betaFeedback

WorksPhysics and Astronomy

Hamilton's Principle with Phase Changes and Conservation Principles for Moist Potential Vorticity

Abstract

Many definitions of moist potential vorticity (PV) have been proposed to extend the dry theory of Ertel PV. None of the moist PV definitions seem to have all of the desirable properties of the dry Ertel PV. For instance, dry PV is not only a globally conserved quantity, but also a material invariant that is conserved along fluid parcel trajectories. Therefore, an open question remains: is there a moist PV that is a material invariant, if clouds and phase changes of water are present? In prior studies, definitions of moist PV have been proposed based on physical and mathematical intuition. Here, a systematic approach is used. In particular, a particle relabeling symmetry is devised for a moist atmosphere and then Noether's theorem is employed to arrive at the associated conservation laws for moist PV. A priori, it is not clear whether this systematic approach will be viable, since it relies on variational derivatives in Hamilton's principle, and phase changes introduce singularities that could potentially prevent derivatives at cloud edge. However, it is shown that the energy and the Lagrangian density are sufficiently smooth to allow variational derivatives, in a moist Boussinesq system with reversible phase transitions between water vapor and liquid cloud water.. From the particle relabeling symmetry, a moist Kelvin circulation theorem is found, along with a moist PV conservation law that applies not for each individual parcel but for parcel-integrated PV, integrated over certain local volumes.

Introduction

Potential vorticity (PV) is one of the fundamental conserved quantities of geophysical fluid dynamics. Conservation of PV was recognized long ago by Rossby [43] and Ertel [14]. One important property of PV is that it is a material invariant, meaning that it is conserved along fluid particle trajectories; this allows it to function as a tracer of fluid particles. This material invariance property is exploited in isentropic PV maps which have been used extensively as an operational diagnostic tool in the development of mid-latitude synoptic weather systems ([24, 22, 23, 52]). Another important property is its invertibility: the instantaneous global distribution of PV with appropriate boundary conditions can be used to recover the balanced parts of the wind and temperature fields.

While the concept of PV is well-established for a dry atmosphere, the concept of PV is less clear for a moist atmosphere with clouds, phase changes, and latent heating. Several definitions of moist PV have been proposed, but a moist counterpart that retains all of the desirable properties of dry PV has been elusive. For example, one moist PV definition (here called PVvPV_v) is defined in terms of the virtual potential temperature θv\theta_v, and it was shown to possess an invertibility principle which produces certain velocity and temperature fields ([47]). However, Wetzel et al. [55] have shown that PVvPV_v is not balanced and therefore does not recover the balanced velocity and temperature fields. Although PVvPV_v is not balanced, it is a conserved quantity in unsaturated regions and therefore can be used to monitor and diagnose latent heating in the atmosphere ([12, 27, 15, 3, 33, 4, 26, 28, 7]). Another widely used PV (here called PVePV_e) is based on the equivalent potential temperature θe\theta_e ([2, 13]). One criticism of PVePV_e is that it fails to possess an invertibility principle in the presence of phase changes ([47, 8]). However, taking advantage of one or more additional balanced quantities MM involving water, extended PVPV–MM inversion can indeed be used to solve for the balanced moist flow using PVePV_e ([49, 54, 55]). As a general principle, for a moist system, a single moist PV variable by itself is not sufficient information to find the balanced flow components, and additional moisture variables (M) need to be retained ([49]).

Furthermore, the moist PV quantities that have been proposed are lacking the fundamental conservation property of dry PV: conservation along fluid parcel trajectories. Some moist PV quantities are conserved in one phase—either inside a cloud or outside a cloud—but none are conserved along fluid parcel trajectories in the presence of clouds and phase changes. Hence, there remains an open question: For a moist system with clouds and phase changes, is there a moist PV quantity that is a material invariant?

While past definitions of moist PV have been proposed based on physical and mathematical intuition, there is in fact a systematic approach to identifying conservation principles. In Hamiltonian mechanics and Lagrangian mechanics, conservation principles are related to symmetries in the action, via Noether’s theorem [36, 19, 37]. Therefore, as a systematic route to identifying a moist PV conservation law, one can first write the moist atmospheric equations in a Hamiltonian or Lagrangian formulation, and then look for symmetries in the action or the Lagrangian density.

The goal of the present paper is to take the systematic approach described above. To first define a Hamiltonian or Lagrangian formulation, we use the piecewise-quadratic energy that was recently identified for the moist Boussinesq equations with phase changes [32]. As the symmetry of interest for moist PV, we will seek a moist analog of the symmetry associated with dry PV conservation: particle relabeling symmetry [6, 42, 48, 35] (Salmon 1988).

The present paper is another contribution of the use of the Lagrangian or Hamiltonian formulation for geophysical fluid dynamics. For example, in other applications, by attaching additional constraints to the Lagrangian/Hamiltonian, the Lagrangian/Hamiltonian formulation provides a systematic way of deriving approximate dynamical equations that retain analogs of exact conservation laws [21, 10] (Salmon 1998). Well-known examples of such approximate models are the quasi-geostrophic, shallow water, KdV and Green-Naghdi equations (see Salmon, 1988, and references therein). Additionally, the Hamiltonian formulation is useful in nonlinear stability theory (Salmon, 1988). Other applications include data assimilation [9, 16], statistical analysis [1, 31, 34] and stochastic partial differential equations (SPDEs) for fluid dynamics from a stochastic variational principle (SVP) [20]. It has also been used in the development of numerical methods [41].

In Section 2, we provide background about the governing moist Boussinesq equations and their conservation of energy principle. Next in Section 3, we show that the energy is differentiable, allowing for a formulation of the dynamics from the perspective of Lagrangian mechanics. We further demonstrate that Hamilton’s principle leads to Euler-Lagrange equations that are equivalent to the original equations. Section 4 explains how to adapt particle-relabeling symmetry to the moist system with phase changes, and thereby derive moist analogues of Kelvin’s circulation theorem and conservation of potential vorticity. Conclusions are given in Section 5.

Background

As background, we describe the model equations in section 2.1 and the conserved energy in section 2.2. The energy will be used later to suggest the form of a Hamiltonian and a Lagrangian.

Governing Equations

We start by presenting the model equations, which are moist Boussinesq equations with phase changes. This type of model has been used for many purposes and with varying degree of idealization (e.g., [25, 50, 5, 11, 51, 32, 56 b] (Pauluis and Schumacher 2010)). While a Boussinesq framework is used here, one would suspect that similar results could be obtained for an anelastic or compressible system, although such cases are beyond the scope of the present paper. A Boussinesq framework is used here instead of an anelastic or compressible framework because it allows for some calculations to be in simpler form and because it allows use of an energy function (section 2.2) that has a simple piecewise quadratic form [32].

When the two thermodynamic variables are chosen to be θe\theta_e and qtq_t, the equations take the form

Du⃗Dt=−∇ϕ+bk^,(1)\frac{D\vec{u}}{Dt}=-\nabla\phi+b\hat{k}, \tag*{(1)}
DθeDt+dθ~edzw=0,(2)\frac{D\theta_e}{Dt}+\frac{d\widetilde{\theta}_e}{dz}w=0, \tag*{(2)}
DqtDt+dq~tdzw=0,(3)\frac{Dq_t}{Dt}+\frac{d\widetilde{q}_t}{dz}w=0, \tag*{(3)}
∇⋅u⃗=0,(4)\nabla\cdot\vec{u}=0, \tag*{(4)}

where x⃗=(x,y,z)\vec{x}=(x,y,z) is the position vector, k^\hat{k} is the unit vector in the zz-direction, u⃗=(u,v,w)\vec{u}=(u,v,w) is the velocity vector, D/Dt=∂/∂t+u⃗⋅∇D/Dt=\partial/\partial t+\vec{u}\cdot\nabla is the material derivative, ϕ=p′/ρ0\phi=p'/\rho_0, p′p' is the pressure anomaly, ρ0\rho_0 is a constant background density, θe\theta_e is the equivalent potential temperature anomaly, and qtq_t is the anomalous total water mixing ratio. Every thermodynamic variable considered here has been decomposed into a background function of height zz and an anomalous part. For example, θetot=θ~e(z)+θe(x⃗,t)\theta_e^{tot}=\widetilde{\theta}_e(z)+\theta_e(\vec{x},t) and qttot=q~t(z)+qt(x⃗,t)q_t^{tot}=\widetilde{q}_t(z)+q_t(\vec{x},t). The background vertical gradients, dθ~e/dzd\widetilde{\theta}_e/dz and dq~t/dzd\widetilde{q}_t/dz, will be assumed to be constants, in analogy with a common setup of the Boussinesq equations in the dry case (e.g., [e.g. 29]).

The buoyancy bb is influenced by phase changes of water, and bb can be expressed as a function of θe\theta_e, qtq_t and zz. To do this, one can start from a definition of

b=g(θθ0+Rvdqv−ql),(5)b=g\left(\frac{\theta}{\theta_0}+R_{vd}q_v-q_l\right), \tag*{(5)}

where θ0≈300 K\theta_0\approx300\ \mathrm{K} is the constant background potential temperature, g≈9.8 m s−2g\approx9.8\ \mathrm{m}\,\mathrm{s}^{-2} is the acceleration due to gravity and Rvd=(Rv/Rd)−1≈0.61R_{vd}=(R_v/R_d)-1\approx0.61, where RdR_d is the gas constant for dry air and RvR_v is the gas constant for water vapor. The three variables in this buoyancy expression are potential temperature θ\theta, water vapor mixing ratio qvq_v, and liquid water mixing ratio qlq_l, and they can be related to qtq_t and θe\theta_e as described below.

The set of variables (θ,qv,ql)(\theta,q_v,q_l) can be related to the set of variables (θe,qt)(\theta_e,q_t) in the following way. The total water qtq_t is defined as the sum of water vapor qvq_v and liquid water qlq_l:

qt=qv+ql.(6)q_t=q_v+q_l. \tag*{(6)}

The second variable, equivalent potential temperature θe\theta_e, is defined here in linearized form as

θe=θ+Lvcpqv,(7)\theta_e=\theta+\frac{L_v}{c_p}q_v, \tag*{(7)}

where the latent heat is a constant with value Lv≈2.5×106 J kg−1L_v\approx2.5\times10^6\ \mathrm{J}\,\mathrm{kg}^{-1} and specific heat is a constant with value cp≈103 J kg−1K−1c_p\approx10^3\ \mathrm{J}\,\mathrm{kg}^{-1}\mathrm{K}^{-1}. From these definitions in (6) and (7), one can find the variables (θe,qt)(\theta_e,q_t) if given the variables (θ,qv,ql)(\theta,q_v,q_l).

For the opposite direction, if given qtq_t, one can partition it into its vapor (qvq_v) and liquid (qlq_l) components by using the saturation mixing ratio qvsq_{vs}, which acts as a threshold. The saturation mixing ratio will be assumed to be a function of only zz by making the assumption that TtotT^{tot} and ptotp^{tot} are close to the background states p~(z)\widetilde{p}(z) and T~(z)\widetilde{T}(z), which in turn depend only on the height, in which case qvstot(Ttot,ptot)≈qvstot(T~(z),p~(z))q_{vs}^{tot}(T^{tot},p^{tot})\approx q_{vs}^{tot}(\widetilde{T}(z),\widetilde{p}(z)) (see, e.g., the appendix of Hernandez-Duenas et al., 2013). Two cases are possible, depending on whether qtq_t is below the threshold (qt<qvsq_t<q_{vs}) or above the threshold (qt>qvsq_t>q_{vs}). If qtq_t is below the threshold (qt<qvsq_t<q_{vs}), then no liquid water is present (i.e., ql=0q_l=0), and the water is all in the form of vapor: qt=qvq_t=q_v. On the other hand, if qtq_t is above the threshold (qt>qvsq_t>q_{vs}), then the water vapor is at its saturation value (i.e., qv=qvsq_v = q_{vs}), and the remaining water is liquid water: ql=qt−qvsq_l = q_t - q_{vs}. To encompass both of these cases in a unified way, the formulas for qvq_v and qlq_l can be written as

qv=min⁡(qt,qvs),(8)q_v = \min(q_t, q_{vs}), \tag*{(8)}
ql=max⁡(0,qt−qvs).(9)q_l = \max(0, q_t - q_{vs}). \tag*{(9)}

The two cases will be called the unsaturated phase (if qv<qvsq_v < q_{vs}) and the saturated phase (if qv=qvsq_v = q_{vs}).

If θe\theta_e and qtq_t are both given, then one can also see that

θ=θe−Lvcpmin⁡(qt,qvs).(10)\theta= \theta_e - \frac{L_v}{c_p}\min(q_t, q_{vs}). \tag*{(10)}

This expression follows from (7) and (8). Together, the three expressions (8), (9), and (10) give a specification of the variables (θ,qv,ql)(\theta, q_v, q_l) if given the variables (θe,qt)(\theta_e, q_t).

Note that many of the expressions above have been written in terms of anomalies, such as θe\theta_e, rather than the total expressions, such as θetot=θ~e(z)+θe(x,y,t)\theta_e^{tot} = \widetilde{\theta}_e(z) + \theta_e(x,y,t), which include background states. In particular, notice that the fundamental condition for saturation should be qttot=qvstotq_t^{tot} = q_{vs}^{tot}. Nevertheless, one can choose the background state to be unsaturated with q~t(z)=q~v(z)\widetilde{q}_t(z) = \widetilde{q}_v(z), and one can also break qvstotq_{vs}^{tot} into background and anomaly as qvstot=q~vs+qvsq_{vs}^{tot} = \widetilde{q}_{vs} + q_{vs}, with the choice of q~vs(z)=q~v(z)\widetilde{q}_{vs}(z) = \widetilde{q}_v(z). With these choices of q~t(z)=q~v(z)=q~vs(z)\widetilde{q}_t(z) = \widetilde{q}_v(z) = \widetilde{q}_{vs}(z), it follows that the saturation condition qttot=qvstotq_t^{tot} = q_{vs}^{tot} can be written in terms of anomalies in a simple way, as qt=qvsq_t = q_{vs}.

It is convenient to rewrite these equations by replacing the thermodynamic variables, θe\theta_e and qtq_t, with the unsaturated and saturated buoyancy variables, bub_u and bsb_s. The variables bub_u and bsb_s are defined so that the buoyancy bb can be written simply as

b=buHu+bsHs,(11)b = b_u H_u + b_s H_s, \tag*{(11)}

where HuH_u and HsH_s are Heaviside functions that are indicators of the unsaturated and saturated regions, respectively:

Hu={1if qt<qvs(z)0if qt≥qvs(z),Hs=1−Hu.(12)H_u = \begin{cases} 1 & \text{if } q_t < q_{vs}(z) \\ 0 & \text{if } q_t \ge q_{vs}(z), \end{cases} \qquad H_s = 1 - H_u. \tag*{(12)}

To define bub_u and bsb_s, the buoyancy definition in (5) is rewritten in terms of equivalent potential temperature θe\theta_e, total water qtq_t, and prescribed saturation mixing ratio qvsq_{vs}, using (6)–(10), which yields

bu=g(θeθ0+(Rvd−Lvcpθ0)qt),(13)b_u = g\left(\frac{\theta_e}{\theta_0} + \left(R_{vd} - \frac{L_v}{c_p\theta_0}\right)q_t\right), \tag*{(13)}
bs=g(θeθ0+(Rvd−Lvcpθ0+1)qvs−qt).(14)b_s = g\left(\frac{\theta_e}{\theta_0} + \left(R_{vd} - \frac{L_v}{c_p\theta_0} + 1\right)q_{vs} - q_t\right). \tag*{(14)}

These types of unsaturated and saturated buoyancy variables have also been used in other work on moist convection (e.g., Kuo, [25]; Bretherton, [5]; Pauluis and Schumacher, 2010; Smith and Stechmann, [49]) and are sometimes called the dry and moist buoyancy variables. Notice that b=bub = b_u in unsaturated regions, and b=bsb = b_s in saturated regions, but the variables bub_u and bsb_s are defined and exist everywhere, since they are defined as functions of θe\theta_e, qtq_t, and qvsq_{vs}.

The governing equations in terms of bub_u and bsb_s are

Du⃗Dt=−∇ϕ+(buHu+bsHs)k^(15)\frac{D\vec{u}}{Dt}=-\nabla\phi+(b_uH_u+b_sH_s)\hat{k} \tag*{(15)}
DbuDt+Nu2w=0(16)\frac{Db_u}{Dt}+N_u^2w=0 \tag*{(16)}
DbsDt+Ns2w=0,(17)\frac{Db_s}{Dt}+N_s^2w=0, \tag*{(17)}
∇⋅u⃗=0,(18)\nabla\cdot\vec{u}=0, \tag*{(18)}

where the unsaturated and saturated buoyancy frequencies, Nu2N_u^2 and Ns2N_s^2, are constants given by

Nu2=gddz(θ~eθ0+(Rvd−Lvcpθ0)q~t),(19)N_u^2=g\frac{d}{dz}\left(\frac{\widetilde{\theta}_e}{\theta_0}+\left(R_{vd}-\frac{L_v}{c_p\theta_0}\right)\widetilde{q}_t\right), \tag*{(19)}
Ns2=gddz(θ~eθ0−(Rvd−Lvcpθ0)(qvs(z)−q~t)).(20)N_s^2=g\frac{d}{dz}\left(\frac{\widetilde{\theta}_e}{\theta_0}-\left(R_{vd}-\frac{L_v}{c_p\theta_0}\right)\left(q_{vs}(z)-\widetilde{q}_t\right)\right). \tag*{(20)}

The evolution equations for bub_u and bsb_s in (16)–(17) follow from the evolution equations for θe\theta_e and qtq_t in (2)–(3), using the definitions of bub_u and bsb_s in (13)–(14); and the definitions of Nu2N_u^2 and Ns2N_s^2 in (19)–(20) also follow from this derivation. Finally, the indicator functions (12) can be written as

Hu={1if bu>bs0if bu≤bs,Hs=1−Hu.(21)H_u= \begin{cases} 1 & \text{if } b_u>b_s\\ 0 & \text{if } b_u\leq b_s, \end{cases} \qquad H_s=1-H_u. \tag*{(21)}

With this reformulation of the equations in terms of buoyancy variables, many aspects of the subsequent discussion will be simplified.

Piecewise Quadratic Energy

In subsequent sections, to explore a possible particle-relabeling symmetry of the Lagrangian, it will first be necessary to define a suitable Lagrangian functional. To help motivate the form of a Lagrangian functional for the model in (1)–(4), we now describe an energy function that was recently derived [32].

The model in (1)–(4) or (15)–(18) has an associated energy that is given by

E=K+V=12u⃗⋅u⃗+12bu2Nu2Hu+12bs2Ns2Hs+12Nu2Ns2Nu2−Ns2M2Hu(22)E=\mathcal{K}+\mathcal{V}=\frac{1}{2}\vec{u}\cdot\vec{u}+\frac{1}{2}\frac{b_u^2}{N_u^2}H_u+\frac{1}{2}\frac{b_s^2}{N_s^2}H_s+\frac{1}{2}\frac{N_u^2N_s^2}{N_u^2-N_s^2}M^2H_u \tag*{(22)}

and evolves according to

∂E∂t+∇⋅[u⃗(E+ϕ)]=0,(23)\frac{\partial E}{\partial t}+\nabla\cdot[\vec{u}(E+\phi)]=0, \tag*{(23)}

so that the domain-integrated energy is conserved [32]. In the energy definition in (22), the first term is the kinetic energy K\mathcal{K}, and the latter three terms are the potential energy V\mathcal{V}. The bu2b_u^2 and bs2b_s^2 terms are the buoyant potential energy terms in the unsaturated and saturated phases, respectively. The new variable MM that appears in the potential energy is actually related to bub_u and bsb_s via

M=buNu2−bsNs2,(24)M = \frac{b_u}{N_u^2} - \frac{b_s}{N_s^2}, \tag*{(24)}

and it is a material invariant—i.e.,

DMDt=0,(25)\frac{D M}{D t} = 0, \tag*{(25)}

which can be seen from (16), (17), and (24). This variable MM is associated with the additional eigenmode that is present in moist systems as opposed to dry systems [18, 49]. The M2M^2 term in the potential energy in (22) is a moist energy term that accounts for the latent heat released during change of phase. Notice that each potential energy term is multiplied by either HuH_u or HsH_s, so the form of the potential energy will be different in different phases. In this way, the energy in (22) is piecewise quadratic, as opposed to the quadratic energy that arises in the dry case.

Since the energy in (22) is piecewise-quadratic, it is positive if the coefficients Nu2N_u^2, Ns2N_s^2, and Nu2−Ns2N_u^2-N_s^2 are positive. Each of Nu2N_u^2 and Ns2N_s^2 will be positive if the fluid is stably stratified in the dry and moist sense, according to (19)–(20). For simpler expressions of Nu2N_u^2 and Ns2N_s^2, approximate forms of (19)–(20) are

Nu2≈gθ0dθ~dz,Ns2≈gθ0dθ~edz,(26)N_u^2 \approx\frac{g}{\theta_0}\frac{d\widetilde{\theta}}{d z}, \qquad N_s^2 \approx\frac{g}{\theta_0}\frac{d\widetilde{\theta}_e}{d z}, \tag*{(26)}

which indicate that Nu2N_u^2 and Ns2N_s^2 will be positive if the background vertical gradients of θ\theta and θe\theta_e are positive. In this case, the difference Nu2−Ns2N_u^2-N_s^2 will be positive if dq~v/dzd\widetilde{q}_v/d z is negative, as can be seen from (7) and (26).

While the energy in (22) is piecewise-defined in terms of Heaviside functions, it was shown by Marsico et al. [32] that the energy is continuous across phase boundaries. In other words, the potential energy in the unsaturated phase,

12bu2Nu2+12Nu2Ns2Nu2−Ns2M2,(27)\frac{1}{2}\frac{b_u^2}{N_u^2}+\frac{1}{2}\frac{N_u^2N_s^2}{N_u^2-N_s^2}M^2, \tag*{(27)}

and the potential energy in the saturated phase,

12bs2Ns2,(28)\frac{1}{2}\frac{b_s^2}{N_s^2}, \tag*{(28)}

will be equal at the phase interface. To see this, recall from (21) that the phase interface can be defined as locations where bu=bsb_u=b_s. Consequently, at the phase interface, one can rewrite the potential energy from the unsaturated phase as

12bu2Nu2+12Nu2Ns2Nu2−Ns2M2=12bu2Nu2+12Nu2Ns2Nu2−Ns2(buNu2−bsNs2)2(29)\frac{1}{2}\frac{b_u^2}{N_u^2}+\frac{1}{2}\frac{N_u^2N_s^2}{N_u^2-N_s^2}M^2 =\frac{1}{2}\frac{b_u^2}{N_u^2}+\frac{1}{2}\frac{N_u^2N_s^2}{N_u^2-N_s^2}\left(\frac{b_u}{N_u^2}-\frac{b_s}{N_s^2}\right)^2 \tag*{(29)}
=12bu2Nu2+12Nu2Ns2Nu2−Ns2(bu2Nu4−2bubsNu2Ns2+bs2Ns4)(30)=\frac{1}{2}\frac{b_u^2}{N_u^2}+\frac{1}{2}\frac{N_u^2N_s^2}{N_u^2-N_s^2}\left(\frac{b_u^2}{N_u^4}-\frac{2b_ub_s}{N_u^2N_s^2}+\frac{b_s^2}{N_s^4}\right) \tag*{(30)}
=12bs2Nu2+12Nu2Ns2Nu2−Ns2(bs2Nu4−2bu2Nu2Ns2+bs2Ns4)(31)=\frac{1}{2}\frac{b_s^2}{N_u^2}+\frac{1}{2}\frac{N_u^2N_s^2}{N_u^2-N_s^2}\left(\frac{b_s^2}{N_u^4}-\frac{2b_u^2}{N_u^2N_s^2}+\frac{b_s^2}{N_s^4}\right) \tag*{(31)}
=12bs2Ns2,(32)=\frac{1}{2}\frac{b_s^2}{N_s^2}, \tag*{(32)}

so that (27) and (28) are equal at the phase interface, where the definition of MM from (24) was also used in the calculation. Intuitively, the buoyant energy bu2/Nu2b_u^2/N_u^2 should be smaller than bs2/Ns2b_s^2/N_s^2 at the phase interface, since Nu2N_u^2 is typically larger than Ns2N_s^2; the moist energy M2M^2 then compensates for the difference and ensures continuity of potential energy.

Beyond this earlier demonstration of continuity, it will be shown in the present paper that the energy is also continuously differentiable, which allows us to take derivatives, which are needed for using the energy in a Hamiltonian or Lagrangian formulation.

Lagrangian Formulation and Euler–Lagrange Equations

The goal of this section is to present a Lagrangian formulation of the model in (1)–(4) or (15)–(18). Then, in section 4, the Lagrangian formulation can be used to investigate potential vorticity and particle-relabeling symmetry of the Lagrangian.

Lagrangian density and its regularity at cloud edge

The piecewise quadratic energy in (22) can be used to construct a Lagrangian density ℓ\ell for our system as the difference between kinetic K\mathcal{K} and potential V\mathcal{V} energy terms:

ℓ=K−V=12u⃗⋅u⃗−(12bu2Nu2Hu+12bs2Ns2Hs+12Nu2Ns2Nu2−Ns2M2Hu),(33)\ell=\mathcal{K}-\mathcal{V}=\frac{1}{2}\vec{u}\cdot\vec{u}-\left(\frac{1}{2}\frac{b_u^2}{N_u^2}H_u+\frac{1}{2}\frac{b_s^2}{N_s^2}H_s+\frac{1}{2}\frac{N_u^2N_s^2}{N_u^2-N_s^2}M^2H_u\right), \tag*{(33)}

where the potential energy V\mathcal{V} was discussed above in section 2.2. For instance, while V\mathcal{V} is defined in a piecewise manner in unsaturated and saturated phases, it is nevertheless a continuous function.

In order to proceed with a Lagrangian formulation of the dynamics including phase changes, it is necessary to demonstrate that the density ℓ\ell given by (33) is differentiable at phase boundaries, which are indicated by the Heaviside functions HuH_u, HsH_s. Thus we proceed to compute derivatives in the unsaturated (bu>bsb_u>b_s) and saturated (bu<bsb_u<b_s) phases separately, and then to show equality of their one-sided limits as the phase boundary (bu=bsb_u=b_s) is approached.

Let us fist consider (33) in an unsaturated flow region with Hu=1H_u=1 and Hs=0H_s=0, and denote qq as a placeholder for xx, yy, zz or tt. A derivative with respect to qq of (33) in the unsaturated phase gives

∂ℓ∂q=u⃗⋅∂u⃗∂q−buNu2∂bu∂q−Nu2Ns2Nu2−Ns2M∂M∂q(34)\frac{\partial\ell}{\partial q}=\vec{u}\cdot\frac{\partial\vec{u}}{\partial q}-\frac{b_u}{N_u^2}\frac{\partial b_u}{\partial q}-\frac{N_u^2N_s^2}{N_u^2-N_s^2}M\frac{\partial M}{\partial q} \tag*{(34)}
=u⃗⋅∂u⃗∂q−buNu2∂bu∂q−1Nu2−Ns2(buNs2Nu2∂bu∂q−bu∂bs∂q+bsNu2Ns2∂bs∂q−bs∂bu∂q)(35)=\vec{u}\cdot\frac{\partial\vec{u}}{\partial q}-\frac{b_u}{N_u^2}\frac{\partial b_u}{\partial q}-\frac{1}{N_u^2-N_s^2}\left(b_u\frac{N_s^2}{N_u^2}\frac{\partial b_u}{\partial q}-b_u\frac{\partial b_s}{\partial q}+b_s\frac{N_u^2}{N_s^2}\frac{\partial b_s}{\partial q}-b_s\frac{\partial b_u}{\partial q}\right) \tag*{(35)}

where we have used the definition of MM given by (24). As the phase interface is approached from the unsaturated side, bub_u tends to bsb_s from above, and therefore

lim⁡bu−bs→0+∂ℓ∂q=lim⁡bu−bs→0+(u⃗⋅∂u⃗∂q−[(Nu2−Ns2)+Ns2Nu2(Nu2−Ns2)bu−Ns2bsNu2(Nu2−Ns2)]∂bu∂q−[buNs2−buNu2Ns2(Nu2−Ns2)]∂bs∂q)(36)\lim_{b_u-b_s\to0^+}\frac{\partial\ell}{\partial q}=\lim_{b_u-b_s\to0^+}\left(\vec{u}\cdot\frac{\partial\vec{u}}{\partial q}-\left[\frac{(N_u^2-N_s^2)+N_s^2}{N_u^2(N_u^2-N_s^2)}b_u-\frac{N_s^2b_s}{N_u^2(N_u^2-N_s^2)}\right]\frac{\partial b_u}{\partial q}-\left[\frac{b_uN_s^2-b_uN_u^2}{N_s^2(N_u^2-N_s^2)}\right]\frac{\partial b_s}{\partial q}\right) \tag*{(36)}
=u⃗⋅∂u⃗∂q−bsNs2∂bs∂q.(37)=\vec{u}\cdot\frac{\partial\vec{u}}{\partial q}-\frac{b_s}{N_s^2}\frac{\partial b_s}{\partial q}. \tag*{(37)}

Starting again from (33), but now in a saturated region, the derivative of ℓ\ell with respect to qq is

∂ℓ∂q=u⃗⋅∂u⃗∂q−bsNs2∂bs∂q.(38)\frac{\partial\ell}{\partial q}=\vec{u}\cdot\frac{\partial\vec{u}}{\partial q}-\frac{b_s}{N_s^2}\frac{\partial b_s}{\partial q}. \tag*{(38)}

From (37) and (38), one can see that

lim⁡bu−bs→0+∂ℓ∂q=lim⁡bs−bu→0−∂ℓ∂q=u⃗⋅∂u⃗∂q−bsNs2∂bs∂q,(39)\lim_{b_u-b_s\to0^+}\frac{\partial\ell}{\partial q}=\lim_{b_s-b_u\to0^-}\frac{\partial\ell}{\partial q}=\vec{u}\cdot\frac{\partial\vec{u}}{\partial q}-\frac{b_s}{N_s^2}\frac{\partial b_s}{\partial q}, \tag*{(39)}

which establishes differentiability of the Lagrangian density ℓ\ell at phase boundaries where bu=bsb_u=b_s.

More generally, while the derivatives in (35) and (38) were calculated by restricting attention to unsaturated and saturated regions, respectively, one might further desire a derivative formula that is valid throughout the entire domain. In this direction, and for a generic scenario, consider a piecewise-defined function g(q)g(q) given by

g(q)=gu(q)Hu+gs(q)Hs.(40)g(q)=g_u(q)H_u+g_s(q)H_s. \tag*{(40)}

For example, the Lagrangian density in (33) is of this form. Then its derivative can be calculated as

dgdq=ddq[gu(q)Hu+gs(q)Hs](41)\frac{dg}{dq}=\frac{d}{dq}\left[g_u(q)H_u+g_s(q)H_s\right] \tag*{(41)}
=gu′(q)Hu+gs′(q)Hs+gu(q)dHudq+gs(q)dHsdq(42)=g'_u(q)H_u+g'_s(q)H_s+g_u(q)\frac{dH_u}{dq}+g_s(q)\frac{dH_s}{dq} \tag*{(42)}
=gu′(q)Hu+gs′(q)Hs+gu(q)dHudq+gs(q)dHudq(1−Hu)(43)=g'_u(q)H_u+g'_s(q)H_s+g_u(q)\frac{dH_u}{dq}+g_s(q)\frac{dH_u}{dq}(1-H_u) \tag*{(43)}
=gu′(q)Hu+gs′(q)Hs+[gu(q)−gs(q)]dHudq(44)=g'_u(q)H_u+g'_s(q)H_s+\left[g_u(q)-g_s(q)\right]\frac{dH_u}{dq} \tag*{(44)}
=gu′(q)Hu+gs′(q)Hs,(45)=g'_u(q)H_u+g'_s(q)H_s, \tag*{(45)}

where the last equality follows from an additional assumption that gg is continuous, which implies that gu(q)−gs(q)=0g_u(q)-g_s(q)=0 at the phase interface, so that the coefficient of the Dirac-delta dHu/dqdH_u/dq above is zero at the phase interface. Consequently, in calculating the derivative of a piecewise-defined function as in (40), if the function is also continuous, then one can simply use

dgdq=gu′(q)Hu+gs′(q)Hs,(46)\frac{dg}{dq}=g'_u(q)H_u+g'_s(q)H_s, \tag*{(46)}

by differentiating the coefficients of HuH_u and HsH_s, without needing to consider derivatives of HuH_u and HsH_s themselves. We will use this result below in § 3.2.

Hamilton’s Principle and the Euler–Lagrange Equations

The ℓ\ell defined in (33) is a candidate for a Lagrangian density, which can potentially be used to define an associated set of Euler–Lagrange equations, via a variational principle—i.e., the principle of least action, or Hamilton’s principle. A priori, it is not immediately clear whether this will work here, due to complications of phase changes. In § 3.1 above, we established that, at the very least, ℓ\ell is differentiable, so it is at least possible to take variational derivatives of ℓ\ell. Now we investigate whether a variational principle, using ℓ\ell as the Lagrangian density, will lead to the model in (1)–(4) or (15)–(18) as the associated Euler–Lagrange equations.

Hamilton’s principle states that the dynamics of a system are determined according to a variational statement, and for fluid systems, the variational statement may be expressed in either the Lagrangian1 description or the Eulerian description (Hill, 1951; Bretherton, 1970; Salmon, 1982, 1988, 1998; Cotter and Holm, 2014). Our goal is to derive the model (15)–(18), which is written in the Eulerian description. Nevertheless, for clarity and for further use below, we shall start now from the Lagrangian description and then briefly recall the steps in the transformation from Lagrangian to Eulerian description. In the Lagrangian description, one considers variations δx⃗\delta\vec{x} in the particle trajectories x⃗=x⃗(a⃗,t)\vec{x}=\vec{x}(\vec{a},t) corresponding to particles labeled by a⃗\vec{a}. In the Eulerian description, the system is described by the velocity u⃗\vec{u} at fixed locations, and thus one must use the inverse flow map a⃗=a⃗(x⃗,t)\vec{a}=\vec{a}(\vec{x},t) (assumed one-to-one) to transform the variational statement accordingly. As in the previous sections, it will be convenient to work in Cartesian coordinates for simplicity.

Following fluid particles labeled by a⃗\vec{a}, the Lagrangian L\mathcal{L} for the system is given by

L(t)=∭dVa ℓ(x⃗(a⃗,t),x⃗t(a⃗,t),a⃗)=∭dVa 12(x⃗t⋅x⃗t−bu2Nu2Hu−bs2Ns2Hs−Nu2Ns2Nu2−Ns2M2Hu)(47)\mathcal{L}(t)=\iiint dV_a\,\ell\left(\vec{x}(\vec{a},t),\vec{x}_t(\vec{a},t),\vec{a}\right)=\iiint dV_a\,\frac{1}{2}\left(\vec{x}_t\cdot\vec{x}_t-\frac{b_u^2}{N_u^2}H_u-\frac{b_s^2}{N_s^2}H_s-\frac{N_u^2N_s^2}{N_u^2-N_s^2}M^2H_u\right) \tag*{(47)}

where a⃗\vec{a} is chosen such that dVadV_a represents a differential mass element, dmdm, in particle-label space. For example, a⃗\vec{a} can be chosen as the initial position x⃗0\vec{x}_0. We make use of a different such choice of a⃗\vec{a} in later sections. The notation x⃗t\vec{x}_t means the derivative of x⃗\vec{x} with respect to time tt, keeping particle label a⃗\vec{a} fixed. Hamilton’s principle is stated in terms of the action A\mathcal{A}:

δA=0,A=∫t1t2dt L(t),(48)\delta\mathcal{A}=0,\qquad\mathcal{A}=\int_{t_1}^{t_2}dt\,\mathcal{L}(t), \tag*{(48)}

for variations δx⃗(a⃗,t)\delta\vec{x}(\vec{a},t) that vanish outside the range t∈[t1,t2]t\in[t_1,t_2]. These variations must also tend to zero as x⃗\vec{x} tends to infinity, or satisfy no-flow through the domain boundary (see for example Bretherton, 1970). To transform (48) to the Eulerian description, one must change variables in the volume element using

dVa=JdVx=∂(a⃗)∂(x⃗)dVx,(49)dV_a=\mathcal{J}dV_x=\frac{\partial(\vec{a})}{\partial(\vec{x})}dV_x, \tag*{(49)}

where dVxdV_x is the volume element in physical space, J\mathcal{J} is the Jacobian determinant, and we adopt the notation J≡∂(a⃗)∂(x⃗)\mathcal{J}\equiv\frac{\partial(\vec{a})}{\partial(\vec{x})} as in some of the previous literature. In Cartesian coordinates, J\mathcal{J} is the determinant of the matrix ∂ai/∂xj\partial a_i/\partial x_j. For labeling coordinates chosen such that equal volumes in a⃗\vec{a}-space have equal masses, ∂(a⃗)∂(x⃗)\frac{\partial(\vec{a})}{\partial(\vec{x})} is the mass density ρ\rho. Here we wish to describe incompressible Boussinesq dynamics, and thus we take the constant ρ=1\rho=1. Using the Eulerian notation u⃗=x⃗t\vec{u}=\vec{x}_t, Hamilton’s principle (48) becomes

δA=0,A=∫t1t2dt∭dVx12(u⃗⋅u⃗−bu2Nu2Hu−bs2Ns2Hs−Nu2Ns2Nu2−Ns2M2Hu),(50)\delta\mathcal{A}=0,\qquad\mathcal{A}=\int_{t_1}^{t_2}dt\iiint dV_x\frac{1}{2}\left(\vec{u}\cdot\vec{u}-\frac{b_u^2}{N_u^2}H_u-\frac{b_s^2}{N_s^2}H_s-\frac{N_u^2N_s^2}{N_u^2-N_s^2}M^2H_u\right), \tag*{(50)}

where now the action A\mathcal{A} must be stationary with respect to arbitrary variations δa⃗\delta\vec{a}. Thus one must express the integrand as ℓ=ℓ(a⃗(x⃗,t),…,x⃗)\ell=\ell(\vec{a}(\vec{x},t),\ldots,\vec{x}), where the ... means derivatives of a⃗\vec{a} with respect to x⃗\vec{x} and/or tt.

For example, let us consider how to express the velocity u⃗\vec{u} in (50) as first derivatives of a⃗\vec{a} with respect to x⃗\vec{x} and tt. The latter can be achieved from conservation of label a⃗\vec{a} following fluid particles:

DDta⃗=a⃗t+u⃗⋅∇a⃗=0,(51)\frac{D}{Dt}\vec{a}=\vec{a}_t+\vec{u}\cdot\nabla\vec{a}=0, \tag*{(51)}

by solving (51) as a 3×33\times3 system for u⃗\vec{u} (Salmon, 1988), and we note that the ∇\nabla without a subscript is the gradient operator ∇=(∂x,∂y,∂z)\nabla=(\partial_x,\partial_y,\partial_z). Alternatively, an equivalent variational statement is found by appending (51) to (50) as constraints (the Lin constraints), and then considering independent variations δu⃗\delta\vec{u} and δa⃗\delta\vec{a} [6] (Salmon 1988). In addition to the Lin constraints, a constraint is required to enforce the incompressibility condition ∇⋅u⃗=0\nabla\cdot\vec{u}=0, thus leading to

A=∫dt∭dVx(12u⃗⋅u⃗−12bu2Nu2Hu−12bs2Ns2Hs−12Nu2Ns2Nu2−Ns2M2Hu−α⃗⋅Da⃗Dt−λ(∇⋅u⃗)),(52)\mathcal{A}=\int dt\iiint dV_x\left(\frac{1}{2}\vec{u}\cdot\vec{u}-\frac{1}{2}\frac{b_u^2}{N_u^2}H_u-\frac{1}{2}\frac{b_s^2}{N_s^2}H_s-\frac{1}{2}\frac{N_u^2N_s^2}{N_u^2-N_s^2}M^2H_u-\vec{\alpha}\cdot\frac{D\vec{a}}{Dt}-\lambda(\nabla\cdot\vec{u})\right), \tag*{(52)}

where λ\lambda, α⃗=(α,β,γ)\vec{\alpha}=(\alpha,\beta,\gamma) are Lagrange multipliers. As will be shown below, the multiplier λ\lambda contributes to the pressure. Next, the potential energy in the integrand can be re-written in terms of the material invariants butot=butot(a⃗)b_u^{tot}=b_u^{tot}(\vec{a}) and bstot=bstot(a⃗)b_s^{tot}=b_s^{tot}(\vec{a}) using the relations

bu=butot−Nu2z,bs=bstot−Ns2z.(53)b_u=b_u^{tot}-N_u^2z,\qquad b_s=b_s^{tot}-N_s^2z. \tag*{(53)}

One can verify that butot,bstotb_u^{tot},b_s^{tot} are invariant following fluid particles according to the conservation equations

DDtbutot=DDtbstot=0,DDt=∂∂t+u⃗⋅∇.(54)\frac{D}{Dt}b_u^{tot}=\frac{D}{Dt}b_s^{tot}=0,\qquad\frac{D}{Dt}=\frac{\partial}{\partial t}+\vec{u}\cdot\nabla. \tag*{(54)}

Using (53), one can also re-write HuH_u, HsH_s, and MM in the potential energy to be functions of butotb_u^{tot}, bstotb_s^{tot}, based on their definitions in (21) and (24). Notice that, via (53), the height zz now appears explicitly in the potential energy terms, along with butotb_u^{tot}, bstotb_s^{tot}. However, the explicit zz-dependence does not have an influence on variations δa⃗\delta\vec{a} for the Lagrangian density ℓ=ℓ(a⃗(x⃗,t),…,x⃗)\ell=\ell(\vec{a}(\vec{x},t),\ldots,\vec{x}), since the explicit zz-dependence is part of the explicit x⃗\vec{x}-dependence in ℓ=ℓ(a⃗(x⃗,t),…,x⃗)\ell=\ell(\vec{a}(\vec{x},t),\ldots,\vec{x}) and is separate from the implicit x⃗\vec{x}-dependence of the a⃗(x⃗,t)\vec{a}(\vec{x},t)-dependence of ℓ\ell. As some physical connection, note that the potential energy is now considered as a function of the three variables butotb_u^{tot}, bstotb_s^{tot}, zz, and the presence of these three variables is analogous to the presence of three thermodynamic variables θetot\theta_e^{tot}, qttotq_t^{tot}, zz or stots^{tot}, qttotq_t^{tot}, pp that are needed to describe the thermodynamic state of a moist system. The quantities butotb_u^{tot} and bstotb_s^{tot} are material invariants that play the same type of role as the material invariants of entropy stots^{tot} and total water mixing ratio qttotq_t^{tot}, and zz plays the role of pressure pp in Boussinesq and anelastic systems (Pauluis, 2008). We continue to write the potential energy in terms of bub_u, bsb_s, MM for simplicity of the expression; however, the transformation to the variables butotb_u^{tot}, bstotb_s^{tot}, zz is henceforth implied. Finally, we can exploit the material invariance of the buoyancy variables butotb_u^{tot}, bstotb_s^{tot} by assigning them as two of the particle labels: a⃗=(a,b,c)=(a,butot,bstot)\vec{a}=(a,b,c)=(a,b_u^{tot},b_s^{tot}), thereby allowing for independent variations δbutot\delta b_u^{tot} and δbstot\delta b_s^{tot}. Using the new labels, and integration by parts in the incompressibility constraint, we arrive at

A=∫dt∭dVx(12u⃗⋅u⃗−12bu2Nu2Hu−12bs2Ns2Hs−12Nu2Ns2Nu2−Ns2M2Hu−αDaDt−βDbutotDt−γDbstotDt−u⃗⋅∇λ).(55)\mathcal{A}=\int dt\iiint dV_x\left(\frac{1}{2}\vec{u}\cdot\vec{u}-\frac{1}{2}\frac{b_u^2}{N_u^2}H_u-\frac{1}{2}\frac{b_s^2}{N_s^2}H_s-\frac{1}{2}\frac{N_u^2N_s^2}{N_u^2-N_s^2}M^2H_u-\alpha\frac{Da}{Dt}-\beta\frac{Db_u^{tot}}{Dt}-\gamma\frac{Db_s^{tot}}{Dt}-\vec{u}\cdot\nabla\lambda\right). \tag*{(55)}

After all of the preceding transformations, Hamilton’s principle requires that the A\mathcal{A} be stationary with respect to variations in u⃗(x⃗,t)\vec{u}(\vec{x},t), a⃗(x⃗,t)\vec{a}(\vec{x},t), butot(x⃗,t)b_u^{tot}(\vec{x},t), bstot(x⃗,t)b_s^{tot}(\vec{x},t), α(x⃗,t)\alpha(\vec{x},t), β(x⃗,t)\beta(\vec{x},t), γ(x⃗,t)\gamma(\vec{x},t) and λ(x⃗,t)\lambda(\vec{x},t). These variations result in the following:

δu⃗: u⃗−α∇a−β∇butot−γ∇bstot−∇λ=0,(56)\delta\vec{u}:\ \vec{u}-\alpha\nabla a-\beta\nabla b_u^{tot}-\gamma\nabla b_s^{tot}-\nabla\lambda=0, \tag*{(56)}
δα: DaDt=0,(57)\delta\alpha:\ \frac{Da}{Dt}=0, \tag*{(57)}
δa: DαDt=0,(58)\delta a:\ \frac{D\alpha}{Dt}=0, \tag*{(58)}
δβ: DbutotDt=0,(59)\delta\beta:\ \frac{Db_u^{tot}}{Dt}=0, \tag*{(59)}
δbutot: DβDt−buNu2Hu−Ns2Nu2−Ns2MHu=0,(60)\delta b_u^{tot}:\ \frac{D\beta}{Dt}-\frac{b_u}{N_u^2}H_u-\frac{N_s^2}{N_u^2-N_s^2}MH_u=0, \tag*{(60)}
δγ: DbstotDt=0,(61)\delta\gamma:\ \frac{Db_s^{tot}}{Dt}=0, \tag*{(61)}
δbstot: DγDt−bsNs2Hs+Nu2Nu2−Ns2MHu=0,(62)\delta b_s^{tot}:\ \frac{D\gamma}{Dt}-\frac{b_s}{N_s^2}H_s+\frac{N_u^2}{N_u^2-N_s^2}MH_u=0, \tag*{(62)}
δλ:∇⋅u⃗=0,(63)\delta\lambda:\nabla\cdot\vec{u}=0, \tag*{(63)}

where we have used (46). To obtain the momentum equation, one can compute Du⃗/DtD\vec{u}/Dt using (56), and then substitute from the remaining equations in (57)–(63) to obtain

Du⃗Dt=DDt(α∇a+β∇butot+γ∇bstot+∇λ)=∂∂t(α∇a+β∇butot+γ∇bstot+∇λ)+u⃗⋅∇u⃗(64)\frac{D\vec{u}}{Dt}=\frac{D}{Dt}\left(\alpha\nabla a+\beta\nabla b_u^{\mathrm{tot}}+\gamma\nabla b_s^{\mathrm{tot}}+\nabla\lambda\right)=\frac{\partial}{\partial t}\left(\alpha\nabla a+\beta\nabla b_u^{\mathrm{tot}}+\gamma\nabla b_s^{\mathrm{tot}}+\nabla\lambda\right)+\vec{u}\cdot\nabla\vec{u} \tag*{(64)}
=α∂∇a∂t+β∂∇butot∂t+γ∂∇bstot∂t+∂∇λ∂t+∇a∂α∂t+∇butot∂β∂t+∇bstot∂γ∂t+u⃗⋅∇u⃗(65)=\alpha\frac{\partial\nabla a}{\partial t}+\beta\frac{\partial\nabla b_u^{\mathrm{tot}}}{\partial t}+\gamma\frac{\partial\nabla b_s^{\mathrm{tot}}}{\partial t}+\frac{\partial\nabla\lambda}{\partial t}+\nabla a\frac{\partial\alpha}{\partial t}+\nabla b_u^{\mathrm{tot}}\frac{\partial\beta}{\partial t}+\nabla b_s^{\mathrm{tot}}\frac{\partial\gamma}{\partial t}+\vec{u}\cdot\nabla\vec{u} \tag*{(65)}
=−(α∇(u⃗⋅∇a)+β∇(u⃗⋅∇butot)+γ∇(u⃗⋅∇bstot)+∇(u⃗⋅∇λ))+∇(u⃗⋅∇λ+∂λ∂t)+HubuNu2∇butot+HuNs2Nu2−Ns2M∇butot+HsbsNs2∇bstot−HuNu2Nu2−Ns2M∇bstot+u⃗⋅∇u⃗−(∇a(u⃗⋅∇α)+∇butot(u⃗⋅∇β)+∇bstot(u⃗⋅∇γ))(66)=-\left(\alpha\nabla(\vec{u}\cdot\nabla a)+\beta\nabla(\vec{u}\cdot\nabla b_u^{\mathrm{tot}})+\gamma\nabla(\vec{u}\cdot\nabla b_s^{\mathrm{tot}})+\nabla(\vec{u}\cdot\nabla\lambda)\right) +\nabla\left(\vec{u}\cdot\nabla\lambda+\frac{\partial\lambda}{\partial t}\right)+H_u\frac{b_u}{N_u^2}\nabla b_u^{\mathrm{tot}}+H_u\frac{N_s^2}{N_u^2-N_s^2}M\nabla b_u^{\mathrm{tot}}+H_s\frac{b_s}{N_s^2}\nabla b_s^{\mathrm{tot}}-H_u\frac{N_u^2}{N_u^2-N_s^2}M\nabla b_s^{\mathrm{tot}} +\vec{u}\cdot\nabla\vec{u}-\left(\nabla a(\vec{u}\cdot\nabla\alpha)+\nabla b_u^{\mathrm{tot}}(\vec{u}\cdot\nabla\beta)+\nabla b_s^{\mathrm{tot}}(\vec{u}\cdot\nabla\gamma)\right) \tag*{(66)}
=−(α∇(u⃗⋅∇a)+β∇(u⃗⋅∇butot)+γ∇(u⃗⋅∇bstot)+∇(u⃗⋅∇λ))−∇(−u⃗⋅∇λ−∂λ∂t−12(bu)2Nu2Hu−12(bs)2Ns2Hs−12Nu2Ns2Nu2−Ns2M2Hu)+(buHu+bsHs)k^+u⃗⋅∇u⃗−(∇a(u⃗⋅∇α)+∇butot(u⃗⋅∇β)+∇bstot(u⃗⋅∇γ)).(67)=-\left(\alpha\nabla(\vec{u}\cdot\nabla a)+\beta\nabla(\vec{u}\cdot\nabla b_u^{\mathrm{tot}})+\gamma\nabla(\vec{u}\cdot\nabla b_s^{\mathrm{tot}})+\nabla(\vec{u}\cdot\nabla\lambda)\right) -\nabla\left(-\vec{u}\cdot\nabla\lambda-\frac{\partial\lambda}{\partial t}-\frac{1}{2}\frac{(b_u)^2}{N_u^2}H_u-\frac{1}{2}\frac{(b_s)^2}{N_s^2}H_s-\frac{1}{2}\frac{N_u^2N_s^2}{N_u^2-N_s^2}M^2H_u\right)+(b_uH_u+b_sH_s)\hat{k}+\vec{u}\cdot\nabla\vec{u} -\left(\nabla a(\vec{u}\cdot\nabla\alpha)+\nabla b_u^{\mathrm{tot}}(\vec{u}\cdot\nabla\beta)+\nabla b_s^{\mathrm{tot}}(\vec{u}\cdot\nabla\gamma)\right). \tag*{(67)}

In moving from (65) to (66), the term ∇(u⃗⋅∇λ)\nabla(\vec{u}\cdot\nabla\lambda) was added and subtracted, and several of the relations from (57)–(63) were used to replace ∂/∂t\partial/\partial t terms. Then, in moving from (66) to (67), the term ∇butot\nabla b_u^{\mathrm{tot}} was split into contributions from bub_u and Nu2zN_u^2z, using the definition in (53), and similarly for ∇bstot\nabla b_s^{\mathrm{tot}}, leading to the buoyancy term (buHu+bsHs)k^(b_uH_u+b_sH_s)\hat{k} in (67). Also used in moving from (66) to (67) are (24), (39), and (46). To proceed further, the main challenge is to identify and collect gradient terms so as to determine an effective pressure to recover the requisite governing equations. With that goal, we use the identity

α∇(u⃗⋅∇a)+∇a(u⃗⋅∇α)=∇(u⃗⋅(α∇a))+u⃗⋅∇(α∇a)−[∇(α∇a)]Tu⃗(68)\alpha\nabla(\vec{u}\cdot\nabla a)+\nabla a(\vec{u}\cdot\nabla\alpha)=\nabla(\vec{u}\cdot(\alpha\nabla a))+\vec{u}\cdot\nabla(\alpha\nabla a)-[\nabla(\alpha\nabla a)]^{T}\vec{u} \tag*{(68)}

where [∇(α∇a)]Tu⃗[\nabla(\alpha\nabla a)]^{T}\vec{u} is a matrix–vector multiplication that can be written in index notation as uj∂xi(α∂xja)u_j\partial_{x_i}(\alpha\partial_{x_j}a), with summation over repeated indices. Using this identity in (67), we obtain

Du⃗Dt=−∇p+(buHu+bsHs)k^,(69)\frac{D\vec{u}}{Dt}=-\nabla p+(b_uH_u+b_sH_s)\hat{k}, \tag*{(69)}

where the pressure pp is related to the Lagrange multiplier λ\lambda via

p=(12∣u⃗∣2−u⃗⋅∇λ−∂λ∂t−12(bu)2Nu2Hu−12(bs)2Ns2Hs−12Nu2Ns2Nu2−Ns2M2Hu)(70)p=\left(\frac{1}{2}|\vec{u}|^2-\vec{u}\cdot\nabla\lambda-\frac{\partial\lambda}{\partial t}-\frac{1}{2}\frac{(b_u)^2}{N_u^2}H_u-\frac{1}{2}\frac{(b_s)^2}{N_s^2}H_s-\frac{1}{2}\frac{N_u^2N_s^2}{N_u^2-N_s^2}M^2H_u\right) \tag*{(70)}
=(ℓ−u⃗⋅∇λ−∂λ∂t).(71)=\left(\ell-\vec{u}\cdot\nabla\lambda-\frac{\partial\lambda}{\partial t}\right). \tag*{(71)}

Therefore we have demonstrated that the governing equations (59), (61), (63), (69) are obtained using Hamilton’s principle, starting from the Lagrangian density (33) [compare to (15)–(18) and use (53)].

Circulation and Potential Vorticity in Moist Systems

Using the Lagrangian formulation from section 3, we can now investigate potential vorticity from the perspective of particle-relabeling symmetry of the Lagrangian. In this section, a moist version of particle-relabeling symmetry is studied, and it is used to identify various versions of potential vorticity conservation.

Background on Noether's Theorem and Possible Outcomes

As a systematic approach to seeking conservation laws, one could identify symmetries of the Lagrangian function and apply Noether’s theorem to arrive at a corresponding conservation law [36]. For conservation of potential vorticity and Kelvin’s circulation theorem, the symmetry of interest for dry dynamics is the particle-relabeling symmetry [e.g. 38, 6] (Salmon 1988). Here, we now seek a particle-relabeling symmetry that is valid for moist dynamics with phase changes. By taking such a systematic approach, we can discover statements for conservation of circulation and moist potential vorticity in a system with phase changes.

Before considering particle-relabeling symmetry and the systematic approach itself, it is worthwhile to consider what the possible outcomes may be. As examples, consider other known symmetries and their associated conservation laws. One well-known symmetry is translation-invariance, whereby the action is invariant under translations of the x⃗\vec{x} coordinates. The corresponding conservation law is conservation of momentum. A second symmetry is the particle-relabeling symmetry, which, for a dry atmosphere, is known to correspond with conservation of PV. As examples of conservation laws, these two examples differ in that it is the total (or global) momentum, integrated over all fluid parcels, that is conserved, whereas the PV is conserved for each individual fluid parcel. A priori, for the moist case of the present paper, either scenario is possible. In other words, it is possible that moist PV will be conserved for each individual fluid parcel, or that moist PV is globally conserved (upon integrating over all fluid parcels), or some other type of conservation law. By finding the moist particle-relabeling symmetry and applying Noether’s theorem, we will discover the corresponding conservation law that arises.

Particle-Relabeling Symmetry and PV

In this section, our goal is to find a particle-relabeling symmetry for the action A\mathcal{A} defined in (47)–(48), whose Euler-Lagrange equations are the moist system (15)–(18) with phase changes. The action for our system may be written as

A=∫dt∭dVa ℓ(x⃗(a⃗,t),x⃗t(a⃗,t),a⃗)=∫dt∭dVa(12x⃗t⋅x⃗t−V(butot(a⃗),bstot(a⃗),z))(72)\mathcal{A} = \int dt \iiint dV_a \, \ell(\vec{x}(\vec{a},t),\vec{x}_t(\vec{a},t),\vec{a}) = \int dt \iiint dV_a \left(\frac{1}{2}\vec{x}_t \cdot\vec{x}_t - \mathcal{V}(b_u^{tot}(\vec{a}),b_s^{tot}(\vec{a}),z)\right) \tag*{(72)}

where x⃗=x⃗(a⃗,t)\vec{x}=\vec{x}(\vec{a},t) is the particle path, x⃗t=x⃗t(a⃗,t)\vec{x}_t=\vec{x}_t(\vec{a},t) is the fluid velocity following fluid particles, and butot,bstotb_u^{tot},b_s^{tot} are the total buoyancies specified by (53)–(54). Henceforth it is understood that the limits of integration and boundary conditions are appropriately defined, as mentioned in § 3.2 and discussed by Bretherton [6].

One may consider a particle relabeling a⃗′=a⃗+δa⃗(a⃗,t)\vec{a}'=\vec{a}+\delta\vec{a}(\vec{a},t) that does not change the particle paths, such that x⃗′(a⃗′,t)=x⃗(a⃗,t)\vec{x}'(\vec{a}',t)=\vec{x}(\vec{a},t). Under the relabeling, (72) must remain invariant, and in fact, we will further restrict the relabeling to those that do not alter the potential energy density V\mathcal{V} itself.

To achieve invariance of potential energy, based on its form V(butot,bstot,z)\mathcal{V}(b_u^{tot},b_s^{tot},z), we see that we should consider variations δa⃗\delta\vec{a} along curves of constant butotb_u^{tot} and constant bstotb_s^{tot}, where a schematic of such curves is shown in figure 1. These are curves at the intersection of two surfaces: one with constant butotb_u^{tot} and the other with constant bstotb_s^{tot}. Since surfaces of constant butotb_u^{tot} and surfaces of constant bstotb_s^{tot} are both material invariant, it follows that their intersection is also material invariant. To simplify the variations of this form, it is convenient to consider the following label coordinate system:

a⃗=(a,b,c)=(a,butot,bstot).(73)\vec{a}=(a,b,c)=(a,b_u^{tot},b_s^{tot}). \tag*{(73)}

One can explicitly construct such curves with a⃗=a⃗(x⃗0)\vec{a}=\vec{a}(\vec{x}_0) depending on the initial particle label x⃗0\vec{x}_0 associated with time t=0t=0 (see Appendix § 10). For a given particle, the first coordinate aa is defined to be an integral along a curve of constant (butot,bstot)(b_u^{\mathrm{tot}},b_s^{\mathrm{tot}}), starting from a reference point x⃗0 r\vec{x}_0^{\,r} on the curve, and ending at the initial location x⃗0\vec{x}_0 of the particle at t=0t=0. The label a=a(x⃗0)a=a(\vec{x}_0) is chosen such that dVa=da db dc=da dbutot dbstotdV_a=da\,db\,dc=da\,db_u^{\mathrm{tot}}\,db_s^{\mathrm{tot}} represents a mass element ρdVx\rho dV_x, and furthermore satisfies the condition for material invariance, Da/Dt=0D_a/Dt=0. Also note that bb had been used to represent buoyancy in some early parts of the paper, and bb is now being used in the label coordinate a⃗=(a,b,c)\vec{a}=(a,b,c), and the meaning of bb should be clear from context in what follows.

Next, we must enforce the condition that the relabeling should not change the mass, such that

da⃗=∂(a⃗)∂(a⃗ ′)da⃗ ′=da⃗ ′,(74)d\vec{a}=\frac{\partial(\vec{a})}{\partial(\vec{a}^{\,\prime})}d\vec{a}^{\,\prime}=d\vec{a}^{\,\prime}, \tag*{(74)}

where

∂(a⃗)∂(a⃗ ′)=∂(a⃗ ′−δa⃗)∂(a⃗ ′)=1−(∂δa∂a+∂δb∂b+∂δc∂c)+O(∣δa⃗∣2)for ∣δa⃗∣→0.(75)\frac{\partial(\vec{a})}{\partial(\vec{a}^{\,\prime})}=\frac{\partial(\vec{a}^{\,\prime}-\delta\vec{a})}{\partial(\vec{a}^{\,\prime})}=1-\left(\frac{\partial\delta a}{\partial a}+\frac{\partial\delta b}{\partial b}+\frac{\partial\delta c}{\partial c}\right)+O(|\delta\vec{a}|^2)\quad\text{for }|\delta\vec{a}|\to0. \tag*{(75)}

Neglecting the O(∣δa⃗∣2)O(|\delta\vec{a}|^2) truncation error leads to the condition

∂δa∂a+∂δb∂b+∂δc∂c=0.(76)\frac{\partial\delta a}{\partial a}+\frac{\partial\delta b}{\partial b}+\frac{\partial\delta c}{\partial c}=0. \tag*{(76)}

By requiring (76), together with δa⃗\delta\vec{a} occurring along curves of constant (butot,bstot)(b_u^{\mathrm{tot}},b_s^{\mathrm{tot}}), leads to δa⃗\delta\vec{a} of the form

δa⃗=(δa(b,c,t),0,0)=(δa(butot,bstot,t),0,0),(77)\delta\vec{a}=(\delta a(b,c,t),0,0)=(\delta a(b_u^{\mathrm{tot}},b_s^{\mathrm{tot}},t),0,0), \tag*{(77)}

where δa(butot,bstot,t)\delta a(b_u^{\mathrm{tot}},b_s^{\mathrm{tot}},t) is an arbitrary scalar function with suitably small amplitude, and that vanishes at the boundaries of the time and space domain. This form of the variation ensures a mass-conserving particle relabeling, up to an asymptotically small error in the limit ∣δa⃗∣→0|\delta\vec{a}|\to0. In brief, the relabeling shift δa\delta a is a change in the label coordinate aa of the location on the curve, and it is a uniform shift for each curve because it is of the form δa(butot,bstot,t)\delta a(b_u^{\mathrm{tot}},b_s^{\mathrm{tot}},t) and does not depend on the label aa itself.

By adopting variations of the form (77), we now proceed to enforce δA=0\delta\mathcal{A}=0, where δA\delta\mathcal{A} arises from variations δx⃗t\delta\vec{x}_t in the kinetic energy,

δA=∫dt∭dVa ∂x⃗t∂t⋅∂δx⃗∂t=∫dt∭dVa x⃗t⋅δx⃗t=0,(78)\delta\mathcal{A}=\int dt\iiint dV_a\,\frac{\partial\vec{x}_t}{\partial t}\cdot\frac{\partial\delta\vec{x}}{\partial t}=\int dt\iiint dV_a\,\vec{x}_t\cdot\delta\vec{x}_t=0, \tag*{(78)}

with the eventual aim of relating the result to conservation of potential vorticity. We remind the reader that x⃗t\vec{x}_t means the derivative of x⃗(a⃗,t)\vec{x}(\vec{a},t) with respect to time tt, keeping a⃗\vec{a} fixed. We will define δx⃗t\delta\vec{x}_t below, and for clarity throughout the rest of this section, we will be explicit about what is held fixed when time derivatives are taken. The steps in the rather lengthy calculation are (i) express δx⃗t\delta\vec{x}_t in terms of (δa⃗)t(\delta\vec{a})_t; (ii) substitute expression (77) for δa⃗\delta\vec{a}; (iii) rearrange the ordering of terms in the integral; (iv) simplify the integral and interpret the result in physical terms. As above in (75), it will be necessary to consider variations with ∣δa⃗∣→0|\delta\vec{a}|\to0, and the final result is valid up to an asymptotically small error of O(∣δa⃗∣2)O(|\delta\vec{a}|^2).

An expression for δx⃗t\delta\vec{x}_t in terms of (δa⃗)t(\delta\vec{a})_t follows from calculus, followed by taking the limit ∣δa⃗∣→0|\delta\vec{a}|\to0. Starting from the definition of δx⃗t\delta\vec{x}_t, one may apply the chain rule:

δ∂x⃗∂t=∂x⃗′(a⃗′,t)∂t∣a⃗′−∂x⃗(a⃗,t)∂t∣a⃗(79)\delta\frac{\partial\vec{x}}{\partial t} = \left.\frac{\partial\vec{x}'(\vec{a}',t)}{\partial t}\right|_{\vec{a}'} - \left.\frac{\partial\vec{x}(\vec{a},t)}{\partial t}\right|_{\vec{a}} \tag*{(79)}
=∂x⃗′(a⃗′,t)∂t∣a⃗′−∂x⃗(a⃗,t)∂t∣a⃗(80)= \left.\frac{\partial\vec{x}'(\vec{a}',t)}{\partial t}\right|_{\vec{a}'} - \left.\frac{\partial\vec{x}(\vec{a},t)}{\partial t}\right|_{\vec{a}} \tag*{(80)}
=∂x⃗(a⃗,t)∂t∣a⃗+∂x⃗∂a∂a∂t∣a⃗′+∂x⃗∂b∂b∂t∣a⃗′+∂x⃗∂c∂c∂t∣a⃗′−∂x⃗(a⃗,t)∂t∣a⃗(81)= \left.\frac{\partial\vec{x}(\vec{a},t)}{\partial t}\right|_{\vec{a}} + \frac{\partial\vec{x}}{\partial a}\frac{\partial a}{\partial t}\bigg|_{\vec{a}'} + \frac{\partial\vec{x}}{\partial b}\frac{\partial b}{\partial t}\bigg|_{\vec{a}'} + \frac{\partial\vec{x}}{\partial c}\frac{\partial c}{\partial t}\bigg|_{\vec{a}'} - \left.\frac{\partial\vec{x}(\vec{a},t)}{\partial t}\right|_{\vec{a}} \tag*{(81)}
=∂x⃗∂a∂a∂t∣a⃗′+∂x⃗∂b∂b∂t∣a⃗′+∂x⃗∂c∂c∂t∣a⃗′,(82)= \frac{\partial\vec{x}}{\partial a}\frac{\partial a}{\partial t}\bigg|_{\vec{a}'} + \frac{\partial\vec{x}}{\partial b}\frac{\partial b}{\partial t}\bigg|_{\vec{a}'} + \frac{\partial\vec{x}}{\partial c}\frac{\partial c}{\partial t}\bigg|_{\vec{a}'}, \tag*{(82)}

where ∇a⃗x⃗\nabla_{\vec{a}}\vec{x} ∂/∂t∣a⃗′\partial/\partial t|_{\vec{a}'} is a matrix-vector multiplication, and where we have used x⃗′(a⃗′,t)=x⃗(a⃗,t)\vec{x}'(\vec{a}',t) = \vec{x}(\vec{a},t) and a⃗′=a⃗+δa⃗(a⃗,t)\vec{a}' = \vec{a} + \delta\vec{a}(\vec{a},t). Furthermore a⃗′=a⃗+δa⃗(a⃗,t)\vec{a}' = \vec{a} + \delta\vec{a}(\vec{a},t) implies that

∂a⃗′∂t∣a⃗′=0=∂a⃗∂t∣a⃗′+∂δa⃗∂t∣a⃗′,(83)\left.\frac{\partial\vec{a}'}{\partial t}\right|_{\vec{a}'} = 0 = \left.\frac{\partial\vec{a}}{\partial t}\right|_{\vec{a}'} + \left.\frac{\partial\delta\vec{a}}{\partial t}\right|_{\vec{a}'}, \tag*{(83)}

One may then re-arrange (83) to find

∂a⃗∂t∣a⃗′=−∂δa⃗∂t∣a⃗′,∼−∂δa⃗∂t∣a⃗+O(∣δa⃗∣2).(84)\left.\frac{\partial\vec{a}}{\partial t}\right|_{\vec{a}'} = -\left.\frac{\partial\delta\vec{a}}{\partial t}\right|_{\vec{a}'}, \sim-\left.\frac{\partial\delta\vec{a}}{\partial t}\right|_{\vec{a}} + O(|\delta\vec{a}|^2). \tag*{(84)}

Using (84) in (82) leads to

δ∂x⃗∂t∼−(∂x⃗∂a∂δa∂t∣a⃗+∂x⃗∂b∂δb∂t∣a⃗+∂x⃗∂c∂δc∂t∣a⃗)+O(∣δa⃗∣2),(85)\delta\frac{\partial\vec{x}}{\partial t} \sim-\left(\frac{\partial\vec{x}}{\partial a}\frac{\partial\delta a}{\partial t}\bigg|_{\vec{a}} + \frac{\partial\vec{x}}{\partial b}\frac{\partial\delta b}{\partial t}\bigg|_{\vec{a}} + \frac{\partial\vec{x}}{\partial c}\frac{\partial\delta c}{\partial t}\bigg|_{\vec{a}}\right) + O(|\delta\vec{a}|^2), \tag*{(85)}

and for the special variations defined in (77) we have,

δ∂x⃗∂t∼−∂x⃗∂a∂δa∂t∣a⃗+O(∣δa⃗∣2).(86)\delta\frac{\partial\vec{x}}{\partial t} \sim-\frac{\partial\vec{x}}{\partial a}\frac{\partial\delta a}{\partial t}\bigg|_{\vec{a}} + O(|\delta\vec{a}|^2). \tag*{(86)}

Thus, asymptotically as ∣δa⃗∣→0|\delta\vec{a}| \to0 and using integration by parts in tt, the variation δA\delta\mathcal{A} from the kinetic energy becomes

∫dt∭dVa ∂x⃗∂t⋅δ∂x⃗∂t∼−∫dt∭dVa ∂x⃗∂t⋅(∂x⃗∂a∂δa∂t∣a⃗)(87)\int dt\iiint dV_a\,\frac{\partial\vec{x}}{\partial t}\cdot\delta\frac{\partial\vec{x}}{\partial t} \sim-\int dt\iiint dV_a\,\frac{\partial\vec{x}}{\partial t}\cdot\left(\frac{\partial\vec{x}}{\partial a}\frac{\partial\delta a}{\partial t}\bigg|_{\vec{a}}\right) \tag*{(87)}
=−∫dt∭dVa ∂x⃗∂t⋅∂x⃗∂a∂∂tδa(b,c,t)(88)= -\int dt\iiint dV_a\,\frac{\partial\vec{x}}{\partial t}\cdot\frac{\partial\vec{x}}{\partial a}\frac{\partial}{\partial t}\delta a(b,c,t) \tag*{(88)}
=∫dt∭dVa δa(b,c,t)∂∂t(∂x⃗∂t⋅∂x⃗∂a)(89)= \int dt\iiint dV_a\,\delta a(b,c,t)\frac{\partial}{\partial t}\left(\frac{\partial\vec{x}}{\partial t}\cdot\frac{\partial\vec{x}}{\partial a}\right) \tag*{(89)}
=∫dt∬db dc δa(b,c,t)∫da ∂∂t(∂x⃗∂t⋅∂x⃗∂a),(90)= \int dt\iint db\,dc\,\delta a(b,c,t)\int da\,\frac{\partial}{\partial t}\left(\frac{\partial\vec{x}}{\partial t}\cdot\frac{\partial\vec{x}}{\partial a}\right), \tag*{(90)}

where the notation ∣a⃗|_{\vec{a}} was dropped after (87) because all time derivatives are now of the same type with a⃗\vec{a} held fixed. Now recall that this variation (90) is set to zero, and since δa(b,c,t)\delta a(b,c,t) is arbitrary, it follows that

∫da ∂∂t(∂x⃗∂t⋅∂x⃗∂a)=0.(91)\int da\,\frac{\partial}{\partial t}\left(\frac{\partial\vec{x}}{\partial t}\cdot\frac{\partial\vec{x}}{\partial a}\right) = 0. \tag*{(91)}

Note that ∂/∂t\partial/\partial t can be brought outside the integral, since it is a time derivative with the label coordinates a⃗=(a,b,c)\vec{a} = (a,b,c) held fixed. Also note that ∂x⃗/∂t\partial\vec{x}/\partial t is the velocity, u⃗\vec{u}, and consequently we arrive at

ddt∫u⃗⋅∂x⃗∂a da=0.(92)\frac{d}{dt}\int\vec{u}\cdot\frac{\partial\vec{x}}{\partial a}\,da=0. \tag*{(92)}

According to our definition of labels, aa parameterizes curves of constant (butot,bstot)(b_u^{tot},b_s^{tot}); see Figure 1. Therefore, returning from label space to x⃗\vec{x}-space and considering closed curves C(butot,bstot,t)C(b_u^{tot},b_s^{tot},t) of constant (butot,bstot)(b_u^{tot},b_s^{tot}), the result (92) may be restated as

Schematic diagrams showing particle labels and level sets

Figure 1. Left: Schematic diagram showing the particle labels, a⃗=(a,butot,bstot)\vec{a}=(a,b_u^{tot},b_s^{tot}). Right: Schematic diagram showing level sets of bstotb_s^{tot} on a surface of constant butotb_u^{tot}, assuming that bstotb_s^{tot} and butotb_u^{tot} are smooth functions.

DDt∮C(butot,bstot,t)u⃗⋅dx⃗=0,(93)\frac{D}{Dt}\oint_{C(b_u^{tot},b_s^{tot},t)}\vec{u}\cdot d\vec{x}=0, \tag*{(93)}

where dx⃗d\vec{x} is the line element tangent to the curve C(butot,bstot,t)C(b_u^{tot},b_s^{tot},t). Statement (93) is the moist circulation theorem for the system (15)-(18) with phase changes. It is the moist analogue of the Kelvin or Bjerknes circulation theorems from the dry case [e.g. 30, 10, 53].

Furthermore, it is possible to obtain conservation principles for potential vorticity. In this direction, using Stokes’ theorem, we may rewrite the moist Kelvin circulation theorem (93) to obtain a conservation law for vorticity, ω⃗=∇×u⃗\vec{\omega}=\nabla\times\vec{u}, on a material surface patch S(x⃗,t)S(\vec{x},t):

DDt∬S(x⃗,t)ω⃗⋅dS⃗=0,(94)\frac{D}{Dt}\iint_{S(\vec{x},t)}\vec{\omega}\cdot d\vec{S}=0, \tag*{(94)}

where the boundary of the patch is the closed curve C(butot,bstot,t)C(b_u^{tot},b_s^{tot},t). For the special case where S(x⃗,t)S(\vec{x},t) is on a surface of constant butotb_u^{tot}, we find

DDt∬S(x⃗,t)ω⃗⋅∇butot∥∇butot∥ dS=0.(95)\frac{D}{Dt}\iint_{S(\vec{x},t)}\vec{\omega}\cdot\frac{\nabla b_u^{tot}}{\lVert\nabla b_u^{tot}\rVert}\,dS=0. \tag*{(95)}

To then move toward a volumetric integral, take an additional integral with respect to butotb_u^{tot} to yield

∫C1C2DDt∬S(x⃗,t)ω⃗⋅∇butot∥∇butot∥ dS dbutot=0.(96)\int_{C_1}^{C_2}\frac{D}{Dt}\iint_{S(\vec{x},t)}\vec{\omega}\cdot\frac{\nabla b_u^{tot}}{\lVert\nabla b_u^{tot}\rVert}\,dS\,db_u^{tot}=0. \tag*{(96)}

Since butotb_u^{tot} is a label coordinate, and the material derivative D/DtD/Dt is a time derivative holding labels fixed, we may rewrite (96) as

DDt∫∬S(x⃗,t)ω⃗⋅∇butot∥∇butot∥ dS dbutot=DDt∬S(x⃗,t)ω⃗⋅∇butot∥∇butot∥ dS(∥∇butot∥dσ)=0,(97)\frac{D}{Dt}\int\iint_{S(\vec{x},t)}\vec{\omega}\cdot\frac{\nabla b_u^{tot}}{\lVert\nabla b_u^{tot}\rVert}\,dS\,db_u^{tot} = \frac{D}{Dt}\iint_{S(\vec{x},t)}\vec{\omega}\cdot\frac{\nabla b_u^{tot}}{\lVert\nabla b_u^{tot}\rVert}\,dS\left(\lVert\nabla b_u^{tot}\rVert d\sigma\right)=0, \tag*{(97)}

where σ\sigma is arc length along curves in the direction of ∇butot\nabla b_u^{tot} (Figure 2). Simplifying, we find conservation of a volume-integrated moist potential vorticity ∫PVu\int PV_u for a material volume which looks like a distorted cylinder or pill-box as in Figure 2:

Figure 2. ‘Pancake’-like local material volume for parcel-integrated potential vorticity.

DDt∭V(x⃗,t)dV PVu=DDt∭V(x⃗,t)dV ω⃗⋅∇butot=0.(98)\frac{D}{Dt}\iiint_{V(\vec{x},t)}dV\,PV_u = \frac{D}{Dt}\iiint_{V(\vec{x},t)}dV\,\vec{\omega}\cdot\nabla b_u^{tot}=0. \tag*{(98)}

In (98), the base and lid of the cylinder are surfaces of constant butotb_u^{tot}, connected by gradient lines of butotb_u^{tot}. Moreover, the cylinder can be viewed as a stack of patches, where each patch is on a different surface of constant butotb_u^{tot}, and the boundary of each patch is a curve of constant bstotb_s^{tot}. In this way, the bstotb_s^{tot} values on the sides of the cylinder can be written as functions bstot(butot)b_s^{tot}(b_u^{tot}). In the supplementary information, we also show that the conservation statement can be extended to a volume with sides given by any arbitrary function bstot(butot)b_s^{tot}(b_u^{tot}).

Similarly, one can also arrive at conservation of volume-integrated moist potential vorticity ∫PVs\int PV_s:

DDt∭V(x,t)dV PVs=DDt∭V(x,t)dV ω⋅∇bstot=0,(99)\frac{D}{Dt}\iiint_{V(\mathbf{x},t)} dV\,PV_s = \frac{D}{Dt}\iiint_{V(\mathbf{x},t)} dV\,\boldsymbol{\omega}\cdot\nabla b_s^{tot}=0, \tag*{(99)}

where the material volume in (99) has base and lid which are surfaces of constant bstotb_s^{tot}, connected by sides which are gradient lines of bstotb_s^{tot} and have butot=butot(bstot)b_u^{tot}=b_u^{tot}(b_s^{tot}).

For the moist Boussinesq system (15)-(18), relations (98) and (99) are the strongest type of conservation statement for potential vorticity. When phase changes are present, conservation applies following local volumes of potential vorticity, enclosed by special surfaces defined in terms of (butot,bstot)(b_u^{tot},b_s^{tot}), rather than following individual fluid particles as in the dry system.

Now that the systematic procedure has produced the form of the circulation theorem and the conservation principle for potential vorticity, it is possible to derive these results directly from the moist Boussinesq evolution equations in (15)-(18). In the supporting information, we present direct derivations for the circulation theorem, globally-integrated PVuPV_u, and parcel-integrated PVuPV_u and PVsPV_s. The direct derivations provide an independent confirmation of these conservation principles. The supporting information also includes special variations of the action leading to global conservation of PVuPV_u and PVsPV_s.

Discussion and Conclusions

Here we have shown that Hamilton’s Principle can be used to derive the moist Boussinesq dynamics (1)–(4), including phase changes between water vapor and liquid water. The Hamiltonian formulation is not obvious a priori because the buoyancy is piecewise, changing its functional form at interfaces between unsaturated and cloudy air. A key observation is that the piecewise potential energy associated with buoyancy is both continuous and differentiable.

After the Hamiltonian formulation was established, we were then able to investigate particle relabeling symmetry for the action associated with the moist dynamics, and thereby provide a systematic derivation of conservation statements for moist potential vorticity. The conservation statements derived herein had not previously been found from other approaches. Furthermore, the symmetry analysis naturally links the moist versions of Kelvin’s circulation theorem and conservation of potential vorticity.

Using particle relabeling symmetry, we uncover a fundamental difference between the dry and moist conservation statements for potential vorticity, following directly from the requirement that potential energy must remain invariant under the relabeling. In the dry case, it is necessary to restrict relabeling to constant entropy, leading to material invariance of dry potential vorticity on constant-entropy surfaces. In the moist case with phase changes, the relabeling is restricted to fluid parcels situated on curves of constant butotb_u^{tot} and constant bstotb_s^{tot}. These curves are not restricted to the unsaturated or saturated regimes since bub_u and bsb_s are globally defined quantities as in (13, 14). Ultimately, the stronger constraint leads to a weaker conservation law, namely, material invariance of parcel-integrated moist potential vorticity, where the local volumes have special surfaces defined in terms of (butot,bstot)(b_u^{tot},b_s^{tot}), as in Figure 2, for example. Table 1 summarizes the dry and moist results, with respective definitions for potential vorticity. Note that one could restate the moist results in terms of any two material invariants for the system (1)–(4), e.g., total potential temperature θetot\theta_e^{tot} and total water qttotq_t^{tot}.

FeaturesDry case (non-isentropic)Moist Boussinesq case
Particle relabeling symmetryOn level surfaces Ss(t)S_s(t) of entropy ss.On curves Cb(t)C_b(t) of constant butotb_u^{tot} and bstotb_s^{tot}
Circulation theorem: ∮C(t)u⃗⋅dx⃗=0\oint_{C(t)} \vec{u} \cdot\mathrm{d}\vec{x}=0C(t)C(t): Any curve on Ss(t)S_s(t)C(t)C(t): Special curves Cb(t)C_b(t)
PV definition1ρω⃗⋅∇s\frac{1}{\rho}\vec{\omega}\cdot\nabla sPVu=ω⃗⋅∇butot\mathrm{PV}_u=\vec{\omega}\cdot\nabla b_u^{tot}; PVs=ω⃗⋅∇bstot\mathrm{PV}_s=\vec{\omega}\cdot\nabla b_s^{tot}
Conserved quantityparcel PV‘pancake’-integrated PV

Table 1. Summary of potential vorticity conservation principles for various regimes

In previous work ([32]), we established conservation of energy for the moist Boussinesq dynamics. Here, we expand our knowledge of the relevant conservation laws to include a moist Kelvin circulation theorem and conservation of parcel-integrated moist potential vorticity, as well as their relationship to each other. Thus, we now have moist analogues for some of the most important theorems for dry Boussinesq dynamics. Future work will explore how these results can be applied to analyze observations and numerical simulations of atmospheric flows. It is straightforward to include the effects of rotation ([e.g. 10]), which were omitted here for simplicity. In addition, we have already generalized the statement for conservation of potential vorticity to the case of compressible moist flows, and these results will be presented elsewhere.

Acknowledgements

Leslie Smith and Sam Stechmann gratefully acknowledge grant support from the Division of Mathematical Sciences of the National Science Foundation, in conjunction with the award DMS-1907667.

Author Contributions

All authors contributed to the problem formulation, mathematical calculations and preparation of the written manuscript.

Conflict of Interest

The authors declare no conflict of interest.

Supporting Information

In the supporting information, we include special variations of the action leading to global conservation of PVuPV_u and PVsPV_s. We also present direct derivations for the circulation theorem, globally-integrated PVuPV_u, and parcel-integrated PVuPV_u and PVsPV_s.

Particle label definition

We can define the particle label a⃗\vec{a} as follows:

a⃗=(a,b,c)=(a,butot,bstot),(100)\vec{a} = (a,b,c) = (a,b_u^{tot},b_s^{tot}), \tag*{(100)}

where the first component aa should be such that it locates particles on curves of constant (butot,bstot)(b_u^{tot},b_s^{tot}). In addition, we require that dVa=da dbutot dbstotdV_a = da\,db_u^{tot}\,db_s^{tot} represents a mass element in physical space with dVa=[∂(a⃗)∂(x⃗)]dVx=ρdVxdV_a = \left[\frac{\partial(\vec{a})}{\partial(\vec{x})}\right]dV_x = \rho dV_x. Keeping in mind that the initial position vector x⃗0\vec{x}_0 is a mass conserving label with ∂(x⃗0)∂(x⃗)=ρ\frac{\partial(\vec{x}_0)}{\partial(\vec{x})} = \rho, we therefore choose the first component aa in (100) such that ∂(a⃗)/∂(x⃗0)=1\partial(\vec{a})/\partial(\vec{x}_0)=1. A third requirement is for aa to be material invariant with Da/Dt=0Da/Dt=0. Below, we show an explicit construction for a⃗\vec{a} satisfying all the above three restrictions.

In order to concretely define the label aa, start by drawing a parallelepiped in terms of aa, butotb_u^{tot}, bstotb_s^{tot}, where the direction of aa is determined by the cross product ∇0butot×∇0bstot\nabla_0b_u^{tot}\times\nabla_0b_s^{tot}, and where ∇0\nabla_0 is a gradient with respect to the initial position vector x⃗0\vec{x}_0 at time t=0t=0. To enforce (49), and using the chain rule, the magnitude of aa is determined by

∂(a⃗)∂(x⃗0)=∣∇0a ∇0butot ∇0bstot∣=1\frac{\partial(\vec{a})}{\partial(\vec{x}_0)}=\left|\nabla_0a\ \nabla_0b_u^{tot}\ \nabla_0b_s^{tot}\right|=1

where ∇0a\nabla_0a etc. are column vectors. Therefore the label aa is chosen to satisfy the condition ∣∇0a ∇0butot ∇0bstot∣=1\left|\nabla_0a\ \nabla_0b_u^{tot}\ \nabla_0b_s^{tot}\right|=1, which can be restated in terms of a scalar triple product as

(∇0butot×∇0bstot)⋅∇0a=1.(101)\left(\nabla_0b_u^{tot}\times\nabla_0b_s^{tot}\right)\cdot\nabla_0a=1. \tag*{(101)}

The relation (101) will be true if we define

∇0a=∇0butot×∇0bstot∣∇0butot×∇0bstot∣2.(102)\nabla_0a=\frac{\nabla_0b_u^{tot}\times\nabla_0b_s^{tot}}{\left|\nabla_0b_u^{tot}\times\nabla_0b_s^{tot}\right|^2}. \tag*{(102)}

Then integrating along a material curve CC of constant (butot,bstot)\left(b_u^{tot},b_s^{tot}\right), the particle label aa is given by

a(x⃗0)=∫C∇x⃗0butot×∇x⃗0bstot∣∇x⃗0butot×∇x⃗0bstot∣2⋅dx⃗0,(103)a(\vec{x}_0)=\int_C\frac{\nabla_{\vec{x}_0}b_u^{tot}\times\nabla_{\vec{x}_0}b_s^{tot}}{\left|\nabla_{\vec{x}_0}b_u^{tot}\times\nabla_{\vec{x}_0}b_s^{tot}\right|^2}\cdot d\vec{x}_0, \tag*{(103)}

where the integration is from reference point x⃗0r\vec{x}_0^{r} to initial point x⃗0\vec{x}_0. Equivalently, we may write

a(x⃗0)=∫0σdσ′∣∇x⃗0butot×∇x⃗0bstot∣,(104)a(\vec{x}_0)=\int_0^\sigma\frac{d\sigma'}{\left|\nabla_{\vec{x}_0}b_u^{tot}\times\nabla_{\vec{x}_0}b_s^{tot}\right|}, \tag*{(104)}

where σ\sigma is the arc length along the curve CC (see Figure 3). Finally, notice that since aa is completely determined by x⃗0\vec{x}_0, then Da/Dt=0Da/Dt=0 by the chain rule, and therefore aa is material invariant as required.

Figure 3. Schematic to illustrate the computation of the label a(x⃗0)a(\vec{x}_0) in terms of initial position labels x⃗0\vec{x}_0, where aa is the first coordinate of a⃗=(a,butot,bstot)\vec{a}=(a,b_u^{tot},b_s^{tot}).

references

References

  1. [1]Abramov, R. V., Kovačič, G. and Majda, A. J. (2003) Hamiltonian structure and statistically relevant conserved quantities for the truncated Burgers-Hopf equation. Communications on Pure and Applied Mathematics: A Journal Issued by the Courant Institute of Mathematical Sciences, 56, 1–46.DOI
  2. [2]Bennetts, D. A. and Hoskins, B. (1979) Conditional symmetric instability-a possible explanation for frontal rainbands. Quarterly Journal of the Royal Meteorological Society, 105, 945–962.DOI
  3. [3]Brennan, M. J. and Lackmann, G. M. (2005) The influence of incipient latent heat release on the precipitation distribution of the 24–25 January 2000 US east coast cyclone. Monthly Weather Review, 133, 1913–1937.
  4. [4]Brennan, M. J., Lackmann, G. M. and Mahoney, K. M. (2008) Potential vorticity (PV) thinking in operations: The utility of nonconservation. Weather and Forecasting, 23, 168–182.
  5. [5]Bretherton, C. S. (1987) A theory for nonprecipitating moist convection between two parallel plates. Part I: Thermodynamics and "linear" solutions. Journal of the Atmospheric Sciences, 44, 1809–1827.doi.org/10.1175/1520-0469(1987)044<1809:atfnmc>2.0.co;2
  6. [6]Bretherton, F. P. (1970) A note on Hamilton’s principle for perfect fluids. Journal of Fluid Mechanics, 44, 19–31.DOI
  7. [7]Büeler, D. and Pfahl, S. (2017) Potential vorticity diagnostics to quantify effects of latent heating in extratropical cyclones. Part I: Methodology. Journal of the Atmospheric Sciences, 74, 3567–3590.DOI
  8. [8]Cao, Z. and Cho, H.-R. (1995) Generation of moist potential vorticity in extratropical cyclones. Journal of the Atmospheric Sciences, 52, 3263–3282.doi.org/10.1175/1520-0469(1995)052<3263:gompvi>2.0.co;2
  9. [9]Cotter, C. (2013) Data assimilation on the exponentially accurate slow manifold. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 371, 20120300.DOI
  10. [10]Cotter, C. and Holm, D. (2014) Variational formulations of sound-proof models. Quarterly Journal of the Royal Meteorological Society, 140, 1966–1973.
  11. [11]Cuijpers, J. W. M. and Duynkerke, P. G. (1993) Large eddy simulation of trade wind cumulus clouds. Journal of the Atmospheric Sciences, 50, 3894–3908.doi.org/10.1175/1520-0469(1993)050<3894:lesotw>2.0.co;2
  12. [12]Davis, C. A. and Emanuel, K. A. (1991) Potential vorticity diagnostics of cyclogenesis. Monthly Weather Review, 119, 1929–1953.doi.org/10.1175/1520-0493(1991)119<1929:pvdoc>2.0.co;2
  13. [13]Emanuel, K. A. (1979) Inertial instability and mesoscale convective systems. Part I: Linear theory of inertial instability in rotating viscous fluids. Journal of the Atmospheric Sciences, 36, 2425–2449.doi.org/10.1175/1520-0469(1979)036<2425:iiamcs>2.0.co;2
  14. [14]Ertel, H. (1942) Ein neuer hydrodynamischer wirbelsatz. Meteorologische Zeitschrift, 59, 277–281.DOI
  15. [15]Gao, S., Wang, X. and Zhou, Y. (2004) Generation of generalized moist potential vorticity in a frictionless and moist adiabatic flow. Geophysical Research Letters, 31.DOI
  16. [16]Hastermann, G., Reinhardt, M., Klein, R. and Reich, S. (2021) Balanced data assimilation for highly oscillatory mechanical systems. Communications in Applied Mathematics and Computational Science, 16, 119–154.DOI
  17. [18]Hernandez-Duenas, G., Smith, L. M. and Stechmann, S. N. (2015) Stability and instability criteria for idealized precipitating hydrodynamics. Journal of the Atmospheric Sciences, 72, 2379–2393.DOI
  18. [19]Hill, E. L. (1951) Hamilton’s principle and the conservation theorems of mathematical physics. Reviews of Modern Physics, 23, 253.DOI
  19. [20]Holm, D. D. (2015) Variational principles for stochastic fluid dynamics. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 471, 20140963.pubmed.ncbi.nlm.nih.gov/27547083
  20. [21]Holm, D. D., Marsden, J. E. and Ratiu, T. S. (1998) The Euler–Poincaré equations and semidirect products with applications to continuum theories. Advances in Mathematics, 137, 1–81.
  21. [22]Hoskins, B. and Berrisford, P. (1988) A potential vorticity perspective of the storm of 15–16 October 1987. Weather, 43, 122–129.DOI
  22. [23]Hoskins, B. J. (1991) Towards a PV-θ view of the general circulation. Tellus A: Dynamic Meteorology and Oceanography, 43, 27–36.
  23. [24]Hoskins, B. J., McIntyre, M. E. and Robertson, A. W. (1985) On the use and significance of isentropic potential vorticity maps. Quarterly Journal of the Royal Meteorological Society, 111, 877–946.
  24. [25]Kuo, H. L. (1961) Convection in conditionally unstable atmosphere. Tellus, 13, 441–459.DOI
  25. [26]Lackmann, G. (2011) Midlatitude Synoptic Meteorology: Dynamics, Analysis, and Forecasting. American Meteorological Society.
  26. [27]Lackmann, G. M. (2002) Cold-frontal potential vorticity maxima, the low-level jet, and moisture transport in extratropical cyclones. Monthly Weather Review, 130, 59–74.doi.org/10.1175/1520-0493(2002)130<0059:cfpvmt>2.0.co;2
  27. [28]Madonna, E., Wernli, H., Joos, H. and Martius, O. (2014) Warm conveyor belts in the era-interim dataset (1979–2010). Part I: Climatology and potential vorticity evolution. Journal of Climate, 27, 3–26.DOI
  28. [29]Majda, A. J. (2003) Introduction to PDEs and Waves for the Atmosphere and Ocean, vol. 9 of Courant Lecture Notes in Mathematics. Providence: American Mathematical Society.DOI
  29. [30]Majda, A. J. and Bertozzi, A. L. (2002) Vorticity and Incompressible Flow, vol. 27 of Cambridge Texts in Applied Mathematics. Cambridge: Cambridge University Press.DOI
  30. [31]Majda, A. J., Moore, M. and Qi, D. (2019) Statistical dynamical model to predict extreme events and anomalous features in shallow water waves with abrupt depth change. Proceedings of the National Academy of Sciences, 116, 3928–3987.DOI
  31. [32]Marsico, D. H., Smith, L. M. and Stechmann, S. N. (2019) Energy decompositions for moist Boussinesq and anelastic equations with phase changes. Journal of the Atmospheric Sciences, 76, 3569–3587.DOI
  32. [33]Martin, J. E. (2013) Mid-latitude Atmospheric Dynamics: A First Course. John Wiley & Sons.
  33. [34]Moore, M., Bolles, C. T., Majda, A. J. and Qi, D. (2020) Anomalous waves triggered by abrupt depth changes: Laboratory experiments and truncated KdV statistical mechanics. arXiv preprint arXiv:2001.00995.arxiv.org/abs/2001.00995
  34. [35]Müller, P. (1995) Ertel’s potential vorticity theorem in physical oceanography. Reviews of Geophysics, 33, 67–97.DOI
  35. [36]Noether, E. (1918) Invariante Variationsprobleme. Königlich Gesellschaft der Wissenschaften Göttingen Nachrichten Mathematik-physik Klasse, 2, 235–267.
  36. [37]Olver, P. J. (2000) Applications of Lie Groups to Differential Equations, vol. 107. Springer Science & Business Media.
  37. [38]Padhye, N. and Morrison, P. (1996) Relabeling symmetries in hydrodynamics and magnetohydrodynamics. Tech. rep., Texas Univ., Austin, TX (United States). Inst. for Fusion Studies.DOI
  38. [41]Pavlov, D., Mullen, P., Tong, Y., Kanso, E., Marsden, J. E. and Desbrun, M. (2011) Structure-preserving discretization of incompressible fluids. Physica D: Nonlinear Phenomena, 240, 443–458.arxiv.org/abs/0912.3989
  39. [42]Ripa, P. (1981) Symmetries and conservation laws for internal gravity waves. In AIP Conference Proceedings, vol. 76, 281–306. American Institute of Physics.DOI
  40. [43]Rossby, C.-G. (1939) Relation between variations in the intensity of the zonal circulation of the atmosphere and the displacements of the semi-permanent centers of action. J. Mar. Res., 2, 38–55.
  41. [47]Schubert, W. H., Hausman, S. A., Garcia, M., Ooyama, K. V. and Kuo, H.-C. (2001) Potential vorticity in a moist atmosphere. Journal of the Atmospheric Sciences, 58, 3148–3157.doi.org/10.1175/1520-0469(2001)058<3148:pviama>2.0.co;2
  42. [48]Shepherd, T. G. (1990) Symmetries, conservation laws, and Hamiltonian structure in geophysical fluid dynamics. In Advances in Geophysics, vol. 32, 287–338. Elsevier.DOI
  43. [49]Smith, L. M. and Stechmann, S. N. (2017) Precipitating quasigeostrophic equations and potential vorticity inversion with phase changes. Journal of the Atmospheric Sciences, 74, 3285–3303.
  44. [50]Sommeria, G. (1976) Three-dimensional simulation of turbulent processes in an undisturbed trade wind boundary layer. J. Atmos. Sci., 33, 216–241.
  45. [51]Spyksma, K., Bartello, P. and Yau, M. K. (2006) A Boussinesq moist turbulence model. J. Turbulence, 7, 1–24.DOI
  46. [52]Thorpe, A. (1985) Diagnosis of balanced vortex structure using potential vorticity. Journal of the Atmospheric Sciences, 42, 397–406.
  47. [53]Thorpe, A. J., Volkert, H. and Ziemiański, M. J. (2003) The Bjerknes’ circulation theorem: A historical perspective. Bulletin of the American Meteorological Society, 84, 471–480.DOI
  48. [54]Wetzel, A. N., Smith, L. M., Stechmann, S. N. and Martin, J. E. (2019) Balanced and unbalanced components of moist atmospheric flows with phase changes. Chin. Ann. Math. Ser. B, 40, 1005–1038.
  49. [55]Wetzel, A. N., Smith, L. M., Stechmann, S. N., Martin, J. E. and Zhang, Y. (2020) Potential vorticity and balanced and unbalanced moisture. Journal of the Atmospheric Sciences, 77, 1913–1931.DOI
  50. [56]Zhang, Y., Smith, L. M. and Stechmann, S. N. (2021a) Effects of clouds and phase changes on fast-wave averaging: A numerical assessment. Journal of Fluid Mechanics, 920, A49.

Paper details

vlnc:W_2wrqkjvm
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 8

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents