Quasi-umbilic points

Pablo Castañeda 1,∗ and Stefan Berres 2

1 Academic Department of Mathematics, ITAM, Ciudad de México, Mexico.
2 Department of Information Systems, Universidad del Bı́o-Bı́o, Concepción, Chile.
Correspondence*:
Pablo Castañeda and Stefan Berres
pablo.castaneda@itam.mx

The most well-studied cases of loss of hyperbolicity are related to umbilic points, which are isolated points
where strict hyperbolicity fails, [1, 2, 3]. A description of new types of isolated points exhibiting loss of
hyperbolicity is provided in [4]. If there is a connected set of points with loss of hyperbolicity, then more
refined characterizations are needed. For instance, a generalization of the umbilic point is the coincidence
point.

D EFINITION 1. A point Φ is called a coincidence point of the PDE

∂t ϕi + ∂x fi (Φ) = 0, fi (Φ) = ϕi vi (Φ), i = 1, 2,

with flux function f (Φ) if the eigenvalues of the Jacobian matrix J (Φ) at that point coincide. We say that a
coincidence point Φ∗ is an umbilic point if it satisfies the following conditions:

(H1) The Jacobian matrix J (Φ∗ ) is diagonalizable.

(H2) There is a neighborhood V of Φ∗ , such that J (Φ) has distinct eigenvalues for all Φ ∈ V \{Φ∗ }.

Thus, umbilic points are expected to be isolated coincidence points. However, the previous definition
of an umbilic point may fail to satisfy either of the two conditions. In such cases, we classify an isolated
coincidence point that does not satisfy condition (H1) as a quasi-umbilic point, as first introduced in [5].
A detailed description of the classification and properties of quasi-umbilic points is provided in [4].

L EMMA 1. There is a unique coincidence point on Φ ∈ ∂ 1 . Upon the choice of parameters, there may
exist up to two coincidence points on Φ ∈ ∂ 2 .

1
P. Castañeda and S. Berres

Proof. By definition, a state Φ is a coincidence point if and only if λa (Φ) = λb (Φ) holds. Namely, from
the eigenvalue expression
(
λa (Φ) := (1 − (1 + ni )ϕ)ui (Φ), for Φ ∈ ∂i , i = 1, 2,
λb (Φ) := u3−i (Φ) − ϕui (Φ), for Φ ∈ ∂i , i = 1, 2,

one has a coincident point on axis Φ ∈ ∂ 1 if there exists some ϕ1 ∈ [0, 1] such that (1 − n1 ϕ1 )u1 (Φ) =
u2 (Φ) holds. Therefore, we must look for such a value ϕ1 = ϕu which is a zero root of the equation

(1 − n1 ϕ)(1 − ϕ)n1 −n2 − v∞2 /v∞1 = 0, (1)

i.e., a zero root of R1 (ϕ) − v∞2 /v∞1 = 0, where

R1 (ϕ) := (1 − n1 ϕ)(1 − ϕ)n1 −n2 .

Notice that by conditions

(S1) v∞1 > v∞2 > 0,

(S2) n1 > n2 > 1,

the ratio v∞2 /v∞1 is smaller than one and that the power n1 −n2 is positive. Since R1 (0) = 1, R1 (1/n1 ) =
0 and R1′ (ϕ) = −n1 (1 − ϕ)n1 −n2 − (n1 − n2 )(1 − n1 ϕ)(1 − ϕ)n1 −n2 −1 is negative for all ϕ ∈ [0, 1/n1 ],
we conclude that R1 (ϕ) = v∞2 /v∞1 occurs at a single point, say ϕu defined as unique zero root of equation
(1). Let us denote such a state as Q1 := (ϕu , 0)T .

Similarly, for the axis Φ ∈ ∂ 2 , we seek a value ϕ2 ∈ [0, 1] (with ϕ1 = 0) that is a zero root of the
equation R2 (ϕ) − v∞1 /v∞2 = 0, where

R2 (ϕ) := (1 − n2 ϕ)(1 − ϕ)n2 −n1 and ϕ ̸= 1.

It should be noted that R2 (0) = 1 and R2 (1/n2 ) = 0. Also, since n1 > n2 > 1, the limit limϕ→1− R2 (ϕ) =
−∞ holds. Moreover, R2 (ϕ) is positive for ϕ ∈ [0, 1/n2 ) and negative for ϕ ∈ (1/n2 , 1). As the ratio
v∞1 /v∞2 is greater than one, no coincidence point exists if R2 (ϕ) is always less than this ratio. For
ϕ ∈ [0, 1) we have

R2′ (ϕ) = −n2 (1 − ϕ)n2 −n1 − (n1 − n2 )(1 − n2 ϕ)(1 − ϕ)n2 −n1 −1
= [−n2 (1 − ϕ) + (n2 − n1 )(1 − n2 ϕ)](1 − ϕ)n2 −n1 −1 .

Thus, a single extremum of R(ϕ) occurs at

n1 − 2n2
ϕm := . (2)
(n1 − n2 − 1)n2

A single coincidence point exists on the axis if R2 (ϕm ) = v∞1 /v∞2 . The existence of two coincidence
points occurs if R2 (ϕm ) is larger than v∞1 /v∞2 ; we denote these points as Q2 := (0, ϕu2 )T , Q3 :=
(0, ϕu3 )T where R2 (ϕu2 ) = R2 (ϕu3 ) = v∞1 /v∞2 . □

Frontiers 2
P. Castañeda and S. Berres

The necessary and sufficient conditions for coincidence points on ∂ 2 are difficult to evaluate. Let us
define the ratio W := v∞1 /v∞2 . On the one hand, we know that W exceeds one due to condition (S1).
On the other hand, the powers n1 and n2 are also free parameters. Some calculations are possible if the
difference between n1 and n2 is a natural number. For example, for n1 − n2 = 1, there are√two coincidence
points if n2 > W ; for n1 − n2 = 2, there are two coincidence points if n2 > 2(W + W 2 − W ); for
n1 − n2 = 3, there are two coincidence points if 4n32 − 27W (n2 − 1)2 > 0. The inequality R2 (ϕm ) < 1
holds for powers satisfying n2 + 1 < n1 < 2n2 , and for ϕm defined in (2). In this case there are not
coincidences on the axis ∂ 2 .

For Φ = (ϕ, 0)T the Jacobian matrix is an upper triangular matrix, and for Φ = (0, ϕ)T a lower triangular
matrix. In view of Definition (1), condition (H2) is satisfied, but not necessarily (H1). The coincidence
point Q1 = (ϕu , 0)T is a quasi-umbilic point since the Jacobian matrix has the form
 
cσ ∗
,
0 cσ

which is a Jordan block with ∗ = ̸ 0 and therefore not diagonalizable. This means that (H1) is violated.
Therefore, based on this matrix structure and Lemma 1, we have the following result.

C OROLLARY 1. Any isolated coincidence point on the edges ∂ 1 or ∂ 2 turns out to be quasi-umbilic.

To illustrate the behavior of the coincidence points, we extend the flux functions

fi (Φ) = ϕi vi (Φ), vi (Φ) = ui (Φ) − ΦT u, i = 1, 2,

beyond the phase space by considering relative velocities ui (Φ) = v∞i |1 − ϕ|ni −1 for any ϕ. The resulting
visualization in Fig. 1 shows that the identified quasi-umbilic point coincides with an elliptic/hyperbolic
boundary, implying that λ1 (Q1 ) = λ2 (Q1 ). This concept was introduced in [5], and here the flux functions
were also extended by allowing the dominated quadratic terms to act outside the physical domain. Fig. 1 (c)
depicts the case where quasi-umbilic points also belong to an elliptic/hyperbolic boundary. It would be
interesting to investigate the relationship between these points and the DRS points (in honor of Dumortier,
Roussarie, and Sotomayor, [6], coined in [7]) in future work. To the best of our knowledge, the first
identification of DRS points on the elliptic/hyperbolic boundary is due to Sunčica Čanić [8]. In all these
cases DRS points are inside the phase space which allows a better analysis without an artificial expansion
of the domain.

We considered two general examples for the bidisperse model with parameter specifications (S1)–(S2).
In addition to Example 1 with parameter setting

v∞1 = 1, v∞2 = 1/2, and n1 = 4, n2 = 3,

we examine Example 2, with the parameters specified as

v1∞ = 1, v2∞ = 1/2, n1 = 4.6, n2 = 1.5.

We also considered the ELD model (Equal-Lighter-Density fluids) case in Rodrı́guez-Bermúdez and
Marchesin (2013) with parameters chosen as ρ1 = 2.0 > 1.0 = ρ2 = ρ3 .

Frontiers 3
P. Castañeda and S. Berres

Figure 1. Extended domains. Concentration and saturation triangles are indicated by bold solid lines,
and the shaded regions correspond to states with complex eigenvalues of the flux Jacobian. In panel (a),
Example 1 exhibits two quasi-umbilic points located at the boundary of elliptic regions, and our analysis
detects one of them at point Q1 . Panel (b) shows Example 2, where three quasi-umbilic points are identified
by the previous analysis, and another one is found at the boundary of the elliptic regions. In panel (c),
the Equal-Lighter-Density fluids (ELD) model is displayed, and two quasi-umbilic points are observed to
belong to the boundary of elliptic regions.

REFERENCES
[1] Isaacson E, Marchesin D, Plohr B, Temple B. The Riemann problem near a hyperbolic singularity: the
classification of solutions of quadratic Riemann problems. SIAM J. Appl. Math. 48 (1988) 1009–1032.
[2] Matos V, Castañeda P, Marchesin D. Classification of the umbilic point in immiscible three-phase flow
in porous media. Ancona F, Bressan A, Marcati P, Marson A, editors, Hyperbolic Problems: Theory,
Numerics, Applications, Proc. of Hyp2012 (Springfield, MO: AIMS Series on Appl. Math.) (2014),
791–799.
[3] Schaeffer DG, Shearer M. The classification of 2 x 2 systems of non-strictly hyperbolic conservation
laws with application to oil recovery. Comm. Pure and Appl. Math. 40 (1987) 141–178.
[4] Rodrı́guez-Bermúdez P, Marchesin D. Loss of strict hyperbolicity for vertical three-phase flow in
porous media. Ancona F, Bressan A, Marcati P, Marson A, editors, Hyperbolic Problems: Theory,
Numerics, Applications, Proc. of Hyp2012 (Springfield, MO: AIMS Series on Appl. Math.) (2014),
881–888.
[5] Rodrı́guez-Bermúdez P, Marchesin D. Riemann solutions for vertical flow of three phases in porous
media: simple cases. J. Hyperbolic Differ. Equ. 10 (2013) 335–370.
[6] Dumortier F, Roussarie R, Sotomayor J. Generic 3-Parameter Families of Vector Fields: Unfoldings of
Saddle, Focus, and Elliptic Singularities with Nilpotent Linear Parts (Berlin-Heidelberg-New York:
Springer-Verlag) (1991).
[7] Azevedo AV, Marchesin D, Plohr B, Zumbrun K. Bifurcation of nonclassical viscous shock profiles
from the constant state. Commun. Math. Phys. 202 (1999) 267–290.
[8] Čanić S. On the influence of viscosity on Riemann solutions. J. Dyn. Differ. Equ. 10 (1998) 109–149.

Frontiers 4