representation. This attributes to Γ ∂S the meaning of temperature displacement, as considered in Green and Naghdi [1991] and introducedinvon Helmholtz [1884]. The secondequationsimplifies to S˙+DIVJ =Σ˙ and S attributes to Σ˙ the meaningofentropy generation rate density. Fromthe firstequationand the constraint, we thus get the system d ∂L ∂L ∂L +DIV −Pfr − =0 dt∂ϕ˙ ∂∇ϕ ∂ϕ (cid:18) (cid:19) T(S˙ +DIVJ )=Pfr :∇ϕ˙ −J ·∇T, S S for the fields ϕ(t,X) and S(t,X). We will discuss the parameterization of the thermo- dynamic fluxes Pfr and J in terms of the thermodynamic forces below in the Eulerian S description. We leave to the reader the computation of the explicit form of these equations for the Lagrangian L of the dry atmosphere in (2.2). We shall only present the computation in Eulerian variables below. Remark 2.1 (Lagrange-d’Alembert principle in nonholonomic mechanics). We now com- ment on the analogy between our variational formulation and the Lagrange-d’Alembert principle used in nonholonomic mechanics. Let us consider a mechanical system with con- figuration variable q and velocity v = q˙. We assume that the system is subject to a linear constraint on velocity, i.e., a constraint of the form ω(q)·q˙ = 0, for a q-dependent linear map ω(q). Typical examples are rolling constraints. The extension of Hamilton’s principle to nonholonomic systems (called the Lagrange-d’Alembert principle) consists in imposing that the action functional is critical with respect to variations δq subject to the linear con- straint ω(q)·δq = 0. It formally follows by replacing the time derivative in the constraint ω(q)·q˙ =0 by a δ-derivative, see, e.g., Bloch [2003]. In our case, the passage from (2.5) to (2.6) can be formally seen as a generalization of this approach, to the case of a nonlinear constraint. We refer to Gay-Balmaz and Yoshimura [2017b] for an extensive discussion of the principle (2.4)–(2.6). 2.2 Variational formulation in Eulerian description In terms of Eulerian fields, the Lagrangian(2.2) reads ℓ(v,ρ,s)= L(v,ρ,s)dx, ZD for the Lagrangiandensity L given by 1 L(v,ρ,s)=ρ |v|2+R·v−Φ−u(s/ρ,1/ρ) . (2.7) 2 (cid:18) (cid:19) The Eulerian quantities γ(t,x) and σ(t,x) associated to Γ(t,X) and Σ(t,X) are defined as Γ(t,X)=γ(t,ϕ(t,X)) and Σ(t,X)=σ(t,ϕ(t,X))|∇ϕ(t,X)|; (2.8) 2.2 Variational formulation in Eulerian description 6 the entropy flux density j in Eulerian representation is related to J as s S ∇ϕ(t,X)·J (t,X)=|∇ϕ(t,X)|j (t,ϕ(t,X)); (2.9) S s finally, the friction stress σfr is related to the Piola-Kirchhofffriction stress Pfr as Pfr(t,X)·∇ϕ(t,X)=|∇ϕ(t,X)|σfr(t,X). (2.10) By using these relations,we canrewrite the variationalformalism(2.4)–(2.6) entirely in terms of Eulerian variables as follows: t2 δ (L+(s−σ)D γ)dxdt=0, (2.11) t Zt1 ZD subject to the phenomenological constraint ∂L D¯ σ =−σfr :∇v+j ·∇D γ (2.12) t s t ∂s and with respect to variations δv=∂ ζ+v·∇ζ−ζ·∇v, δρ=−div(ρζ), δs, δσ, and δγ, (2.13) t that are subject to the variational constraint ∂L D¯ σ =−σfr :∇ζ+j ·∇D γ (2.14) δ s δ ∂s with ζ(t )=0 and δγ(t )=0, i=1,2. i i Thefirsttwoexpressionsin(2.13)areobtainedbytakingthevariationswithrespecttoϕ, u,andρ,oftherelationsϕ˙(t,X)=u(t,ϕ(t,X))and̺ (X)=ρ(t,ϕ(t,X))|∇ϕ(t,X)|,respec- 0 tively, and defining ζ(t,x) as δϕ(t,X) = ζ(t,ϕ(t,X)). These formulas can be also directly justified by employing the Euler-Poincar´e reduction theory, see Holm, Marsden and Ratiu [2002]. In (2.12) and (2.14), we introduced the notations D f := ∂ f +v·∇f, D¯ f := ∂ f + t t t t div(fv), D f :=δf +ζ·∇f and D¯ f :=δf +div(fζ) for the Lagrangiantime derivatives δ δ and variations of scalar fields and density fields. By applying (2.11) and using the expression for the variations δv and δρ, we find the condition t2 ∂L ∂L +(s−σ)∇γ ·(∂ ζ+v·∇ζ−ζ·∇v)− div(ρζ) ∂v t ∂ρ Zt1 ZD(cid:20)(cid:18) (cid:19) ∂L + +D γ δs−D¯ (s−σ)δγ−δσD γ dxdt=0. t t t ∂s (cid:18) (cid:19) (cid:21) Using the variational constraint (2.14) and collecting the terms proportional to ζ, δs, and δγ, we obtain the three conditions ∂L ∂L ∂L−1 (∂t+£v) ∂v +(s−σ)∇γ =ρ∇∂ρ −σ∇Dtγ−div Dtγ∂s σfr (cid:18) (cid:19) (cid:18) (cid:19) ∂L−1 +div D γ j ∇γ, t s ∂s (cid:18) (cid:19) ∂L−1 ∂L D¯ (s−σ)=div D γ j , +D γ =0, t t s t ∂s ∂s (cid:18) (cid:19) 2.2 Variational formulation in Eulerian description 7 where we introduced the Lie derivative notation£vm:=v·∇m+∇vT·m+mdivv for a one-form density m along a vector field v. Further computations finally yield the system ∂L ∂L ∂L (∂t+£v)∂v =ρ∇∂ρ +s∇∂s +divσfr ∂∂Ls(D¯ts+divjs)=−σfr :∇v−js·∇∂∂Ls (2.15) D¯ ρ=0, whoselastequation,themastsconservationequation,followsfromthedefinitionofρinterms of ̺ . These are the general equations of motion, in Lagrangian form, for fluid dynamics 0 subject to the irreversible processes of viscosity and heat conduction. From one of the above conditions, we note that D γ = −∂L =: T, which attributes to γ the meaning of t ∂s temperature displacement in Eulerian variables. From the above conditions, we also note that the variable σ verifies 1 D¯ σ =D¯ s+divj = (σfr :∇v−j ·∇T). (2.16) t t s s T Therefore, D¯ σ corresponds to the total entropy generation rate density of the system. At t this stage, the expressions of σfr and j are still not specified. s In absence of irreversible processes (i.e., if σfr =0 and j =0), equations (2.15) recover s thegeneralformoftheEuler-Poincar´eequations derivedinHolm, Marsden and Ratiu[2002] by Lagrangianreduction. We now write this system in the case of the Lagrangian L of the dry atmosphere given in (2.7), but keeping a general expression for the internal energy. The partial derivatives are ∂L ∂L 1 ∂L ∂u =ρ(v+R), = |v|2+R·v−Φ−u+ηT −vp, =− =−T. ∂v ∂ρ 2 ∂s ∂η On one hand, we note that ∂L ∂L 1 ρ∇ +s∇ =ρ∇ |v|2+R·v −ρ∇Φ−∇p, ∂ρ ∂s 2 (cid:18) (cid:19) on the other hand, using D¯ ρ=0, we have t T (∂t+£v)(ρ(v+R))=ρ(∂tv+v·∇(v+R)+∇v ·(v+R)). From (2.15) we thus get, as desired, the equations of motion for a heat conducting gas subject to the Coriolis, gravitational, and viscous forces: ρ(∂ v+v·∇v+2Ω×v)=−ρ∇Φ−∇p+divσfr t T(D¯ s+divj )=σfr :∇v−j ·∇T (2.17) t s s D¯ ρ=0. t The second equation in (2.17) can be rewritten in terms of the temperature as 1 D T =−ρc2Γdivv+ σfr :∇v−div(Tj ) , (2.18) t s ρC s v (cid:0) (cid:1) whereC (v,T)=T ∂η(T,v)isthespecificheatatconstantvolume,c2(v,η)= ∂p(v,η)isthe v ∂T s ∂ρ square of the speed of sound, and Γ(p,η)= ∂T(p,η) is the diabatic temperature gradient. ∂p 2.3 The case of dry air and ideal gases 8 For meteorologicalapplications, the potential temperature θ is preferred as the temper- aturetodescribethethermodynamicequation,sinceitturnsouttobeconservedinabsence ofviscosityandheatconduction. Ageneraldefinitionofthe potentialtemperaturevalidfor any state equation is given by p0 θ(p,T):=T + Γ(p′,η(p,T))dp′, (2.19) Zp where p is a given reference pressure. One obtains, from (2.18), the potential temperature 0 equation 1 ∂θ D θ = σfr :∇v−div(Tj ) . (2.20) t s ρC ∂T p (cid:0) (cid:1) 2.3 The case of dry air and ideal gases The above development is applicable to any state equation. For the case of a perfect gas, suchasthedryair,thereareseveralsimplifications. ThespecificheatcoefficientC in(2.18) v is a constant and from (2.1), we have p = ρR T in (2.17) and ρc2 = p in (2.18), where d s ρCv R is the gas constantfor dry air. In this case, the potential temperature in (2.19) recovers d its usual expression θ = T/π, where π := (p/p0)Rd/Cp is the Exner pressure associated to p . One obtains, from (2.20), the potential temperature equation 0 θ D θ = (σfr :∇v−div(Tj )). (2.21) t s ρTC p It is the system (2.17) with the second equation replaced by (2.21), together with the state equationfortheidealgasp=ρR T,thatistraditionallyusedtodescribethedynamicsofa d dryatmosphere,see,forinstance,Gill[1982]. Ourvariationalapproachishowevernaturally expressed in terms of the entropy density rather than the potential temperature. In some atmospheres, the perfect gas hypothesis may no longer be made. For example, for the atmosphere of Venus, while the state equation of an ideal gas can still be used, the specific heat depends on the temperature (Seiff et al. [1985]). An analytic approximation for this temperature dependence is givenby C (T)=C (T/T )ν, where C ,ν,T are con- p p0 0 p0 0 stants, see Lebonnois et al [2010], which yields a correspondingexpressionfor the potential temperature,accordingtothegeneraldefinitionabove. Ofcourse,ourvariationalformalism does apply to this case, as it does for any one component gas. 2.4 Heat exchanges So far, we have considered the atmosphere as an isolated system, i.e., with no exchange of work,heatormatterwithitssurroundings(space,oceanorearth’ssurface). Therighthand side of the entropy equation in (2.17) thus only consists of a net production of entropy due to the irreversible processes. Heatexchanges between the atmosphere andits surroundings,such as radiativeheating andcooling,surfacesensibleheatfluxandsurfacelatentheatflux,canbeeasilyincorporated in our variational formulation, as long as these are considered as external processes. We incorporate the heating into our variational formalism (2.4)–(2.6) by modifying the phenomenological constraint (2.5) to ∂L Σ˙ =−Pfr :∇ϕ˙ +J ·∇Γ˙ −R, (2.22) S ∂S whereRdenotestheheatingratedensityinLagrangianrepresentation. Thevariationalcon- straint (2.6) is however kept unchanged. This follows a general principle on the variational 2.5 Potential vorticity and circulation theorem 9 formulation of thermodynamics stated in Gay-Balmaz and Yoshimura [2017a,b], namely, that external effects only affect the phenomenological constraint and not the variational constraint. Schematically, one passes from the phenomenological constraint to the varia- tional constraint as J Λ˙ +P J δΛ , where P denotes the power density α α α ext α α α ext associated to heat transfer between the system and the exterior. P P In Eulerian representation, the variational formalism (2.11)–(2.14) is thus modified by adding the contribution of the heating in (2.12) as ∂L D¯ σ =−σfr :∇v+j ·∇D γ−r, (2.23) t s t ∂s whereRandrarerelatedasR(t,X)=r(t,ϕ(t,X))|∇ϕ(t,X)|. Thisresultsinamodification of the entropy equation in (2.15) into T(D¯ s+divj )=σfr :∇v−j ·∇T +r (2.24) t s s and corresponding modifications in the temperature and potential temperature equations. 2.5 Potential vorticity and circulation theorem Potentialvorticityis a scalarfield constructedby projecting the absolute vorticityonto the gradientofthepotentialtemperature(or,moregenerally,thegradientofanadvectedscalar field) anddividing the resultbythe totalmassdensity. Potentialvorticityis a fundamental quantity since it allows one to understand many processes taking place in the atmosphere and the ocean (e.g., Hoskins, McIntyre, and Robertson [1985]). In absence of irreversible processes, it turns out to be a conserved quantity. We first recall here how the irreversible processes affect the evolution of the potential vorticity. Then we extend the definition of potential vorticity to the general setting of the Lagrangiansystem (2.15) and write its evolution equation. Having such a general equation is useful for a unified treatment of potential vorticity for various approximations of the equations of atmospheric dynamics, with the irreversible processes included. TheRossby-Ertelpotentialvorticityq isdefinedbyρq =ζ ·∇θ,whereζ =curlv+2Ω a a is the absolute vorticity and θ is the potential temperature defined in (2.19). Its evolution equation ∂θ ρD q =div ρ−1divσfr×∇θ+ Qζ , t ∂T a (cid:18) (cid:19) follows from the evolution of the absolute vorticity ∂ ζ +curl(ζ ×v) = ρ−2∇ρ×∇p+ t a a curl(ρ−1divσfr), the conservation of mass, and the potential temperature equation (2.21) (see e.g., Haynes and McIntyre [1986]). In absence of viscosity and heat conduction, q is conserved. In the more general situation of the system (2.15) associated to a general expression of the Lagrangian density L, we can consider a potential vorticity q˜defined as ρq˜= ζ ·∇ψ, a where ζ :=curl ρ−1∂L/∂v and ψ is a scalar field satisfying an evolution equation of the a type D ψ =ψ˙. The first equation in (2.15) implies t (cid:0) (cid:1) ∂L ∂ ζ +curl(ζ ×v)=∇η×∇ +curl(ρ−1divσfr). (2.25) t a a ∂s From this equation and D ψ =ψ˙, one obtains the evolution of potential vorticity as t ∂L ρD q˜=div η∇ ×∇ψ+ρ−1divσfr×∇ψ+ψ˙ζ . (2.26) t ∂s a (cid:18) (cid:19) 2.6 Phenomenological relations and entropy production 10 If, for example, ψ is chosen as the specific entropy η, then (2.26) simplifies to ρD q˜=div ρ−1divσfr×∇η+η˙ζ , t a where, from the second equation in (2.(cid:0)15), η˙ = 1 (σfr :∇v−(cid:1)div(Tj )) with T =−∂L. Tρ s ∂s LetusnowconsiderKelvin’scirculationtheorem. Forthesystem(2.17)itdirectlyfollows from the balance of momentum as d (v+R)·dx= ρ−1divσ·dx, σ =−pδ+σfr, dt Ict Ict where c is a loop advected by the wind flow. Its more general version for system (2.15) is t found to be d 1∂L ∂L ·dx= η∇ +ρ−1divσfr ·dx. dt ρ∂v ∂s Ict Ict(cid:18) (cid:19) In absence of viscosity it recovers the general form of Kelvin’s circulation theorem in the Euler-Poincar´esetting derived in Holm, Marsden and Ratiu [2002]. 2.6 Phenomenological relations and entropy production Thephenomenologicalconstraints(2.5)and(2.12)tobeusedinthevariationalformulations are determined by the expressions of the thermodynamic fluxes σfr and j in terms of the s thermodynamic forces ∇v and ∇T. The phenomenological expression of these fluxes must be in agreement with the second law of thermodynamics, which requires that the internal entropy production, given in equation (2.16), is positive. When only viscosity and heat conduction are considered, we have the well-known relations 2 σfr =2µ(Defv)+ ζ− µ (divv)δ and Tj =−k∇T (Fourier law), (2.27) s 3 (cid:18) (cid:19) with Defv= 1(∇v+∇vT) the deformation tensor, and where µ≥0 is the first coefficient 2 of viscosity (shear viscosity), ζ ≥ 0 is the second coefficient of viscosity (bulk viscosity), and k ≥ 0 is the thermal conductivity. In general, these coefficients may all depend on thermodynamic variables. For the atmosphere below 100 km, µ is so small (µ = 1.7× 10−5kg m−1s−1 for standard atmospheric conditions) that molecular viscosity is negligible except in a thin layer within a few centimeters of the earth’s surface where the vertical shear is very large. The second coefficient of viscosity is irrelevant in incompressible flows and notoriouslydifficult to measure in compressible flows. A common assumptionis Stokes hypothesis ζ =0. 3 Variational derivation of the thermodynamic of a moist atmosphere Moist air consists of dry air and water substance in its different phases, namely, water vapor, liquid water, and solid water. The gas component of moist air is thus a mixture of dry air and water vapor. The liquid and solid phases of water both consist of an airborne component (cloud) that follows the motion of the gas component, and of a precipitating component (rain or snow) that has relative motion with respect to the gas component. Weshallnotconsiderthe precipitatingcomponentinthis paper. Also,forsimplicity,we do not consider the ice phase, but it can be easily added. We thus assume that moist air