Strong Sard Conjecture for analytic sub-Riemannian structures ...trelat/SRGI/Belotto.pdfStrong Sard...

25
Strong Sard Conjecture for analytic sub-Riemannian structures in dimension 3 Andr´ e Belotto Aix-Marseille Universit´ e Sub-Riemannian Geometry and Interactions September 07, 2020

Transcript of Strong Sard Conjecture for analytic sub-Riemannian structures ...trelat/SRGI/Belotto.pdfStrong Sard...

  • Strong Sard Conjecture for analyticsub-Riemannian structures in dimension 3

    André Belotto

    Aix-Marseille Université

    Sub-Riemannian Geometry and Interactions

    September 07, 2020

  • Collaborators

    Alessio Figalli Adam Parusiński Ludovic Rifford

  • Preliminaries

    In this talk, we consider:

    A smooth connected manifold M of dimension n ≥ 3;A totally nonholonomic regular distribution ∆ of rank k < n;

    A Riemannian metric g over M.

    Local meaning: For every point x ∈ M, there is an openneighbourhood V where ∆ is locally parametrized by k linearlyindependent vector fields:

    {X 1x , . . . ,X kx }

    satisfying the Hörmander bracket generating condition

    Lie{X 1x , · · · ,X kx

    }(y) = TyM ∀y ∈ V.

    During the seminar: V = M and ∆ = Span{X 1, . . . ,X k}.

  • Preliminaries

    In this talk, we consider:

    A smooth connected manifold M of dimension n ≥ 3;A totally nonholonomic regular distribution ∆ of rank k < n;

    A Riemannian metric g over M.

    Local meaning: For every point x ∈ M, there is an openneighbourhood V where ∆ is locally parametrized by k linearlyindependent vector fields:

    {X 1x , . . . ,X kx }

    satisfying the Hörmander bracket generating condition

    Lie{X 1x , · · · ,X kx

    }(y) = TyM ∀y ∈ V.

    During the seminar: V = M and ∆ = Span{X 1, . . . ,X k}.

  • Horizontal paths

    An absolutely continuous curve γ : [0, 1]→ M is called horizontalwith respect to ∆ if it satisfies

    γ̇(t) ∈ ∆(γ(t)) for a.e. t ∈ [0, 1].

    Trajectories of a control system: Fix x ∈ M and denote byx(·; x , u) : [0, 1]→ M the solution of:

    ẋ(t) =k∑

    i=1

    ui (t)Xi (x(t)) for a.e. t ∈ [0, 1] and x(0) = x

    where u = (u1, · · · , uk) ∈ Ux ⊂ L2([0, 1],Rk).

    γ is horizontal ⇐⇒ it is a solution of the Cauchy problem.

  • Endpoint Mapping and singular controls

    The End-Point Mapping from x is defined as

    Ex : Ux −→ Mu 7−→ x(1; x , u).

    A control u ∈ Ux ⊂ L2([0, 1],Rk) is said to be singular if it is acritical point of E x . We consider the set of singular controls

    Sx = {u ∈ Ux ; u is singular} ⊂ L2([0, 1],Rk)

    Consider the set of critical values of Ex :

    X x := Ex (Sx) ⊂ M.

  • The Sard Conjecture

    Sard’s Conjecture

    For every x ∈ M, the set X x has Lebesgue measure zero.

    Difficulty: In infinite dimension, Sard’s Theorem is known tofail (Bates and Moreira)

    If dim(M) ≥ 4: Partial results for certain Carnot groups (ofprescribed ranks and steps):

    Le Donne, Montgomery, Ottazzi, Pansu, and Vittone (2016).Boarotto and Vittone (2019)

  • The Sard Conjecture in dimension 3 (dimM = 3 andrank ∆ = 2)

    The Martinet surface is the set:

    Σ :={x ∈ M |∆(x) + [∆,∆](x) 6= TxM

    },

    Every singular horizontal path must be contained in Σ, that is:

    X x ⊂ Σ, ∀x ∈ M.

    Sard’s Conjecture in dimension 3

    ∀x ∈ M, the set X x has 2-dimensional Hausdorff measure zero.

    Strong Sard’s Conjecture in dimension 3

    ∀x ∈ M, the sets X x have Hausdorff dimension at most one.

  • Main results when dimM = 3.

    Theorem (Zelenko and Zhitomirskii, 1995)

    Assume that ∆ is generic (in respect to the Whitney C∞-topology)than all the sets X x have Hausdorff dimension at most one.

    Furthermore, whenever ∆ is generic, they prove that the Martinetsurface Σ is smooth.

    Theorem (Belotto da Silva and Rifford, 2018)

    Suppose that the Martinet surface Σ is smooth. Then for everyx ∈ M the set X x has 2-dimensional Hausdorff measure zero.

  • Main results when dimM = 3 and M and ∆ are analytic.

    Theorem (Belotto da Silva and Rifford, 2018)

    Under additional hypothesis on the singular set of Σ (which werecall later), for every x ∈ M the set X x has 2-dimensionalHausdorff measure zero.

    Theorem (Belotto da Silva, Figalli, Parusinski and Rifford)

    For every x ∈ M, the set X x is a countable union of semianalyticcurves and it has Hausdorff dimension at most 1.

    Combining this result with (Hakavuori and Le Donne 2016) we get:

    Corollary

    Every singular minimizing geodesic γ : [0, 1]→ M is of class C 1.

  • Characteristic line foliation

    Singular horizontal paths are tangent to TΣ and to ∆.

    Lemma

    Singular horizontal paths are concatenation of leaves (of finitelength) of the characteristic line foliation F .

  • Generic characteristic line foliation.

    (Zelenko, Zhimtomirskii, 1995) shows that for ∆ generic:

    (i) The Martinet surface Σ is smooth;

    (ii) The singular points of F are isolated;

    (iii) The singular points of F can be of the following types:

    During the seminar: We will assume (i) and (ii).

  • Characteristic line foliation (local models)

    We may consider the following two local models:

    Vector-fields: We assume that ∆ = Span{∂x1 ; ∂x2 + A(x)∂x3}.

    [X 1,X 2] = h(x)∂x3

    and we conclude that:

    Martinet : Σ = {h(x) = 0}Foliation :Z = X 1(h)X 2 − X 2(h)X 1.

    One-form: We assume that ∆ = Span{µ} ⊂ Ω1M .

    µ ∧ dµ = h · volM , iΣ : Σ→ M

    and we conclude that:

    Martinet : Σ = {h(x) = 0}Foliation : η = i∗Σ(µ).

  • Controlled divergence (preliminary)

    Let S be a surface. We have the equivalence:

    Z ←→ η if η = iZvolS .

    and the divergence of Z may be computed from η:

    dη = divS(Z) volS .

    Example: if S = R2 and volS = dx ∧ dy :

    Z = A(x , y)∂x + B(x , y)∂yη = A(x , y)dy − B(x , y)dx

    And we easily verify that:

    dη = (Ax + By )dx ∧ dy

  • Controlled divergence (key property)

    Since the generator µ of ∆ is regular, so:

    dµ = α ∧ µ+ h 〈µ, ∗µ〉−1 (∗µ)

    so if i : S → Σ ⊂ M is a map (gen. max. rank) into Σ:

    dη = i∗(dµ) = i∗(α ∧ µ) = i∗(α) ∧ η = i∗(α) ∧ (iZvolS)

    This allows us to conclude that Z has controlled divergence:

    divS(Z) ∈ Z(OS).

    Example: Recalling that Z = A∂x + B∂y :

    divS(Z)dx ∧ dy = (fdx + gdy) ∧ (Ady − Bdx)= (f · A + g · B)dx ∧ dy .

  • Blowing-up

    For the talk: We consider polar coordinate changes

    σ : S1 × R≥0 → R2(θ, r) 7→ (r cos(θ), r sin(θ))

    and we denote E = σ−1(0) = S1 × {0} the exceptional divisor.

    Formally: we work with blowings-up

    σ : (S,E )→ (R2, 0)

    where E = σ−1(0) = P1.

  • Elementary singular points.

    A singularity p ∈ R2 of Z = A∂x + B∂y is elementary if:

    Jac(Z) =[∂xA ∂yA∂xB ∂yB

    ]evaluated at p admits one eigenvalue with non-zero real part.

  • Example: Degenerate focus point

    After performing two blowings-up, all singular points are saddles.

  • Reduction of singularities of characteristic foliations.

    Theorem (Bendixson-Seidenberg)

    Let S be an analytic smooth surface and F be a line foliation onS . There exists a proper analytic morphism

    π : (S̃, Ẽ )→ (S,Sing(Z))

    which is (locally) given by a finite composition of blowings-up,such that the strict transform F̃ of the line foliation only hasisolated elementary singularities.

    For the characteristic line foliation, moreover:

    (i) F̃ only has saddle points;

    (ii) If Σ is smooth, F̃ is non-dicritical, i.e. tangent to Ẽ .

  • Monodromic convergent singular trajectories.

    Proposition

    If Z is analytic and p is a focus point, then either all trajectorieshave finite length, or all trajectories have infinite length.

  • Characteristic convergent singular trajectories.

    Lemma

    If there is one characteristic singular trajectory whose limit is thethe singular point p, then there are only a finite number of singulartrajectories with finite length (all of them characteristic), whoselimit is the singular point p.

  • Proof of Proposition: no singular point after blowing-up

    We consider the return map T : Λ→ Λ.

    Denote by γp,T (p) the trajectory between p and T (p) and:

    L : Λ → Rp 7→ length(γp,T (p))

    We can show that L is an analytic function and w.l.g. monotone.

    length(γp) =∑n∈N

    length(γT n(p),T n+1(p)) =⇒ length(γq) ≤ length(γp)

  • Proof of Proposition: general case

    In general, we consider two types of transitions and length maps:

    T1 :Λ1 → Λ2, T2 :Λ2 → Λ3L1 :Λ1 → R L2 :Λ2 → R

    T2 belongs to a Hardy field (Speissegger, 2018);Combining with reduction of singularities of metrics:

    ∃K > 0, s.t. length(γq,T2(q)) ≤ K · length(γp,T2(p))

  • General case: singular Martinet surface

    There are 2N distinct singular horizontal paths from z to x .

    Resolution of singularities: If one path has finite length, all pathswith the same “jumps” have finite length.

    Symplectic geometry: We obtain a contradiction.

  • Thank you for the attention!