A definition of the background state of the atmosphere using optimal transport
summary
The gist
The paper proves that for any snapshot of an atmospheric flow in a single hemisphere, there exists a unique energy-minimising MLM state in geophysical coordinates (latitude and pressure) given the…
In short
The paper defines an energy-minimizing Modified Lagrangian Mean (MLM) state for atmospheric flow snapshots in a single hemisphere using optimal transport theory. It proves that for any given mass distribution, there exists a unique state that minimizes an energy cost, making it suitable for quantifying disturbances and analyzing atmospheric instability.
Key concepts
- MLM State
- An MLM state is defined by three components: surface pressure (p), potential temperature (Θ), and zonal angular momentum (Z). It represents the physical configuration of the atmosphere that minimizes a specific energy cost when transporting mass from a source domain to a target distribution.
- Optimal Transport Formulation
- The problem is framed as finding an optimal transport map that moves mass from one coordinate system (geophysical coordinates like latitude/pressure) to another (conserved variables like Z and Θ). The cost of this transport is shown to be equal to the energy of the resulting MLM state.
- Twisted Cost Function
- The paper uses a specific cost function that satisfies certain assumptions, including being 'twisted.' This property ensures that optimal transport maps exist, meaning there is a unique way to rearrange the mass distribution while minimizing the defined energy cost.
- Optimal Transport Map
- This map describes how mass from the source distribution (defined by pressure) is transported into the target distribution (defined by Z and Θ). The existence and uniqueness of this map are crucial, as it identifies the specific physical state that minimizes the overall energy.
Terminology used across episodes
This episode discusses
The paper
A definition of the background state of the atmosphere using optimal transport · Read on arXiv
Institute of Computer Science, University of Goettingen · Department of Meteorology, University of Reading · Maxwell Institute for Mathematical Sciences and Department of Mathematics, Heriot-Watt University · Met Office
Transcript
Introduction to the show: ident: Mathematics. The day in mathematics: new theorems, new proofs, and what they open.
Ada: I'm Ada.
Emmy: And I'm Emmy, here with Kolya.
Ada: We're looking at the paper: "A definition of the background state of the atmosphere using optimal transport".
Emmy: Here's the gist of the paper. The paper proves that for any snapshot of an atmospheric flow in a single hemisphere, there exists a unique energy-minimising MLM state in geophysical coordinates (latitude and pressure) given the distribution of mass in conserved variable coordinates (Z, Θ) Since the state is an energy minimum, it is suitable for quantification of finite amplitude disturbances and examining atmospheric instability Definition of the MLM State and Optimal Transport Formulation The problem of finding energy-minimising MLM states is phrased as the minimisation of an optimal transport cost over a suitable space of source measures (µ) for a given target measure (ν) The source space corresponds to geophysical coordinates (latitude and pressure) and the target space is defined by the conserved variable coordinates (Z, Θ) In physical terms, the optimal transport solution maps the mass from the target space (distribution taken as given) into the source space, while also minimising a cost which is shown to be equal to the energy of the state Definition 1.2 defines an MLM state by three maps: surface pressure p, potential temperature Θ, and zonal angular momentum Z The physical interpretation is that (Z, Θ) is an admissible transport map from µp to ν Mathematical Framework and Existence Proof The main result is that for a large class of cost functions c (Assumption 2.1), the minimisation problem (1) has a unique solution ¯p, which has continuous density (Theorem 3.13) The existence and uniqueness of solutions are established by deriving a dual functional Gν, and showing that its maximiser corresponds to the optimal transport map Theorem 3.8 states that for each ν ∈ P(Y), there exists a unique minimiser of Fν This minimiser is denoted by p¯ψ, and the optimality condition implies that ψ is a Kantorovich potential from µp¯ to ν Stability and Numerical Tractability The results establish stability of solutions with respect to input data (Theorem 4.1) Furthermore, the problem of finding energy minimising MLM states reduces to a standard semi-discrete optimal transport problem when the target measure ν is discrete (Corollary 5.3) This reduction makes the problem numerically tractable Cost Function Properties and Caveats The cost function c defined by (61) satisfies Assumptions 2.6, which ensures that the cost function is twisted, guaranteeing the existence of optimal transport maps rather than merely optimal transport plans The results come with three main caveats: first, the exclusion of the extreme latitudes in the given hemisphere to ensure regularity, second, an arbitrary minimum pressure is assumed and built into the cost function, and third, zonal angular momentum is assumed to be positive and bounded away from zero to ensure that the cost function is twisted The exclusion of the equator is also mentioned as ensuring Assumption 2.5 and Assumption 2.6 In this work, we have given a rigorous definition of a Modified Lagrangian Mean (MLM) state in a single hemisphere in latitude-pressure coordinates (Definition 1.1) Such a state consists of a surface pressure function p determining the extent of the source domain, and an admissible transport map T = (Z, Θ) between the uniform density on the source domain and a mass distribution ν in the space of zonal angular momentum and potential temperature We have proved that if ν has compact support in (0, +∞)2, then there exists a unique energy-minimising MLM state defined on a domain excluding the equator and the pole (Theorem 3.13) Optimality conditions imply that this state is in fact the solution of a single optimal transport problem (Theorem 5.1) Our results hold not only for the cost function defining the energy of an MLM state (11) but for a wide class of cost functions satisfying mild assumptions The exclusion of the extreme latitudes in the given hemisphere is necessary to ensure that the cost function is sufficiently regular (locally Lipschitz; Assumption 2.1.A.) The equator is excluded to ensure that the cost function is twisted (Assumption 2.5) and admits a twisted extension (Assumption 2.6) An arbitrary minimum pressure is assumed and built into the cost function (11) The zonal angular momentum is assumed to be positive and bounded away from zero to ensure that the cost function is twisted This reduction to a standard semidiscrete optimal transport problem, which reduces Problem 2.4 to a standard semidiscrete optimal transport problem, is numerically tractable The relationship between Problem 2.4 and the optimal transport problem to which it can be reduced is illustrated in Figure 2 The cost function (6) defines the energy of an MLM state The set S is uniformly bounded and equicontinuous by Lemma 3.6 The final result establishes that the map F: Pac(B) ∩ L∞(B),∥ ·∥L∞(B) × P(Y) → R given by F(¯p, ν):= Fν(¯p) = Tc(µp¯, ν) is continuous The cost function (61) satisfies Assumptions 2.1 and 2.6 The final proof shows that for all n ≥ N2, µp¯(∆n,ε) = µp¯(∆n,ε) − µp¯n(∆n,ε) + µp¯n(∆n,ε) ≤ δ The final proof shows that for all n ≥ N2,∥T − Tn∥r Lr(X)=Z X∥T − Tn∥r Rd dx= 2ε3 Page 33 ---
5: M.
Ada: J.
Emmy: P.
Ada: Cullen, The Mathematics of Large-Scale Atmosphere and Ocean, World Scientific, 2021
6: M.
Title and authors: Ada: So, we're looking at this paper today titled "A definition of the background state of the atmosphere using optimal transport," and it deals with how we can rigorously define what we consider to be the background state of an atmospheric flow.
Emmy: It’s really about moving away from just picking some heuristic method for that background state and instead using optimal transport theory to select a unique one based on energy minimization.
Ada: Exactly. The paper sets up the problem by saying that for any snapshot of an atmospheric flow in a single hemisphere, there is a unique energy-minimizing modified Lagrangian mean, or MLM, state in geophysical coordinates—that means latitude and pressure—given the distribution of mass in conserved variable coordinates like zonal angular momentum and potential temperature.
Kolya: That sounds like we're trying to find the most stable way to describe a flow snapshot by minimizing some kind of energy cost over all possible ways to map those variables.
Ada: Right, Kolya. The core idea is framing the search for this MLM state as a minimization problem of an optimal transport cost over a space of source measures for a given target measure, which is what they call Problem one point four in their setup <ref:2504.11141#pg2>.
Emmy: And they define the source space by geophysical coordinates like latitude and pressure, and the target space by those conserved variables, which are zonal angular momentum and potential temperature. The optimal transport solution maps mass from that target space distribution into the source space while minimizing a cost that turns out to be equal to the energy of that state <ref:2504.11141#pg2>.
Ada: That’s a big conceptual leap, moving the definition of a background state from some ad hoc choice to an explicit minimization problem over all possible source measures.
Emmy: And they define this MLM state by three maps: surface pressure, potential temperature, and zonal angular momentum. Physically, it means that the conserved variables are an admissible transport map from the distribution on the source domain to the distribution of mass in those conserved variables <ref:2504.11141#pg3>.
Kolya: So what’s the actual result here? What do they prove about this minimization problem?
Ada: The main result is that for a large class of cost functions, specifically those satisfying Assumption two point one, the minimization problem has a unique solution for the surface pressure function p, and it has a continuous density <ref:2504.11141#pg3>.
Emmy: And they establish this existence and uniqueness by looking at a dual functional called Gnu, showing that its maximizer is precisely the optimal transport map <ref:2504.11141#pg3>.
Ada: This means that for every distribution of mass in those conserved variables, there exists one and only one minimizer of the functional Fnu, which they denote as p bar psi <ref:2504.11141#pg3>. The optimality conditions then imply that this state is the solution to a standard optimal transport problem <ref:2504.11141#pg3>.
Kolya: That sounds promising for numerical work, but what about the practical side? How does this actually help us with real atmospheric data?
Emmy: Well, they’ve made things tractable because when the target measure ν is discrete—meaning you have a finite set of conserved variable distributions—the problem reduces to a standard semi-discrete optimal transport problem
two thousand five hundred four point one one one four one#corollaries five point three: .
Ada: Which means instead of solving this incredibly complex continuous optimization problem, we can maximize the Kantorovich dual functional, which they say is a concave, finite dimensional, unconstrained maximization problem
two thousand five hundred four point one one one four one#corollaries five point three: .
Kolya: That reduces the complexity significantly. But what about stability? If we feed this into real-world data that isn't perfectly clean, does the solution hold up?
Emmy: They’ve established stability with respect to input data in Theorem four point one <ref:2504.11141#pg3>. Plus, they show convergence properties in Theorem four point two, which proves that minimizers of Problem two point four converge uniformly to the true minimizer when the target measure converges in probability—that is, as your data gets closer to something theoretical <ref:2504.11141#pg3>.
Ada: That convergence proof is key for making this useful in a real setting, because it guarantees that if our input data fluctuates slightly, the resulting background state won't jump around wildly.
Kolya: I noticed they have some restrictions on the cost function themselves. They exclude extreme latitudes and the equator to ensure regularity and to keep things twisted <ref:2504.11141#pg3>. Are those exclusions necessary for something more fundamental?
Emmy: The exclusion of the equator is specifically required to ensure that the cost function is twisted, which they call Assumption two point five, and it admits a twisted extension <ref:2504.11141#pg3>. Without that twist, you only get optimal transport plans instead of actual maps <ref:2504.11141#pg3>.
Ada: And they also assume the zonal angular momentum is positive and bounded away from zero to ensure that the cost function remains twisted <ref:2504.11141#pg3>. These assumptions are important for guaranteeing that we actually find a map, not just a plan, which is essential for defining that unique background state.
Kolya: So if we only look at the wins, what’s the actual limitation here? What does this method not cover?
Emmy: They mention three main caveats upfront. First, they exclude the extreme latitudes in the hemisphere to ensure regularity, which is required for Assumption 2 point 1A <ref:2504.11141#pg3>. Second, there's an assumption about a minimum pressure being built into the cost function <ref:2504.11141#pg3>. And third, they assume the zonal angular momentum is positive and bounded away from zero for that twist condition <ref:2504.11141#pg3>.
Ada: They also explicitly mention that excluding the equator ensures Assumption two point five and two point six, but they also note that the exclusion of extreme latitudes is necessary for Assumption 2 point 1A to hold <ref:2504.11141#pg3>. It’s a trade-off for mathematical regularity.
Emmy: And finally, they state that the reduction to a standard semi-discrete optimal transport problem is only possible when the target measure ν has compact support in two dimensions, specifically in (zero plus infinity) squared <ref:2504.11141#pg3>.
Kolya: So if we want to apply this to a general atmospheric snapshot that doesn't fit those compact support conditions, does the whole framework fall apart?
Ada: The paper focuses on cases where the target measure ν has compact support in (zero plus infinity) squared <ref:2504.11141#pg3>, so if we look outside those conditions, this specific result doesn't directly apply to that general case.
Emmy: But the broader implication is that even with these specific constraints on the input distribution, we get a rigorous definition for the background state <ref:2504.11141#pg2>. It gives us a way to select a state based on energy minimization, which is much more principled than just picking one from a set of possible states.
Ada: So, to wrap up this part of the paper, the definition of the background state using optimal transport allows us to rigorously select a unique configuration—the MLM state—that minimizes an energy functional over all possible source measures <ref:2504.11141#pg2>.
Kolya: I’m just thinking about how this relates to what we usually do when we look at instabilities in the atmosphere. Does this minimization problem tell us anything concrete about finite amplitude disturbances?
Emmy: Yes, because since this state is an energy minimum, it makes it suitable for quantifying finite amplitude disturbances and examining atmospheric instability <ref:2504.11141#pg1>. It gives us a baseline that we can compare against the actual flow to see how much energy is needed to move away from that minimum state.
Ada: It’s a way of saying that the background state isn't just any arbitrary choice; it’s the most energetically favorable one, which makes analyzing deviations from it much more meaningful <ref:2504.11141#pg2>.
Emmy: So, we’ve seen how they define this MLM state using optimal transport and how the mathematical framework leads to a unique energy-minimizing solution for the background state of the atmosphere in geophysical coordinates <ref:2504.11141#pg2>.
Kolya: It feels like a solid mathematical foundation, especially with that convergence theorem showing stability when we have noisy data <ref:2504.11141#pg3>.
Ada: Exactly. It grounds the definition in optimization theory rather than just physics intuition, which is what makes this approach useful for understanding atmospheric instability and disturbances <ref:2504.11141#pg2>.
Emmy: It’s a powerful way to move from describing a flow by its displacements relative to some background state to actually defining that background state through the lens of energy minimization <ref:2504.11141#pg3>.
Kolya: We’ll be looking at how this reduction to optimal transport problems helps us handle the numerical side when we have discrete data
two thousand five hundred four point one one one four one#corollaries five point three: . That seems like a practical path forward for researchers working with observational data.
Ada: It does, Kolya, because it shows that even though the continuous version is tough, the discrete version becomes solvable via maximizing that dual functional
two thousand five hundred four point one one one four one#corollaries five point three: .
Emmy: So in short, we've looked at how they define this background state using optimal transport and established a unique energy-minimizing solution for atmospheric flows in latitude and pressure coordinates <ref:2504.11141#pg2>.
Kolya: And we’ve seen how that framework is robust enough to handle some noise, thanks to the convergence results, even though they have those specific limitations on the input distribution <ref:2504.11141#pg3>.
Ada: Indeed. The definition of the background state of the atmosphere using optimal transport gives us a principled way to select a unique background state based on energy minimization for atmospheric snapshots <ref:2504.11141#pg2>.
Emmy: That means we can start quantifying finite amplitude disturbances and examining instability with much more mathematical rigor than before <ref:2504.11141#pg3>.
The paper's summary: Ada: So, we're looking at how they’ve defined this Modified Lagrangian Mean state using optimal transport theory for atmospheric flows in a single hemisphere, and what that actually means for us.
Emmy: Basically, they take a snapshot of the atmosphere—latitude and pressure—and they try to find one unique energy-minimizing background state by mapping it into conserved variables like zonal angular momentum and potential temperature.
Kolya: So the takeaway is that we’re not just picking some arbitrary starting point for our models; we’re using a mathematical method to select the most energetically favorable configuration available for that snapshot.
Ada: Exactly. They prove this state exists uniquely for a huge class of cost functions, and they show how to find it by maximizing a dual functional, which is really the optimal transport map.
Emmy: It’s structured beautifully because it turns what looks like a messy fluid dynamics problem into a standard optimal transport problem when you look at the discrete case.
Kolya: And that reduction to discrete problems is actually what makes this approach useful for computation, because we can use well-understood methods for those cases.
Ada: Right. But they’ve put some guardrails on the math, which is where things get interesting—they have to exclude certain latitudes and the equator just to keep the cost function regular and twisted.
Emmy: That twist condition is important because without it, you only get an optimal transport plan instead of a true map connecting the source space to the target space.
Kolya: And they also need zonal angular momentum to be positive and bounded away from zero just so that twist condition holds true, which means those assumptions aren't totally arbitrary.
Ada: The stability analysis is also solid; they prove that these minimizers won't jump around wildly if the input data itself is changing in a predictable way.
Emmy: It’s like they built a very sturdy mathematical foundation for defining what the atmosphere looks like at any given moment, based purely on minimizing energy.
Kolya: So, this gives us a rigorous way to quantify those finite amplitude disturbances you mentioned earlier because we have this clear baseline state to measure them against.
Ada: Precisely. It means that when you want to study how atmospheric disturbances evolve, you start by defining the background state through this energy minimization process before you even look at the actual noise in the data.
Emmy: It shifts our focus from just describing changes in flow to describing how much energy it takes to get away from its lowest possible state.
Kolya: And that’s a big shift for anyone working on long-term climate or instability studies—it gives you a principled starting point.
Ada: So, this paper shows that the background state isn't just a guess; it's an energy minimum found through optimal transport theory, and it sets up the framework for analyzing atmospheric instability based on that minimization.
Emmy: We’ll be diving into how they handle those numerical issues next.
The paper's improvements: Tom: Okay, so we’re moving on to what the authors suggest they should do next with this work on defining background states using optimal transport.
Ada: Basically, they are suggesting ways to make the mathematical framework even more robust and applicable to real data situations.
Emmy: They want to show how the definition of that MLM state holds up when you introduce some noise into your measurements, which is something we talked about earlier.
Kolya: So they’re looking at how the minimizers behave when the input measure, that target distribution, starts to look a bit messy or noisy instead of being perfectly defined.
Ada: Yes, they are focusing on proving convergence results for the optimal transport map itself when your input data converges in probability.
Emmy: That’s about showing that if you have a sequence of atmospheric snapshots that are getting closer and closer to some true distribution, the resulting MLM states will also get closer and closer to each other.
Kolya: That convergence proof is really important because it tells us the method is stable enough for real-world observations where things aren't perfect.
Ada: It provides a rigorous way to analyze how atmospheric disturbances evolve if your boundary conditions or initial snapshots are changing slightly over time, which is something we need when we look at climate change.
Emmy: They’re also hinting at expanding the class of cost functions they can use, not just sticking to the specific ones they had to assume earlier.
Kolya: That makes sense because those assumptions—like positivity and bounds on zonal angular momentum—are quite restrictive for real atmospheric data that might have weird fluctuations.
Ada: They are looking into how to relax those constraints while still maintaining the uniqueness of the solution, which is a big challenge in these kinds of minimization problems.
Emmy: It’s about finding a broader range of cost functions where the structure behind the computation still lets you find that unique energy-minimizing state.
Kolya: If they can broaden that range, it opens up more possibilities for applying this to different atmospheric phenomena, not just one specific type of flow.
Ada: Right. The next step is making sure this theory isn't too narrowly defined by the assumptions they had to make just to get the initial existence and uniqueness proofs down.
Emmy: It’s about building a more general structure where you can plug in different physical constraints without losing the core idea of energy minimization.
Kolya: So, they’re pushing for a more universal method, moving away from needing those very specific assumptions on the angular momentum term.
Ada: The implication is that this framework could become a much more general tool for atmospheric modeling, provided they can successfully generalize that cost function structure.
Conclusion: Ada: So we’re wrapping up this discussion on "A definition of the background state of the atmosphere using optimal transport."
Emmy: We've covered how they use optimal transport to select a unique, energy-minimizing MLM state for atmospheric flows in latitude and pressure coordinates.
Kolya: And we touched on the stability results and the caveats about those cost function assumptions.
Ada: Exactly. The core idea is that this method gives us a principled way to define what the background state of a flow is by finding its lowest energy configuration through optimization theory.
Emmy: It’s like they’re taking all those messy fluid dynamics equations and turning them into a clean, structured problem in optimal transport geometry.
Kolya: And the practical part is that they showed this reduces to standard optimal transport when the target distribution is discrete, which makes it numerically solvable.
Ada: That reduction is key because it means we can actually compute these states without getting bogged down by infinite-dimensional continuous problems.
Emmy: So, this gives us a way to define the atmospheric background state based on minimizing energy across all possible source measures in that space.
Kolya: It’s a solid foundation for quantifying disturbances because we have a mathematically rigorous baseline to compare against later.
Ada: And it sets up the framework for analyzing instability by looking at how much energy is needed to move away from that minimum state.
Emmy: It shifts our focus from just describing changes in flow to describing how much energy it takes to get away from its lowest possible state.
Kolya: I just want to say that the convergence proofs they have, especially with the input data converging in probability, are what really make this method trustworthy for real observations.
Ada: True. It shows that even when your data isn't perfect, you can still trust the result because it will converge to a true minimizer.
Emmy: So that we can use this structure to analyze things like finite amplitude disturbances with more confidence than before.
Kolya: And I just want to say that while they have those constraints on the cost function, their work opens up a new way of thinking about defining these states in atmospheric science.
Ada: It’s a solid piece of math applied directly to our atmosphere, showing how energy minimization can define a physical state.
Emmy: We’ll be looking at how this framework helps us tackle those more complex numerical problems next.
More episodes
- 2508.20749-Rates of convergence for extremal spacings in Kakutani's random interval-splitting process
- 2509.05860-Concentration Inequalities for Branching Random Walks
- 2509.19309-Small Prime Primitive Roots in Arithmetic Progressions
- 2509.21830-Noncollapsing for Curvature Flows with Inhomogeneous Speeds
- 2510.08346-Separability and Submetrizability in Locally Convex Spaces
- 2510.27290-Borel Combinatorics of Schreier Graphs of Z-actions
- 2511.13500-Sampling Density for Gabor Phase Retrieval
- 2511.19161-Linear dynamics of random products of weighted shifts
- 2511.21548-Metastability of diffusion processes in narrow tubes
- 2512.01536-Abelian approximations for bicrossed products