EXTENDIBILITY AND BOUNDEDNESS OF INVARIANTS ON

SINGULARITIES OF WAVEFRONTS

T. A. MEDINA-TEJEDA

Abstract. We investigate necessary and sufficient conditions for the extendibil-

ity and boundedness of Gaussian curvature, Mean curvature and principal cur-
arXiv:2011.09511v1 [math.DG] 18 Nov 2020

vatures near all types of singularities on fronts. Also we study the divergence of
these geometrical invariants and we show how this is tightly related to a property
of approximation of fronts by parallel surfaces.

1. Introduction
Frontal is a class of surfaces with singularities that have been intensely studied
in these last years within the field of differential geometry and singularities, in
particular wave fronts [1, 2, 4, 5, 6, 7, 8, 9, 10, 11]. They admit a normal vector
field along their parametrizations defined even on singularities of these. More
precisely, a smooth map x : U → R3 defined in an open set U ⊂ R2 is called a
frontal if, for all p ∈ U there exists a unit normal vector field np : Vp → R3 along
x (i.e xu , xv are orthogonal to n), where Vp is an open set of U , p ∈ Vp . If the
singular set Σ(x) = {p ∈ U : x is not immersive at p} has empty interior we call
x a proper frontal and if (x, np ) : U → R3 × S 2 is an immersion for all p ∈ U we
call x a wave front or simply front.
The behavior of Gaussian curvature, Mean curvature and principal curvatures
near non-degenerated singularities on wave fronts have been widely studied in [8, 5,
9, 11]. However, in the degenerated case this is unknown, as well as the divergence
of these invariants has been little explored. For this reason it is natural to wonder
which properties of wave fronts determine one behavior or another on general types
of singularities. Also, there is a lack of literature about the geometry of singularities
of rank 0 (or corank 2) on wave fronts and our approach here allow us to study
them. It is known that a wavefront x has a neighborhood Ul of a singularity p
in which yl = x + ln is an immersion for l sufficiently small. This neighborhood
Ul may shrink as l is smaller, then is natural to ask when Ul can be hold fixed
for l arbitrarily small. We will see that, this last behavior is determined by the
divergence of the invariants at each type of singularity and also is related with the
extendibility of the principal curvatures.

2010 Mathematics Subject Classification. Primary 57R45; Secondary 53A05, 53A55 .

Key words and phrases. singular surface, frontal, front, relative curvature.
The author is supported by CAPES Grant no. PROEX-10359340/D.
1
2 T. A. MEDINA-TEJEDA

In section 2, we establish the notation, terminology and basic results that we

use. Most of it is introduced and obtained in [6]. In section 3 we introduce the rel-
atives principal curvatures which give us geometrical information near singularities
and are defined even on them. In section 4, we study singularities of rank 1 both
degenerated and non-degenerated. The theorems 4.4, 4.6 gives equivalent condi-
tions for boundedness and extendibility of the Gaussian curvature which generalize
the one found in [8] and similarly theorem 4.17 for the principal curvatures. By
last in section 5, we study the behavior of the invariants at singularities of rank 0,
obtaining results quite different from those obtained in the rank 1 case.

2. Fixing notation, definitions and some basic results

In this paper, all the maps and functions are of class C ∞ . We denote U and V
open sets in R2 when is not mentioned anything about them. Let x : U → R3 be
a smooth map, we call a tangent moving base (tmb) of x a smooth map Ω :
U →
M3×2 (R) in which the columns w1 , w2 : U → R3 of the matrix Ω = w1 w2 are
linearly independent smooth vector fields and xu , xv ∈ hw1 , w2 i, where h, i denotes
the linear span vector space. It is known that a smooth map x : U → R3 is a
frontal if and only if there exist tangent moving bases of x locally. Since we are
interested in exploring local properties of frontals, we always assume that we have
w1 ×w2
a global tmb Ω for x. We denote by n := kw 1 ×w2 k
the normal vector field induced
∂fi
by Ω. Let f : U → Rn be a smooth map, we denote by Df := ( ∂x j
), the differential
of f and we consider it as a smooth map Df : U → Mn×2 (R). We write Dfx1 ,
∂fi
Dfx2 the partial derivatives of Df and Df (p) := ( ∂x j
(p)) for p ∈ U . Also, vectors
in Rn are identified as column vectors in Mn×1 (R) and if A ∈ Mn×n (R), A(i) is
the ith -row and A(j) is the j th -column of A. The trace and adjoint of a matrix are
denoted by tr() and adj() respectively.
Let x : U → R3 be a frontal, Ω a tmb of x. Denoting ()T the operation of
transposing a matrix, we set the matrices of the first and second fundamental
forms:    
E F T e f
I= := Dx Dx, II = := −DxT Dn.
F G f g
The Weingarten matrix α := −IIT I−1 is defined in Σ(x)c . Also, we set the matri-
ces:    
EΩ FΩ T eΩ f1Ω
IΩ = := Ω Ω, IIΩ = := −ΩT Dn,
FΩ GΩ f2Ω gΩ
µΩ := −IITΩ I−1 T −1
Ω , ΛΩ := Dx Ω(IΩ ) , αΩ := µΩ adj(ΛΩ ).
If x is a frontal and Ω a tmb of x, we write simply Λ = (λij ) and µ = (µij ) instead
of ΛΩ and µΩ when there is no risk of confusion, λΩ := det(Λ) and TΩ (U ) as
the principal ideal generated by λΩ in the ring C ∞ (U, R). The matrix Λ and µ
satisfy Dx = ΩΛT and Dn = ΩµT (see [6]), thus Σ(x) = λ−1 Ω (0) and rank(Dx) =
rank(Λ).
 EXTENDIBILITY AND BOUNDEDNESS OF INVARIANTS ON WAVEFRONTS 3

Definition 2.1. Let x : U → R3 be a frontal, Ω a tmb of x, p ∈ Σ(x), we say

that p is a non-degenerated singularity if DλΩ (p) 6= (0, 0), in another case is called
degenerated.
Remark 2.2. This definition does not depend on the chosen tmb Ω because with
another tmb Ω̄, we have λΩ̄ = ρλΩ with ρ : U → R being a C ∞ -function that never
vanish.
The following propositions were proved in [6] and we are going to use these
frequently.
Definition 2.3. Let x : U → R3 be a frontal, Ω a tmb of x, the Ω-relative
curvature and the Ω-relative mean curvature are defined on U by KΩ = det(µΩ )
and HΩ = − 21 tr(αΩ ) respectively.
Proposition 2.4. [6, Proposition 3.17] Let x : U → R3 be a proper frontal, Ω
a tangent moving base of x, KΩ , HΩ , K and H the Ω-relative curvature, the
Ω-relative mean curvature, the Gaussian curvature and the mean curvature of x
respectively.Then,
(1) for p ∈ Σ(x)c , KΩ = λΩ K and HΩ = λΩ H,
(2) for p ∈ Σ(x), KΩ = lim λΩ K and HΩ = lim λΩ H,
(u,v)→p (u,v)→p
where the right sides are restricted to the open set Σ(x)c .
Proposition 2.5. [6, Theorem 3.5] Let x : U → R3 be a frontal and Ω a tmb of
x, then first and second fundamental forms have the following decomposition:
I = ΛIΩ ΛT , II = ΛIIΩ
which satisfy,
Λ(1)u IΩ ΛT(2) − Λ(1) IΩ ΛT(2)u + Ev − Fu ∈ TΩ
Λ(1)v IΩ ΛT(2) − Λ(1) IΩ ΛT(2)v + Fv − Gu ∈ TΩ

Proposition 2.6. [6, Theorem 3.22] Let x : U → R3 be a frontal, Ω a tangent

moving base of x and p ∈ Σ(x). Then,
(1) x : U → R3 is a front on a neighborhood V of p with rank(Dx(p)) = 1 if
and only if HΩ (p) 6= 0.
(2) x : U → R3 is a front on a neighborhood V of p with rank(Dx(p)) = 0 if
and only if HΩ (p) = 0 and KΩ (p) 6= 0.
Corollary 2.7. A frontal x : U → R3 is a front if and only if (KΩ , HΩ ) 6= (0, 0)
on Σ(x) for whatever tangent moving base Ω of x.

3. The relative principal curvatures

Proposition 3.1. Let x : U → R3 be a proper frontal, Ω a tmb of x, we have the
following equality, αΩ = αλΩ on Σ(x)c , in particular αΩ is symmetric on U .
4 T. A. MEDINA-TEJEDA

Proof. By proposition 2.5, I = ΛIΩ ΛT and II = ΛIIΩ , then for p ∈ Σ(x)c , α =

−IIT I−1 = −IITΩ ΛT (ΛT )−1 I−1
Ω Λ
−1 = µ Λ−1 . Thus, we have αλ = µ adj(Λ) =
Ω Ω Ω
αΩ . Since the Weingarten matrix α is symmetric, then αΩ as well. By density of
Σ(x)c and the smoothness of αΩ , this is symmetric on the entire U . 
As αΩ is symmetric, the eigenvalues −k1Ω , −k2Ω are real with k1Ω , k2Ω sat-
isfying the equation k 2 + tr(αTΩ )k + det(αTΩ ) = 0. Since KΩ = det(µΩ ) and
HΩ = − 21 tr(αΩ ), we have k 2 − 2HΩ k + KΩ = 0. Then,
q
k = HΩ ± HΩ2 − λΩ KΩ
3
q Let x : U → R be a frontal, q
Definition 3.2. Ω a tmb of x, we call the functions
k1Ω := HΩ − HΩ − λΩ KΩ and k2Ω := HΩ + HΩ2 − λΩ KΩ the relative principal
2

curvatures. We also define the following functions on U − Σ(x):

( √
H − H 2 − K if λΩ > 0,
k1 := √
H + H 2 − K if λΩ < 0.
( √
H + H 2 − K if λΩ > 0,
k2 := √
H − H 2 − K if λΩ < 0.
We√clarify that, the principal√curvatures of x are the functions defined by κ− :=
H − H 2 − K and κ+ := H + H 2 − K on U − Σ(x).
Remark 3.3. k1 and k2 are smooth functions on U − Σ(x) and they have similar
properties to the classical principal curvatures. Also their definitions do not depend
on the chosen tmb Ω inducing the same orientation of the normal vector field n.
If another tmb Ω̂ induces an opposite orientation of n, then the signs of these
functions are opposite as well. Observe that k1 k2 = K and k1 +k 2
2
= H on U −
Σ(x). In the case of non-degenerated singularities, if we make a suitable change of
coordinates k1 , k2 coincides with those functions defined in ([9], equation (2.6)).
Proposition 3.4. Let x : U → R3 be a proper frontal, Ω a tangent moving base
of x. Then,
(1) for p ∈ Σ(x)c , k1Ω = λΩ k1 and k2Ω = λΩ k2 ,
(2) for p ∈ Σ(x), k1Ω = lim λΩ k1 and k2Ω = lim λΩ k2 .
(u,v)→p (u,v)→p
q √
Proof. We have that k1Ω = λΩ H − λ2Ω H 2 − λ2Ω K = λΩ H − |λΩ | H 2 − K = λΩ k1
and similarly k2Ω = λΩ k2 on Σ(x)c . For p ∈ Σ(x), by smoothness of k1Ω , k2Ω and
density of Σ(x)c , k1Ω = lim λΩ k1 and k2Ω = lim λΩ k2 . 
(u,v)→p (u,v)→p

In [6](proof of proposition 3.19) was observed that making change of coordinates

h on a frontal x and taking Ω̂ := Ω ◦ h as tmb of x ◦ h, it results with new different
relative curvatures det(h)(KΩ ◦ h) and det(h)(HΩ ◦ h). However, if we choose the
 EXTENDIBILITY AND BOUNDEDNESS OF INVARIANTS ON WAVEFRONTS 5

tmb Ωh := (Ω ◦ h)Dh instead of Ω̂ when we make a change of coordenates, they

remain invariant.
Proposition 3.5 (Invariance property). Let x : U → R3 be a frontal, Ω a tmb
of x, h : V → U diffeomorphism, then the new relative curvatures of x ◦ h are
KΩh = KΩ ◦ h and HΩh = HΩ ◦ h. In particular, k1Ωh = k1Ω ◦ h, k2Ωh = k2Ω ◦ h
and λΩh = λΩ ◦ h.
Proof. Observe that, the matrix ΛΩh induced by Ωh is (Dh)−1 (ΛΩ ◦ h)Dh, then
λΩh = λΩ ◦ h and since D(n ◦ h) = Ωh µTΩh , we have (Dh)−1 µTΩ (Dh) = µTΩh and
therefore KΩh = KΩ ◦h. Also, µΩh adj(ΛΩh ) = det(Dh)Dh−1 µΩ adj(ΛΩ )adj(Dh−1 )
and using that tr(ABC) = tr(CAB) we get HΩh = HΩ ◦ h. Since, k1Ωh and k2Ωh
are written in terms of KΩh and HΩh , the result follows for them. 
Lemma 3.6. Let x : U → R3 be a wavefront, Ω a tmb of x, for each p ∈ Σ(x)
there exist locally an embedding yl : V → R3 , p ∈ V , such that Dyl is a tmb of
x, the matrix ΛDyl determined for this tmb is I − lαl , where l ∈ R+ and αl is the
Weingarden matrix of yl and I is the identity matrix.
Proof. For each t ∈ R, consider yt = x+tn, as Dx = ΩΛT and Dn = ΩµT we have
Dyt = ΩΛT +tΩµT , then yt has a singularity at q if and only if det(ΛT +tµT )(q) =
0. Making a direct computation det(ΛT +tµT ) = λΩ −2tHΩ +t2 KΩ and now taking
p ∈ Σ(x), by corollary (2.7), there exist l ∈ R+ such that det(ΛT + lµT )(p) =
−2lHΩ (p) + l2 KΩ (p) 6= 0. Thus, there exist a neighborhood V of p such that
yl : V → R3 is an embedding. Since, Dyl = Ω(Λ + lµ), Dyl is a tmb of x. We can
assume Dyl and Ω induce the same normal vector n (i.e det(ΛT + lµT ) > 0 on V ),
otherwise we can change the order of column in Ω from the beginning. Therefore,
we have Dyl (I − lαTl ) = Dyl − lDn = Dx = Dyl ΛTDyl , thus ΛDyl = I − lαl . 

Lemma 3.7. Let x : U → R3 be a wavefront, p ∈ Σ(x) and Ω = Dyl a tmb of x

as above, then
(1) IΩ = Il , IIΩ = IIl , µ = αl and µadj(Λ) = αl − lKl I.
(2) KΩ = Kl , HΩ = Hl + Kl l and λΩ = 1 + 2Hl l + Kl l2 .
(3) k1Ω = k1l (1 + lk2l ) and k2Ω = k2l (1 + lk1l ).
where Il , IIl , Kl , Hl , k1l , k2l are first fundamental form, second fundamental form,
Gaussian curvature, mean curvature and principal curvatures of yl respectively.
Additionally, rank(Dx(p)) = 1 if and only if yl is free of umbilical point on a
neighborhood of p. Similarly, rank(Dx(p)) = 0 if and only if yl has a umbilical
point at p of positive Gaussian curvature.
Proof.
(1) Applying the definition directly we get the first three equalities. By lemma
3.6 Λ = I − lαl , then µadj(Λ) = αl (I − ladj(αl )) = αl − lKl I.
(2) Using item (1), KΩ = det(µ) = det(αl ) = Kl , HΩ = − 12 tr(µadj(Λ)) =
− 12 tr(αl −lKl I) = Hl +Kl l and λΩ = det(Λ) = det(I−lαl ) = 1+2Hl l+Kl l2 .
6 T. A. MEDINA-TEJEDA

(3) q
Using the formulas in definition
q 3.2, item (2) and knowing that k1l = Hl −
Hl2 − Kl , k2l = Hl + Hl2 − Kl and Kl = k1l k2l , a simple computation
leads to item (3).
For the last part, by proposition 2.6 rank(Dx(p)) = 0 if and only if HΩ (p) = 0,
λΩ (p) = 0 and KΩ (p) 6= 0. On the other hand these conditions are equivalent to
Hl (p) = −Kl (p)l and 0 = 1 − 2l2 Kl (p) + l2 Kl (p) which is equivalent to Kl (p) =
1
l2
and Hl (p) = − 1l . Then yl has an umbilical point at p of positive Gaussian
curvature. Conversely, we have 0 < Kl (p) = Hl (p)2 , then 0 = 1 + 2lHl (p) +
l2 Hl (p)2 which imply Hl (p) = − 1l and therefore HΩ (p) = − 1l + l12 l = 0, by
proposition 2.6 rank(Dx(p)) = 0. Equivalently rank(Dx(p)) = 1 if and only if p
is not a umbilical point which is equivalent to have yl free of umbilical point on a
neighborhood of p. 

4. Singularities of rank 1
In the following propositions, we study the behavior at a singular point of rank
1 of the invariants function defined in previous section. The non-degenerated case
was investigated in [8, 5, 9, 11].

Proposition 4.1. Let x : U → R3 be a proper wavefront, Ω a tmb of x, then for

every p ∈ Σ(x) with rank(Dx(p)) = 1, the following always is satisfied:
(1) (k1Ω (p), k2Ω (p)) 6= (0, 0). In particular, if k1Ω (p) 6= 0 (resp. k2Ω (p) 6= 0),
then k2Ω (p) = 0 (resp. k1Ω (p) = 0). Also, HΩ (p) < 0 (resp. HΩ (p) > 0)
if and only if k1Ω (p) 6= 0 (resp. k2Ω (p) 6= 0).
(2) There is an open neighborhood V ⊂ U of p in which one of the functions
k1 , k2 has a C ∞ extension to V . More precisely, k1 (resp. k2 ) has a C ∞
extension if only if k1Ω (p) = 0 (resp. k2Ω (p)).
(3) One of the functions k1 , k2 in module diverge to ∞. More precisely,
lim |k1 | = ∞ (resp. |k2 |) if and only if k1Ω (p) 6= 0 (resp. k2Ω (p)).
(u,v)→p
(4) lim |H| = ∞.
(u,v)→p
(5) If KΩ (p) 6= 0 then lim |K| = ∞.
(u,v)→p

Proof.
(1) Observe that, for p ∈ Σ(x), k1Ω (p) = HΩ (p) − |HΩ (p)| and k2Ω (p) =
HΩ (p) + |HΩ (p)|. By proposition 2.6, HΩ (p) 6= 0, then just one of k1Ω (p),
k2Ω (p) is different of zero. Thus, the sub index of kiΩ (p) corresponding to
the non-zero value is determined bijectively by the sign of HΩ (p).
(2) By item (1), without loss of generality, we can assume that k1Ω (p) 6= 0. Let
V a neighborhood of p such that k1Ω 6= 0 on V , then by proposition 2.4,
k2 = λλΩΩkk1 k1 2 = kK1Ω
Ω
on V − Σ(x). Thus, kK1Ω
Ω
is a C ∞ extension of k2 to V .
 EXTENDIBILITY AND BOUNDEDNESS OF INVARIANTS ON WAVEFRONTS 7

(3) By item (1), without loss of generality, we can assume that k1Ω (p) 6= 0.
Let V be a neighborhood of p such that k1Ω 6= 0 on V , then k1 = kλ1Ω
Ω
on
V − Σ(x). Thus, for every p ∈ Σ(x) ∩ V , lim |k1 | = lim |k1Ω | = ∞.
(u,v)→p (u,v)→p |λΩ |
(4) Since HΩ (p) 6= 0 and by proposition 2.4, lim |H| = lim |HΩ | = ∞.
(u,v)→p (u,v)→p |λΩ |
|KΩ |
(5) Since KΩ (p) 6= 0 and by proposition 2.4, lim |K| = lim = ∞.
(u,v)→p (u,v)→p |λΩ |

Lemma 4.2. Let x : U → R3 be a proper wavefront, W ⊂ U a compact set, V ⊂ U
an open set, a : W → R a continuous function, Ω1 and Ω2 tangent moving bases
of x, then we have:
(1) If there exist a constant C1 > 0 such that |a| ≤ C1 |λΩ1 | on W then there
exist a constant C2 > 0 such that |a| ≤ C2 |λΩ2 | on W .
(2) TΩ1 (V ) = TΩ2 (V )
Proof.
(1) Setting A = I−1 T
Ω1 Ω1 Ω2 (change of basis matrix) and ρ = det(A), we have
Ω2 = Ω1 A, therefore ΛΩ1 = AΛΩ2 and λΩ1 = ρλΩ2 . Since |a| ≤ C1 |λΩ1 | =
C1 |ρ||λΩ2 | and choosing C2 as the maximum of C1 |ρ| on W , we get the
result.
(2) Using the proof of item (1), λΩ1 = ρλΩ2 with ρ 6= 0, then we have the
equality.

Lemma 4.3. Let x1 : U → R3 be a proper wavefront with U open connected,
W ⊂ U a compact set, V ⊂ U an open set, h : Z → U a diffeomorphism and Ω
a tmb of x. Setting x2 := x1 ◦ h and choosing Ωh = (Ω ◦ h)Dh as tmb of x2 we
have:
(1) There exist a constant C1 > 0 such that |e1 | ≤ C1 |λΩ |, |f1 | ≤ C1 |λΩ |,
|g1 | ≤ C1 |λΩ | on W if and only if there exist a constant C2 > 0 such that
|e2 | ≤ C2 |λΩh |, |f2 | ≤ C2 |λΩh |, |g2 | ≤ C2 |λΩh | on h−1 (W ). Where e1 , f1 ,
g1 and e2 , f2 , g2 are the coefficients of the second fundamental form of x1
and x2 respectively.
(2) e1 , f1 , g1 ∈ TΩ (V ) if and only if e2 , f2 , g2 ∈ TΩh (h−1 (V )).
Proof. Let us denote by II1 , II2 the matrices of the second fundamental forms
of x1 and x2 respectively. If det(Dh) > 0, then n2 = n1 ◦ h (in the case
det(Dh) < 0, n2 = −n1 ◦ h and it is analogous) therefore II2 = −DxT2 Dn2 =
−DhT DxT1 Dn1 Dh = DhT II1 Dh. This last equality expresses the coefficients
e2 , f2 , g2 as sum of multiples of the coefficients e1 , f1 , g1 and vice versa. Since
λΩh = λΩ ◦ h (see proposition 3.5) we get items (1) and (2) easily. 
Theorem 4.4. Let x : U → R3 be a proper wavefront with just singularities of
rank 1, Ω a tmb of x, p ∈ Σ(x). Let W ⊂ U be a compact neighborhood of p
8 T. A. MEDINA-TEJEDA

in which the relative principal curvature kiΩ 6= 0 does not vanish. Let kj be the
function that admit a C ∞ extension to W and K the Gaussian curvature, then the
following statements are equivalent:
(1) K is bounded on W − Σ(x).
(2) There exist a constant C > 0 such that |KΩ | ≤ C|λΩ | on W .
(3) There exist a constant C > 0 such that |kj | ≤ C|λΩ | on W .
(4) There exist a constant C > 0 such that |e| ≤ C|λΩ |, |f | ≤ C|λΩ | and
|g| ≤ C|λΩ | on W .
Proof.
• (1 ⇔ 2) As |K| = |K Ω|
|λΩ | on W − Σ(x) and by density of W − Σ(x) in W , it
follows the equivalence.
• (2 ⇔ 3) Since W is compact, kj , KΩ , kiΩ are continuous on W and kj = K kiΩ ,
Ω

from this last equality follows the equivalence.

• (4 ⇒ 1) we have that |eg − f 2 | ≤ 2C 2 |λΩ |2 on W , but by proposition 2.5,
2
EG − F 2 = (EΩ GΩ − FΩ2 )λ2Ω , then |K| ≤ E G2C −F 2 on W − Σ(x). Since
Ω Ω Ω
2C 2
2
EΩ GΩ −FΩ
is continuous on U and W is compact, K is bounded on W −Σ(x).
• (2 ⇒ 4) if we prove (4) locally on W , we can choose an open covering Bk
(open sets with the induced topology) of W in which (4) is satisfied in each
compact B¯k with constants Ck . Reducing this covering to a finite one, we
have finite constants Ck1 , .., Ckn and choosing C as the maximum of these
constants, (4) is satisfied globally on W .
To prove this locally, first, for each q ∈ W let us take a tmb Dyl as
in lemma 3.6 on a neighborhood V of q with yl free of umbilical point
(3.7) on V . Shrinking V if it is necessary, there exist a diffeomorphism
h : V 0 → V , such that yl ◦ h has derivatives as principal directions. By
lemmas 4.2 and 4.3, we can assume that Ω = Dyl being yl an embedding
with derivatives as principal directions. Thus, by lemmas 3.6 and 3.7 IΩ ,
IIΩ , αl and Λ = (λij ) = I − lαl are diagonal matrices. If rank(Λ(q)) = 1,
without loss of generality shrinking V to a compact neighborhood, we can
suppose that λ22 (q) = 1 − lαl22 (q) = 0, with λ11 6= 0 and − GgΩΩ = αl22 6= 0
on V . By proposition 2.5, f = 0, e = λ11 eΩ , g = λ22 gΩ = λΩ λg11 Ω
and by
gΩ eΩ 0 EΩ GΩ
hypothesis (2) | EΩ GΩ | ≤ C|λΩ | on V = V ∩W , thus |e| ≤ C|λ11 || gΩ ||λΩ |.
If we choose C 0 as the biggest maximum of the functions C|λ11 || EΩgΩGΩ | and
| λg11
Ω
| on V 0 , we get that |e| ≤ C 0 |λΩ |, |f | ≤ C 0 |λΩ | and |g| ≤ C 0 |λΩ | on V 0 .
On the other hand, if rank(Λ(q)) = 2, shrinking V to a compact neigh-
borhood, we can suppose that λ11 6= 0 and λ22 6= 0 on V . Thus, f = 0,
e = λ11 eΩ = λΩ λe22 Ω
, g = λ22 gΩ = λΩ λg11Ω
, then choosing C 0 as the biggest
eΩ gΩ
maximum of the functions | λ22 | and | λ11 | on V 0 = V ∩ W , we have |e| ≤
C 0 |λΩ |, |f | ≤ C 0 |λΩ | and |g| ≤ C 0 |λΩ | on V 0 .

 EXTENDIBILITY AND BOUNDEDNESS OF INVARIANTS ON WAVEFRONTS 9

Example 4.5. The wavefront x(u, v) = (u, 2v 3 + u2 v, 3v 4 + u2 v 2 ) (cuspidal lips)

1
with normal vector n = (2uv 2 , −2v, 1)(4u2 v 4 +4v 2 +1)− 2 has a singularity at (0, 0),
2
Gaussian curvature K = − (4u2 v4 +4v24v+1)2 (u2 +6v 2 )
with |K| ≤ 1. Observe that K
does not converge when (u, v) −→ (0, 0), then it is not extendable.

Figure 1. A wavefront with degenerated singularity of rank 1 at

the origin and Gaussian curvature bounded but non-extendable.

Theorem 4.6. Let x : U → R3 be a proper wavefront with just singularities

of rank 1, Ω a tmb of x, p ∈ Σ(x) with rank(Dx(p)) = 1. Let V ⊂ U be a
open neighborhood of p in which the relative principal curvature kiΩ 6= 0 does not
vanish. Let kj be the function that admit a C ∞ extension to V and K the Gaussian
curvature, then the following statements are equivalent:
(1) The Gaussian curvature K admits a C ∞ extension to V .
(2) KΩ ∈ TΩ (V ).
(3) kj ∈ TΩ (V ).
(4) e, f, g ∈ TΩ (V ).
where TΩ (V ) is the principal ideal generated by λΩ in the ring C ∞ (V, R)
Proof.
• (1 ⇔ 2) As KΩ = KλΩ on V − Σ(x) and by density of V − Σ(x) in V , it
follows the equivalence.
• (2 ⇔ 3) Since kj = KkiΩ , from this last equality follows the equivalence.
Ω

• (4 ⇒ 1) we have that eg −f 2 = φλ2Ω with φ ∈ C ∞ (V, R), but by proposition

2.5, EG − F 2 = (EΩ GΩ − FΩ2 )λ2Ω , then K = E Gφ −F 2 on V − Σ(x). Since
Ω Ω Ω
φ
2
EΩ GΩ −FΩ
is smooth on V , K has a C ∞ extension to V .
• (2 ⇒ 4) if we prove (4) locally on V , we can choose an open covering locally
finite Bk ⊂ V (open balls) of V , k ∈ N with a partition of the unity ψk
subordinated to this open cover in which e, f, g ∈ TΩ (Bk ) for every Bk .
For each k ∈ N there exist f1k , f2k , f3k ∈ C ∞ (V, R) such that e = f1k λΩ ,
f = f2k λΩ , g = f3k λΩ on Bk . Since that the supports of fsk ψk form families
10 T. A. MEDINA-TEJEDA

locally finite for s = 1, 2, 3, we have that fs := k fsk ψk ∈ C ∞ (V, R) for

P
s = 1, 2, 3, therefore e = f1 λΩ , f = f2 λΩ , g = f3 λΩ on V .
To prove this locally, first, for each q ∈ V let us take a tmb Dyl as in
lemma 3.6 on a neighborhood Z ⊂ V of q with yl free of umbilical point on
Z. Shrinking Z if it is necessary, there exist a diffeomorphism h : Z 0 → Z,
such that yl ◦ h has derivatives as principal directions. By lemmas 4.2 and
4.3, we can assume that Ω = Dyl being yl an embedding with derivatives
as principal directions on Z. Thus, by lemmas 3.6 and 3.7 IΩ , IIΩ , αl and
Λ = (λij ) = I − lαl are diagonal matrices. If rank(Λ(q)) = 1, without loss
of generality shrinking Z to a open neighborhood V 0 , we can suppose that
λ22 (q) = 1 − lαl22 (q) = 0, with λ11 6= 0 and − GgΩΩ = αl22 6= 0 on V 0 . By
proposition 2.5, f = 0, e = λ11 eΩ , g = λ22 gΩ = λΩ λg11 Ω
and by hypothesis
gΩ eΩ ∞ EΩ GΩ
(2) EΩ GΩ = φλΩ for some φ ∈ C (V, R), then e = φλ11 gΩ λΩ . Thus, we
get that e ∈ TΩ (V 0 ), f ∈ TΩ (V 0 ) and g ∈ TΩ (V 0 ).
On the other hand, if rank(Λ(q)) = 2, shrinking Z to a open neigh-
borhood V 0 , we can suppose that λ11 6= 0 and λ22 6= 0 on V 0 . Thus,
f = 0, e = λ11 eΩ = λΩ λe22Ω
, g = λ22 gΩ = λΩ λg11
Ω
, then we have e ∈ TΩ (V 0 ),
0
f ∈ TΩ (V ) and g ∈ TΩ (V ). 0

Let E2,p denote the local ring of smooth function germs at the point p. By [g]p we
denote the function germ at p with g ∈ C ∞ (U, R) a representative of the germ de-
fined on a neighborhood U of p. We define Jp as the ideal in E2,p generated by [λΩ ]p ,
Jˆp := {[g]p ∈ E2,p
1 : there exist C > 0 such that |g| ≤ C|λ | on some neighborhood of p}
Ω
1 : for some neighborhood U of p, g vanish on U ∩ λ−1 (0)}.
and JΣp := {[g]p ∈ E2,p Ω
These ideals satisfy Jp ⊂ Jˆp ⊂ JΣp , their definitions do not depend on the chosen
tmb Ω and when p is a non-degenerated singularity these three ideals are equal.
To see that, we can assume that p = 0 and making a change of coordinates, we
can assume that λΩ is equal to u or v, then applying the Hadamard lemma [3, Ch
IV, page 100], we obtain the result. From this and theorems 4.4 and 4.6 we have
the following corollary.

Corollary 4.7. Let x : U → R3 be a proper wavefront, Ω a tmb of x, p ∈ Σ(x)

a non-degenerated singularity with rank(Dx(p)) = 1. Let kj be the function that
admit a local C ∞ extension at p and K the Gaussian curvature, then the following
statements are equivalent:
(1) The Gaussian curvature K admits a local C ∞ extension at p.
(2) The Gaussian curvature K is locally bounded on some neighborhood of p.
(3) [KΩ ]p ∈ JΣp .
(4) [kj ]p ∈ JΣp .
(5) [e]p , [f ]p , [g]p ∈ JΣp .
Where JΣp = Jˆp = Jp in this case.
 EXTENDIBILITY AND BOUNDEDNESS OF INVARIANTS ON WAVEFRONTS 11

The equivalence between (1), (2) and (5) was obtained by K. Saji, M. Umehara,
and K. Yamada in [8].
k+1 k+2
Example 4.8. The wavefront (u, sin(ku) vk+1 , sin(ku) vk+2 ), k ∈ N, has as a tangent
moving base:
 
1 0  
cos(ku)k vk+1 1 1 0
Ω= , Λ =
k+1
k+2 0 sin(ku)v k
cos(ku)k vk+2 v
1 1
sin(ku)k 2 v k+2 ( k+1
 
− k+2 ) 0 1
IIΩ = √
0 1 
1 1 2
where  = 1 + v 2 + cos2 (ku)k 2 v 2k+4 ( k+1 − k+2 ) . Then, λΩ = sin(ku)v k , (eΩ gΩ −
f1Ω f2Ω ) ∈ TΩ therefore KΩ ∈ TΩ and by the proposition above the Gaussian
curvature K admit a C ∞ extension to R2 . Observe that, since II = ΛIIΩ (see
proposition 2.5), we have e, f, g ∈ TΩ .
Remark 4.9. The boundedness and extendibility of Gaussian curvature are con-
served under changes of coordinates in the domain, however they are not con-
served making changes at co-domain. The wavefront x(u, v) = (u, 2v 3 + u2 v, 3v 4 +
u2 v 2 + u2 ) with Gaussian curvature unbounded can be obtained from example 4.5
whose Gaussian curvatures is bounded, applying at co-domain the diffeomorphism
F (X, Y, Z) = (X, Y, Z + X 2 ). The same situation occurs with the wavefronts from
k+1 k+2
example 4.8 and (u, sin(ku) vk+1 , sin(ku) vk+2 + u2 ) which have extendable and non-
extendable Gaussian curvatures respectively.
Definition 4.10. Let x : U → R3 be a wavefront, Ω a tmb of x and p ∈ Σ(x).
We say that x is parallelly smoothable at p if there exist  > 0 and an open
neighborhood V of p such that rank(D(x + ln)(q)) = 2 for every (q, l) ∈ V × (0, )
or every (q, l) ∈ V × (−, 0).
Proposition 4.11. Let x : U → R3 be a wavefront, Ω a tmb of x, p ∈ Σ(x)
with rank(Dx(p)) = 1, KΩ (p) 6= 0 and HΩ (p) < 0 (resp. HΩ (p) > 0) then the
following statements are equivalents:
(1) x is parallelly smoothable at p.
(2) lim H = ±∞.
(u,v)→p
(3) lim k1 = ±∞ (resp. k2 ).
(u,v)→p
(4) lim K = ±∞.
(u,v)→p
(5) There exist an open neighborhood V of p in which λΩ does not change sign.
(6) There exist an open neighborhood V of p in which k2Ω (resp. k1Ω ) does not
change sign.
Proof.
12 T. A. MEDINA-TEJEDA

• (1 ⇒ 5) if x is parallelly smoothable at p if and only if there exist  > 0

and an open neighborhood V of p such that yt = x + tn|V is an immersion
for every t ∈ (0, ) (or every t ∈ (−, 0)) if and only if det(ΛT + tµT )(q) =
λΩ (q) − 2tHΩ (q) + t2 KΩ (q) 6= 0 for every t ∈ (0, ) (or every t ∈ (−, 0))
and q ∈ V . Shrinking V we can suppose this is connected, then we have
that λΩ (q) − 2tHΩ (q) + t2 KΩ (q) > 0 (or < 0) for every (q, t) ∈ V × (0, ),
thus taking the limit in both sides of this inequality when t tends to 0, we
get that λΩ ≥ 0 on V .
• (5 ⇔ 2) as H = H λΩ on U − Σ(x) and HΩ (p) 6= 0, follows the equivalence.
Ω

• (5 ⇔ 4) as K = K λΩ on U − Σ(x) and KΩ (p) 6= 0, follows the equivalence.

Ω

• (2 ⇔ 3) by proposition 4.1, there exist a neighborhood V of p such that k2

has a C ∞ extension and since k1 = 2H − k2 follows the equivalence.
• (3 ⇔ 6) there exist a neighborhood W of p such that HΩ < 0 and KΩ 6= 0
on W , then k1 6= 0 on W − Σ(x). Since Kk1Ω = k2Ω on W − Σ(x) and k2Ω = 0
on Σ(x)(by proposition 4.1), follows the equivalence.
• (6 ⇒ 1) there exist a neighborhood W of p such that HΩ < 0 and KΩ 6= 0
on W . By proposition 4.1 k1Ω (p) 6= 0, thus kK1Ω (p)
Ω (p)
6= 0, then using item
(6) there exist  > 0 and an open connected V of p such that kK2Ω Ω
does
k1Ω k2Ω k2Ω
not change sign and | KΩ | >  on V . Thus, if KΩ ≥ 0 (resp. KΩ ≤ 0)
then λΩ (q) − 2tHΩ (q) + t2 KΩ (q) 6= 0 for every (q, t) ∈ V × (−, 0) (resp.
V × (0, )) because λΩ (q) − 2tHΩ (q) + t2 KΩ (q) = 0 if and only if t = kK1Ω
Ω
(q)
k2Ω
or t = KΩ (q).

Proposition 4.12. Let x : U → R3 be a wavefront, Ω a tmb of x, p ∈ Σ(x)
with rank(Dx(p)) = 1, KΩ (p) = 0 and HΩ (p) < 0 (resp. HΩ (p) > 0) then the
following statements are equivalents:
(1) x is parallelly smoothable at p.
(2) lim H = ±∞.
(u,v)→p
(3) lim k1 = ±∞(resp. k2 ).
(u,v)→p
(4) There exist an open neighborhood V of p, in which λΩ does not change sign.
Proof. The proof of (1 ⇒ 4), (2 ⇔ 3), (2 ⇔ 4) is equal to corresponding ones
in proposition 4.11 because this does not use the hypothesis of KΩ (p) 6= 0. To
prove (3 ⇒ 1), let us define A := {q ∈ U : KΩ (q) 6= 0}. If p ∈ / Ā, there exist
a neighborhood W of p in which KΩ ≡ 0, thus λΩ (q) − 2tHΩ (q) + t2 KΩ (q) =
λΩ (q) − 2tHΩ (q) on W and since HΩ (p) 6= 0, using that λΩ does not change
λΩ
sign on a neighborhood of p, shrinking W we have that H Ω
does not change sign,
λΩ λΩ
then if HΩ ≥ 0 (resp. HΩ ≤ 0) we have that λΩ (q) − 2tHΩ (q) 6= 0 for every
(q, t) ∈ W × (−, 0) (resp. W × (0, )) with an arbitrary  > 0 and it follows the
result. If p ∈ Ā, by hypothesis there exist a open neighborhood W of p such that
 EXTENDIBILITY AND BOUNDEDNESS OF INVARIANTS ON WAVEFRONTS 13

k1 > 0 and H >  0 (resp. or < 0) on W − Σ(x) and as k1Ω (p) 6= 0, we have
k1Ω 
that lim | KΩ | A = ∞. Let  > 0 be given, there exist a open ball B such that
(u,v)→p
| kK1Ω
Ω
| >  on B ∩ A and k1 > 0 (resp. < 0), H > 0 (resp. < 0), H λΩ
Ω
= H1 > 0 (resp.
< 0) on B − Σ(x). Since kK2Ω Ω
= k11 on (B ∩ A) − Σ(x), we have that kK2Ω Ω
≥ 0 (resp.
≤ 0) on B ∩ A. Now, if (q, t) ∈ B × (−, 0) (resp. B × (0, )) and KΩ (q) = 0 then
λΩ
λΩ (q) − 2tHΩ (q) + t2 KΩ (q) = λΩ (q) − 2tHΩ (q) 6= 0 because H Ω
≥ 0 (resp. ≤ 0)
on B. The another option is that KΩ (q) 6= 0, then λΩ (q) − 2tHΩ (q) + t2 KΩ (q) 6= 0
because this is 0 if and only if t = kK1Ω
Ω
(q) or t = kK2Ω
Ω
(q) which is impossible since
k1Ω k2Ω
that | KΩ (q)| >  and KΩ (q) ≥ 0 (resp. ≤ 0). It follows the result. 

Corollary 4.13. Let x : U → R3 be a wavefront, Ω a tmb of x, p ∈ Σ(x) with

rank(Dx(p)) = 1, we have that x is parallelly smoothable at p if and only if λΩ
does not change sign on a neighborhood of p.
Corollary 4.14. Let x : U → R3 be a wavefront, Ω a tmb of x, p ∈ Σ(x)
with rank(Dx(p)) = 1, if x is parallelly smoothable at p, then p is a degenerated
singularity.
Proof. If we suppose that p = (p1 , p2 ) is a non-degenerated singularity, then
λΩu (p) 6= 0 or λΩv (p) 6= 0 and therefore λΩ (u, p2 ) or λΩ (p1 , v) is strictly mono-
tone as function of one variable on every sufficient small neigborhood of p, which
is contradictory, because λΩ (p) = 0 and this does not change sign by corollary
4.13. 
Corollary 4.15. Let x : U → R3 be a wavefront, Ω a tmb of x, p ∈ Σ(x) with
rank(Dx(p)) = 1, if p is an isolated singularity then x is parallelly smoothable at
p.
Proof. if p is an isolated singularity, then there exist an open connected neighbor-
hood V of p, such that λΩ 6= 0 on V − {p} and since that V − {p} is arc-connected,
λΩ does not change sign on V . By corollary 4.13, it follows the result. 
Example 4.16. The wavefront (u, 2v 3 + u2 v, 3v 4 + u2 v 2 ) (cuspidal lips) has a
isolated singularity at (0, 0), then it is parallelly smoothable at (0, 0). On the other
hand, (u, 2v 3 − u2 v, 3v 4 − u2 v 2 ) (cuspidal beaks) is not parallelly smoothable at
(0, 0), because taking as a tangent moving base:
 
1 0  
1 0
Ω =  −2uv 1  , Λ =
0 6v 2 − u2
−2uv 2 2v
we get λΩ = 6v 2 −u2 , which changes of sing on every neighborhood of (0, 0). By the
same argument, x(u, v) = (u, v 2 , v 3 ) (cuspidal edge) and x(u, v) = (3u4 +u2 v, 4u3 +
2uv, v) (swallowtail) are not parallelly smoothable at (0, 0), because can be chosen
tmb’s Ω in which λΩ is v and 12u2 + 2v respectively.
14 T. A. MEDINA-TEJEDA

Figure 2. A wavefront (cuspidal beaks) non-parallelly smoothable

at (0, 0) by rank 2

Theorem 4.17. Let x : U → R3 be a wavefront, Ω a tmb of x and p ∈ Σ(x) with

rank(Dx(p)) = 1. One of the principal curvatures κ− , κ+ has a C ∞ -extension to
an open neighborhood of p if and only if x is parallelly smoothable at p.
Proof. if x is parallelly smoothable at p, then by corollary 4.13 λΩ ≥ 0 (or λΩ ≤ 0,
this case is analogous) on an open neighborhood V of p, thus k1 = κ− , k2 = κ+ on
V − Σ(x) and by proposition 4.1 one of these function has a C ∞ -extension to an
open neighborhood of p. Conversely, without loss of generality let us suppose that
κ− has a C ∞ -extension to an open neighborhood W of p, then λΩ does not change
sign on some neighborhood of p, otherwise there are sequences an −→ p, bn −→ p
such that λΩ (an ) > 0 and λΩ (bn ) < 0 for every n ∈ N. Thus, lim |k1 (an )| =
n→∞
lim |κ− (an )| = |κ− (p)| = lim |κ− (bn )| = lim |k2 (bn )| which is contradictory,
n→∞ n→∞ n→∞
because by proposition 4.1 one of the limits lim |k1 (an )|, lim |k2 (bn )| is ∞. 
n→∞ n→∞

Remark 4.18. Observe that, if κ− (or κ+ ) have a C ∞ -extension locally at p, then

the another one diverge to ±∞ at p by items (3)s of propositions 4.11 and 4.12.
In fact, reasoning similarly as in theorem 4.17, we have that one of the principal
curvatures κ− , κ+ diverge to ±∞ at p if and only if x is parallelly smoothable at
p. Also can be proved similarly that κ− (resp. κ+ ) is bounded locally at p if and
only if κ− (resp. κ+ ) have a C ∞ -extension locally at p.

5. Singularities of rank 0
Proposition 5.1. Let x : U → R3 be a proper wavefront, Ω a tmb of x, p ∈ Σ(x)
with rank(Dx(p)) = 0, then:
(1) (k1Ω , k2Ω )(p) = (0, 0).
(2) lim |K| = ∞.
(u,v)→p
 EXTENDIBILITY AND BOUNDEDNESS OF INVARIANTS ON WAVEFRONTS 15

1
(3) and k12 have continuous extensions on a neighborhood V of p, which are
k1
of class C ∞ except possibly at umbilical points and singularities of rank 0
of x.
(4) lim |k1 | = ∞ and lim |k2 | = ∞.
(u,v)→p (u,v)→p
(5) 1 1
k1 + k2 has a C ∞ -extension on a neighborhood V of p.

Proof.
q
(1) By proposition 2.6, HΩ (p) = 0, then k1Ω (p) = HΩ (p)− HΩ2 (p) − λΩ (p)KΩ (p) =
0. Similarly k2Ω (p) = 0.
(2) By proposition 2.6, KΩ (p) 6= 0 and therefore lim |K| = lim | K λΩ | = ∞
Ω
(u,v)→p (u,v)→p
(3) There exist a neighborhood V of p such that KΩ (p) 6= 0 and hence k1 6=
0, k2 6= 0 on V − Σ(x), then k11 = kk1 k2 2 = kK2Ω
Ω
and similarly k12 = kK1Ω
Ω
which
are well defined on V . Notice that k1Ω and k2Ω may not be differentiable
at umbilical points and singularities of rank 0 of x.
(4) By item (2), there exist a neighborhood V of p such that k1 6= 0, k2 6= 0 on
V − Σ(x), therefore k1Ω 6= 0, k2Ω 6= 0 as well. Then, k1 = kK2Ω Ω
and k2 = kK1Ω
Ω

on V − Σ(x) and using that KΩ (p) 6= 0 and item (1) we get (3).
(5) Since KΩ (p) 6= 0, there exist an open neighborhood V of p such that KΩ
2HΩ 2HΩ
does not vanish on V , then k11 + k12 = 2H K = KΩ on V − Σ(x) and as KΩ
∞
is well defined on V , this is a C -extension.


Proposition 5.2. Let x : U → R3 be a proper wavefront, Ω a tmb of x, p ∈ Σ(x)

with rank(Dx(p)) = 0, then the following statements are equivalent:
(1) x is parallelly smoothable at p.
(2) λΩ KΩ ≥ 0 and HΩ does not change sign on a neighborhood V of p.
(3) lim K = ∞ and lim H = ±∞.
(u,v)→p (u,v)→p
(4) lim k1 = lim k2 = ∞ or lim k1 = lim k2 = −∞.
(u,v)→p (u,v)→p (u,v)→p (u,v)→p

Proof.
• (1 ⇒ 2) If x is parallelly smoothable at p, then there exist  > 0 and an open
connected neighborhood V of p such that λΩ (q) − 2tHΩ (q) + t2 KΩ (q) 6= 0
for every (q, t) ∈ V × (0, ) (or V × (−, 0), this case is analogous). As
KΩ 6= 0 and does not change sign on V (shrinking V if it is necessary),
k1Ω = k2Ω = 0 and since λΩ (q) − 2tHΩ (q) + t2 KΩ (q) 6= 0 if and only
if t = kK1Ω
Ω
(q) or t = kK2ΩΩ
(q), we have that kK1Ω
Ω
≤ 0 and kK2Ω Ω
≤ 0 on V .
Then, k1Ω k2Ω ≥ 0 on V , but −k1Ω and −k2Ω are the eigenvalues of αTΩ =
(µΩ adj(Λ))T , then λΩ KΩ = k1Ω k2Ω ≥ 0 on V . Observe that k1Ω and k2Ω
do not change sing on V , then HΩ neither.
16 T. A. MEDINA-TEJEDA

• (2 ⇒ 3) Since λΩ KΩ = λ2Ω K on U −Σ(x) and using that lim |K| = ∞ we

(u,v)→p
get that lim K = ∞. On the other hand, H2 ≥ K, then lim |H| = ∞.
(u,v)→p (u,v)→p
As HΩ and λΩ do not change sign on a neighborhood of p, H = H neither
λΩ
Ω

and we get the result.

• (3 ⇒ 4) As K is positive near to p then λΩ KΩ = λ2Ω K ≥ 0 and KΩ 6= 0
on a neighborhood Z of p. Shrinking Z, λΩ does not change sing and H
neither on Z − Σ(x).√ Without loss of generality,
√ let us suppose λΩ ≥ 0 on
2 2
Z, then k1 = H − H − K and k2 = H + H − K and since that H does
not change sing on Z − Σ(x), one of the function k1 , k2 neither. By this
last and since that K > 0, we have that k1 > 0, k2 > 0 or k1 < 0, k2 < 0 on
Z − Σ(x), then using item (3) of proposition 5.1 we get the result.
• (4 ⇒ 1) There exist a neighborhood V of p such that k1 > 0, k2 > 0 (or
k1 < 0, k2 < 0, this case is analogous) on V − Σ(x) and KΩ 6= 0 on V ,
then kK1Ω
Ω
= k12 > 0 and kK2ΩΩ
= k11 > 0 on V − Σ(x), thus by density of
V − Σ(x) kK1Ω , k2Ω ≥ 0 on V . Choose  > 0 arbitrary and we have that
Ω KΩ
λΩ (q) − 2tHΩ (q) + t2 KΩ (q) 6= 0 for every (q, t) ∈ V × (−, 0). It follows
(1).

Corollary 5.3. Let x : U → R3 be a wavefront, Ω a tmb of x, p ∈ Σ(x) with
rank(Dx(p)) = 0. If there exist a neighborhood V of p in which the only singularity
of rank 0 is p, then x is parallelly smoothable at p if and only if λΩ KΩ ≥ 0 on a
neighborhood W of p.
Proof. If λΩ KΩ ≥ 0 on W , shrinking if it is necessary we can suppose that K 6= 0
on W − Σ(x) and since λΩ KΩ = λ2Ω K, then H 2 > K > 0 on W − Σ(x). As HΩ 6= 0
on singularities of rank 1, then HΩ has a isolated zero on W ∩ V and therefore HΩ
does not change sing. Applying the last proposition we get the result. 

Observe that, if we have a wavefront x : U → R3 , Ω a tmb of x, p ∈ Σ(x)

with rank(Dx(p)) = 0 and this is parallelly smoothable at p, since that KΩ (p) 6=
0 and λΩ KΩ ≥ 0 on a neighborhood of p, then λΩ does not change sign on a
neighborhood of p. However, this condition is not sufficient as it happened in the
case of singularities of rank 1. The next example shows this.
k k
Example 5.4. The wavefront x := (uk , ±v k , k+1 uk+1 ± k+1 v k+1 ), with k ∈ N,
k ≥ 2 has as tmb:
 
1 0  k−1 
ku 0
Ω =  0 1 , Λ = ,
0 ±kv k−1
u v
then λΩ = ±k 2 uk−1 v k−1 , KΩ = (1+u21+v2 )2 . By corollary 5.3, x is parallelly smooth-
able at (0, 0) when we choose k odd and the sign +. If k is even or the sign is −,
 EXTENDIBILITY AND BOUNDEDNESS OF INVARIANTS ON WAVEFRONTS 17

this is not parallelly smoothable at (0, 0), even when k is odd with sing − in the
expression, in which λΩ does not change sign.
Corollary 5.5. Let x : U → R3 be a proper wavefront, Ω a tmb of x, p ∈ Σ(x)
with rank(Dx(p)) = 0 and Σ(x)0 = {q ∈ Σ(x) : rank(Dx(q)) = 0}. If x is
parallelly smoothable at p then:
(1) There exist a open neighborhood V of p in which one of the functions k1 ,
k2 has a C ∞ extension to V − Σ(x)0 . More precisely, k1 (resp. k2 ) has a
C ∞ extension to V − Σ(x)0 if only if HΩ ≤ 0 (resp. HΩ ≥ 0) on V .
(2) There exist a open neighborhood V of p in which one of the functions k1 ,
k2 diverge to ±∞ (just one sign globally) near the singularities to p. More
precisely, lim k1 = ±∞ (resp. k2 ) if and only if HΩ ≤ 0 (resp.
(u,v)→Σ(x)∩V
HΩ ≥ 0) on V .
Proof.
(1) Since that HΩ does not change sign on a neighborhood V of p by item (2)
of proposition 5.2 and applying proposition 4.1 we get the result.
(2) By items (2) and (4) of proposition 5.2 HΩ and k1 do not change sign on a
neighborhood V of p and applying proposition 4.1 we get the result.

Proposition 5.6. Let x : U → R3 be a proper wavefront, Ω a tmb of x and
p ∈ Σ(x) with rank(Dx(p)) = 0. If H is bounded on a neighborhood of p then we
have:
(1) There exist a neighborhood V of p such that K < 0 on V − Σ(x) and
lim K = −∞.
(u,v)→p
(2) λΩ does not change sign on a neighborhood V of p.
(3) One of the function k1 , k2 diverge to ∞ and the another one to −∞ at p.
More precisely, if λΩ ≥ 0 (resp. λΩ ≤ 0) on a neightborhhod V of p then
lim k1 = −∞ (resp. ∞) and lim k2 = ∞ (resp. −∞).
(u,v)→p (u,v)→p
(4) There is no singularities of rank 1 on a neighborhood of p.
k1
(5) There exist a neighborhood V of p such that lim k2 = −1
(u,v)→Σ(x)∩V

Proof.
(1) We know that K 6= 0 near to p. If there exist a sequence an −→ p with
K(an ) > 0 for every n ∈ N, as H 2 ≥ K and lim K(an ) = ∞ we have
(u,v)→p
that lim |H(an )| = ∞ witch is contradictory, then we have (1).
(u,v)→p
(2) By (1) λΩ KΩ = λ2Ω K < 0 near to p and since that KΩ (p) 6= 0, then λΩ
does not change sign on a neighborhood
√ V of p. √
(3) If λΩ ≥ 0 near to p, k1 = H − H 2 − K and k2 = H + H 2 − K on a
neighborhood of p and since that K < 0 near to p, then k2 > 0 > k1 on a
neighborhood of p. By (3) of proposition 5.1 we have the result.
18 T. A. MEDINA-TEJEDA

(4) If H is bounded on a neighborhood V of p and suppose that exist a singu-

larity q of rank 1 in V , by (4) of proposition 4.1 lim |H| = ∞ witch is
(u,v)→p
contradictory.
(5) Let V a bounded neighborhood of p with just singularities of rank 0 with
H bounded. There exist C > 0 such that |k1 + k2 | < C, then |1 + kk12 | < |kC2 |
k1
and by (3) of proposition 5.1 lim k2 = −1 for every q ∈ Σ(x) ∩ V . Since
(u,v)→q
that Σ(x) ∩ V is compact we have the result.


Proposition 5.7. Let x : U → R3 be a proper wavefront, Ω a tmb of x, p ∈ Σ(x)

with rank(Dx(p)) = 0 and let us choose W a compact neighborhood of p in which
KΩ 6= 0. The following statements are equivalents:
(1) The Mean curvature H is bounded on W − Σ(x).
(2) There exist C > 0 such that |HΩ | ≤ C|λΩ | on W .
(3) There exist C > 0 such that |eG + gE − 2f F | ≤ C|λ2Ω | on W .
(4) There exist C > 0 such that | k11 + k12 | ≤ C|λΩ | on W .
1 1
(5) k1Ω + k2Ω is bounded on W .

Proof.
• (1 ⇔ 2) Using that HΩ = λΩ H on W − Σ(x) which is dense in W , we get
the equivalence.
• (1 ⇔ 3) Using that H(EG − F 2 ) = eG + gE − 2f F on W − Σ(x) and
EG − F 2 ∈ T2Ω (W ) (see proposition 2.5), by compactness of W we get the
equivalence.
• (2 ⇔ 4) by proposition 5.1 k11 + k12 has a C ∞ -extension to W and this is
HΩ
equal to K Ω
. From this equality follows the equivalence.
• (4 ⇔ 5) Since that k1Ω = λΩ k1 and k2Ω = λΩ k2 on W − Σ(x) which is
dense in W , we get the equivalence.


Proposition 5.8. Let x : U → R3 be a proper wavefront, Ω a tmb of x, p ∈ Σ(x)

with rank(Dx(p)) = 0 and let us choose V an open neighborhood of p in which
KΩ 6= 0. The following statements are equivalents:
(1) The Mean curvature H has a C ∞ -extension to the neighborhood V of p.
(2) HΩ ∈ TΩ (V )
(3) eG + gE − 2f F ∈ T2Ω (V ).
(4) k11 + k12 ∈ TΩ (V )
1
(5) k1Ω 1
+ k2Ω has a C ∞ -extension to the neighborhood V of p.

Proof. The proof of proposition 5.7 can be reproduced here to prove the corre-
sponding equivalences. 
 EXTENDIBILITY AND BOUNDEDNESS OF INVARIANTS ON WAVEFRONTS 19

2
Example 5.9. The wavefront x := ( 12 log v 2 + 1 − 12 log u2 + 1 , v2uv , uv −



+1 v 2 +1
u + tan−1 (u)) has as tmb:
 
−(1+v 2 )
  uv
1 0 −u v
!
1+u2 1+v 2  (1+u2 +v2 ) 32 (1+u2 +v2 ) 32 
Ω =  0 1 , Λ = v u(1−v 2 ) , µ =  uv −(1+u2 )  ,
u v 1+v 2 (1+v 2 )2 2 2
3 3
(1+u +v ) 2 (1+u2 +v 2 ) 2
−(u2 +v 2 )
then λΩ = (1+u2 )(1+v 2 )2
, HΩ = − 12 (λ22 µ11 − λ21 µ12 + λ11 µ22 − λ12 µ21 ) = 0 and
therefore the Mean curvature is extendable, with H = 0 on R2 .

Acknowledgement
I am grateful to professors Kentaro Saji and Keisuke Teramoto for the fruitful
discussions about the extendibility of the Gaussian Curvature, during their visit
to the ICMC-USP on September 2019 in which they gave the idea of express this
extendibility property in terms of the ideal generated by λΩ and the second funda-
mental form.

References
[1] V. I. Arnol0 d, S. M. Gusein-Zade and A. N. Varchenko, Singularities of differentiable maps,
Vol.1, Monographs in Mathematics 82, Birkhäuser, Boston, 1985.
[2] V. I. Arnol0 d, Singularities of caustics and wave fronts, volume 62 of Mathematics and its
Applications (Soviet Series). Kluwer Academic Publishers Group, Dordrecht, 1990.
[3] C.G. Gibson, Singular Points of Smooth Mappings. London, Pitman, 1979.
[4] G. Ishikawa, Singularities of frontals, Adv. Stud. Pure Math., 78, 55–106, Math. Soc. Japan,
Tokyo, 2018.
[5] L. F. Martins, K. Saji, M. Umehara and K. Yamada, Behavior of Gaussian curvature and
mean curvature near non-degenerate singular points on wave fronts, Geometry and Topology
of Manifold, Springer Proc. Math. & Statistics, 2016, 247–282.
[6] T.A. Medina-Tejeda, The fundamental theorem for singular surfaces with limiting tangent
planes, arXiv:1908.04821v2.
[7] S. Murata and M. Umehara, Flat surfaces with singularities in Euclidean 3-space, J. Differ-
ential Geom. 82 (2009), 279-316.
[8] K. Saji, M. Umehara, and K. Yamada, The geometry of fronts, Ann. of Math. 169 (2009),
491–529.
[9] K. Teramoto, Parallel and dual surfaces of cuspidal edges, Differential Geom. Appl. 44 (2016),
52–62.
[10] K. Teramoto, Focal surfaces of wave fronts in the Euclidean 3-space, Glasg. Math. J. 61,
(2019), no. 2, 425–440.
[11] K. Teramoto, Principal curvatures and parallel surfaces of wave fronts, Adv. Geom. 19 (2019),
no. 4, 541–554.

Instituto de Ciências Matemáticas e de Computação - Universidade de São Paulo,

Av. Trabalhador são-carlense, 400 - Centro, CEP: 13566-590 - São Carlos - SP,
Brazil
Email address: tamedinat@usp.br