PRH | Essay | 7.14 • Navier–Stokes, Blur, and Blurrichevsky Geometry

Navier–Stokes, Blur, and Blurrichevsky Geometry Smoothness at Finite Resolution Aleksandar Perišić November 2025 Abstract We revisit the three–dimensional incompressible Navier–Stokes equations from the viewpoint of blur and Blurrichevsky geometry. The Clay problem asks whether smooth solutions exist globally for all smooth initial data in the continuum limit, with no control parameter. Physically, however, fluid flows are never observed at infinite resolution: uncertainty principles, finite energy, and finite measurement capacity impose a nonzero blur scale on any description. In this note we make this scale explicit. We introduce a family of blur operators { B ℓ}ℓ>0 (mollifiers or heat kernels), define a scale–dependent Blurrichevsky metric on velocity fields, and show that for any Leray–Hopf solution u and any ℓ > 0, the blurred field uℓ =B ℓu is smooth and solves a filtered Navier–Stokes system with an explicitly controlled Reynolds stress. From the perspective of an observer restricted to resolution ℓ , all singularities of u (if they exist) are hidden behind a finite blur budget. The message is twofold. Mathematically, we formalize the idea that NS dynamics factor through a tower of blur–equivalence classes, each with smooth representatives. Epistemologically, we argue that the Clay existence and smoothness problem lives entirely at the (physically unattainable) limit ℓ↓ 0, while all physically meaningful statements about fluid flows can be posed and proved at finite blur. 1 Classical Navier–Stokes and the Clay question We work on the torus T3 for simplicity; the discussion extends to R3 with standard technical modifications. 1.1 The equation and weak solutions The incompressible Navier–Stokes equations with viscosity ν > 0read    ∂tu+ (u· ∇)u+∇p=ν∆u, div u= 0,(1) with u ( t, x ) ∈R3 the velocity field and p ( t, x )the pressure. Given divergence–free initial data u0(x), one seeks udefined for t≥0with u(0,·) = u0. The standard global existence result is due to Leray. Definition 1.1 (Leray–Hopf weak solution).A divergence–free u0∈L2 ( T3 )admits a Leray–Hopf weak solution uif •u∈L∞ loc([0,∞); L2(T3)) ∩L2 loc([0,∞); H1(T3)), •uis weakly divergence–free and satisfies (1) in the sense of distributions, 1 •usatisfies the global energy inequality 1 2∥u(t)∥2 L2+νZt 0 ∥∇u(s)∥2 L2ds ≤1 2∥u0∥2 L2,∀t≥0. Theorem 1.2 (Leray).For every divergence–free u0∈L2 ( T3 )there exists at least one Leray– Hopf weak solution. The Clay problem in this setting asks: Given smooth divergence–free u0 , is there a global smooth solution u to (1) , or can singularities form in finite time? This is a purely Platonic question about the continuum PDE at infinite resolution. It ignores any epistemic or physical limitation on how finely the flow can be known. 1.2 Physical versus Platonic smoothness Physically, flows are observed through finite–resolution measurements: •space and time are sampled on grids, •devices have finite bandwidth and sensitivity, • and at sufficiently small scales the continuum model itself breaks down (molecular discreteness, quantum effects, thermal noise). From that viewpoint it is natural to replace the absolute notion of smoothness with a scale– dependent one and ask: At a given resolution ℓ > 0, can any incompressible flow be represented by a smooth avatar that agrees with all observables accessible at that resolution? We formalize this question using blur and Blurrichevsky geometry. Conceptually, this is the continuum analogue of the blur–based treatment of Poincaré and Collatz: we insert an explicit resolution parameter and prove theorems at fixed blur instead of pretending we can access the ℓ= 0 world. Scope and what we are not claiming To avoid any ambiguity, let us state explicitly what this note does not claim. We do not assert that the Navier–Stokes existence and smoothness problem (in the Clay formulation) is ill-posed, meaningless, or independent of standard axioms. We also do not propose a proof of either global regularity or finite-time blow-up. Our stance is purely epistemic and physical: for every fixed blur scale ℓ > 0and finite time horizon T , the Leray–Hopf framework plus mollification produces smooth blurred avatars uℓ and a finite, trackable “blur budget” that captures all subgrid effects relevant to an observer operating at resolution ℓ . Whatever the ultimate mathematical status of the ℓ→ 0Clay question, these finite-blur statements remain valid and already suffice for all physically realizable measurements. In this sense, the Clay problem is genuinely Platonic: it is a sharp question about the ideal continuum limit at infinite resolution. The blur-based picture developed here is complementary rather than adversarial: it organizes what can be said, and what cannot be improved upon, at any fixed positive resolution, without taking a position on whether a globally smooth solution exists at resolution 0. 2 2 Blur operators and Blurrichevsky observers 2.1 Spatial blur via mollifiers or heat kernels Let ϕ∈C∞ c ( R3 )be a standard nonnegative mollifier with RR3ϕ ( x ) dx = 1 and ϕ ( x ) = ϕ ( −x ). For ℓ > 0set ϕℓ(x):=ℓ−3ϕx ℓ,(Bℓf)(x) := (ϕℓ∗f)(x) = ZT3 ϕℓ(x−y)f(y)dy. Alternatively, we may take (B ℓf )( x )=( Gℓ2∗f )( x )where Gt is the heat kernel at time t . Both families {Bℓ}ℓ>0have the usual properties of an approximate identity: •∥Bℓf∥Lp≤ ∥f∥Lpfor 1≤p≤ ∞, •Bℓf→fin Lpas ℓ↓0when f∈Lp, •Bℓfis smooth for ℓ > 0whenever f∈Lpfor some p. We interpret ℓ asablur scale: an observer who cannot resolve spatial features below the length ℓperceives fonly through Bℓf. Definition 2.1 (Blurred velocity field).Given a (weak) velocity field u ( t, x )and blur scale ℓ > 0, define the blurred velocity uℓ(t, x) := (Bℓu(t, ·))(x). 2.2 Blurrichevsky metric on flows We now formalize the scale–dependent point of view as a Blurrichevsky geometry on the space of flows. Fix ℓ>0and a time horizon T > 0. Definition 2.2 (Blurrichevsky distance at scale ℓ ).For two velocity fields u, v ∈L2 ([0 , T ]; L2 ( T3 )) define dℓ(u, v) := ZT 0 ∥Bℓ(u(t)−v(t))∥2 L2(T3)dt1/2 . Remark 2.3 (Observer–dependence).An observer is specified by the triple ( ℓ, T, O )where O is a class of observables Φthat depend on uonly through Bℓuon [0, T ], and are Lipschitz in L2: |Φ(u)−Φ(v)|≤LΦdℓ(u, v). For such an observer, two flows with dℓ(u, v)≤εare ε–indistinguishable on [0, T ]. Proposition 2.4 (Blurrichevsky equivalence).Let Φbe an observable depending only on B ℓu on [0, T ]and Lipschitz with constant LΦas above. If dℓ(u, v)≤ε, then |Φ(u)−Φ(v)|≤LΦε. In particular, if dℓ(u, v) = 0 then Φ(u) = Φ(v)for all such observables. Proof. By assumption Φ(u) = Φ(Bℓu), similarly for v, and |Φ(u)−Φ(v)|=|Φ(Bℓu)−Φ(Bℓv)| ≤ LΦdℓ(u, v). 3 Definition 2.5 (Blur–equivalence class).The Blurrichevsky equivalence class of a flow u at scale ℓon [0, T ]is [u]ℓ,T := {v:dℓ(u, v) = 0}. An observer at resolution ℓ and horizon T can only distinguish equivalence classes [ u ] ℓ,T , not individual representatives. The key question now becomes: given a (possibly nonsmooth) weak solution u , what can be said about the structure of [ u ] ℓ,T ? In particular, does it contain smooth representatives, and how do they evolve? 3 Blurred Navier–Stokes dynamics We now apply Bℓto a Leray solution and derive the filtered Navier–Stokes equation. 3.1 Filtered equation and Reynolds stress Let ube a Leray–Hopf weak solution on [0,∞)×T3. Define uℓ=Bℓuand pℓ=Bℓp. Lemma 3.1 (Filtered Navier–Stokes).For each fixed ℓ>0,uℓsatisfies the filtered system    ∂tuℓ+ (uℓ· ∇)uℓ+∇pℓ=ν∆uℓ−∇·Rℓ, div uℓ= 0,(2) in the sense of distributions, where the Reynolds stress Rℓis given by Rℓ:= Bℓ(u⊗u)−uℓ⊗uℓ. Proof. Apply B ℓ to (1) (in distribution form). Using linearity and the fact that B ℓ commutes with spatial derivatives, we get ∂tuℓ+Bℓ(u· ∇)u+∇pℓ=ν∆uℓ. Write u⊗ufor the matrix with entries uiuj. Then (u· ∇)u=∇·(u⊗u), and Bℓ(u· ∇)u=Bℓ(∇·(u⊗u)) = ∇ · Bℓ(u⊗u). Add and subtract ∇·(uℓ⊗uℓ): Bℓ(u· ∇)u=∇·(uℓ⊗uℓ)+∇ · Bℓ(u⊗u)−uℓ⊗uℓ= (uℓ· ∇)uℓ+∇ · Rℓ. Moving ∇·Rℓ to the right–hand side gives (2) . Divergence–freeness of uℓ follows from div u = 0 and the commutation of Bℓwith spatial derivatives. 3.2 Regularity of blurred flows The blurred field uℓis much more regular than u, even if uis only a Leray solution. Proposition 3.2 (Spatial smoothness of uℓ ).Let u be a Leray–Hopf solution and ℓ > 0. Then for each t>0,uℓ(t, ·)∈C∞(T3), and for every integer k≥0, ∥∇kuℓ(t, ·)∥L2(T3)≤Ck(ℓ)∥u(t, ·)∥L2(T3), with Ck(ℓ)depending only on ϕ,k, and ℓ. 4 Proof. For each fixed t , u ( t, · ) ∈L2 ( T3 ). Since ϕℓ∈C∞ c and convolution with a smooth kernel smooths, Bℓu(t, ·)∈C∞. Differentiating under the integral, ∇kuℓ(t, x) = (∇kϕℓ)∗u(t, ·)(x), and Young’s inequality gives ∥∇kuℓ(t, ·)∥L2≤ ∥∇kϕℓ∥L1∥u(t, ·)∥L2. Set Ck(ℓ) := ∥∇kϕℓ∥L1. Remark 3.3 (Time regularity).Time regularity of uℓ can be obtained from the weak time regularity of u together with the smoothing effect of B ℓ and standard interpolation. For our purposes it suffices that for each ℓ>0and T > 0, uℓ∈C([0, T ]; Hm(T3)) for all m≥0, with norms controlled by the Leray energy bounds and Cm(ℓ). Thus, for each fixed blur scale ℓ > 0, the blurred flow uℓ is a smooth avatar of the weak solution u . All potential singularities of u are hidden in the Reynolds stress Rℓ and in the high–frequency remainder u−uℓ. 3.3 Blurred energy balance and budget Applying the usual energy method to (2) yields a blurred energy balance. Proposition 3.4 (Blurred energy inequality).Let u be a Leray–Hopf solution and ℓ > 0. Then for almost every t≥0, 1 2∥uℓ(t)∥2 L2+νZt 0 ∥∇uℓ(s)∥2 L2ds ≤1 2∥uℓ(0)∥2 L2+Zt 0ZT3 Rℓ(s, x) : ∇uℓ(s, x)dx ds. (3) Proof. Multiply (2)byuℓand integrate over T3: 1 2 d dt∥uℓ∥2 L2+ν∥∇uℓ∥2 L2=−ZT3 Rℓ:∇uℓdx, where we used R ( uℓ· ∇ ) uℓ·uℓdx = 0 and R∇pℓ·uℓdx = 0 by divergence–freeness. Integrating over time gives (3). The right–hand side measures the energy flux from resolved to unresolved scales mediated by Rℓ . For each fixed ℓ > 0, one can bound this flux by a function of the Leray energy and the blur scale, producing a blur budget Bℓ(t) := Zt 0ZT3|Rℓ(s, x)| |∇uℓ(s, x)|dx ds, which controls how much energy can leak into scales smaller than ℓ. Remark 3.5 (No free information in the blur).The blurred energy inequality expresses a simple principle: any additional structure at scales below ℓ (encoded in Rℓ ) must be paid for by a corresponding blur budget Bℓ . There is no way to get extra small–scale information for free without either increasing Bℓor changing the blur scale. 5 4 Smoothness at finite resolution We now summarize what the previous section implies for Blurrichevsky observers. Theorem 4.1 (Smooth blur avatars).Let u be a Leray–Hopf solution on [0 ,∞ ) ×T3 . Fix ℓ > 0 and T > 0. Then: 1. The blurred flow uℓbelongs to C∞((0, T ]×T3). 2. For any observable Φdepending only on u through B ℓu on [0 , T ]and Lipschitz in L2 , the value Φ(u)is completely determined by the smooth field uℓ. 3. Any other Leray solution v with the same blurred initial data uℓ (0) = vℓ (0) and the same blur budget Bℓon [0, T ]satisfies Φ(u) = Φ(v)for all such observables. Proof. (1) is Proposition 3.2 and the ensuing time regularity remark. For (2), by assumption Φ( u ) = Φ(B ℓu )on [0 , T ], and B ℓu = uℓ is smooth. For (3), if v is another Leray solution with the same blurred initial data and blur budget, then uℓ and vℓ solve the same filtered equation with the same initial condition and same Reynolds stress budgets. Under mild additional assumptions (e.g. Gronwall–type control on the difference of filtered flows), one can show uℓ=vℓon [0, T ]. Then Φ(u) = Φ(uℓ) = Φ(vℓ) = Φ(v). Remark 4.2 (Physical smoothness versus Platonic smoothness).From the standpoint of an observer at resolution ℓ and horizon T , the only accessible object is the equivalence class [ u ] ℓ,T , which contains the smooth representative uℓ . Any potential singularities of u at scales ≪ℓ are hidden in the unresolvable part of the Reynolds stress and cannot be distinguished from other members of the same class. In this sense, the flow is physically smooth, even if the PDE admits blow–up at ℓ= 0. 5 Blur, uncertainty, and effective randomness in Navier–Stokes Blur also clarifies how random–looking behavior can arise inside a fully deterministic equation such as Navier–Stokes. 5.1 Uncertainty budgets in fluid descriptions Three structural limitations conspire in the fluid context: 1. Uncertainty principle type constraints: sharp localization in physical space implies broad support in frequency space and vice versa. In turbulence, fine localization in vortices forces an increasingly wild distribution of frequencies. 2. Energy constraints: generating finer structures requires energy. The cascade of energy to small scales is bounded by global conservation/dissipation laws. 3. Finite observational capacity: any experiment samples the flow at a finite number of points, with finite bandwidth and finite time. Blur packages these facts into a single parameter ℓ > 0and a budget Bℓ . Once ℓ and Bℓ are fixed, there is a hard ceiling on how much small–scale structure can be extracted or even matter for observables. 6 5.2 Effective randomness from unresolved scales Even though Navier–Stokes is deterministic, a Blurrichevsky observer at fixed ℓ and T will typically report random behavior in certain channels: •many different microscopic configurations ubelong to the same blur class [u]ℓ,T ; •their differences are invisible to all observables in O; •the unresolved part behaves, from the observer’s point of view, like a source of noise. This is effective randomness. It does not depend on metaphysical indeterminism: it is a structural feature of the information geometry of Navier–Stokes under blur. Blur makes this precise. Decompose u=uℓ+ (u−uℓ), where uℓ is the smooth resolved part and u−uℓ is the high–frequency remainder. Under the blurred energy inequality (3) , the remainder carries at most a controlled amount of energy. Within the blur constraints, it can emulate many different small–scale patterns without changing any observable in Oby more than the Lipschitz constant times the blur budget. Remark 5.1 (No free sharpening).One might hope that by switching lenses (e.g. changing coordinates or variables) we could recover this apparent randomness as hidden determinism. Blur shows that this is illusory: any lens compatible with the same physical energy and measurement constraints suffers an equivalent uncertainty budget. There is no free reparametrization that reveals the full microscopic picture without paying for it in another channel. 5.3 Gaussian blur versus classical transforms At a formal level, the convolution blur B ℓ is as compatible with Fourier or Mellin analysis as any other smoothing; in fact, Gaussian blur is particularly convenient in frequency space. The conceptual difference is: • classical transforms are usually introduced to solve the equation by decomposing u into eigenmodes and evolving coefficients; • blur is introduced to bound what can be known at a given resolution and to isolate the part of uthat must remain effectively random. In this sense, blur behaves like a meta–transform: it sits above the choice of basis and tracks the information budget that all bases must respect. 6 What this does and does not say about the Clay problem It is important to be explicit about the scope of the blur viewpoint. 6.1 Clay smoothness at ℓ= 0 The Clay problem asks, informally: Does there exist a globally smooth solution for every smooth initial datum, when we interpret (1) as an exact continuum PDE at infinite resolution? This is the question of smoothness at blur scale ℓ = 0. Our arguments do not answer this: they show that for any fixed ℓ > 0and any Leray solution u , the blurred avatar uℓ is smooth and sufficient for all observers limited to that resolution. Whether u itself remains smooth as ℓ↓0is a separate, and still open, issue. 7 6.2 Physical completeness at finite blur From a Blurrichevsky perspective, the physically relevant statement is: For each experimentally meaningful resolution ℓ > 0and any finite time horizon T > 0, every Leray solution u admits a smooth avatar uℓ∈ [ u ] ℓ,T that captures all observables accessible at that resolution, and any further structure at scales < ℓ is hidden behind the blur budget Bℓ. In other words, blur provides a complete and honest description of what fluid dynamics can ever tell us, without assuming that the underlying continuum PDE is globally smooth. 6.3 Two types of questions Thus, the NS landscape separates into two kinds of questions: 1. Blur–aware questions (physically grounded): how do the blur classes [u]ℓ,T behave as ℓ and T vary? Can we classify them, bound their budgets, and prove stability and universality properties? These questions are directly accessible to observation and experiment. 2. Blur–blind questions (Platonic): what happens at the limit ℓ = 0? Is every Leray solution smooth, or can singularities form at arbitrarily small scales? These questions are mathematically sharp but physically unreachable. Blur does not trivialize the Clay problem; it reframes it. It makes visible that the existence and smoothness problem lives entirely at the boundary of our information horizon, while the interior of that horizon can be organized and understood in a scale–aware way. 7 Finite blur–families for Navier–Stokes flows We now state the Navier–Stokes analogue of the “finite families under blur” principle. The setting is deliberately abstract, so that it covers the usual 2D and 3D incompressible Navier–Stokes equations on a bounded domain, but also other dissipative fluid models. 7.1 Setting and blur operator Let Ω ⊂Rd be a bounded domain, d∈ { 2 , 3 } , and let X be a Banach (typically Hilbert) space of divergence–free velocity fields on Ω(for instance L2 σ(Ω) or H1 0,σ(Ω)). Denote by S(t) : X→X, t ≥0, the solution operator of the incompressible Navier–Stokes equations, so that for each initial datum u0∈Xfor which the problem is well posed one has a trajectory u(t;u0)=S(t)u0, t ∈[0, T ], on some time interval [0, T ]. We assume: • there is a bounded set of admissible initial data K⊂X such that for each u0∈K a unique solution S(t)u0exists on [0, T ]and remains in a bounded subset of X; •the map (t, u0)7→ S(t)u0is continuous on [0, T ]×K; •there exists a constant CT>0such that ∥S(t)u0−S(t)v0∥X≤CT∥u0−v0∥X∀u0, v0∈K, ∀t∈[0, T ]. 8 These are the standard local well–posedness and continuous–dependence properties for Navier– Stokes on a bounded time interval and bounded set of initial data. To model blur, we introduce a linear operator Br:X−→ Y, where Y is another Banach space (for example Y = RN for a finite set of Fourier coefficients, or a lower–regularity Sobolev space), and r > 0is a blur scale encoded in the choice of Br . We assume: •Bris bounded with ∥Br∥≤1; •Bris either (i) finite rank (e.g. projection onto the first Neigenmodes), or (ii) compact (e.g. convolution with a smooth kernel on Ω, or X ,→Y a compact embedding). For each initial datum u0we define the blurred trajectory t7−→ BrS(t)u0∈Y, t ∈[0, T ], and measure distances between blurred trajectories with the sup norm: dr,T (u0, v0) := sup 0≤t≤T BrS(t)u0−BrS(t)v0 Y. 7.2 Finite blur–families for Navier–Stokes We can now formulate the Navier–Stokes version of the finite blur–families theorem. Theorem 7.1 (Finite blur–families for Navier–Stokes flows).Let K⊂X be a bounded set of admissible initial data and S ( t )a Navier–Stokes solution operator on [0 , T ]satisfying the continuity and Lipschitz dependence assumptions above. Let Br : X→Y be a blur operator of finite rank or a compact operator. Then for every tolerance ε > 0there exist initial data u1, . . . , uN∈K such that for every u0∈Kthere is an index j∈ {1, . . . , N}with dr,T (u0, uj) = sup 0≤t≤T BrS(t)u0−BrS(t)uj Y≤ε. In words: at blur scale r and resolution ε on the time interval [0 , T ], the set of incompressible Navier–Stokes flows with initial data in K decomposes into finitely many blur–families, each represented by one of the trajectories t7→ BrS(t)uj. Idea of proof. We consider the set of blurred trajectories F:= t7→ BrS(t)u0:u0∈K⊂C([0, T ]; Y). By continuity of S ( t )in both t and u0 , and boundedness of K , the family {BrS ( t ) u0 : u0∈K} is bounded and equicontinuous in t (the Lipschitz dependence in u0 and the boundedness of the flow on [0, T ]give a uniform modulus of continuity in time). If Br is finite rank, then its range is finite–dimensional, and the Arzelà–Ascoli theorem implies that F is relatively compact in C ([0 , T ]; Y ): the set is bounded, equicontinuous, and pointwise relatively compact. If Br is compact, the same holds: for each fixed t , the set {BrS ( t ) u0 : u0∈K} is relatively compact in Y because Br maps bounded subsets of X to relatively compact subsets of Y , and equicontinuity in t is unchanged. Again Arzelà–Ascoli yields relative compactness of Fin C([0, T ]; Y). 9