You are on page 1of 12

The cut locus of a C 2 surface in the Heisenberg

group
Alessandro Socionovo, Supervisor: Nicola Arcozzi

My master thesis deals with some fine properties of the natural Carnot
distance in the Heisenberg group. In particular, we are interested in properties
related to the cut locus of a smooth surface. The cut locus of a closed subset
S of Rn , denoted by cut(S), is the set containing the endpoints of maximal
segments that minimize the distance to S (i.e. if a segment starting at S locally
minimize the distance from S, then the last distance-minimizing point of such
segment is a point of cut(S)). In the Euclidean case many properties of cut(S)
are well known if S is the C 2 boundary of an open set of Rn . For example the
cut loci of such surfaces are closed (this is the fact we will be most interested
in). Such results are still valid if the Euclidean metric is replaced with any
Riemannian metric with coefficients of class C 2 (see [1], [2]).
Not all the known properties of the cut locus in Riemannian geometry are
also known in the sub-Riemannian case. So our goal is to generalize and prove
some of them in the Heisenberg group, which has the simplest sub-Riemannian
structure. In particular we are very interested in proving that the cut loci of
surfaces which are the C 2 boundary of open sets in the Heisenberg group are
closed sets. We choose a regularity of class C 2 because there is already a
counterexample in the Riemannian C 1,1 case (see [6]).
At the moment we are not able to giva a proof of the closure of the cut
locus. So our current plan is to find some new properties of the Carnot distance
which may be related with the cut locus of a C 2 surface in the Heisenberg group
and that may be useful to prove its closure. Precisely we are investigating in
points conjugate to the surface S, since they are strictly connected with the
cut locus in the Riemannian case. In particular, in the Euclidean case such
points are related with the principal curvatures of S.

Now we enter into a more detailed description in which we explain our


approach to the problem. We refer the reader to the titles in the bibliography
for complete and rigorous arguments. Let’s start with a quick description of
the Heisenberg group and of its geodesics.

Definition 1. The Heisenberg group H is the unique analytic, nilpotent Lie


group whose background manifold is R3 and whose Lie algebra h has the
following properties:

1
1. h = V1 ⊕ V2 , where V1 has dimension 2 and V2 has dimension 1
2. [V1 , V1 ] = V2 , [V1 , V2 ] = 0 and [V2 , V2 ] = 0
Definition 1 is well posed thanks to the following
Proposition 2. Let G be a simply connected, nilpotent Lie group and let g
be the Lie algebra of G. Then the exponential map exp : g → G is a (global)
diffeomorphism.
Infinite explicit representations of the Heisenberg group can be obtained
by identifying, with a slight abuse of notation, H as (C × R, ·) = (R3 , ·). Here
points of H are denoted by x = (z, x3 ) = (x1 + ix2 , x3 ) = (x1 , x2 , x3 ) and the
group law · is given by
(z, x3 ) · (w, y3 ) = (z + w, x3 + y3 + αIm(zw)), α∈R (1)
In this coordinates left translation by a point x is denoted by Lx y := x · y. A
base of left invariant vector fiels for the first layer of h is
X1 = ∂x1 − αx2 ∂x3
(2)
X2 = ∂x2 + αx1 ∂x3
By Definition X3 = [X1 , X2 ] = 2α∂x3 is a left invariant base for V2 . This
provides a complete description of h, depending on α. In literature the most
used choices of α are α = 1/2, α = 1 or α = 2. A canonical isomorphism (see
[5]) allows passing from one representation to another. We choose α = 2.
In the following, points in H can also be denoted as P = (x, y, t) = (z, t) =
(x + iy, t). Consequently the base for V1 becomes X = ∂x − 2y∂t and Y =
∂y + 2x∂t .
The group has an important homogeneous structure given by the dilations
δλ (P ) = (λx, λy, λ2 t), λ > 0, as a metter of fact (δλ )∗ (dx∧dy∧dt) = λ4 dx∧dy∧
dt. So the homogeneous dimension of H with respect to the group dilatations
δλ is 4. The vector fields X and Y are homogeneous of order 1 with respect to
the dilations δλ , that is X(δλ (P )) = λXP .
Definition 3. The subbundle generated by the left invariant frame X, Y of
the tangent bundle T H of H is called horizontal bundle. We denote it with
HH or simply with H. Every frame given by a linear combination of X and Y
is called horizontal and every frame having no component neither long X and
long Y is called vertical. Similarly the subspace HP := span{XP , YP } of the
tangent space TP H is called horizontal section at the point P . Every vector in
HP is called horizontal and every vector having no components in HP is called
vertical.
Definition 4. Let Ω be an open set of H and let φ : Ω → R be a C 1 function,
in the euclidean sense. We define the horizontal gradient of φ to be
∇H φ := (Xφ)X + (Y φ)Y (3)

2
Let h·, ·i0 be the inner product in H that makes X and Y an orthonormal
frame. The Carnot-Carathéodory metric in H is the sub-Riemannian metric
induced by h·, ·i0 .

Definition 5. An absolutely continuous path γ : [a, b] → H is called horizontal


if there exist measurable functions a, b : [0, 1] → R such that

γ̇(t) = a(t)(X1 )γ(t) + b(t)(X2 )γ(t) , a.e in [0, 1] (4)

For an horizontal curve γ we define its horizontal length as


Z bq
LH (γ) := hγ̇(t), (X1 )γ(t) i20 + hγ̇(t), (X2 )γ(t) i20 dt
a
Z bp (5)
= a(t)2 + b(t)2 dt
a

Then the Carnot-Carathéodory distance between P and Q is

dC (P, Q) = inf LH (γ) (6)


γ

and the infimum is taken on horizontal curves that joining P and Q.


Actually the infimum is a minimum, which is realized by the length of a
geodesic. The unit-speed geodesics passing through the origin 0=(0,0,0) are
−iφt
 
iα(W ) 1 − e φt − sin(φt)
γ0,φ,W (t) = e ,2 (7)
iφ φ2

Here W is a unitary vector in H0 and α(W ) ∈ [0, 2π) is unique, determined


by γ̇0,φ,W (0) = W . Instead φ ∈ R has the property that the geodesic is length-
minimizing over any interval of length 2π/φ.
We observe some fundamental properties of the geodesics

i) γP,φ,W := LP (γ0,φ,W ) is a (unit-speed) geodesic passing through the point


P at t = 0, due to the left invariance of the vector fiels X1 and X2 under
left translation.

ii) δλ (γP,φ,W )(t) = γP,φ/λ,W (λt), λ > 0

iii) φ > 0 ⇐⇒ x3 -coordinate of γP,φ,W (t) increase with t.

iv) γ0,φ,W is uniquely determined by the two parameters W = γ̇(0) and φ.


Unlike the Riemannian case, the only initial velocity is not sufficient to
describe the geodesic.

v) γP,−φ,−W (−t) is the geodesic oriented in the opposite verse.

vi) If γP,φ,W and γP 0 ,φ0 ,W 0 have an arc in common, then φ = ±φ0 .

3
Now we are ready to study the cut locus of a suface in the Heisenberg
group. From now on S will denote a surface in H. If it is not specified, S
satisfies this two hypothesis
H1. S is the boundary of an open and connected set Ω and of H r Ω

H2. S is of class C 2 in the euclidean sense


Many of the results we will discuss later have been proved in [3] with less
restrictive hypotheses. However, due to our goals discussed in the introductory
part, we consider it appropriate to suppose H1 and H2.
The distance from a point P to S is

dS (P ) = inf dC (P, Q) (8)


Q∈S

Thanks to H1, we can also define the so-called signed distance from a point P
to S, that is (
−dS (P ), if P ∈ Ω
δS (P ) = (9)
dS (P ), if P ∈
/Ω
Property (iv) of H’s geodesics leads to a serious problem. Indeed, unlike
the Riemannian case, the normal geodesic to a point P ∈ S is not uniquely
determined by the normal vector of S at P . For this, it is not clear which is the
(locally) distance-minimizing geodesic which minimizes (locally) the distance
from S. This fact suggests us to introduce some new differential geometric
notions about S, which will be used to define the metric normal to S.
Definition 6. Let P ∈ S. Then TP S denote the euclidean tangent space
to S at P and ΠS P denote the euclidean plane in H tangent to S at P in
the euclidean sense. The direction tangent to S at P is VP S := TP S ∩ HP ,
while the direction normal to S at P is NP S := HP VP S. Finally let ν be
the euclidean exterior normal to S at P and let h·, ·i be the euclidean inner
product. The Pansu exterior normal to S at P , denoted by NP+ S, is the unique
vector v ∈ NP S such that hv, νi > 0.
A direct calculus provides the next result.
Proposition 7. Let S be implicitly defined by g(x, y, t) = 0. Then

VP S = span{(Y g)X − (Xg)Y }, NP S = span{∇H g}

Definition 8. A point C ∈ S is said to be characteristic if TC S = HC . We


denote by Char(S) the set of all characteristic points of S.
Remark 9. Each euclidean plane P of the form t = ax + by + c, a, b, c ∈ R, has
one and only one characteristic point, which is

C = (−b/2, a/2, c). (10)

4
Indeed the tangent space of P at (x, y, t) is spanned by (1, 0, a) and (0, 1, b).
They are both horizontal iff a = 2y and b = −2x. Any other euclidean plane
which is not of this form, has equation ax + by + c = 0. Such a plane is called
vertical and has no characteristic points. Sometimes, with a slight abuse of
notation, we said that a vertical plane P has characteristic point C at infinity
and that dC (P, C) = ∞ for all P ∈ P.

Definition 10. Let P ∈ S. The metric normal to S at P , denoted by NP S,


is the set of the points Q ∈ H such that dS Q = dC (Q, P ).

So the metric normal describe which points of H r S are “closest” to the


points of S. Similarly for P, Q ∈ S, one can ask what is the fastest path to
go from P to Q staying within S. This is the sub-Riemannian metric induced
by h·, ·i0 in S. Note that if such a path is realized by a geodesic γ of H, as a
consequence of (6) and Definition 6 it needs γ̇(0) ∈ VP S.

Example 11. Let’s compute VP S and NP S if S is the plane t−ax−by −c = 0,


a, b, c ∈ R and P ∈ S is the point P = (zP , tP ) = (xP , yP , tP ) = (xP , yP , axP +
byP + c). We use the results guven by Proposition 7.

g(x, y, t) = t − ax − by − c
Xg(P ) = −a − 2yP , Y g(P ) = −b + 2xP
XP = (1, 0, −2yP ), YP = (0, 1, 2xP )

By (10) the S’s characteristic point is C = (−b/2, a/2, c). Then we have

1
C − P = (b − 2xP , −a − 2yP , −2axP − 2byP )
2
NP S = span{Xg(P )XP + Y g(P )YP } = span{(yC−P , −xC−P , tC−P + 2|zP |2 )}
VP S = span{Xg(P )YP − Y g(P )XP } = span{C − P }

Lemma 12. Let S be the plane t − ax − by − c = 0, a, b, c ∈ R, with character-


istic point C = (−b/2, a/2, c). Let P = (xP , yP , tP ) ∈ S be noncharacteristic.
Then the CC distance between P and the characteristic point C of S is

b a
dC (P, C)2 = |xP + |2 + |yP − |2 (11)
2 2
that is the euclidean distance between the projections of P and C in C.

Proof. VP S = span{C − P }, then the euclidean segment joining C and P is


an H’s geodesic. Its length is the euclidean length of the projection in C.

Note that (8) and Definition 10 remain the same for any closed subset E
of H. The results below gives a geometic description of NP S and explain why
we have introduced the notions of Definitions 6 and 8.

5
Lemma 13. Let E be a closed subset of H and let P ∈ E. Let Q ∈ NP E
and γ : I → H be a length minimizing geodesic from P to Q, with I ⊂ R an
interval. Then γ(I) ⊂ NP E.

Theorem 14. Let Q ∈ S, then NP S is a nontrivial arc of the unique geodesic


γ passing through Q and such that

i) γ̇(0) ∈ NQ S

ii) the maximal length l > 0 over which γ is length minimizing in H is


l = πd(P, C), where C is the characteristic point of ΠQ S.

If ΠQ S has no characteristic points, then (ii) holds with l = ∞.

Now we give an improvement of the description of the normal metric to


a suface given by Theorems 13 and 14. These facts allow us to define the
exponential map. The exponential map, in turn, is the basis of the proof of a
fundamental regularity Theorem concerning the distance function.

Theorem 15. Let P be a plane in H and let P ∈ P be non-characteristic.


If P has a characteristic point C and Ω is one of the half-spaces having P as
boundary, then
h π π i
NP P = γP, 2 ,N + P − dC (P, C), dC (P, C) (12)
dC (P,C) P 2 2
and NC P = {C} is degenerate. If P is a vertical plane, then NP P is the
straight geodesic through P , in the direction NP P.

By Theorem 15 and (7) we have an explicit expression for the metric normal
to a plane.

Proposition 16. The metric normal to P = {t = 0} at P = (z, t), z = x + iy,


is the support of the geodesic arc
2
 
1 −iφσ |z|
γP (σ) = (w(σ), s(σ)) = z(1 + e ), (φσ + sin(φσ)) (13)
2 2

with φ = 2/|z| and |σ| ≤ π/φ.

Observe that by (13) we have


(
|w(σ)| = |z|(1 + cos(φσ))
2 (14)
s(σ) = |z|2 (φσ + sin(φσ))

which is, for σ fixed, |σ| ≤ π/φ, a parametrization of the set of points having
distance σ from P = t = 0.

6
Theorem 17. Let P ∈ S be noncharacteristic and let ΠP S be the euclidean
tangent plane to S at P . Then NP S is a nontrivial geodesic arc having end-
points in Ω and H r Ω. Moreover, NP S ∩ NP ΠP S is an nontrivial geodesic
arc containing P .
Theorem 17 leads us to define the exponential map of S.
Definition 18. Let P ∈ S. The oriented metric normal to S at P , denoted
by NP+ S, is the unique parametrization of NP S such that δS (NP+ S(σ), P ) = σ,
∀σ ∈ D(NP S(σ)).
This means that, if NP+ S is nontrivial, NP+ S(σ) ∈ Ω for σ < 0, NP+ S(σ) ∈
H r Ω for σ > 0 and NP+ S(0) = NP+ S.
Now if C is characteristic for S, then by Theorems 15 and 17 we have
NC S = NC+ S = {C}. Therefore suppose that P ∈ S is noncharacteristic and
that, locally near P , S has equation g(x, y, t) = 0, where g : H → R is C 2
in the euclidean sense (according to H2) and ∇g 6= 0 pointwise on S. The
euclidean tangent plane to S at P , ΠP S, has equation
∂x g(P )(x − xP ) + ∂y g(P )(y − yP ) + ∂t g(P )(t − tP ) = 0 (15)
If ∂t g(P ) 6= 0, thanks to (10) we know the explicit expression of the character-
istic point C of ΠP S. Then by Definition 6, Example 11 and Lemma 12, by a
direct calculus we have
|∇H g(P )| ∇H g(P )
dC (P, C) = , NP+ S =
2|∂t g(P )| |∇H g(P )|
Here ∇H g(P ) 6= 0 since P in noncharacteristic. So Theorems 15 and 17 allow
us to write NP+ S in terms of g’s partial derivatives: NP+ S(σ) = P · η(σ), where
−iφσ
 
iα 1 − e φσ − sin(φσ)
η(σ) = e ,2 (16)
iφ φ2
and φ = 4∂t g(P )/|∇H g(P )|, cos α = Xg(P )/|∇H g(P )|, sin α = Y g(P )/|∇H g(P )|.
If ∂t g(P ) = 0 ΠP S is a vertical plane, with characteristic point at infinity. Then
dC (P, C) = ∞ and (16) becomes
η(σ) = (eiασ , 0) (17)
Definition 19. Let C := {(P, s) : P ∈ S, s ∈ D(NP+ S)} ⊂ S × R. The
exponential map associated with S is the map
expS : C → H, expS (P, s) := NP+ S(s) (18)
We call this map exponential because it extends the notion of exponential
map in the Riemannian manifolds. Clearly S = S × {0} ⊂ C and expS (P, 0) =
P , ∀ P ∈ S. Then {expS (P, 0) : P ∈ S} = expS (S, 0) = S.
Let’s define
Unp(S) := {P ∈ H : ∃! Q ∈ S s.t. dS (P ) = dC (P, Q)}. (19)
Then we have

7
Theorem 20. The map expS is a homeomorphism of int(C) onto an open
subset of int(Unp(S)). Moreover, S ∩ int(U np(S)) = S ∩ expS (int(C)).

Theorem 20 can be easily improved. Indeed it is not difficult to show


that expS is a diffeomorphism in an open neighborhood of Srchar(S). To
do this suppose that P ∈ S is noncharacteristic and that U ⊂ S is an open
neighborhood of P free of characteristic points, parametrized by

U = {(u, v, f (u, v)) : (u, v) ∈ A}. (20)

Here A ∈ C is open and f : A → H is C 2 . In particular P is described by


P = (uP , vP , f (uP , vP )). Moreover the map

F : A × R → H, F (u, v, τ ) := expS ((u, v, f (u, v)), τ ) (21)

is an expression of expS in local coordinates and we have

Lemma 21. The matrix representing JF (u, v, 0) with respect to the basis
{∂u , ∂v , ∂τ } of A × R and the basis {X, Y, ∂t } of H is
 
Xf
1 0 |∇H f |
Yf 
 0 1 (22)

|∇H f | 
−Xf −Y f 0

Here we have denoted Xf = Xg and Y f = Y g with g(u, v, t) = t − f (u, v).

This shows that |JF (u, v, 0)| = 2dC (P, C) 6= 0, where C is the character-
istic point of ΠP S. Then we have the desired improvement of Theorem 20.
The proof of Lemma 21 is completely given in [3]. The main fact (it will be
fundamental from now on!!) is that from (16) we have an explicit expression of
the map F . Given its importance, we want to write this explicit expression of
F . To do this let’s fix the coordinates (z, t) = (x, y, t) and (z 0 , t0 ) = (x0 , y 0 , t0 )
for H and the coordinates (u, v, τ ) = (u + iv, τ ) = (w.τ ) for A × R. By (16)
we obtain

(z, t) = F (u, v, τ ) = N(u,v,f (u,v)) S(τ ) = (u + iv, f (u, v)) · η 0 (τ ), (23)

where
|∇H f |2
 
0 0 0 1 −iφτ
η (τ ) = (z (τ ), t (τ )) = (1 − e ), (φτ − sin(φτ )) (24)
4 8

and φ = 4/|∇H f |. Actually z 0 (τ ) = z 0 (u, v, τ ), since z 0 contain f in its expres-


sion. So
(z, t) = (u + iv + z 0 , t0 + f + 2Im((u + iv)z 0 )). (25)
Then to prove Lemma 21 just calculate the deivatives of F .
From Lemma 21 (and other minor facts) it follows

8
Theorem 22. If S is C k in the euclidean sense, k ≥ 2, then δS and ∇H δS are
of class C k−1 in the euclidean sense, in an open neighborhood of S r Char(S).

In particular, according to our hypothesis H1 and H2, we have always that


δS and ∇H δS are at least of class C 1 .
Now we are ready to define the cut locus of S.

Definition 23. Let P ∈ S and define Q ∈ Ω to be the endpoint, other than


P , of NP S. When NP S reduces to the only point P , we set Q = P . The set
of such points Q as P varies over S, denoted by KS or cut(S), is the cut locus
of S in Ω. The definition of the cut locus of S in H r Ω is the same, with
Q ∈ (H r Ω).

In other words cut(S) is the set of the metric normals’ endpoint of S. Below
KS denote the cut locus of S in Ω and NP S refers to the portion of metric
normal at P which lies in Ω.

Proposition 24. The cut locus KS of S has the following properties

i) KS has empty interior

ii) Char(S) ⊂ KS and each characteristic point of S is an accumulation


point of KS

iii) Let R ∈ (H r KS ). Then there is a unique geodesic γ from R to S such


that LH (γ) = dS (R), i.e. Unp(S) ⊂ KS

As we have said at the beginning of this paper, we think that KS is a


closed subset of H, but we are still unable to prove it. The first step we
would like to take is to find a link between Theorem 20 (actually, between its
“improvement”) and the cut locus KS . Clearly, by (iii) of Proposition 24, the
exponential map is differentiable in an open subset of H r KS ).

Definition 25. Let (P, s) ∈ C such that expS is not a local diffeomorphism
around (P, s) and, ∀ 0 < ε ≤ s, expS is a local diffeomorphism around (P, s−ε).
Then we call expS (P, s) conjugate point of S (or S-conjugate).

For example, characteristic points of S are conjugate points of S. We use


the name ”conjugate” for the similarity that is noted with the Riemannian
case. The next Lemma gives a limit condition in order that a point of H is
S-conjugate.

Lemma 26. Let Sτ denote the surface expS (S, τ ). Let P ∈ S be noncharac-
teristic and suppose that there is a σ > 0 such that expS (P, σ) is characteristic
for Sσ . Then expS (P, σ + ε) can not be S-coniugate, ∀ ε > 0.

9
Before giving the proof we need some preparation. Suppose that A is an
open subset of C and that S has implicit equation, around a noncharacteristic
point P
S = {(α(u, v), β(u, v), γ(u, v)) : (u, v) ∈ A} (26)
We want compute the tangent plane, ΠP S, of S at P and the distance between
P and the characteristic point C of ΠP S.
Since P ∈ S, we can write P = (α(u, v), β(u, v), γ(u, v)). So the tangent
plane is given by
x − α y − β t − γ

αu βu γu = 0 (27)

αv βv γv
Equation (27) becomes
1
t= ((γu βv − γv βu )(x − α) + (αu γv − αv γu )(y − β) + γ) (28)
αu βv − αv βu
if αu βv − αv βu 6= 0, while we have the vertical plane

(γu βv − γv βu )(x − α) + (αu γv − αv γu )(y − β) = 0 (29)

if αu βv − αv βu = 0. In (28) the characteristic point is


1
C= (αv γu − αu γv , γu βv − γv βu , c) (30)
2(αu βv − αv βu )

where c contains everything that in (28) does not multiply x and y. The
distance between P and C is (in the nonvertical case)
 2  2
2 αu γv − αv γu βu γv − βv γu
dC (P, C) = α+ + α+ (31)
2(αu βv − αv βu ) 2(αu βv − αv βu )

Proof of Lemma 26. We denote Pτ := expS (S, τ ) and we use the same nota-
tions of Lemma 21. Then we have a local parametrization of S around P of the
type (u, v, f (u, v)), with (u, v) ∈ A ⊂ C. Since expS is a local diffeomorphism,
we can restrict A just enough to have Sτ of class C 1 for all τ . This allow us
to derive F at each time τ .
Let τ > 0 be fixed and suppose that ΠPτ Sτ is nonvertical. This is sure
possible for a certain time τ < σ, otherwise we have that the distance from
Pτ and the characteristic point of ΠPτ Sτ should be ∞ for all τ ≤ σ. But this
contradicts our hypothesis. With these assumptions the parametrization of Sτ
is as in (26), with α = u + x0 , β = v + y 0 and γ = t0 . Then the Jacobian of the
map F at the time τ is

1 + x0u x0v x0τ


 

JF (u, v, τ ) =  yu0 1 + yv0 yτ0  (32)


0 0 0
tu tv tτ

10
and
|JF (u, v, τ )| 0 yu0 t0v − t0u (1 + yv0 )
=x τ +
(1 + x0u )(1 + yv0 ) − x0v yu0 (1 + x0u )(1 + yv0 ) − x0v yu0
(33)
x0v t0u − t0v (1 + x0u )
+ yτ0 + t0τ
(1 + x0u )(1 + yv0 ) − x0v yu0

For τ = σ we have, by hypothesis, d(Pσ , C) = 0, where C is the characteristic


point of ΠPτ Sτ . So by (31) we have

yu0 t0v − t0u (1 + yv0 )


= −2(u + x0 ) (34)
(1 + x0u )(1 + yv0 ) − x0v yu0
x0v t0u − t0v (1 + x0u )
= 2(v + y 0 ) (35)
(1 + x0u )(1 + yv0 ) − x0v yu0

Putting (34) and (35) into (33) and calculating all the derivatives (the explicit
calculus is just made in [3]) we obtain |JF (u, v, σ)| = 0.
Remark 27. As in Lemma 21 one have |JF (u, v, τ )| = 2dC (P, Cτ ) for all τ < σ.
Here Cτ is the characteristic point of the tanget plane at the time τ . Indeed
we have seen that at the time τ , Sτ has a C 1 implicit expression as in (26).
So we can restrict around the point Pτ to find an explicit expression of the
kind (u, v, f (τ ) (u, v)). Then just repeat the proof of Lemma 21 (see [3]) by
replacing f with f (τ ) . Since we have supposed that the point expS (P, τ ) is
noncharacteristic, the Jacobian of F must be other than 0 for all τ < σ. So
the characteristic points which S develops at a certain time σ (if they exist)
are all and only the conjugate points of the domain of F .
The notion of conjugate point, Lemma 26 and its consequences are the the
starting point for our study of the Carnot distance and the cut locus in the
Heisenberg group.
Now I am studying the topics covered in [8] to understand if, using similar
arguments, it is possible to extend such results to more general one. The aim
is to describe the global behavior of the conjugate points of S. Morever I am
looking for an analytic expression for the “local” conjugated points of S.

11
References
[1] P. Albano. On the cut locus of closed sets. Nonlinear Anal. 125 (2015), 398-405.

[2] P. Albano. On the stability of the cut locus. Nonlinear Anal. 136 (2016), 51-61.

[3] N. Arcozzi and F. Ferrari. Metric normal and distance function in the Heisenberg
group. Math. Z. 256 (2007), no. 3, 661-684.

[4] N. Arcozzi and F. Ferrari. The Hessian of the distance from a surface in the
Heisenberg group. Ann. Acad. Sci. Fenn. Math. 33 (2008), no. 1, 35-63.

[5] L. Capogna, D. Dainelli, S. D. Pauls and J. T. Tyson. An introduction to the


Heisenberg and the sub-Riemannian isoperimetric problem, volume 259 of Progress
in Mathematics. Birkhäuser Verlag, Basel (2007).

[6] C. Mantegazza, A. C. Mennucci. Hamilton-Jacobi equations and distance func-


tions on Riemannian manifolds. Appl. Math. Optim. 47 (2003), no. 1, 1-25.

[7] R. Montgomer. A tour of sub-Riemannian geometries, their geodesics and appli-


cations, volume 91 of Mathematical Surveys and Monographs. American Mathe-
matical Society, Providence, RI (2002).

[8] M. Ritoré. Tubular neighborhoods in the sub-Riemannian Heisenberg groups.


44 pages, accepted version to appear in Adv. Calc. Var, https://www.ugr.es/ ri-
tore/preprints/pr10.pdf, https://arxiv.org/abs/1703.01592 (2017).

12