On the Mathematical Analysis and Physical Implications of the Principle of Minimum Pressure Gradient

On the Mathematical A nalysis and Ph ysical Implications of the Principle of M inim um Pressure Gradien t By Haithem E. T aha Abstract In this pap er, we esta blish a t w o-w a y equiv alence b et w een the incompressible Na vier- Stok es equatio n (INSE) and the principle of m in im um p ressure gradien t (PMPG). W e pro v e that a candidate smo oth flo w fi eld is a solution of the INSE if and only if its instan taneous ev olution min imizes, at ev er y instan t, th e norm of the pressur e force, required to enforce incompr essibilit y . W e sho w that the PMPG is p recisely the mini- mization form ulation of the Lera y-Helmholtz pro jection. An y admissible i nstan taneous ev olution (e.g., onset of separation) resulting from the INSE n ecessarily minimizes the PMPG cost. Con v ersely , any other kinematical ly admissible evolutio n, r equiring a strictly larger pressure force to ensure the same constrain ts, d o es not satisfy the INSE. Th us, th e PMPG offers a v ariational p ersp ectiv e through whic h intricate in compress- ible flow b eha viors may b e inte rpreted. In a finite-dimensional setting with diverge nce-free mo d es, we sho w that the PMPG yields the same dynamics as classical Galerkin pro jection. Moreo v er, the P MPG p ro- vides a natural generalization of classical Galerkin pro jection b eyo nd linear mod al expansions, accommo dating nonlinear and n on-mo dal representati ons. W e then exam- ine the r elation b et we en instan taneous dynamical minimization and steady v ariational selection, including its connection to the v ariational theory of lift. Motiv ated by these observ ations, we formulate conjectures concernin g necessary conditions for stabilit y and the con vergence of Navier–Sto k es solutions to Eu ler’s in the v anishing-viscosit y limit. 1 In tro duct ion Gauss’s principle of least constrain t tr a nsforms a constrained dynamics pro blem in to pure minimization. It asserts that a mec hanical sys tem ev olv es from one time instan t to another in the closest p ossible manner to its fr e e motion —the motio n that w ould o ccur in the ab- sence of constrain ts [1 – 4]. More precisely , among all kinematically admissible motions, the actual motio n is the one for whic h the deviation from the free motion is minimal at eac h instan t. This deviation is directly prop ortional to the constrain t forces, required to ensure the kinematical and geometrical constraints. In this sense, Gauss’s principle asserts that a mec hanical system ev olv es b y minimizing the mag nitude of the constraint force. An y alter- nativ e ev olution would require an unnecessarily lar ger constrain t force to main tain the same constrain ts. It can b e sho wn that the first-order necessary condition of optimalit y for the Gaussian cost is New ton’s equations of motion [5]. That is, If the acce leration of a tra jectory minimizes the Gaussian cost at ev ery instan t, among all kinematically admissible acce lerations, then the tra jectory neces sarily satisfies Newton’s equations of motion. This result can b e strengthened further for mec hanical systems sub ject to smo ot h forcing functions, fo r whic h solutions of Newton’s equations are unique. In this case, a candidate admissible tra jectory is a solution of Newton’s equations o f motion if and only if its acceleration minimizes the G a ussian cost at ev ery instan t. 1 In this pap er, w e demonstrate this minimization framew ork of Gauss’s principle using the double-p endulum as a concrete example. At a giv en instant, f o r whic h the t w o angles ( θ , φ ) and angular v elo cities ( ˙ θ , ˙ φ ) ar e presc rib ed, there ex ist infinitely many p o ssible ev olutions; i.e. accelerations ( ¨ θ , ¨ φ ); all of whic h are kinematically admissible. Eac h suc h ev olution, how eve r, requires a differen t magnitude of constrain t fo rce to main ta in the p endulum constrain ts. In other w ords, on the t w o- dimensional configuration manifold (the torus in this case), there are infinitely man y admissible directions in the tangent bundle, y et only one evolution o ccurs in realit y . It is t he one t hat satisfies Newton’s equations of motion. Equiv a len tly , according to G auss’s principle, it is the ev olution that requires the smallest p ossible constrain t fo r ce to maintain the p endulum constrain ts. An y alternativ e ev olution w ould necess arily require a strictly larger constraint for ce to ensure the same constraints. In the absence of impresse d fo r ced (e.g., p endulum in the horizon tal plane), this mini- mization framew ork implies motion a long ge o desics (i.e., straight lines) on the configuration manifold. That is, the tra jectories of motion are the least curv ed paths on the configura- tion manifold, followin g Hertz’s principle of least curv ature [3]. If the syste m is sub ject t o impressed forces, then G a uss’s principle b ecomes the minimization framew ork of pro jecting the impressed forces on to the configuratio n manifold. Gauss’s principle w as recently extended from particle mec hanics to con tinu um mec hanics of incompressible flo ws, leading to the principle of minim um pressure gradien t (PMPG) [6]. Recognizing that, for incompressible flows, the pressure force plays the role of a constraint force that main tains the con t inuit y constrain t, t he PMPG asserts that an incompressible flo w ev olve s from one instan t to the next b y minimiz ing the L 2 -norm o f the pres sure force, requ ired to ens ure incompressibilit y . Similar to the fact that Newton’s equ ations of motion arise as the first-order necessary condition of optimality for the Gaussian cost in particle mec hanics, it w as sho wn that the Na vier-Sto k es equation is the first-or der necessary condition of o ptimalit y for the PMPG cost. That is, if the lo cal acceleration o f an incompressible flo w minimizes the L 2 -norm of t he pr essure force at ev ery instant, among all div ergence-free accelerations, then the resulting flow field necessarily satisfies the Na vier-Stokes equation. In this pap er, w e strengthen this result b y prov ing that a candidate smo oth flo w field is a solution of the Navie r-Stok es equation if and only if its lo cal acceleration minimizes the L 2 -norm of the pressure force at eve ry instant. This mathematical fact establishes a t wo- w ay equiv alence b et w een the Navier-Stok es equation (NSE) and the PMPG minimization framew ork. In p articular, if the acceleration of a flo w field minimizes the PMPG cos t at ev ery instan t, then the flow necessarily satisfies the NSE; con v ersely , if a t any instan t the lo cal acceleration requires a larger pressure-gradient fo rce than another kinematically a dmissible ev olution, then the corresp onding flow field cannot b e a solution of the NSE. This t wo-w ay equiv alence suggests that the PMPG ma y b e view ed a s a causal mechanis m for incompressible flo w phenomena. If a smo oth incompressible flow field b eha v es a certain w ay , w e may attribute such a behavior to the minimization of the pressure gradien t, since an y alternativ e b eha vior w ould require an unnecessarily larger pr essure force to main t a in incompressibilit y a nd t herefore, b y virtue of t he tw o-w a y equiv alence, could not b e a solutio n of the NSE. Noting that any o rthogonal pro jection can b e form ulated as a minimization problem, w e sho w that the PMPG pro vides the minimization framew ork underlying the Lera y pro jection: pro jecting the NSE onto the space of divergenc e-free fields [7 – 11]. W e also clarify sev eral 2 common p oints of confusion a rising from conflating the PMPG with the principle of sta- tionary action or with the Diric hlet principle asso ciated with t he pressure P oisson equation in pro jection metho ds. W e conclude b y p osing conjectures that ma y b e of in t erest for fur- ther mathematical analysis, including an invisc id stabilit y criterion a nd the conv ergence of Na vier-Sto k es solutions to Euler’s. 2 Gauss’s Prin ciple of Le ast Cons train t In his four-page philosophical note, published in a journal that still exists to da y , Gauss [1] p ostulated one of the fundamen tal principles of mechanics . Conside r a particle of mass m sub ject to an impr esse d force F , as illustrated in Fig . 1. In the absence of constrain ts, it w ould accelerate in the direction of the f o rce with acceleration a free = F m . Gauss re- ferred to this motion as the fr e e motion —the motio n tha t would o ccur in the absence of constrain ts. Ho we v er, if the par t icle is constrained to accelerate within some instan taneous plane of admissible motion defined by the constraint [ A ] a = b , then the actual motion m ust necessarily deviate from the free motion. Since this deviation arises solely due to the con- strain t, Gauss’s profo und insigh t was that it m ust b e the le ast dev iation compatible with the constrain t. Nature will not o verdo it. He wrote: “ The motion of a system of N material p oin ts takes plac e in every moment in maximum ac c or danc e with the fr e e movement or under le ast c onstr aint, the me asur e of c o nstr aint, is c onsider e d as the sum of p r o ducts of mass and the squar e of the deviation to the fr e e motion. ”  − ∗  ∗ −     Plane of Admissible Motions:     Figure 1: Sc hematic illustration of the in- stan ta neous plane of admissible motion, de- fined by (2), a t a pa rticular configuratio n. In- finitely man y instan ta neous motions lie this plane (dotted v ectors), eac h satisfying the lin- ear constrain t (2) a t the exp ense of a con- strain t f orce R . Gauss’s principle asserts that, among all kinematically admissible motions, Nature selects the one that requires the least constrain t force. This minimization is simply equiv alen t to pro jecting the impressed force F on t o the pla ne of admissible motions. Gauss’s principle admits an equally illumi- nating in terpretation. On the instantaneous admissible plane of motion, there exist in- finitely many candidate a ccelerations, as il- lustrated in Fig . 1. Each candidat e requires a sp ecific c onstr aint force R suc h that, when com bined with the impressed for ce F , the resulting motion m a = F + R satisfies the constraint. The fo rce R exists solely to enfor ce the constrain t; its r aison d’ˆ etr e is the constraint itself—if t he con- strain t is remo v ed, R v anishes. Gauss then asserted that Nature selects the motio n re- quiring the smallest p ossible magnitude of the constrain t force necessary to enfor ce the constrain t, hence the term L e ast Cons tr aint . An y alternativ e candidate would demand an unnecess arily la r g er constraint fo rce to en- sure the same constrain t, and is therefore not r ealized. 3 Jacobi [12] later gav e Gauss’s principle an explicit mathematical formulation b y in tro - ducing the quadratic cost function: Z = 1 2 N X i =1 m i     a i − F i m i     2 , (1) where N denotes the num b er of particles and a i is t he inertial acceleration of the i th par- ticle. According to Gauss’ principle, Z m ust b e minim um at ev ery instant, prov ided that the constrain ts are satisfied. Assume the pa r ticles ev olv e in a d - dimensional space so that the stac k ed acceleration ve ctor a = [ a T 1 , ..., a T N ] T ∈ R dN collects all inertial accelerations. Supp ose the system is sub ject to c ≤ dN constrain ts, p ossibly nonlinear in p ositions and v elo cities. D iff erentiation of the constrain t equations with resp ect t o time yields a relation linear in the accelerations: A ℓj a j = b ℓ (2) for some A ∈ R c × dN , b ∈ R c . Under Newtonian mec ha nics, solving the dynamics requires dN + c equations in dN + c unkno wns. O ne has the dN equations of mot io n m i a i = F i + R i ∀ i = 1 , .., N , (3) together with the c constraint equations (2). The unkno wns are the dN accelerations a and the c indep enden t comp onen ts of the constraint forces R i . Gauss’s principle transforms this dynamic s problem in to the follo wing minimization prob- lem: min a Z ( a ) s.t. [ A ( x , v )] a = b , (4) In our recent effort [5], we sho w ed that this minimization pro blem is a strongly conv ex quadratic programming problem and therefore admits a uniq ue solution. Moreo v er, its first-order necess ary conditions of optimalit y coincide precisely with Newton’s equations of motion. Th us, the unique solution of G auss’s minimization problem (4) necessarily satisfies Newton’s equation of motion.             ℓ 1 ℓ 2 Figure 2: A Sche matic for a double- p endulum oscillating in a ho rizon ta l plane. Double P endulum Example: Because Gauss’s principle is rarely cov ered in engineering graduate cur- ricula, it is instructiv e to illustrate it through a simple example t ha t also clarifies its minimization structure. Consider a double p endulum constrained to mov e in a horizon ta l plane, as sho wn sc hematically in Fig. 2. The masses m 1 and m 2 are constrained to remain at fixed distances ℓ 1 and ℓ 2 from their resp ectiv e hinges. These geometric constrain ts giv e rise to constrain t forces R 1 and R 2 . In the absence of impresse d forces, the fr e e mo tion (that w ould o ccur in the absence of the constrain t) corresp onds to zero acceleration. The Gaussian cost therefore reduces to Z = 1 2  m 1 a 2 1 + m 2 a 2 2  . (5) 4 F o r simplicit y , assume unit masses and unit lengths. The inertial accelerations a 1 and a 2 can then b e expressed in terms of the generalized co ordinates q = ( θ , φ ) as a 1 = ¨ θ  cos θ − sin θ  − ˙ θ 2  sin θ cos θ  , a 2 = a 1 + ¨ φ  cos φ − sin φ  − ˙ φ 2  sin φ cos φ  . Substituting a 1 and a 2 in to the G aussian cost (5), o ne obtains a quadratic for m in the accelerations ( ¨ θ , ¨ φ ): Z = 1 2  ¨ θ ¨ φ   2 cos( θ − φ ) cos( θ − φ ) 1   ¨ θ ¨ φ  + sin( θ − φ )  ˙ φ 2 − ˙ θ 2   ¨ θ ¨ φ  + h, (6) where h dep ends on q = ( θ , φ ) and ˙ q = ( ˙ θ , ˙ φ ) but is independen t of the accelerations ¨ q = ( ¨ θ , ¨ φ ). The dynamics problem is therefore transformed, via Gauss’s principle, to the minimiza- tion problem min ( ¨ θ , ¨ φ ) Z ( ¨ θ , ¨ φ ) whose first-order necessary conditions fo r optimalit y are simply: ∂ Z ∂ ¨ θ = 0 and ∂ Z ∂ ¨ φ = 0 , yielding  2 cos( θ − φ ) cos( θ − φ ) 1   ¨ θ ¨ φ  = sin( θ − φ )  − ˙ φ 2 ˙ θ 2  , (7) whic h coincide precisely with Newton’s equations of motio n for the double p endulum. It ma y b e imp or tan t to emphasize the insigh t gained from Gauss’s principle as it p ertains to the core discussion in this pap er. W hile Gauss’s principle and Newton’s equations of motion are mathematically equiv alen t , the minimization f o rm ula t ion of Gauss’s princip e ma y pro vide additional insigh t. F or example, at a given p oint ( q , ˙ q ) in the tangent bundle, there are infinitely many kinematically-admissible ev olutions ( ¨ θ , ¨ φ ), a s sho wn in Fig. 3(a). In the co ordinate-system of q = ( θ , φ ), the p endulum constrain ts are automatically satisfied. F o r a g iv en p oint ( q , ˙ q ) at a particular instant, there a re infinitely man y p ossible ev olut io ns that resp ect the pendulum constraints. How ev er, eac h kinematically-admissible ev olution requires certain constraint fo rces R 1 , R 2 to maintain tang ency of the v elo cit y ve ctor to the configuration manifold. Figure 3(b) sho ws con tours of the Gaussian cost Z in Eq. (6) in the acceleration space ( ¨ θ , ¨ φ ) at the p oin t q = (0 ◦ , 5 ◦ ), ˙ q = (1 , 1). Each admissible ev olutio n, represen ted by a candidate a cceleration ( ¨ θ , ¨ φ ), corresp o nds to a certain magnitude Z o f the constrain t forces, required to main tain the p endulum constrain ts. The minimizing a cceleration ( ¨ θ ∗ , ¨ φ ∗ ) co- incides precisely with that obtained from Newton’s equations of motio n. An y alternativ e candidate ( ¨ θ , ¨ φ ) requires constrain t forces o f strictly larger magnitude. Th us, G a uss’s prin- ciple is not merely predictiv e; it also suggests a causal mec hanism underlying the motion. F r om G a uss’s p ersp ectiv e, the motion is driv en b y minimizing the magnitude of the con- strain t forces, required to main tain t he constrain ts. Only this candidat e ev olution satisfies 5 ( ∗  ∗ ) ( ,  ) Plane of Admissible Accelerations (  ,  ) (  ,  ) (a) The plane of admissible accelera tions. -1 -0.5 0 0.5 1 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 2.5 3 3.5 4 4.5 Contours of Z Newtonian Evolution Gaussian Evolution Minimum of |a| 4 (b) V ariation o f Z with ¨ θ , ¨ φ . Figure 3: A Sc hematic for t he plane of admissible accelerations at a g iv en p oint ( q , ˙ q ) in the tangen t bundle, along with t he con tours of the Gaussian cost Z in Eq. (6) with the accelerations ¨ θ , ¨ φ at t he p oint q = (0 ◦ , 5 ◦ ), ˙ q = (1 , 1). The minimizing a ccelerations ( ¨ θ ∗ , ¨ φ ∗ ) satisfy Newton’s equations o f motion. Ho w ev er, the ones that minimize the 4-norm P i m i a 4 i do not necessarily coincide with the Newtonian ev olution. Newton’s equations of motion. An y alternativ e ev olution would demand an unnecess ar- ily larger constrain t force to ensure the same constrain ts, and w o uld not satisfy Newton’s equations. In the absence of impressed f o rces, as in the horizon tal-plane double p endulum, Gauss’s principle yields least-curv ed tra jectories. These a r e the closest tra jectories to straight lines in the a mbien t Euclidean space; equiv alen tly , they are geo desics on the configuration manifold itself [13, 14]. If the double p endulum is instead considered in the v ertical plane, where it is sub ject to an impressed gr avitational force, G auss’s principle reduces to the orthogonal pro jection of the impressed force onto the plane of admissible accelerations. 3 The Principl e of Minim um Pressure Gradien t F o r incompressible flo ws, the G aussian cost admits a natural extension to the contin uum setting [6]: Z = 1 2 Z Ω ρ | a − F | 2 d x , (8) where Ω denotes the spatial domain, F is the impressed f orce p er unit mass, and a = u t + u · ∇ u is the total inertial acceleration of a fluid particle. Incompressible flows are sub ject to pressure forces, viscous forces, and other b o dy forces (e.g., gra vit y or electromagnetism), and are constrained to satisfy the con tin uity equation ∇ · u = 0 together with the no-p enetration bo undar y conditio n. As discussed in the pre ceding section, ev ery constrain t giv es rise to a corresp o nding constrain t force whose sole role is to enforce that constrain t. F or the con tin uity constrain t, the Helmholtz decomposition rev eals the underlying geometry of incompressible flo ws and the nature of the asso ciated constrain t 6   Space of Divergence - Free Fields  ∙  0  Gradient Fields (a) Sc hematic of the Helmholtz decomp osition.     Space of Divergence - Free Fields  ∙  =   Gradient Fields   −        (b) Geometry of incompr essible flows. Figure 4: A Sche matic diagram illustrating the Helmholtz-Leray pro jection and geometry o f incompressible flo ws. force [11, 15, 16]. Giv en a square in tegrable v ector field v ( x ) o v er a smo oth domain Ω ⊂ R n , it can b e uniquely decomp osed into tw o orthogonal components: (i) a div ergence-free field w that satisfies the no-p enetration b oundary condition w · n = 0 o n ∂ Ω, and (ii) a gradien t field ∇ g , for some scalar function g , as illustrated sch ematically in Fig. 4 (a). Accordingly , v ( x ) = w ( x ) + ∇ g ( x ) ∀ x ∈ Ω . The orthogonality is understo o d in the L 2 sense: Z Ω ( ∇ g · w ) d x = 0 . This orthogonality follows directly from integration b y pa r t s: Z Ω ( ∇ g · w ) d x = − Z Ω   g ∇ · w | {z } =0 in Ω   d x + I ∂ Ω g w · n | {z } =0 o n ∂ Ω ! d x = 0 . (9) This p ersp ectiv e is particularly illuminating for the geometry of incompressib le flows . F r om this viewpoint, the Na vier-Stok es equation ma y b e rear r a nged a s F − u · ∇ u | {z } u free t = u t + 1 ρ ∇ p, (10) Equation (10) rev eals that the dynamic ev olution of an incompressible flo w from one in- stan t to the next may b e in t erpreted as a Helmholtz–Leray pro j ection on to the space of div ergence-free vec tor fields [11, 15 , 16]. F or a giv en smo oth velocity field u at some instan t, b oth the con ve ctiv e acceleration u · ∇ u and the impresse d forces (e.g., viscous or gravita- tional) are kno wn. T og ether, they define a v ector field u free t that admits a unique Helmholtz decomp osition into t wo or t hogonal comp onen ts: (i) a div ergence-free comp onen t satisfying the no-p enetration b o undary condition, represen ted b y the lo cal acceleration u t , and (ii) a 7 gradien t comp onent, represen ted by the pressure gr a dien t ∇ p , as illustrated in the sche matic diagram of Fig. 4(b). The sc hematic illustrates that, on the instantaneous subspace o f admissible motions de- fined b y the div erg ence-free constrain t, there exist infinitely many candidates. Each candi- date deviates f rom the fr e e motion u free t and requires a constraint force to ensure incompress- ibilit y; i.e., to pro ject u free t on t o the space of dive rgence-free fields. According to Ga uss’s principle, Nature selects the motio n w ith the le ast deviation from the free one; equiv alently , it is the motion that requires the smallest p ossible magnitude of the constrain t force required to ensure incompressibilit y . G eometrically , this selection corresp onds to the orthogonal pro jec- tion of the free motion u free t on t o the space of div ergence-free fields. The associated constraint force m ust therefore lie in the orthogonal complemen t of that subspace, namely the space of gradien t fields. Hence, the pressure gradien t emerges as the constrain t f orce enforcing the con tinuit y equation and the no-p enetration b oundary condition. The free motion is precisely the motion that w ould o ccur in t he absence of this constraint fo rce, as giv en b y the left-hand side of (10 ). On the basis of the preceding discussion, the Gaussian cost for incompressible flo ws (ignoring grav itational and electromagnetic forces) can b e written as: A = 1 2 Z Ω ρ   u t + u · ∇ u − ν ∇ 2 u   2 d x , (11) where ν denotes the kinematic viscosit y . This cost was prop o sed b y T a ha et al. [6] as the basis fo r the Principle of Minimum Pr ess ur e Gr adient (PMPG), whic h is the contin uum- mec hanics analogue of Ga uss’s principle of least constrain t. It asserts that an incompressible flo w ev olve s from one instant to the next b y minimizing the mag nitude of the pressure force required to ensure the contin uit y constrain t. An y alternativ e evolution would require an un- necessarily larger pressure force to ensure contin uit y , whic h is against ph ysical considerations as conceiv ed b y G a uss. Similar to the fact that the first-order optimalit y condition of the Gaussian cost Z in (1) r eco ve rs Newton’s equation of motion in particle mec hanics, the first-o rder optimality condition of the PMPG cost functional A in (11) yields the Na vier-Stok es equation. In par- ticular, any incompress ible flow whose evolution minimizes A at ev ery instan t is g uaran teed to satisfy t he Na vier-Sto k es equation, as established in Theorem 1 in [6 , 17] and Prop osition 2 in [5]. Par  cle Mechanics: Newton’ s Equa  on (Principia 1687) Gauss’ s Principle (1828, 1848) Navier-Stok es’ (1822, 1845) The PMPG (2023) Con  nuum Mechanics: Incompressible Figure 5: Sc hematic illustrating the equiv a - lence betw een Newton’s eq uations and Gauss’s principle in particle mec hanics, and the cor- resp onding equiv alence b et we en the Navie r- Stok es equation and the Principle o f Minim um Pressure Gradien t (PMPG) in incompressible con tinuum mec hanics. In this sense, the Principle of Minim um Pressure Gradien t pla ys, within incompress- ible contin uum mec hanics, a role directly analogous to that pla y ed b y Gauss’s prin- ciple in particle mechanics . Figure 5 il- lustrates t his corresp ondence b y sho wing the equiv alence b et w een Newton’s equations and Gauss’s principle in part icle mec hanics, and the analogous relationship b etw een the Na vier-Sto k es equation and the PMPG in the con tinuum mec hanics of incompressible flo ws. 8 In the absence of a ll impressed f o rces suc h as viscous forces and b o dy forces (i.e., for an ideal fluid), the PMPG, similar to Gauss’s principle in particle mec hanics, reduces to a purely inertial minimization. In this case, the resulting ev olution corresp o nds to geo desic flo ws on the configuratio n manifold. This geometric in terpretation is f ully consisten t with Arnold’s seminal result: the tra jectories o f ideal flows are geo desics with resp ect to the righ t-in v aria n t L 2 (kinetic energy) Riemannian metric on the manifold of v olume-preserving diffeomorphisms [18, 19]. 4 Tw o-W a y Equiv alenc e Bet ween th e Na vier-Sto k es and the Princip le of Minim u m Pressure Gradient This section con ta ins one of the cen tral con tributions of the presen t pap er. W e extend the one-w ay implication established in Theorem 1 in [6, 17] and Prop osition 2 in [5] to a t wo- w ay ( if a n d only if ) equiv a lence. W e b egin by stating the assumptions that will b e used throughout this section. General Assumptions : Let ρ > 0 b e a constant and let Ω ⊂ R 3 b e a b o unded domain with smoo t h b oundary ∂ Ω. Assume there exists T > 0 suc h that for all t ∈ [0 , T ], the v elo cit y field u ( x , t ) satisfies: (i) u · ∇ u ∈ L 2 (Ω), (ii) a square in tegrable viscous f orce: ∇ · τ ∈ L 2 (Ω). In addition, assume a square-in tegrable impressed b o dy fo rce f ∈ L 2 (Ω) for all t ∈ [0 , T ]. 4.1 Necessary Condition for Optimalit y Lemma 1 (cf. Theorem 1 in [6]): Recall the General Ass umptions stated ab o ve. In addition, supp ose that the initial conditio n u ( x , 0) is kinematically admissible: ∇ · u ( x ; 0) = 0 ∀ x ∈ Ω and u ( x ; 0) · n = h ( x ) ∀ x ∈ ∂ Ω , (12) where n denotes the out w ard unit normal and h is a prescrib ed smoo th b oundary function. Supp ose that for ev ery t ∈ [0 , T ], the time deriv ativ e u t ( · , t ) ∈ L 2 (Ω) exists and minimizes the functional A ( u t ) = 1 2 Z Ω ρ     u t + u · ∇ u − 1 ρ ( ∇ · τ + f )     2 d x (13) o v er the space of admissible ev olutions: Θ = { v ∈ L 2 | ∇ · v = 0 ∀ x ∈ Ω , and v · n = 0 ∀ x ∈ ∂ Ω } . (14) Then u ( x ; t ) mus t satisfy the Navier-Stok es equations ρ ( u t + u · ∇ u ) = − ∇ λ + ∇ · τ + f and ∇ · u = 0 ∀ x ∈ Ω , t ∈ [0 , T ] (15) for some λ ∈ H 1 (Ω), together with the no-p enetration b oundary condition u ( x ; t ) · n = h ( x ) ∀ x ∈ ∂ Ω t ∈ [0 , T ] . (16) 9 Pro of: Fix t ∈ [0 , T ]. T o minimize A sub ject to the constraint ∇ · u t = 0 for all x ∈ Ω, we construct the Lagra ng ian L ( u t ) = A ( u t ) − Z Ω λ ( x ) ( ∇ · u t ( x ; t )) d x , where λ is a La grange m ultiplier. A necessary condition for the constrained minimization problem is then: the first v aria tion of the L a grangian v anishes with respect to v ariations in u t ( x ) that b elong to Θ . The first v ariation of L with resp ect to u t is written as δ L = Z Ω [ ρ ( u t + u · ∇ u − ∇ · τ − f ) · δ u t − λ ∇ · δ u t ] d x = 0 . (17) Using the iden tity λ ∇ · δ u t = ∇ · ( λδ u t ) − ∇ λ · δ u t , in tegra ting the div ergence term, and applying the div ergence theorem we obtain Z Ω ∇ · ( λδ u t ) d x = Z ∂ Ω λδ u t · n d x . Since admissible v ariations m ust satisfy δ u t · n = 0 on ∂ Ω, the b o undar y term v anishes: R Ω ∇ · ( λδ u t ) d x = 0. Hence, we hav e δ L = Z Ω [ ρ ( u t + u · ∇ u ) + ∇ λ − ∇ · τ − f ] · δ u t d x = 0 Since this holds for all admissible δ u t ∈ Θ, w e obtain ρ ( u t + u · ∇ u ) = − ∇ λ + ∇ · τ + f ∀ x ∈ Ω . This argumen t holds for eac h t ∈ [0 , T ]. In a dditio n, b ecause the initial condition u ( x ; 0) is div ergence-free for all x ∈ Ω and the lo cal acceleration u t ( x ; t ) is div ergence-free f o r all x ∈ Ω and t ∈ [0 , T ], then it implies that ∇ · u ( x , t ) = 0 ∀ x ∈ Ω and t ∈ [0 , T ] . Similarly , since the initial condition u ( x ; 0 ) satisfies the no- p enetration b oundary condition u ( x ; 0) · n = h ( x ) ∀ x ∈ ∂ Ω , and the lo cal acceleration u t ( x ; t ) satisfies a homog enous normal b oundary condition u t ( x ; t ) · n = 0 ∀ x ∈ ∂ Ω and t ∈ [0 , T ] , then it implies that u ( x ; t ) · n = h ( x ) ∀ x ∈ ∂ Ω and t ∈ [0 , T ] , whic h completes the pro of.  10 V erbal Statemen t of Lemma 1: A mo n g al l kinematic al ly-admissible flows—namel y, those that satisfy the c ontinuity c onstr aint a n d the pr escrib e d normal b oundary c ondition (12)—the flow whose evolution u t minimizes the PMPG c ost functional (13) at e a c h instant is guar a n te e d to satisfy the Navier-Stokes e quations . Remark 1: The a b o ve result is indep enden t of spatial dimension and applies to b o t h in viscid a nd viscous flow s. It accommo dates arbitrary viscous stress mo dels that do not dep end explicitly on the lo cal acceleration u t , as w ell as arbitrary square-in tegra ble b o dy forces f . The essen t ial structural requiremen ts are incompressibilit y and a prescrib ed normal b oundary condition (e.g., no-p enetration). Remark 2: The ab o ve r esult extends to time-dep endent domains Ω( t ) ⊂ R 3 with suffi- cien tly smoo t h b oundary motio n. In this setting, the no -p enetration condition b ecomes u ( x , t ) · n ( x , t ) = V b ( x , t ) · n ( x , t ) , x ∈ ∂ Ω( t ) , (18) where V b denotes the prescrib ed b oundary velocity . Differen tiating this condition along the mo ving bo undary yields a prescrib ed normal comp onen t of the lo cal acceleration, u t ( x , t ) · n ( x , t ) = γ ( x , t ) , x ∈ ∂ Ω( t ) , (19) where γ is a kno wn function determined b y the b oundary motion and the curren t v elo cit y field. The admissible set therefore b ecomes Θ( t ) =  v ∈ L 2 (Ω( t ))   ∇ · v = 0 in Ω( t ) , v · n = γ on ∂ Ω( t )  . (20) Since the normal comp onen t o f u t is prescribed, admissible v ariations satisfy δ u t · n = 0 on ∂ Ω( t ) , and the v ariational argumen t pro ceeds unchanged. Remark 3: The ab ov e result do es not require a sp ecific tangen tial b oundary condition (e.g., no-slip). The equiv alence holds for arbitrary tangen tial b oundary conditions compati- ble with the admissible function space. See Corollary 1 b elo w. Corollary 1 : Recall the General Assumptions ab o ve , and assume that the initial condi- tion is kinematically admissible with a no- slip b oundary condition: ∇ · u ( x ; 0) = 0 ∀ x ∈ Ω and u ( x ; 0) = 0 ∀ x ∈ ∂ Ω . (21) Assume that for e v ery t ∈ [0 , T ], the lo cal acceleration u t ( · , t ) ∈ L 2 (Ω) min imizes the functional (13) ov er the admissible set Θ 0 = { v ∈ H 1 | ∇ · v = 0 ∀ x ∈ Ω , and v = 0 ∀ x ∈ ∂ Ω } , (22) Then u necessarily satisfies the Na vier–Stok es equations (15) for some λ ∈ H 1 (Ω), together with the no-slip b oundary conditio n u ( x ; t ) = 0 ∀ x ∈ ∂ Ω t ∈ [0 , T ] . (23) 11 Pro of. The pro of is iden tical t o tha t of Lemma 1, with the admissible set Θ replaced b y Θ 0 . In part icular, admissible v ariations satisfy δ u t = 0 on ∂ Ω, hence δ u t · n = 0 on ∂ Ω, and the b oundary term arising from integration b y parts v anishes.  Remark 4: The corollary extends v erbatim to non-homogeneous no-slip b o undary con- ditions u = u b on ∂ Ω, provide d u b is prescribed and sufficien tly regular. Remark 5: Within this form ulation, the pressure arises naturally as the Lagrange mul- tiplier λ enforcing the kinematic constraints of incompress ibilit y and no-p enetration. This in terpretatio n is consisten t with the classical pro jection and v ariationa l for mulations of in- compressible flo w [7, 8, 10, 11 , 13, 15, 20 – 2 5]. Consequen tly , in t he la ng ua ge of analytical mec hanics, the pressure force represen ts the constrain t fo r ce required to maintain these con- strain ts. 4.2 Existence of Unique Minimizers Lemma 2: Fix t ∈ [0 , T ] and assume the General Assumptions hold at time t . Then there exists a unique minimizer u ∗ t ( · , t ) ∈ Θ o f the functional A in (13) ov er the admissible set Θ defined in (14). Pro of: Fix t ∈ [0 , T ] and define the kno wn v ector field u free t ( t ) := − u · ∇ u + 1 ρ  ∇ · τ + f  ∈ L 2 (Ω) . (24) Then the functional (13) can b e written as A ( v ) = 1 2 ρ Z Ω | v − u free t ( t ) | 2 d x = 1 2 ρ k v − u free t ( t ) k 2 L 2 (Ω) . Hence, minimizing A ov er Θ is equiv alen t to finding the elemen t of Θ closest to u free t ( t ) in the L 2 (Ω) norm. If Θ ⊂ L 2 (Ω) is a c losed subspace, the result follo ws directly from the Pro j ection Theorem in Hilb ert spaces (Theorem I I.3 in [26]): if H is a Hilb ert space and M ⊂ H is a closed subspace, then for ev ery x ∈ H there exists a unique m ∈ M minimizing J ( m ) = k m − x k 2 H o v er M . Θ is a linear subspace of L 2 (Ω): for v 1 , v 2 ∈ Θ and α, β ∈ R , w e ha ve α v 1 + β v 2 ∈ Θ. It remains to sho w b elow that Θ is closed in L 2 (Ω). Let { v k } ⊂ Θ con v erge to v in L 2 (Ω). W e show t hat v ∈ Θ. First, for an y ψ ∈ C ∞ 0 (Ω), con vergenc e in L 2 (Ω) implies Z Ω v · ∇ ψ d x = lim k →∞ Z Ω v k · ∇ ψ d x = 0 , since eac h v k ∈ Θ is div erg ence-free in the we ak sense. Hence ∇ · v = 0 in the distributional sense, and therefore ∇ · v = 0 ∈ L 2 (Ω). In particular, v ∈ H (div; Ω), where H (div; Ω) = { w ∈ L 2 (Ω) | ∇ · w ∈ L 2 (Ω) } . 12 Moreo ver, since ∇ · v k = 0 for all k , k v k − v k 2 H (div;Ω) = k v k − v k 2 L 2 (Ω) + k∇ · ( v k − v ) k 2 L 2 (Ω) = k v k − v k 2 L 2 (Ω) → 0 , so v k → v in H (div; Ω). Since the normal trace map is con tin uous fro m H (div; Ω) to H − 1 / 2 ( ∂ Ω) [27, 28], it follow s that v k · n → v · n in H − 1 / 2 ( ∂ Ω) . Because v k · n = 0 on ∂ Ω f o r all k , w e conclude v · n = 0 as w ell. Therefore v ∈ Θ, and Θ is closed in L 2 (Ω). The pro jection theorem now implies tha t there exists a unique minimizer u ∗ t ∈ Θ of A at eac h t ∈ [0 , T ].  V erbal Stat emen t of Lemma 2: At any given instant whe r e the L 2 -norms of the c onve ctive and visc ous ac c eler ations as wel l as the b o dy for c e ar e wel l-define d, ther e exists a unique evol ution u t that minim i z es the PMPG c ost functional (13) over al l kinematic al ly- admissible e volutions . Remark 6: Lemma 2 is, in essence , a statemen t ab out the Helmholtz-Lera y pro jection (e.g., [29]) of t he free a cceleration u free t , give n in Eq. (24), on to the space of div ergence-free, no-p enetration fields Θ. The minimizer u ∗ t is precisely the L 2 -orthogonal pro jection o f this free acceleration. F urthermore, the Sob o lev regula r ity and smo othness of the minimizer u ∗ t matc hes those of the free acceleration since the Helmholtz-Lera y pro jection is con tin uo us in the corresp onding Sob o lev spaces (see, e.g., [30]). Remark 7: Lemma 2 extends to the no-slip admissible set Θ 0 defined in (2 2 ). In this case, one ma y define V =  w ∈ C ∞ 0 (Ω) 3   ∇ · w = 0  . It is w ell known that V is dense in the space of solenoidal L 2 (Ω) fields satisfying the homo- geneous Dirichle t b oundary condition (see, e.g., [27]). Conseque n tly , Θ 0 = V L 2 (Ω) is a closed subspace of L 2 (Ω), a nd t he pro jection theorem applies verbatim to yield existence and uniqueness of the minimizer within Θ 0 . 4.3 Lo c al Existence and Uniqueness of Smo oth Solutions of the Na vier-Stok es Equation W e recall t he fo llo wing classical lo cal well-posedness result for t he three-dimensional Na vier– Stok es equations (see, e.g., Theorem 6.2 in [31, 32]). Theorem 1 (Lo cal existence and uniqueness [31]): Let Ω ⊂ R 3 b e a domain with a smo oth b oundary ∂ Ω (o f class C ∞ ). Let u 0 b e a smo oth initial condition satisfying the 13 compatibilit y condition (21 ). Then, there exists a time T > 0 a nd a unique pair ( u , p ) of smo oth functions t hat satisfy the Nav ier-Stok es equations ( 1 5) with a Newtonian viscous stress ∇ · τ = ρν ∇ 2 u for given ρ > 0, ν > 0 a nd f ≡ 0, as w ell as the b o undary condition (23) o v er the time in terv al t ∈ [0 , T ], matching the initial data: u ( x ; 0) = u 0 . Pro of: See Galdi [31] and Heyw o o d [32 ]. V erbal Statemen t of Theorem 1: Given a smo oth d i ver genc e-fr e e initial c ondition satisfying the no-slip b o und a ry c ondition, the thr e e-d i m ensional Navier–S tokes e quations ad- mit a unique smo oth solution on a short time interval [0 , T ] . Remark 8: The lo cal existence and uniquenes s result of T heorem 1 extends to sufficien tly smo oth non-homogeneous Dirichle t b oundary conditio ns, provid ed that the b oundary data satisfy the usual compatibility conditions with the initial data (see, e.g., [31]). 4.4 Main Theorem: Tw o-W a y Equiv alence Bet w een the PMPG and the Na vier-Stok es Equation Theorem 2: Let u ( x ; t ) b e a smoo t h ( C ∞ ) flow field ov er a b ounded domain Ω ⊂ R 3 and a time in terv al [0 , T ]. Then u is a solution of t he incompress ible Na vier-Sto k es sys tem ρ ( u t + u · ∇ u ) = − ∇ p + ρν ∇ 2 u , ∀ x ∈ Ω , t ∈ [0 , T ] ∇ · u = 0 , ∀ x ∈ Ω , t ∈ [0 , T ] u = 0 , ∀ x ∈ ∂ Ω , t ∈ [0 , T ] if and only if , at ev ery instan t t ∈ [0 , T ], its ev olution u t ( · ; t ) minimizes the functional A ( v ) = 1 2 Z Ω ρ   v + u · ∇ u − ν ∇ 2 u   2 d x (25) o v er the space of admissible solutions Θ 0 , defined in Eq. (22). Pro of: The sufficiency (minimization implies Na vier-Stokes ) follows from Corollary 1. The con v erse statemen t fo llo ws from L emma 2 and Theorem 1, a s fo llo ws. Let u ( s ) b e a smo oth solution of the Nav ier-Stok es with the prescribed initial and b oundary data, whose existence is gua r a n teed b y Theorem 1. Since u ( s ) is smo ot h on a b ounded domain, the General Assumptions are satisfied for all t ∈ [0 , T ]. Then, for eac h fixed t ∈ [0 , T ], Lemma 2 (together with Remark 7) gua r a n t ees the existence of a unique minimizer u ∗ t ( · , t ) = a rgmin Θ 0 A ( v ( · , t )) , whic h is t he Helmholtz-pro jection of the free acceleration u free t = − u · ∇ u + ν ∇ 2 u . Since u ∈ C ∞ ⇒ u free t ∈ C ∞ , then the resulting minimizer u ∗ t is also smo o th. Starting from the same init ia l data u ( s ) ( x , 0), the optimal ev olution u ∗ t ∈ C ∞ giv es rise to a smo oth v elo cit y field u ∗ ( x , t ) = u ( s ) ( x , 0) + Z t 0 u ∗ t ( x , τ ) dτ . 14 Moreo ver, this solution is guarantee d to satisfy the Na vier-Stokes equation and b oundary conditions b y Lemma 1. But Theorem 1 ensures uniqueness of Na vier-Stokes ’ solutions with the same initial and b o undary data. Hence, u ( s ) ≡ u ∗ , whic h concludes the pro of.  V erbal Statemen t of Theorem 2: A c andid ate smo oth flow field over a smo oth b ound e d domain Ω ⊂ R 3 is a solution of the inc o m pr essible Navier-Stokes system on the time interval [0 , T ] if and only if , at every instant o f time, its evo lution u t minimizes the PMPG c ost functional o ver al l kine matic al ly-admissible evol utions . 5 Ph ysical Impli cations of th e Tw o- W a y Equ iv alence Theorem 2 establishes a strict t w o-wa y equiv alence ( i f and only if ) b etw een the Principle of Minim um Pressure Gradient and the Navie r-Stok es equation within t he class of smo oth incompressible flows . An incompressible flo w whose lo cal acceleration u t minimizes the PMPG cost f unctional at ev ery instan t is guaranteed to satisfy the Na vier-Sto k es system. Con ve rsely , if a t some instan t the lo cal a cceleration o f a smo oth flow field r equires a larger pressure force than a nother kinematically-admissible ev olution, then that flow cannot b e a solution of the Na vier-Stok es equation. This t wo-w a y equiv alence can b e used to shed light on the causal mec hanisms b ehind incompressible-flo w phenomena (e.g., separatio n, transition, etc). While the Na vier–Stok es form ulation a nd the PMPG form ulatio n are ma t hematically equiv a lent and therefore yield iden tical flow fields for the same smo o th initial and b oundar y dat a , the minimization f rame- w ork can pro vide fur t her insigh t into the flo w b eha vior. In the classical formulation, the go v erning equations describe a balance of forces. In the PMPG for m ulatio n, the same dy- namics can b e interpreted as the selection, at eac h ins tan t, of the admissible ev olution requiring the smallest L 2 -magnitude o f the pressure force. In this sense, if a flo w exhibits a particular instan taneous b eha vior, that b eha vior can b e view ed as the result of minimizing the constraint force among a ll kinematically-admissible a lternativ es. Such an in terpretation parallels familiar reasoning in design optimization: when the o pt imizatio n algorithm con- v erges to a specific design , it is selected b ecause an y admissible alternative w o uld incur a strictly lar g er v alue of the cost functional. The equiv a lence theorem ensures that this in- terpretation is not heuristic, but a mathematically exact restatemen t of the Na vier-Stokes dynamics. F o r example, consider a smoo t h incompressible flow field u ( x , t ) o ve r a curv ed surface at some instan t t . Supp ose that the Na vier–Stok es acceleration u t pro duces, a t the subsequen t instan t t + ∆ t , a configur a tion u ( t + ∆ t ) that b egins to separate from some lo catio n x ∗ s on the surface. The t wo-w a y equiv alence then implies that the obtained acceleration u t , whic h dictates the subsequen t separating b eha vior, is realized b ecause it is the evolution that requires the smallest mag nitude of the pressure for ce t o main tain incompressibilit y . An y alternativ e admissible ev olution— o ne tha t w ould main tain an attac hed flo w or another that would induce separation fro m a differen t lo cat io n—w o uld demand a pressure for ce with a strictly larg er magnitude to enforce the same constraints , and therefore would not satisfy the Nav ier–Stok es equation at that instant. In this sense, the separating motion is generated b y the minimal-pressure ev olution compatible with incompressibilit y . The same reasoning applies to o ther incompressible-flo w phenomena (e.g., v o rticit y generation, shedding, transi- 15 tion, etc). -0.5 0 0.5 98 98.5 99 99.5 100 100.5 101 Figure 6: V a r ia tion of the normalized cost ˆ A = A ρU 4 lid with the size ǫ of p erturba t ion from the true evolution u ∗ t . The pressure gra- dien t cost attain ts its minim um prec isely at ǫ = 0. That is, the ev o lutio n u ∗ t , obtained from the numerical simulation, minimizes the cost A o v er all kinematically admissible ev o- lutions u t = u ∗ t + ǫ η . T o illustrate this p erspectiv e, w e revisit the unsteady lid-drive n ca vit y simulation presen ted in [5]. W e select a represen ta- tiv e instan t during the sim ulation. A t this time, the v elo city field u and the corre- sp onding ev olution u ∗ t (obtained from the Na vier–Stokes dynamics) are av ailable fr o m the n umerical solution. Let η b e an y kinematically-admissible p erturbation; i.e., div ergence-free: ∇ · η = 0 in Ω, and v anishes at the b oundaries of the cav it y: η | ∂ Ω = 0 . W e then construc t a f a mily of legitimate (kinematically admissible) evolutions: u t = u ∗ t + ǫ η , where ǫ is the size of p erturbatio n from the true ev olution u ∗ t in the direction of η . Since each mem b er of this family satisfies the incompressibilit y constraint and b o und- ary conditions, they represen t legitimate al- ternativ e ev olutions. The PMPG cost functional A at this in- stan t dep ends on the curren t v elo cit y field u and the candidate ev o lution u t , as defined in Eq. (11). W e then compute A ( u t ; u ) fo r eac h mem b er of this family . Figure 6 shows the v a r ia tion of the normalized PMPG cost with ǫ . The functional att a ins its minim um precisely at ǫ = 0, corresp onding to the Nav ier–Stok es ev olution u ∗ t , while any nonzero p erturbation pro duces a strictly lar g er cost. A similar b eha vior is obtained for other p erturbation directions. F or example, let η 1 and η 2 b e tw o kinematically-admissible perturba tions satisfying ∇ · η i = 0 in Ω and η i | ∂ Ω = 0 for i = 1 , 2. W e construct the tw o-parameter family o f admissible instan taneous ev olutions: u t = u ∗ t + ǫ 1 η 1 + ǫ 2 η 2 , where ǫ 1 and ǫ 2 con tro l t he perturba tion magnitudes. W e then ev aluate the PMPG cost functional A ( u t ; u ) for each mem b er of this family . Figure 7 sho ws con tours of u ∗ t , η 1 , and η 2 in addition to the con t ours of the normalized cost in the plane ǫ 1 - ǫ 2 . The PMPG cost attains its minim um precisely at ǫ 1 = ǫ 2 = 0, confirming that u ∗ t is the unique minimizer within this p erturbat io n subspace . This p ersp ectiv e also suggests a p ossible diagnostic criterion for ass essing n umerical solv ers for incompressible flo w sim ulation. Supp ose tw o solv ers enforce incompressibilit y to comparable to lerance lev els and are applied to the same discretization and b oundary- v a lue problem. If one solv er consisten tly pro duces a smaller PMPG cost A than the ot her, then it ma y be regarded a s more faithfully replicating the underlying flo w dynamics. In 16 -1.00 -0.50 0.00 0.50 1.00 X -1.00 -0.50 0.00 0.50 1.00 Y 2 4 6 8 10 12 14 16 18 20 (a) Nondimensional η 1 . -1.00 -0.50 0.00 0.50 1.00 X -1.00 -0.50 0.00 0.50 1.00 Y 2 4 6 8 10 12 14 16 18 20 (b) Nondimensional η 2 . -1.00 -0.50 0.00 0.50 1.00 X -1.00 -0.50 0.00 0.50 1.00 Y 0.5 1 1.5 2 2.5 3 3.5 (c) Nondimensional ˙ U ∗ . -0.5 0 0.5 -0.5 0 0.5 100 102 104 106 108 (d) Normalized Cos t ˆ A . Figure 7: Contours of the magnitude of the nondimensional ev olution u t T r ef U lid corresp onding to (a, b) the p erturbations η 1 , η 2 , and ( c) the true ev o lut io n u ∗ t . The subfigure (d) shows con to urs of the nondimensional cost ˆ A in the ǫ 1 - ǫ 2 plane. The PMPG cost attains its minim um precisely a t ǫ 1 = ǫ 2 = 0, confirming the optimality of u ∗ t o v er the family u t = u ∗ t + ǫ 1 η 1 + ǫ 2 η 2 . 17 this sense, the PMPG cost prov ides not only a theoretical principle but also a quan tita tiv e measure of dynamical consistency . 6 Finite-Dimen sional A ppro xi mation and Connectio n to Galerkin Pro jection It is instructiv e t o observ e that applying the PMPG in a finite-dimensional mo dal setting, with div ergence-free mo des, is equiv alen t to class ical Galerkin pro jection. Consider the standard mo dal expansion of the v elo city field u ( x , t ) = n X i =1 α i ( t ) ψ i ( x ) , (26) where the mo des ψ i are div ergence free: ∇ · ψ i ( x ) = 0 ∀ x ∈ Ω , i ∈ { 1 , ..., n } . The PMPG form ulation then reduces to the finite-dimensional optimization problem: min ˙ α i A ( ˙ α i ) = 1 2 Z Ω ρ   ˙ α i ψ i + α j α k ψ j · ∇ ψ k − ν α j ∇ 2 ψ j   2 d x , (27) where Einstein summation is used. This minimization problem is unconstrained (assuming that the mo des satisfy the geo- metric b oundary conditions), b ecause the represen tation (26) a uto matically enforces incom- pressibilit y . The optimization is therefore tak en directly with respect to the co efficien ts ˙ α i , and the first-order necessary condition for optimalit y is: ∂ A ∂ α i = 0 ⇒ Z Ω ψ i ·  ˙ α j ψ j + α j α k ψ j · ∇ ψ k − ν α j ∇ 2 ψ j  d x = 0 , whic h leads to the finite-dimensional system of or dina r y differen tial equations: M ij ˙ α j + C ij k α j α k − ν L ij α j = 0 , (2 8 ) where M ij = Z Ω ψ i · ψ j d x , C ij k = Z Ω ψ i ·  ψ j · ∇ ψ k  d x , L ij = Z Ω ψ i · ∇ 2 ψ j d x . The ODE system (28) is precise ly the standard Galerkin pro j ection of t he Na vier–Stok es equation on to the function space spanned by t he mo des ψ i (e.g., [ ? ]). If the mo des, how ev er, are not div ergence-free, the t wo form ulations (PMPG and classical Galerkin) may not lead to iden tical dynamical sys tems. In this case, the standard mixed Galerkin metho d introduces a pressure expansion p ( x , t ) = m X ℓ =1 β ℓ ( t ) q ℓ ( x ) , 18 where { q ℓ } span a suitable pressure space. Incompressibilit y is then enforced in the w eak sense b y requiring Z Ω q ℓ ( ∇ · u ) d x = 0 ∀ ℓ ∈ { 1 , ..., m } , whic h yields B ℓj α j :=  Z Ω q ℓ  ∇ · ψ j  d x  α j = 0 . The mixed Galerkin formu lation of the momen tum equation then giv es the coupled system M ij ˙ α j + C ij k α j α k = − G iℓ b ℓ + ν L ij α j , (29) together with the discrete incompressibilit y constraint B ℓj α j = 0, where G iℓ = Z Ω ψ i · ∇ q ℓ d x . In contrast, in the PMPG framew ork, there is no a priori pressure expansion; rather, the pressure arises as the Lagra nge m ultiplier enforcing incompressibilit y in the instan taneous minimization pro blem. T o imp ose suc h a constrain t in a finite-dimensional setting, intro duce m linear functionals {E ℓ } m ℓ =1 acting on scalar fields o v er Ω . W e enforce the discrete con tinuit y constrain ts as E ℓ ( ∇ · u t ) = 0 , ℓ = 1 , . . . , m. Using u t = ˙ α j ψ j , these constrain ts b ecome D ℓj ˙ α j = 0 , D ℓj := E ℓ  ∇ · ψ j  , (30) or, in matrix form, [ D ] ˙ α = 0. Ho wev er, for the constrained minimization problem to b e w ell-p o sed, the admissible set m ust b e non-empt y; equiv alen tly , the n ull space o f D m ust b e non-trivial. This requires that t he span of the chos en v elo cit y mo des contain discrete div ergence-free com binations, i.e., there exist ˙ α 6 = 0 suc h that [ D ] ˙ α = 0 . Under this natural compatibilit y conditio n, the strictly conv ex quadratic functional A admits a unique minimizer within the constrained subspace. The PMPG minimization problem (2 7 ) is therefore carried out sub ject to the discrete con tinuit y constrain t (30 ). Ove r the set of admissible ev o lutio ns parameterized b y ˙ α j that satisfy [ D ] ˙ α = 0, the PMPG selects the unique ev o lutio n that minimizes the magnitude of the pressure force required to enforce the discrete contin uit y constrain ts. T o imp ose these m constrain ts, in tro duce m Lagrange m ultipliers λ ℓ , and the constrained Lagrangia n is written as L ( ˙ α j , λ ℓ ) = 1 2 Z Ω ρ   ˙ α i ψ i + α j α k ψ j · ∇ ψ k − ν α j ∇ 2 ψ j   2 d x − λ ℓ D ℓj ˙ α j . The first-o r der necessary condition for optimalit y with resp ect to ˙ α i is t hen ∂ L ∂ ˙ α i = 0, which yields Z Ω ψ i ·  ˙ α j ψ j + α j α k ψ j · ∇ ψ k − ν α j ∇ 2 ψ j  d x − λ ℓ D ℓi = 0 . (31) 19 Th us, when the v elo cit y mo des are div ergence-free, the classical Galerkin pro jection and the PMPG yield iden tical dynamical systems. How ev er, with arbitrary (non-div ergence-free) mo des, the tw o formu lations ma y result in different dynamics, dep ending o n the w a y the incompressibilit y constraint is discretely imp osed in the PMPG for mulation. In the mixed Galerkin metho d, incompressibilit y is enforced in the w eak sense thro ugh prescrib ed pressure mo des ( B ℓj α j = 0). In con trast, the PMPG framew ork allows more flexibilit y in imp osing the discrete con tinuit y constrain ts ( D ℓj ˙ α j = 0). Differen t choice s of the functionals {E ℓ } (e.g., p oin t wise ev aluat ion) lead to different admissible ev olution spaces. Consequen tly , the resulting discrete dynamics may differ from those obtained b y mixed Galerkin pro jection. Moreo ver, the PMPG formulation naturally extends to non-mo dal (nonlinear) parame- terization of the v elo cit y field (e.g., neural-net w ork represen tations): u ( x , t ) = N ( α i ( t ) , x ) , (32) where α i denote the pa r a meters (e.g., weigh ts and biases of the neural netw o r k, see [33– 35]). In this setting, the v elo city field do es not lie in a linear subspace but rather on a nonlinear manifold in function space. Consequen tly , class ical Ga lerkin pro jection is not directly applicable. How eve r, the PMPG form ula t io n pro ceeds in tuitiv ely , as follo ws. The time deriv ativ e of the v elo city field is written as u t = ˙ α i N α i , N α i := ∂ N ∂ α i , so that the PMPG cost b ecomes A ( ˙ α i ) = 1 2 Z Ω ρ   ˙ α i N α i + N · ∇ N − ν ∇ 2 N   2 d x . Minimization is p erfo r med ov er ˙ α i sub ject to discrete contin uit y constraints : E ℓ ( ∇ · u t ) = E ℓ  ∇ · N α j  ˙ α j = D ℓj ˙ α j = 0 . Geometrically , the PMPG p erforms an o rthogonal pro jection of the free acceleration on to the tangen t space of the nonlinear manifold defined b y the para meterization (32). 7 Concep tual D istinct ions from Class ical V ariational Principles Because the PMPG is a relativ ely recen t dev elopment in the fluid mec hanics literature, and b ecause it builds up on Gauss’s principle of least constrain t—a concept not widely empha- sized in standard engineering curricula— it is natural that questions ma y arise regarding its in terpretatio n and relation to more familiar v ariational principles. In this section, we clar if y sev eral recurring p oints of confusion surrounding the PMPG. 20 7.1 Gauss’s Principle and the PM PG are F und amen tally Distinct from Stationary Action It is imp ortant to emphasize t ha t Gauss’s principle of least constrain t, and its contin uum ana- logue, the PMPG, are not time-in tegra l principles. They differ fundamen tally fro m Hamil- ton’s principle of stationary (or least) action, in whic h the cost functional is the t ime in tegral of a Lagr angian o v er an en tire tra jectory . In contrast, G a uss’s principle and the PMPG are instan taneous principles. They select, at eac h fixed time, the admissible acceleration that minimizes a quadratic cost sub ject to kinematic constrain ts. No optimization is p erformed o v er tra j ectories in t ime, and no time in tegral app ears in the formulation. The minimiza- tion is lo cal in time and determines the instantaneous ev olution, from whic h the tra jectory subseque n tly f ollo ws via in tegra tion. A related distinction concerns the role of endpo int conditions. In Hamilton’s principle, the v ariational form ula tion prescrib es b oundary conditions in t ime (typic ally fixing init ia l and final configurations). In con trast, G auss’s principle and the PMPG require only the instan taneous configuration (p osition and v elo city). They determine the corresp onding ev o- lution without reference to a final state. In this sense, the formulation is naturally aligned with f orw a rd dynamical ev olutio n and do es not presupp ose know ledge of future configura- tions [36]. Moreo ver, Hamilton’s principle yields a stationa rit y condition for the action integral, whic h do es not necessarily imply a strict minim um. In contrast, G a uss’s principle of least constrain t, and the P MPG in the con tin uum setting, are strict quadratic minimization princi- ples. The asso ciated functional is strongly conv ex, and the minimizing a cceleration is unique. Accordingly , the in terpretiv e statemen ts made in the previous section—namely , that the r e- alized ev o lution is the o ne requiring the smallest magnitude of the pressure force—are not teleological. Ra ther, they follo w from strict instan taneous minimalit y and orthogo nal pro - jection in a Hilb ert space, not fro m a time-global optimality condition. In this sense, the reasoning is en tir ely dynamical and forward in time. 7.2 The PMPG Do es not F ollo w from the Diric hlet Principle Despite its name, the Pri n ciple of Minimum Pr e s sur e Gr adie n t do es not inv olv e minimizing the functional k∇ p k 2 L 2 = Z Ω | ∇ p | 2 d x o v er scalar pressure fields. In other w o rds, the PMPG is not a v ariationa l principle p osed fo r p itself. Ra t her, it seeks the acceleration v ector field u t that minimizes the cost functional A , defined in Eq. (13). Although t he tw o cost functionals may app ear similar—b oth reflecting the L 2 -norm of the pressu re gradien t—they are equiv a len t only at the minimizing solution. F o r a general kin ematically-admissible candidate u t , the integrand o f the cost A is not necessarily a gradien t field. It b ecomes a gradien t only at the minimizer, as a conseq uence of or t hogonality to the div ergence-free subspace. F rom this viewp oin t, the terminology “Minim um Pressure Gra dien t” should not b e interpreted as a minimization o v er the space of gradien t fields. 21 Indeed, the t wo v ariational form ulations—the PMPG and minimizing k∇ p k 2 L 2 o v er p — are fundamen tally differen t. Classical p oten tial theory show s, via the Diric hlet principle (e.g., [37]), that minimizing k∇ p k 2 L 2 o v er scalar fields p with prescrib ed Diric hlet b oundary data leads to the Laplace eq uation: ∇ 2 p = 0. In con tra st, Lemma 1 establishes that minimizing t he PMPG cost functional A with r esp ect to the acceleration field u t yields the Na vier-Stok es equation. The asso ciat ed pressure p arises as the La grange m ultiplier enforcing incompressibilit y . T aking the div ergence of the Na vier-Stoke s equation thus give s the familiar pressure Pois son equation ∇ 2 p = ρ ∇ · u free t = ∇ · [ − ρ u · ∇ u + ∇ · τ + ρ f ] , (33) whic h is generally inhomogeneous and sub ject to b oundary conditions deriv ed from the v elo cit y field a nd momen tum balance, not f r o m prescribed Diric hlet dat a on p . Hence, t he t wo v ariational fo rm ula t io ns differ not only in their optimization v ar ia bles ( p vers us u t ) but also in their admissible spaces, b oundary conditions, and resulting gov erning equations. On the o ther hand, if one b egins with the pressure P oisson equation (33) and applies the Diric hlet principle for P oisson problems, it f ollo ws that p minimizes the functional Z Ω  1 2 | ∇ p | 2 + ρp ∇ · u free t  d x , whic h is fundamen t ally differen t from the PMPG cost functional A . The ab ov e functional is defined ov er scalar fields p and includes a linear source term inv o lving ∇ · u free t (whic h is gen- erally non-zero), whereas A is a strictly con vex quadratic f unctional defined o ver admissible acceleration fields u t . 7.3 The Instan taneous Nature of the PMPG and Its Relation to T u rbulence One recurring ques tion r egarding the PMPG is its relation to turbulence. F or example, one ma y ask: if the ev olution at ev ery instan t minimizes the magnitude of the pressure gradient required to enforce incompressibilit y , ho w can turbulence dev elop? In canonical examples suc h as fully dev elop ed laminar channe l flow, the steady parab o lic profile corresp onds to a state in whic h the instan taneous PMPG cost v anishes; i.e., it cannot decrease any further. F r om this viewpoint, it ma y appear that a tra nsition to turbulence con tradicts the minimizing philosoph y underlying Gauss’s principle and the PMPG. T o clarify this p oint, it is ess en tial emphasize that the mathematical statemen ts presen ted here do not imply that the PMPG cost functional A is minimized with resp ect to the v elo cit y field u ( x ; t ) itself. Indeed, if o ne were to set the first v ariation of A with resp ect to u ( x ; t ) to zero, the Na vier-Stok es equation w ould not b e reco v ered a s a necessary condition. Rather, the Nav ier-Stok es equation arises as the neces sary condition for minimizing t he PMPG func- tional A with resp ect to the lo c al ac c eler ation u t ( x ; t ), with the instantaneous v elo city field u ( x ; t ) held fixed. The v aria t ion is therefore t ak en ov er spatial fields at a fixed time; and time ente rs only as a parameter. In this sense, the PMPG is a n instan taneous principle go v erning the admissible accelerations at a giv en configuration. This in terpretation is fully 22 consisten t with the philosophy of G auss’s principle as discussed in Sec. 2 a nd illustrated through the double-p endulum example. F r om this p ersp ective , starting at a g iv en flo w field, the PMPG determines the instanta- ne ous evo lution from that configuration. Hence, if the giv en configuration is an equilibrium of the Na vier–Stokes system (suc h as the laminar parab olic profile in c hannel flow ), the op- timal acceleration is expected to b e zero: the b est ev o lut io n is no ev olution. By the tw o- w ay equiv alence established in Theorem 2, this conclusion necessarily coincides with the Na vier– Stok es dynamics. Indeed, if the flo w is exactly at an equilibrium configura tion, it will remain there under Na vier–Stok es dynamics in the absence of p erturbatio ns. It is only when the equilibrium is p erturb ed that the flo w transitions in to turbulence. Similarly , the PMPG will not result in a non-tr ivial ev olut io n fro m the laminar profile u L unless a p erturbation δ u is intro duced. A t the p erturb ed state u = u L + δ u , the PMPG cost a sso ciated with zero acceleration is generally non-zero: A ( u t = 0; u L + δ u ) > 0 and there may exist a dmissible ev olutions that reduce the instantaneous cost; i.e., there exists u ∗ t suc h that A ( u ∗ t ; u L + δ u ) < A ( u t = 0; u L + δ u ) . The minimizing acceleration u ∗ t then determines the subsequen t ev olution, which ma y am- plify the p erturbation if the equilibrium is unstable. Moreo ver, while the optimal u ∗ t ( x , t ) at a given instant t minimizes the instan ta neous cost for the fixed configuration u ( x , t ) A ∗ ( t ) := A ( u ∗ t ( x , t ); u ( x , t )) ≤ A ( u t ; u ( x , t )) o v er all kinematically admissible evolutions u t , it do es not necessarily lead t o a smaller cost A at the next time step; i.e., A ∗ ( t + ∆ t ) ma y b e larger than A ∗ ( t ). Indeed, ev o lving the flow field for an infinitesimal time step ∆ t a ccording to the optimal acceleration u ( x , t + ∆ t ) = u ( x , t ) + u ∗ t ( x , t )∆ t yields a new configuration at which a new minimization problem is p osed. The resulting minimal v a lue A ∗ ( t + ∆ t ) := A ( u ∗ t ( x , t + ∆ t ); u ( x , t ) + u ∗ t ( x , t )∆ t ) need not b e smaller than A ∗ ( t ). In other w ords, minimizing the pressure-gradien t norm at eac h instan t do es not imply that this norm is a Ly apuno v functional o r that it decreases monotonically in time. This b ehavior is en tirely consisten t with Navier–Stok es dynamics, for whic h the pressure-gradien t norm is not, in general, a monotonically decreasing quan tity . In summary , the tw o-w a y equiv alence established in Theorem 2 implies that the PMPG pro vides an alternativ e fo r m ulatio n of the Na vier–Stokes dynamics for smoo t h flo w fields. It recasts the same dynamics in a minimization framew ork. Consequ en tly , if the Na vier–Stok es equations do not generate turbulence from an equilibrium configurat ion, neither do es the PMPG. Con ve rsely , whenev er the Nav ier–Stok es dynamics pro duces temp or al fluctuations in the pressure-gradien t norm, the PMPG necessarily repro duces t he same b eha vior. 23 8 Relation to th e V ariational Theo ry of Lift The preceding discuss ion indicates that there is no dir e ct mathematical implication b et w een the PMPG (Lemma 1 or Theorem 2) and the v ariational theory of lift [3 8 ]. This the ory was in tro duced to address the classical closure issue in t w o- dimensional p oten tial flo w. Giv en a smo oth b o dy B ⊂ R 2 with exterior fluid domain Ω = R 2 \ B , the steady , incompressible , Euler equation admits a one-parameter fa mily of smo oth solutions u ( x ; Γ) = u 0 ( x ) + Γ u 1 ( x ) , Γ ∈ R , (34) where u 0 is t he no n-circulatory flo w (i.e., it has zero circulation H ∂ B u 0 · d x = 0) ; u 1 is the circulatory flow with unit circulation ( H ∂ B u 1 · d x = 1); and the scalar parameter Γ therefore represen ts the total circulation around t he b o dy asso ciated with eac h mem b er of the family (34). The family (34) satisfies the steady incompressible Euler system for ev ery v alue of Γ. Indeed, for ev ery Γ: (i) It satisfies t he steady momen tum equation ρ u · ∇ u = − ∇ p for some pressure field p , since b oth u 0 , u 1 are irrotationa l, and therefore u ( x ; Γ) is irro- tational for ev ery Γ. (ii) It satisfies incompressibilit y ∇ · u = 0 b ecause b oth u 0 , u 1 are div ergence-free. (iii) It satisfies the no-p enetration b oundary condition u · n = 0 on ∂ B since b o th u 0 , u 1 are tang ential to the b o dy surface. (iv) It satisfies t he far-field condition lim | x |→∞ u ( x ; Γ) = ( U, 0 ) b ecause u 0 approac hes the prescrib ed freestream v elo cit y U and u 1 deca ys at infinit y . Consequen tly , the f amily (34) constitutes a legitimate one-parameter family of steady Euler solutions fo r arbitrary circulation Γ. When the b o dy p ossesses a sharp trailing edge, all mem b ers of the f a mily (34) exhibit singular behav ior at that edge, except for a unique mem b er u K that remains b ounded ev erywhere in the domain. This singular structure prov ides a nat ur a l selection criterion within t he family: the ph ysically relev an t solution is the one that eliminates the trailing- edge singularity . This singularity -remo v al criterio n is kno wn as the Kutta condition [39] and has serv ed as the cornerstone of classical air foil t heory for o v er a cen tury . How ev er, when the b o dy is smoot h (i.e., without a sharp edge), there is no generally accepted phy sical o r mathematical criterion tha t uniquely selects a mem b er from the family . The steady Euler problem therefore remains non-unique in this setting. Although each mem b er in the fa mily (34) satisfies the Euler system, they generally yield differen t v alues of the PMPG cost. In the steady , in viscid setting, the functional reduces to the L 2 -norm of the con vec tiv e a cceleration: A s (Γ) = 1 2 Z Ω | u ( x ; Γ) · ∇ u ( x ; Γ) | 2 d x . (35) Since eac h mem b er satisfies the steady Euler equation, it follows t hat, a lo ng this solution family , A s (Γ) = k∇ p k 2 L 2 (Γ) . Th us, eac h circulation level Γ r equires a distinct pressure field to enforce the constrain t s of incompressibilit y and no-p enetration. So, inspired b y the philosoph y of Gauss’s principle, 24 Gonzalez and T aha [38] prop o sed a v aria t io nal closure criterion: Among all admissible steady solutions (34), select the mem b er that minimizes the mag nitude of the pressure gradien t required to enforce the constraints: Γ ∗ = argmin Γ A s (Γ) . (36) F o r airfoils with a sharp trailing edge, the minimiz ing solution u ∗ con verges , in the limit of v anishing trailing-edge radius, to Kutta’s solution u K . In this sense , the classical Kutta condition is reco v ered as a sp ecial case of the v ariational closure criterion (36) for sharp-edged geometries. F or smo oth airfoils, T aha and G onzalez [40] rep o rted quan titative agreemen t b etw een the minimizing circulation Γ ∗ predicted by the v ariational f ramew ork and that obtained from Reynolds-a v eraged Nav ier–Stok es sim ulat ions. Although the v ariational closure (36) follows the same philosophy as Gauss’s principle and the PMPG, it do es not follow directly from the mathematical statemen ts presen ted in Lemma 1 or Theorem 2 . The latter establish minimization with resp ect to the lo cal acceler- ation u t (or its finite-dimensional parameters, as discussed in Sec. 6) for a fixe d velocity field u . In con trast, the minimization (36) is carr ied o ut with resp ect to the parameter Γ, whic h parameterizes the v elo city field u itself. Moreo v er, the minimization is p erformed o v er a family o f steady Euler solutions. These candidates are not merely kinematically a dmissible; they already satisfy the dynamical equations of motion. In this sense, the v ariational crite- rion (36 ) do es not select a dynamical solution out of kinematically-admissible ones. Rather, it acts as an additional selection criterion within a fa mily of dynamically v alid steady solu- tions, mo t iv ated b y the same philosophical fo undat io n underlying Gauss’s principle a nd the PMPG. Ha ving clarified the fundamen tal distinction b et ween the PMPG and the v ariational closure (36), it is nev ertheless conceiv able that a deep er, indirect connection exists b et w een the tw o framew orks, whic h motiv ates the in v estigation of some mathematical questions, presen ted in the next section. Ho w eve r, b efore form ulating those conjectures, we presen t another example in whic h a Gauss-inspired steady minimization repro duces a classical visc ous asymptotic res ult. This ex ample ma y pro vide insigh t in to the p ossible structural link b et w een the instan ta neous PMPG dynamics and steady v ariational selection. It has b een kno wn since the early exp erimen ts of Prandtl [41] that sufficien tly rapid rotation ( κ = ω b U ≫ 1 ) of a circular cylinder suppresses separation and v ortex shedding, leading to an attac hed flo w. Ra yleigh’s seminal analysis [42] clarified that viscosit y t r ansfers the cylinder’s rotat io n in to circulation in the outer flo w. Outside the b oundary lay er, the motion b ecomes irro tational with circulation Γ. The cen tral question b ecomes: what is the v alue of Γ generated by a cylinder of radius b rotating with angular v elo cit y ω in a free stream U ? As Rayleigh remark ed, “ friction is the imme diate c ause of the wh i rlp o o l mo tion ,” indicating that circulation cannot b e determined without a ccoun ting fo r viscous effects. The difficult y extends b ey ond the absence of a Kutta- lik e condition: in an in viscid f orm ulation, the no-p enetration b oundary condition sp ecifies only the normal velocity . The tangen tia l surface v elo city induced b y rot a tion cannot be imp osed. Th us, for a rotating cylinder, where t he b oundary motion is purely tangen tial, the ideal-fluid for mulation is insensitiv e to the rotat io n. As noted explicitly b y Mo o r e [43], “ it is not p ossible to determine Γ w i thout solving the Navier–Stokes e quations .” 25 Glauert [44] provide d the first comprehensiv e theoretical treatmen t. He mo deled the outer flow as irrotational with unkno wn circulation; i.e., s imilar to the family (3 4). T o determine the v alue of this circulation, he solved Prandtl’s b oundary-lay er equations under the no-slip condition. Exploiting the large- κ assumption and expanding in ǫ = 1 /κ , he obtained Γ Γ ω = 1 − 3 κ 2 + O  1 κ 4  , (37) where Γ ω = 2 π b 2 ω is the circulation asso ciated with the cylinder’s surface motio n. The r atio Γ / Γ ω ma y b e in terpreted as the transfer efficiency of circulation from the rotating b oundar y la y er to the outer flo w. The expansion (37) rev eals a well-defin ed in viscid limit: Γ Γ ω = 1 − 3 κ 2 . Shorbagy and T aha [45] prop osed a G auss-inspired v ariational closure that reco vers t his in viscid limit without solving the nonlinear b oundary-la yer equations. They f ollo w ed Batch- elor’s t wo-stage description [46] of the problem “ in two stages. First the c ylin d er is given a s te ady angular vel o c ity ω in fluid initial ly at r est; ... vorticity is gener ate d in the fluid and diffuses to infinity, le aving a ste a d y irr otational motion with cir culation 2 π b 2 ω . The cylinder is then given a tr an s l a tional vel o c i ty U .” In the language of Gauss, they treated the irrotational flow induced by Γ ω as the impressed (free) motion. This free motion has known acceleration a free = Γ 2 ω 4 π 2 r 3 e r . When a uniform stream U is sup erp osed, the actual outer flo w is selected, in the spirit of Gauss’s principle, as the mem b er o f the f a mily (34) that deviates least fro m this impresse d motion. Accordingly , the G a ussian deviation cost A s, rot (Γ) = 1 2 ρ Z Ω  u ( x ; Γ) · ∇ u ( x ; Γ) − a free  2 d x (38) is minimized with resp ect to Γ. 3 4 5 6 7 8 9 10 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 / Variational Theory Glauert Boundary Layer Solution Figure 8: Comparison for the v ariation of the normalized circulation Γ Γ ω with the normalized rotational sp eed κ = ω b U b et w een Glauert’s viscous b oundar y lay er solution (37) and the PMPG solution (39). The necessary condition for minimization d A s, rot d Γ = 0 yields: Γ ∗ Γ ω = r 1 − 6 κ 2 . (39) Figure 8 c ompares (39) w ith Glauert’s asymptotic result (37). Expanding (39) for large κ giv es Γ ∗ Γ ω = 1 − 3 κ 2 + O  1 κ 4  , whic h coincides with Glauert’s in viscid limit to second order. Th us, a one-dimensional v a r ia tional minimization repro duces the asymptotic circulation obt a ined fro m the viscous b oundary-lay er analysis. 26 9 Dynamical Minimizati on V ersus Ste ady Optimalit y The results obtained from v ariational closure, discussed ab ov e, motiv ate the follo wing math- ematical question: sinc e the flo w ev olve s a t ev ery instant b y minimizing the cost A ( u t ; u ), supp ose that the ev olutio n con ve rges to a steady state equilibrium ¯ u . Do es it necess arily imply that ¯ u minimizes the steady cost A s ( u ) := A ( 0 ; u )? F or a g eneral dynamical system, the answ er is negativ e. The following example pro vides a counterex ample. Consider the t w o- dimensional dynamical system with state v ariables χ = ( x, y ), whose instan taneous ev olution ˙ χ is determined b y minimizing the cost S ( ˙ χ ; χ ) =  ˙ x + x 3  2 +  ˙ y + y x 2  2 + ( y − 3 ) 2 . This minimization results in the dynamical system ˙ x = − x 3 and ˙ y = − y x 2 , (40) whic h has the en t ire y -axis ( x = 0) as a contin uous family of equilibrium p oints. Ho wev er, only the origin is stable; starting f r om any initial condition with x 6 = 0, the tr a jectory con verges to the origin, as illustrated in the pha se p ortrait sho wn in Fig. 9.   Figure 9: Phase por trait of the t w o- dimensional dynamical system (40). It has the en tire y -axis ( x = 0) as a line of equilib- ria. How ev er, only the origin is stable; starting from any p oin t in the plane (except on the y - axis), the system con v erges to the origin. In this example, the steady cost is give n by S s ( χ ) := S ( 0 ; χ ) = x 6 + y 2 x 4 + ( y − 3 ) 2 , whic h is minim um at (0 , 3). In contrast, the stable equilibrium p oint selected by the dy- namics (i.e., the origin) is not a lo cal mini- mizer of the steady cost. This example w a s devised to mirror, in a simplified se tting, the structure of the p oten tial-flow lift problem. Similar to the y - axis in this system, ideal-fluid dynamics ad- mits a con tin uous family of steady solutions, parameterized b y the Γ-a xis. Y et, in phys - ical realit y , only one mem b er of this fa m- ily is realized; fo r a sharp-edged airf oil, this is Kutta’s circulation. The v ariational the- ory of lift asserts that this ph ysically realized equilibrium cor r esp o nds to the minimizer o f the asso ciated steady cost. The coun terex- ample ab ov e sho ws that suc h a conclusion do es not hold for general dynamical sy stems: instan taneous minimization of a cost do es not, in general, imply that the dynamically selected equilibrium minimizes the corresp onding steady cost. Nev ertheless, one m ust exer- cise caution in drawing parallels. The tw o-dimensional dynamical system intro duced ab o ve, although structurally analogous to the lift-selection pr o blem, is f undamen tally differen t in 27 nature; it is not derive d from a mec hanical or con tinuum framew ork. The analogy is therefore suggestiv e, but not conclusiv e. If additional structural prop erties ar e impo sed on the dynamics, the implication that a stable equilibrium ¯ u minimizes the steady cost A s ( u ) can b e ensured. F or instance, supp ose that in a neigh b orho o d of a stationary solutio n ¯ u , the pressure-gradien t norm (equiv alen tly , the PMPG cost) is monotonically non-increasing under the dynamics. Then ¯ u mus t nec- essarily b e a lo cal minimizer of the steady cost A s . The following lemma f o rmalizes this observ ation. Lemma 3 : Consider a dynamical sys tem with stat e v ariables χ ∈ X that ev olv es at ev ery instant b y minimizing a cost S ( ˙ χ ; χ ) , whic h dep ends smo othly on the instan t a neous ev olution ˙ χ and the curren t configuration χ . That is, the dynamics is defined according to the unique optimal ev olution ˙ χ ∗ ( χ ) := a rgmin ˙ χ S ( ˙ χ ; χ ) . A t eac h configuration χ , the o ptimal cost is then giv en by S ∗ ( χ ) := S ( ˙ χ ∗ ( χ ); χ ) . Assume that ¯ χ is an asymptotically stable equilibrium of t he induced dynamics suc h that there exists a neigh b orho o d N ⊂ X o f ¯ χ where, along tra jectories of the system, d S ∗ dt ( χ ( t )) ≤ 0 ∀ χ ( t ) ∈ N . Then, ¯ χ m ust b e a lo cal minimizer of the steady cost S s ( χ ) := S ( 0 ; χ ). In particular, w e ha ve S s ( ¯ χ ) ≤ S s ( χ ) ∀ χ ∈ N . Pro of: Since ¯ χ is asymptotically stable and S ∗ is contin uous in χ (whic h follow s from the assumed smo othness of S in b oth arg uments and uniqueness of the minimizer), w e ha ve lim t →∞ χ ( t ) = ¯ χ ⇒ lim t →∞ S ∗ ( χ ( t )) = S ∗ ( ¯ χ ) . Since ¯ χ is asymptotically stable, w e ma y restrict N , if necessary , so tha t it is con t a ined in the region of attraction of ¯ χ ; i.e., t r a jectories star t ing at an y χ ∈ N conv erge to ¯ χ . Moreo v er, b y assumption, S ∗ is non-increasing along tra jectories in N . Th us, fo r an y initial condition χ ∈ N , S ∗ ( ¯ χ ) = lim t →∞ S ∗ ( χ ( t )) ≤ S ∗ ( χ ) . Equiv alen tly , S ( ˙ χ ∗ ( ¯ χ ); ¯ χ ) ≤ S ( ˙ χ ∗ ( χ ); χ ) ∀ χ ∈ N . Since ¯ χ is an equilibrium, its optimal evolution v anishes: ˙ χ ∗ ( ¯ χ ) = 0 . Hence, S ( 0 ; ¯ χ ) ≤ S ( ˙ χ ∗ ( χ ); χ ) ∀ χ ∈ N . 28 Finally , b y definition of o pt imality , w e hav e S ( ˙ χ ∗ ( χ ); χ ) ≤ S ( 0 ; χ ), whic h yields S ( 0 ; ¯ χ ) ≤ S ( 0 ; χ ) ∀ χ ∈ N . Equiv alen tly , S s ( ¯ χ ) ≤ S s ( χ ) ∀ χ ∈ N , whic h pro ve s that ¯ χ is a lo cal minimizer of the steady cost S s .  The ab o ve Lemma establishes that, if the L 2 -norm of the pressure gradien t (equiv alen tly , the optimal PMPG cost A ∗ ) is non- increasing along the flo w dynamics in a neighborho o d of a steady solution, then that steady solution must b e a lo cal minimizer of the steady PMPG cost A s ( u ) = 1 2 Z Ω | u · ∇ u − ν ∇ 2 u | 2 d x . (41) This b eha vior is observ ed in monotonic al ly stable flo ws at sufficien tly lo w Reynolds n um- b ers, where the p erturbation kinetic energy deca ys monoto nically tow ard the steady state, without ov ersho ot o r non- mo dal transien t grow th [47]. This observ ation provide s a p ossible explanation f or why minimization of the steady cost A s can successfully recov er steady solu- tions in certain cases, suc h as the lid- driv en cavit y at low Reynolds n um b ers [33]. Ho w ev er, extending this reasoning b eyond suc h regimes requires further mathematical analysis. Another observ ation supp orting the hypothesis that stable steady solutions ma y mini- mize the steady cost A s arises from a time-discrete fo rm ulat io n of the PMPG framew ork. Discretizing the instan taneous minimization in time with a forw ard Euler step yields the problem min u k +1 A ( u k +1 ) = 1 2 Z Ω ρ     u k +1 − u k ∆ t + u k · ∇ u k − ν ∇ 2 u k     2 d x (42) o v er the admissible set U = { u ∈ L 2 | ∇ · u = 0 ∀ x ∈ Ω , and u · n = h ( x ) ∀ x ∈ ∂ Ω } . (43) Under the same conv exity and admissibilit y assum ptions discussed earlier, follo wing t he reasoning of Theorem 2, one can sho w a t wo-w ay equiv alence b etw een this discrete-time explicit minimization problem a nd the corresponding Na vier-Stok es syste m: u k +1 = u k − ∆ t  u k · ∇ u k − ν ∇ 2 u k + ∇ p k  , ∇ · u k +1 = 0 . Th us, at eac h time step, adv ancing the Na vier–Stok es solution is equiv alen t to solving the instan taneous discrete PMPG minimization problem (43). A standard strategy in computational practice for obta ining steady-state solutions is to emplo y implicit time integrators with large time steps [48]. Heuristically , if the time step ∆ t is sufficien tly larg e, a single implicit adv ancemen t ma y driv e the solution tow ard a stable steady state ¯ u . W rit ing the implicit analogue o f the discrete PMPG minimization problem (42) min u k +1 A ( u k +1 ) = 1 2 Z Ω ρ     u k +1 − u k ∆ t + u k +1 · ∇ u k +1 − ν ∇ 2 u k +1     2 d x , (44) 29 and taking the limit ∆ t → ∞ , one obtains the steady-state minimization pro blem: min ¯ u A ( ¯ u ) = 1 2 Z Ω ρ   ¯ u · ∇ ¯ u − ν ∇ 2 ¯ u   2 d x ≡ A s ( ¯ u ) . That is, adv ancing the flow field using the implicit PMPG form ulatio n (44) and taking the large–∆ t limit reduces the ev olution to a direct minimization of the steady functional A s . Consequen tly , if the implicit PMPG dynamics con v erges to a steady-state solution ¯ u , then ¯ u m ust b e a lo cal minimizer of the steady cost A s o v er the admissible class. 10 Mathematical C onjectu res In the t w o examples discussed earlier (the airfoil and the rotating cylinder), the minimization of t he invisc id ve rsion of the steady cost A s (e.g., the L 2 -norm of the conv ective accelera- tion) succes sfully selects the ph ysically relev ant solution from the infinitely many members of Euler’s family . This observ a tion suggests that suc h minimization may provid e a princi- pled selection criterion for resolving the non- uniqueness inheren t in steady Euler dynamics. T o elev ate this idea from heuristic observ ation to mathematical conjecture, one m ust first clarify what is mean t b y a physic al solution of the Euler equations. F rom a mathematical standp oin t, a steady Euler solution may b e regarded as phy sically relev an t if it satisfies at least one of the follo wing prop erties: (i) it is a stable solutio n; or (ii) it arises as the invisc id limit of a Nav ier–Stok es solution, and therefore represen ts the limiting b eha vior of ph ysically realizable viscous flow s at sufficien tly high Reynolds num b ers. That is, Kutta ’s solution for the flo w ov er a sharp-edged airfoil ma y b e regar ded as ph ys- ical b ecause it corresp onds either to a stable steady configuration, to the in viscid limit of the Nav ier-Stok es solution, or to b oth. The fact that minimization of A s reco ve rs Kutta’s solution as a special case suggests that this v ariatio na l selection criterion may represen t a necessary condition for (i) stabilit y of steady Euler solutions, or (ii) iden tification of the in viscid limit of the Na vier-Stoke s equation within Euler’s family . It is no tew orthy that, in the rota ting-cylinder problem, minimization of A s repro duces precise ly the invisc id limit obtained fro m the viscous b oundary-lay er analysis. This result further supp orts the p o ssi- bilit y tha t minimization of the con vec tiv e-acceleration no r m encodes structural infor mation traditionally deriv ed from viscous dynamics. These observ ations motiv ate the follo wing t w o conjecture drafts, prop osed as directions for future mat hematical in ves tigation. The qualifier dr afts is used delib erately to emphasize that the statemen ts b elo w are preliminary and not y et express ed in fully rigorous terms. They ar e off ered in the hop e that mathematical ana lysts may refine and formalize them into precise and v erifiable conjectures. Conjecture Draft 1 (A Necessary Condition for Stabilit y) : A stable stationary solution of the inc ompr essible Euler e quation m ust lo c al ly minimize the L 2 -norm of the c on- ve ctive ac c eler ation . More precisely , if ¯ u ∈ C k (Ω), k > 1 is ( i) div ergence-free ∇ · ¯ u = 0, (ii) satisfies the no- p enetration b oundary condition ¯ u · n = h ( x ) on a smo oth b o undar y ∂ Ω, for some prescrib ed smo ot h function h , (iii) ¯ u go es to a constant at infinity , and (iv) is a stable 30 stationary solution o f Euler’s equation, then it m ust be a lo cal minimizer of the functional A s ( u ) = 1 2 Z | u · ∇ u | 2 d x o v er the family of admissible stationary solutions Φ = n u ∈ C k (Ω) | u · ∇ u ∈ L 2 , ∇ · u = 0 , u · n = h ( x ) , lim r →∞ u → constant . o The ab ov e conjecture draft is presen ted f or the in viscid case. A corresp o nding in v esti- gation fo r viscous flows , with the steady cost defined by Eq. (41) is a natural extension and app ears equally w orthy of rigo rous analysis. If the in viscid ve rsion w ere to b e estab- lished, it would suggest a nece ssary condition for stabilit y expressed purely in terms o f the con vectiv e-a cceleration nor m—a p ersp ectiv e that differs fundamen tally fro m classical energy-based criteria. Such a result w o uld pro vide a new structural lens on in viscid stabilit y distinct fr o m classical energy–Casimir or eigen v alue- ba sed criteria, though it w ould not ap- ply to configurations suc h as pure shear flow s, for whic h the conv ectiv e acceleration v anishes iden tically . The rotating-cylinder example suggests that minimization of the steady functional A s ma y enco de information ab out the invisc id limit of viscous flow s. In particular, the v aria- tional closure recov ers the asymptotic circulation obtained from the Na vier–Stok es b oundar y- la y er analysis. This observ ation raises the p ossibilit y that steady minimization of A s ma y act as a selection criterion for iden tifying, within Euler’s family , the member corresponding to the inv iscid limit of Na vier–Stok es solutions. This p ossibilit y is presen ted in the followin g conjectural statemen t. Conjecture D raft 2 (Irrotational Inviscid Limit) : Let ¯ u ν b e a s mo oth steady solution o f the t w o- dimensional incompres sible Na vier–Stokes equations around a smo oth stationary b o dy B , with ¯ u ν → ( U, 0) at infinit y . Assume that, fo r sufficien tly small viscosit y ν , t he viscous b oundary la yer remains confined to a thin neigh b orho o d of ∂ B , and that outside this b oundar y lay er the flow is well-appro ximated by an irrotational field. Let U outer denote the family of admissible irrotational flow s around the b o dy , typ ically expresse d in the form (34). Supp ose that, aw ay from the b o undary lay er, ¯ u ν con verges (in a n appropriate sens e) to an irr otational flow ¯ u 0 as ν → 0. Then ¯ u 0 m ust b e a lo cal minimizer of the steady in viscid cost A s = 1 2 R Ω | u · ∇ u | 2 d x o v er the family U outer . In particular, t he in viscid limit tak es the form ¯ u 0 = u 0 + Γ ∗ u 1 , where Γ ∗ is the minimizer of A s in Eq. (35) o v er U outer. Finally , we mus t em phasize that the t w o-w a y equiv a lence established b y The orem 2 relies on uniqueness of sufficien tly smo oth solutions of Na vier-Stok es and Euler equations. In regimes where only w eak or ro ugh solutions are av ailable and uniqueness is forfeited, the PMPG minimization problem ma y no longer b e we ll-defined in its classical f o rm: the in tegra nd ma y fail t o lie in L 2 , and the t ime deriv a t ive u t ma y not b e defined p oin t wise. Ev en if the minimization problem w ere reform ula ted in a w eak er functional setting—suc h as restricting to flow s that are absolutely contin uous in time and redefining the cost in a weak er 31 norm (e.g., H − 1 )—the tw o- w ay equiv alence need no t p ersist. The Na vier–Stoke s or Euler equations may admit multiple admissible solutions, whereas the asso ciated minimization problem could pro duce a unique ev olution. In suc h a situation, it b ecomes nat ura l to ask whether a suitably defined minimization framew ork might serv e as a selec tion principle for iden tifying ph ysically relev ant solutions. This question is closely related to the longstanding problem of conv ergence of Nav ier–Stok es solutions to Euler solutions a s viscosit y v anishes [49 – 51]. As stated by Eyink [52], “ unde r very mo dest ass ump tions, the visc ous Navier-Stokes sol ution must tend in the limit Re → ∞ to a we ak solution of the in c o mpr essible Euler e quations.” How ev er, w eak solutions of Euler are generally non-unique [53]. So, a selection criterion is needed [54]. Among the f amily of w eak solutions of Euler’s equation, whic h solution arises as the zero-viscosit y limit? The form ulation of a minimization principle capable of prov iding suc h a selection mec hanism constitutes a c hallenging problem in mat hematical fluid dynamics. The presen t work do es not resolv e this question; rather, it merely p oints to a p ot ential cost functional for suc h a se- lection principle: a suitable norm of the pressure fo r ce, required to enforce incompressibilit y , follo wing the philosoph y of Gauss. 11 Concl usion In this pap er, w e presen t a mathematical analysis of the principle of minim um pressure gradien t (PMPG). First, we establish t w o- w ay equiv alence b etw een the PMPG and t he incompressible Na vier–Stok es equations (INSE). W e prov e that a candidate smo oth flow field is a solution of the INSE if and o nly if its instan taneous ev olution minimizes , at ev ery instan t, the L 2 -norm of the pressure force r equired to ensure incompressibilit y . W e show that the PMPG is precisely the minimization form ulat io n of the Lera y-Helmholtz pro jection on t o the space of div ergence-free vec tor fields. An y admissible instantaneous ev olution (e.g., onset of separation or transition) resulting from the INSE m ust necessarily minimize the PMPG cost. Conv ersely , any other kinematically admissible ev olutio n, requiring a strictly larger pressu re fo r ce to ensure the same constrain ts, fails to satisfy the INSE. This tw o- w ay equiv alence sho ws that the PMPG presen ts a minimization p ersp ectiv e through whic h in tricat e incompress ible flo w b eha viors ma y b e in terpreted. In a finite-dimensional setting with div ergence-free mo des, we show that the PMP G yields the same dynamics as classical Galerkin pro jection. How ev er, whe n non-divergenc e-free mo des are adopted, the mixed Galerkin form ula t ion and the PMPG ma y pro duce differen t dynamical systems. Moreo v er, the PMP G pro vides a natural gene ralization of classical Galerkin pro jection b ey ond linear mo dal expansions, a ccommo dating nonlinear and non- mo dal represen tations. W e then examine the relation b etw een instantaneous dynamical minimization and steady v a r ia tional selec tion, including its connection to the v ariational theory of lif t. The cen tral question is: since the flo w evolv es at ev ery instan t b y minimizing t he PMPG cost A ( u t ; u ), do es conv erg ence to a steady equilibrium ¯ u imply that ¯ u minimizes the steady cost A s ( u ) := A ( 0 ; u )? W e show that the answe r is negativ e for a general dynamical system by providing a coun terexample. Ho w ev er, if additional structural prop erties are imp osed (e.g., monotonic non-increase of the pressure-gradien t norm in a neighborho o d of ¯ u ), the conclusion follow s. 32 W e also presen t a discrete-time formulation of the PMPG. In particular, w e sho w that if the implicit discrete-time PMPG sc heme con v erges to a steady equilibrium, then the steady limit m ust minimize the steady cost A s (the L 2 -norm of t he pr essure g radien t) . Motiv ated b y these observ ations, we formulate t w o conjectures that w arra n t further mathematical in- v estigation: (i) whether a stable equilibrium neces sarily minimizes the steady cost A s , and (ii) whether the irro tational inv iscid limit of Na vier-Stokes solutions o v er a tw o- dimensional smo oth b o dy minimizes A s (i.e., the L 2 -norm of the con v ectiv e acceleration in the in viscid setting). Ac knowledgmen ts The author is gra teful to the fruitful discussions with Professors Ka rthik Duraisamy , Neelesh P ata nk ar, Jorn Lo viscac h, Anton y Jameson, Ahmed Roman, and Scott Bollt. The author w ould also lik e to a c knowle dge the supp ort of the Nat io nal Science F oundation gran t num b er CBET-2332556. References [1] C. F. G auß. ¨ Ub er ein neues allgemeines grundgesetz der mec hanik. Journal f¨ ur die r eine und ange w andte Mathematik , 18 29(4):232 – 235, 1829. [2] E. Ramm. Principles of least action and of least constrain t. GAMM-Mitteilungen , 34(2):164– 182, 2011 . [3] J.G. P apastavridis. Analytic a l me chanics: a c om pr ehensive tr e atise on the dynamics of c onstr aine d systems – R eprint e di tion . W ord Scien tific Publishing Compan y , 2 014. [4] F. E. Udw adia and R. E. Kalaba . Analytic al dynamics . 1996. [5] K. Anand and H. T aha. Mec hanics a s a con ve x quadratic programming problem with application to incompressible flow s. IEEE Contr ol Systems L etters , 2025 . [6] H. E. T aha, C. Gonzalez, and M. Shorbagy . A minimization principle f or incompressible fluid mec hanics. Physics of Fluids , 35(12) , 2023. [7] A. J. Chorin. Numerical solution of the Na vier-Sto kes equations. Mathematics of c om- putation , 2 2(104):74 5 –762, 196 8. [8] R. T emam. Sur l’appro ximation de la solution des ´ equations de na vier-stok es par la m ´ etho de de s pas fractionnaires ( ii). A r chi v e for r ational me chanics and analysis , 33:377– 385, 1 969. [9] P . Moin and J. Kim. On t he numerical solution of time-dep enden t viscous incompressible fluid flo ws in v olving solid b o undaries. Journal of c omputational p hysics , 35(3):3 8 1–392, 1980. 33 [10] P . M. Gr esho and R. L. Sani. On pressure b oundary conditions for the incompress- ible navier-stok es equations. International journal fo r numeric al metho ds in flu ids , 7(10):1111 –1145, 1987. [11] A. J. C horin and J. E. Marsden. A mathematic al intr o duction to fluid m e chanics , v olume 3. Springer, 1 990. [12] C. G . J. Jacobi. Ha milton-jacobi-theorie. V orlesunge n ¨ ub er analytische Me chanik: Berlin 1847 /48 Nac h ein e r Mitschrift von Wilhelm Scheibner , pages 216–264, 1996. [13] C. Lanczos. The variational principles o f me chanics . Courier Corp oration, 1970 . [14] F. Bullo and A. D. Lewis. Ge ometric c o n tr o l of me chanic al systems . Applied Mathe- matics. Springer-V erlag , Berlin, Germany , 2004. [15] T. Ka m b e. Geometrical theory of fluid flo ws and dynamical systems . Fluid dynamics r ese ar ch , 30 ( 6 ):331–378 , 200 2 . [16] H. Bhatia, G. Norgard, V. P a scucci, and P .-T. Breme r. The helmholtz-ho dge dec omp osi- tion—a surv ey . IEEE T r an sactions o n visualiza tion and c omputer gr aphics , 19(8 ):1386– 1404, 2012. [17] H. E. T aha and K. Anand. V a riational pro jection of na vier-stok es: Fluid mec ha nics as a quadratic programming problem. Physics R evi e w F luids , 2025. [18] V. I. Arnold. Sur la g´ eom ´ etrie diff ´ eren tielle des gr oup es de lie de dimension infinie et ses applications ` a l’h ydro dynamique des fluides parf a its. In Annales de l’institut F ourier , v olume 16, pages 319–361, 1966. [19] T. Kam b e. Ge ometric al the ory of dynamic al systems and fluid flows , volume 23. W orld Scien tific Publishing Co Inc, 2009. [20] A. Prohl. Pr oje ction and quasi-c om pr essibility m etho ds fo r solving the inc ompr essib l e Navier-Stokes e quations . Springer, 1997 . [21] J.-L. G uermond, P . Minev, and J. Shen. An ov erview of pro jection metho ds for in- compressible flo ws. Computer metho ds in applie d me chan i c s and engine erin g , 19 5 (44- 47):6011– 6 045, 2006 . [22] C. Hirsc h. Numerical computation of internal and external flo ws. Computational meth- o ds fo r invis c id and visc ous flows , 2, 1 990. [23] G. Badin and F. Crisciani. V ariat io nal form ulation of fluid and geophy sical fluid dy- namics. Me ch . Symme tries Con servation L aws , 1 2, 2018. [24] P . J. Morrison, T. Andreussi, and F. Pe goraro. Lagrangian and dirac constraints for the ideal incompressible fluid and magnetoh ydro dynamics. Journal of Plasma Physics , 86(3):8358 60301, 2020. 34 [25] J. W. Sanders, A. C. D eV oria, N. J. W ash uta , G. A. Elamin, K. L. Sk enes, and J. C. Berlinghieri. A canonical hamiltonian form ulat io n of t he navier–stok es problem. Journal of F luid Me chani c s , 984:A27, 2024. [26] M. R eed and B. Simon. Metho ds of mo d ern m a thematic al physics. vol . 1 . F unctional analysis . Academic New Y ork, 1980. [27] V. Gira ult and P .-A. Ra viart. Finite element metho ds for Nav i e r- S tokes e quations: the ory a nd a l g orithms , v olume 5. Springer Science & Business Media, 2012. [28] P . Monk. F inite eleme n t metho d s for Maxwel l’s e quations . Oxford univ ersit y press, 2003. [29] H. Sohr. The Navier-Stokes e quations: An elementary functional analytic appr o ac h . Springer Science & Business Media, 201 2. [30] R. T emam. Navier–Stokes e quations: the ory and numeric al analysis , v olume 34 3 . Amer- ican Mathematical So ciet y , 2024. [31] G. P . Galdi. An in tro duction to the navier-stok es initial-b oundary v alue problem. In F undamental d i r e ction s in mathematic al fluid me chani c s , pages 1–7 0. Springer, 2000. [32] J. G . Heyw o o d. The na vier-stok es equations: on the existence, regularit y and deca y of solutions. Indiana University Mathematics Journal , 29(5):639 – 681, 1980. [33] H. Alh ussein and M. Daqaq. The princ iple of minim um pressure gradien t: An alternativ e basis for phys ics-informed learning of incompressible fluid mec hanics. AIP A dvanc e s , 14(4), 2024. [34] A. Elmaradny , A. A talla h, and H. E. T aha. Minimizing nature’s cost: Exploring data- free phy sics-informed neural netw ork solv ers fo r fluid mech anics applications. Physics of F luids , 2025. [35] H. Sababha, A. Elmaradny , H. T a ha, and M. Da qaq. A v ariational computatio nal-based framew ork fo r unsteady incompressible flow s. Computer Metho ds in Applie d Me chanics and En g ine ering , 2025. [36] E. T o nti. V ariatio nal form ulat io n for eve ry nonlinear problem. In ternational Journal of Engine ering Sci e nc e , 22(11-12 ):1343–13 71, 19 84. [37] B. Da cor o gna. Intr o duction to the Calculus of V ariations . W o r ld Scien tific Publishing Compan y , 2014. [38] C. Gonzalez and H. E. T aha. A v ariat ional theory of lift. Journal of Fluid Me chanics , 941, 2 022. [39] D. G. Crigh ton. The Kutta condition in unsteady flo w. Annual R e view of Fluid Me- chanics , 17(1):411–4 4 5, 1985. 35 [40] H. E. T aha and C. G onzalez. Refining kutta’s flo w o v er a flat plate: Necessary conditions for lift. AIAA Journal , 61 (5):2060– 2 068, 2023 . [41] L. Prandtl and O. Tietjens. Hydro-und aeromec hanik. b d, 1931. [42] L. Ra yleigh. On the ir r egula r fligh t of a t ennis ball. Messenger of Mathematics , 7 :1 4–16, 1877. [43] D. W. Mo ore. The flo w past a rapidly rotating circular cylinder in a uniform stream. Journal of Fluid Me chanics , 2(6):54 1 –550, 195 7. [44] M. B. Gla uert. The flo w past a rapidly rotating circular cylinder. Pr o c e e dings of the R oyal So ciety of L ond on. Seri e s A. Mathematic al and Physic al Scienc es , 242(1 228):108– 115, 1 957. [45] M. Shorbagy and H. T a ha. Magnus force estimation using gauss’s principle of least constrain t. AIAA Journal , pages 1–8 , 202 4. [46] G. K. Batc helor. An intr o duction to fluid dynamics . Cam bridge univ ersit y press, 2000. [47] P . J. Sc hmid, D. S. Henningson, and D. F. Jank o wski. Stabilit y and transition in shear flo ws. applied mathematical sciences, v o l. 142 . Appl. Me ch. R ev. , 55 ( 3 ):B57–B59, 2002. [48] Charles Hirsc h. Numeric al c omputation of internal and ex ternal flows: c omputational metho ds f o r invi s cid a n d vis c ous flows , v olume 2 . Wiley , 1997. [49] N. Masmoudi. The euler limit of the na vier-stok es equations, and rota ting fluids with b oundary . Ar chi v e for r ational me chanics and analysis , 142:375–394 , 1998. [50] A. J. Ma jda, A. L. Bertozzi, and A. Oga w a. V orticit y and incompressible flo w. cambridge texts in applied mathematics. Appl. Me ch. R ev. , 55(4 ):B77–B78, 2002. [51] N. Masmoudi. Remarks ab out the in viscid limit of the na vier–stok es system. C o m mu- nic ations in mathematic al physics , 270:777 – 788, 2007 . [52] G. L. Eyink. Josephson-anderson relation and the classical d’alem b ert paradox . Physic al R eview X , 1 1(3):03105 4, 2021. [53] C. De Lellis and L. Sz´ ek elyhidi Jr. The Euler equations as a differen tial inclusion. A nnals of mathematics , pages 14 1 7–1436, 2 009. [54] G. Eyink. Onsager’s ‘ideal turbulence’theory . Journal o f Fluid Me cha n ics , 98 8:P1, 2024. 36