Vector Elds

Vector fields
Chantal Oberson Ausoni
6.8.2014
ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 1
Vector fields: Outline

1. VECTOR FIELDS: MOTIVATION, DEFINITION
2. INTEGRAL LINES, SINGULARITIES, VECTOR FIELD TOPOLOGY
• Integral lines, phase portrait, separatrices, flow
• Classification of the integral lines
• Hyperbolic singularities and linearization
• 2D classification of hyperbolic stationary points
• 3D classification of hyperbolic stationary points
• Index of a singularity
• Poincaré-Hopf theorem
• Higher order singularities
• Orbits
• Vector field topology
3. DIVERGENCE, CURL, HELMOLTZ-HODGE DECOMPOSITION
4. VISUALIZATION OF VECTOR FIELDS: SMALL CATALOGUE
• Sampled vector field
• Streamlines, streamsurfaces, streamribbons
• The problem of seeding
• Line integral convolution
Vector fields: motivation, definition
Vector fields: why, where?
A vector field arises in a situation where,
for some reason, there is a direction and
a magnitude assigned to each point of the
space or of a surface, typically examples are
fluid dynamics, wheather prediction, ...
A classical example would be to represent
the velocity of the wind with a vector.

Examples of vector fields

Vector fields: definitions
DEFINITION 1. A vector field is a smooth function f : U ⇢ Rn ! Rn. In this frame, the
vector field is steady: it does not depend on the time.
EXAMPLE 1. Any problem implying the gradient of a scalar function0defines
1
a vector field; if
@g
B @x1 C
B @g C
B C
g : U ⇢ Rn ! R is a smooth function, then the gradient rg = B @x2 C is a vector field.
B···C
@ A
@g
@xn
DEFINITION 2. Let S be a surface. A vector field on S is an assignment to each point p 2 S
of a tangent vector X(p) at p, such that these vectors vary in a smooth manner.
On a surface patch given by a parametrization f , we can express the vector field by the
expression
@f @f
X = X1(u, v)
(u, v) + X2(u, v) (u, v)
@x @x
where X1 and X2 are smooth functions of the parameters (u, v).
Integral lines, singularities, vector field topology

Correspondance between vector fields and ODEs
DEFINITION 3. An autonomous, ordinary differential equation, defined on an open set
U ⇢ Rn, is an equation of the form: dx/dt = f (x), where x : I ! U denotes an
unknown curve, f : U ⇢ Rn ! Rn is C 1.
THEOREM 1. [Existence and uniqueness theorem for solutions of autonomous ODEs]
Given any point x0 2 U , there exists an interval I containing 0, and a solution x : I ! U ,
satisfying x0(t) = f (x(t)) and the condition x(0) = x0 (called initial condition), and it
is unique on the interval I.
In a mathematical sense, vector fields and differential equations may be considered
to be the same thing: Given a vector field defined by a function f : U ! Rn,
one may write the differential equation given by dx/dt = f (x). And conversely,
given a differential equation dx/dt = f (x), one may extract the vector field f (x).
REMARKS. • Think of f (x) as representing the velocity of a steady wind at the point
x. Then the solution curve x(t) represents the hypothetical trajectory of a massless
particle released at time t = 0 at the point x0.
• Non autonomous differential equations dx/dt = f (x, t) correspond in the same way
to unsteady vector fields. Example: the real wind is varying with time.

Integral lines, phase portrait, separatrices
DEFINITION 4. • Maximal solutions of
these ODEs are called integral lines,
trajectories or streamlines, they are
tangent in every point to the vector field.
• Given an initial condition x(t0) = x0,
the streamline is given by
Z t
x(t) = x0 + f (x(u))du.
t0
• The set of integral curves is called phase
portrait.
• It gives a partition of the domain in re-
gions of similar behaviour, where partic-
ular integral lines (the separatrices) make
the separation between the regions.
REMARK 1. Let g : U ⇢ Rn ! R be a function, the level lines g( (t)) = c are orthogonal
to the gradient rg, hence to the streamlines defined by this vector field.

Flow
We define a mapping ', called flow, giving all trajectories at the same time:
• (x, t) is the trajectory from point x (in time 0, the position is x: (x, 0) = x).
• ( (x, s), t) = (x, s + t), for all s and t where it is defined (Consistency condition).
REMARKS. • The existence and uniqueness theorem for ODEs tells that this function exists
and is C 1.
• We can also consider the family of transformations: t : U ! U , obtained by keeping
t constant. The consistency is then stated as s t ⌘ s+t , for all s and t where it is
defined.

Classification of the integral lines
• A regular trajectory never returns to where it has been.
• If the curve x(t) has a period t0, i.e., x(t + t0) = x(t) for all t, it is a simple closed
curve and is called closed orbit.
• If the trajectory stays put in p, i.e., x(t) = p for all t, p is a stationary point, critical
point or singularity. It means that v vanishes in p.
Regular trajectory, two kind of stationary points, orbit [Frey]

Counter-example of the uniqueness
EXAMPLE 2. Let us consider the vector field v(x, y) = (1, 3 y 2/3). The derivative @y
@v (x, y) =
2 y 1/3 does not exist in zero, meaning that the existence and unicity theorem cannot be
applied in this context.
Actually, for any point (x, 0) of the x-axis, there are two trajectories meeting:
u1(t) = (x + t, 0) and u2(t) = (x + t, t3).

The Jacobian
✓ ◆
matrix (reminder)
@vi
The Jacobian matrix Dv(p) = @x is the matrix of all first-order partial derivatives of v:
j
• in 2D,
v(x1, x2) = (v1(x1, x2), v2(x1, x2)):
0 1
@v1 @v1
@ @x1 (p) @x2 (p)
A
Dv(p) = @v2 @v2
@x1 (p) @x2 (p)
• in 3D,
v(x1, x2, x3) = (v1(x1, x2, x3), v2(x1, x2, x3), v3(x1, x2, x3)):
0 1
@v1 @v1 @v1
@x (p) @x2 (p) @x3 (p)
B 1 C
B @v2 @v2 @v2 C
Dv(p) = B @x (p) @x2 (p) @x3 (p) C
@ 1 A
@v3 @v3 @v3
@x1 (p) @x2 (p) @x3 (p)

Eigenvalues/eigenvectors (reminder)
DEFINITION 5. A matrix A having a special vector u 6= 0 such that Au = u is said to have
an eigenvector u. The scaling factor is called eigenvalue of A for u.
The eigenvalue equation for a matrix A is Av v = 0.
For a 3 ⇥ 3-matrix, it is equivalent to (A I3)v = 0, where I3 is the 3 ⇥ 3 identity
matrix. It is a fundamental result of linear algebra that an equation M v = 0 has a non-zero
solution v if, and only if, the determinant det(M ) of the matrix M is zero. It follows that
the eigenvalues of A are precisely the real numbers that satisfy the equation
cA( ) = det(A I3) = 0.

This polynomial is called the characteristic polynomial of A. Since the characteristic poly-
nomial always has three roots in C, there are always three eigenvalues.

Eigenvalues/Eigenvectors (symmetric matrix)
When A is a 3 ⇥ 3 symmetric matrix, i.e., when AT = A, there are 3 real eigenvalues,
say 1, 2, 3. Furthermore, all corresponding ⇣eigenvectors u1,⌘u2, u3 are orthogonal. If
we normalize them and build the matrix U = u1 u2 u3 , U will be orthonormal:
U · U T = I3. We have the equation
0 1
1 0 0
B C
@0 2 0 A = U T AU
0 0 3
(corresponding to a change of basis, for an linear map having matrix A in the standard basis,
to a new basis constituted of the eigenvectors).

Eigenvalues/Eigenvectors
0 1
in 3D(Example)
1 2 0
B C
EXAMPLE 3. For the matrix @ 1 4 0A,
0 0 3
0 1
1 2 0
B C
• the characteristic polynomial is the determinant of the matrix @ 1 4 0 A:
0 0 3
cA( ) = (3 )[(1 )(4 ) + 2] = (3 )(3 )(2 )
• the eigenvalues are the numbers 2, 3, 3
⇣ ⌘T ⇣ ⌘T
• the vectors 1 1 0 and 0 0 1 are eigenvectors for the value 3
⇣ ⌘T
• the vector 2 1 0 is an eigenvector for the value 2.

Hyperbolic singularities and linearization
Near a singularity v(p0) = 0, one can consider a local linearization (Taylor):
v(p) = Dv(p0)(p p0) + 0((p p 0 )2 )

DEFINITION 6. We call the singularity p0 non-degenerate if the Jacobian matrix Dv(p0) is
invertible, i.e. if its eigenvalues are all non-zero.
A non-degenerate singularity is isolated, as a consequence of the local inversion theorem:
on a small neighborhood of p0, v is a bijection, hence v(p) 6= 0.
DEFINITION 7. A singularity p0 is called hyperbolic (linear) singularity if the real parts of the
eigenvalues of Dv(p0) are all non-zero.
Hyperbolic singularities are stable (i.e., a local perturbation doesn’t change the topology of
the trajectories near the point). Singularities are nearly always hyperbolic.
The eigenvalues of Dv(p0) allow a classification of the hyperbolic singularities.
This classification corresponds to a geometric description of the behaviour of inte-
gral lines in the vicinity of the singularity.

2D classification of hyperbolic stationary points
Let us consider an hyperbolic singularity and the Jacobian matrix in this point. The eigen-
values of the Jacobian are the roots of the characteristic polynomials and are either both real
or both complex (in this case, they are conjugates):
1. Both eigenvalues are real:
• both eigenvalues are positive: it is a source (repelling node)
• both eigenvalues are negative: it is a sink (attracting node)
• the eigenvalues have opposite signs: it is a saddle
2. Both eigenvalues are complex: 1 = a + ib and 2 = a ib
• a is positive: it is a spiral source (repelling focus)
• a is negative: it is a spiral sink (attracting focus)
The 2D hyperbolic singularities, and the non-hyperbolic center [Theisel]

Let us consider an hyperbolic singularity and the Jacobian matrix in this point. Either the
three roots are real, or there is a real root and a pair of complex conjugate ones.
1. All three eigenvalues are real:
• all eigenvalues are positive: it is a source (repelling node)
• all eigenvalues are negative: it is a sink (attracting node)
• one eigenvalue is positive, two are negative: it is a saddle (1 dimension out, 2 dimen-
sions in)
• one eigenvalue is negative, two are positive: it is a saddle (2 dimensions out, 1 di-
mension in)
2. One real eigenvalue 1 and two complex ones: 2 = a + ib and 3 = a ib
• 1 is positive:
if a is positive: spiral source
if a is negative: spiral saddle (2 dimensions in, 1 dimension out)
• 1 is negative:
if a is positive: spiral saddle (2 dimensions out, 1 dimension in)
if a is negative: spiral sink

The 3D hyperbolic singularities [Scheuermann]

Counter-example for the linearization
EXAMPLE 4. If the singularity is
8 not hyperbolic, there is no stability, and the linearization is
< dx = y
dt x (x2 + y 2)
not valid. Consider the ODE dy .
:
dt = x y (x2 + y 2)
• The point (0, 0) is a singularity. !

3x2 y 2 1 2xy
• The Jacobian matrix is .
1 2xy x2! 3y 2
0 1
• In (0, 0), the Jacobian matrix is , with eigenval-
1 0
ues ±i. As a consequence, the singularity is not hyperbolic.
8
< dx = y
dt
• The linearization of the system around (0, 0) is dy
:
dt = x
The singularity of the linearized system is a centre.
• On the figure, we see that the singularity is not a centre but a sink.

Index of an isolated singularity of a vector field
This (not completely formal) definition uses the angular variation of the vector field around
the singularity. It is useful to classify higher order singularities.
DEFINITION 8. Let v = (v1, v2) be a vector field on a surface S, x an isolated singularity of
v, ✓ be the angular coordinate of the vector field, the Poincaré index of v in x is the integral
Z Z
1 1 v1dv2 v2dv1
indx,v = d✓ =
2⇡ 2⇡ v12 + v22
where is a simple closed curve on S, containing x as only singularity of v and making
one turn counterclockwise around it.
The first field makes one counterclockwise turn along the curve, so the star has index +1.
The second one makes a clockwise turn along the curve, the saddle point has index -1.
[Scheuermann]
REMARKS. • ✓ is the arctangent of ( vv2 )
1
• We see that it measures the number of revolutions of v along the curve .
• It can be generalized in higher dimensions: for a singularity in a 3D-vector field, for
example, we would take a sphere, instead of a closed curve, around it.
Example of the index
EXAMPLE 5. Planar vector fields
We can also represent all (normalized) vectors

along the curve from the origin, hence defin-
v( (t))
ing a path '(t) = kv( (t))k
on S 1 and count
the number of turns made by this path around
the origin.
v(x)
On the left the vector field v in the plane, on the right the normalized kv(x)k in S 1.[Mann]

Poincaré-Hopf theorem
THEOREM 2. (Poincaré-Hopf). Let M be a compact oriented surface and v : M ! T M a
smooth vector field with isolated zeros. The sum of the indices at the zeros equals the Euler
characteristic of M :
X
indx,v = (M )
x zero of v
In particular, if the Euler characteristic is non-zero, then

the vector field must have singularities. This result,
when stated for the sphere ( = 2), is called hairy
ball theorem.
If you consider the atmospheric wind circulation to
happen in the same air layer around the earth and ne-
glect the non-tangential component of it, it means that
there is at any time a cyclone somewhere.
”You cannot comb a coconut. Can you comb a doughnut?”

Higher order critical points, their indices
A singular point for which the Jacobian matrix is degenerate is said higher order singular-
ity. Such singularities can be classified through their index or described by the succession
of sectors.
Higher order singularities, hyperbolic, elliptic and parabolic sectors [Scheuermann, Theisel]
If the goal is to visualize singularities of the right index, complex analysis can be used:
Complex functions presenting singularities of index +1 and +2 [Scheuermann]

Orbits
We consider the 3D-case: the 2D case is simpler and follows from it.
Considering an orbit C and a point p 2 C we can define its associated Poincaré map:
choosing around p a small disk D as a local section of the vector field (meaning that the
vector field is nowhere tangent to this disk), there exists a smaller disk D0 ⇢ D with:
for every q 2 D0, you can define a smallest value t(q), such that 't(q)(q) = q 0 is
in D.
DEFINITION 9. The mapping q 7! 't(q)(q) is called Poincaré map.
An orbit is said hyperbolic if the eigenvalues of the Jacobian matrix are off the unit circle
of the complex plane. Orbits are structurally stable iff they are hyperbolic. In this case,
the Jacobian matrix of the Poincaré map contains all necessary information to classify the
different orbits.

Classification of orbits
In 3D,
1. both roots 1 and 2 are real: both eigenvalues
have the same sign (since Euclidean space is ori-
Closed orbits on surfaces, sources entable).
and sinks orbits([Chen]) • | 1| > 1, | 2| > 1: source closed orbit
• | 1| < 1, | 2| < 1: sink closed orbit
In 2D, the Jacobian matrix is 1 ⇥ 1, • | 1| > 1, | 2| < 1:
there is only one root : both positive: saddle closed orbit
1. if | | > 1, it is a source closed both negative: twisted saddle closed orbit
orbit 2. both roots 1 and 2 are complex conjugates:
2. if | | < 1, it is a sink closed or- • | 1| > 1, | 2| > 1: spiral source closed orbit
bit • | 1| < 1, | 2| < 1: spiral sink closed orbit
REMARK 2. Closed orbits can be knotted or even linked with other orbits.

Vector field topology
The vector field topology (or topological skeleton) is the giving of
• the critical points (with type)
• the separatrices
• the orbits (with type) of the given vector field.
The separatrices partition the vector field in areas of similar behavior.
Topological skeleton of a vector field

Vector field topology: an example
Vector field topology: two sources, one saddle [Scheuermann]


Vector field topology: non-linear topology
Non-linear topology [Scheuermann]

See the higher-order singularities.

Divergence, curl, Helmoltz-Hodge decomposition
Divergence and volume-preserving vector fields
DEFINITION 10. A vector field is volume-preserving if its flow 't carries any open set S in
its domain to a set 't(S) of the same volume, for all times t.
DEFINITION 11. The divergence of a vector field v of coordinates (v1, v2, v3) is a scalar-
valued function:
@v @v @v
div v = 1 + 2 + 3 .
@x @y @z
The notation div v = r · v is often used.
The condition for a vector field v to preserve volume is that its divergence div v is
equal to zero.
For this reason, the term divergence free is used as a synonym for volume-preserving. In
such a vector field, no open set may ever reduce its volume, whether in positive or negative
time; as a consequence, there are no sinks and no sources.

Curl
In vector calculus, the curl is a vector operator that describes the infinitesimal rotation of a
3-dimensional vector field. It cannot be extended easily to higher dimensions.
DEFINITION 12. If you have a 3D vector field v with cartesian coordinates (v1, v2, v3), its
curl is defined as
! ✓ ◆ !
@v3 @v2 @v1 @v3 @v2 @v1
curl v = e1 + e2 + e3 .
@y @z @z @x @x @y
The curl of v is often written r ⇥ v.
Intuitive explanation: If you imagine that your vector field is the velocity field for a fluid, in
every point you can imagine a small ball, that the fluid brings to rotate. The direction of the
curl is the axis of this rotation. The magnitude of the curl is twice the angular speed.

Helmoltz-Hodge decomposition for vector fields
THEOREM 3. (Helmholtz-Hodge Decomposition) A smooth vector field v, defined on a
bounded or an unbounded domain, can be uniquely decomposed into three components:
1. an irrotational component d,
2. an incompressible component r,
3. an harmonic component h, representing the translation part.
The components d and r can be calculated as the gradient of a scalar potential D and the
curl of a vector potential R, respectively,
v = rD + r ⇥ R + h = d + r + h.
REMARK 3. The harmonic part being at the
same time irrotational and incompressible can
be put together with the first or the second
term, giving a version of the decomposition
with only two parts.
Example of a HHD decomposition [Polthier]:
d, with scalar potential D, r with scalar po-
tential R (2D case), h, and the original vector
field v.
Visualization of vector fields: small catalogue

Sampled vector field
Suppose that we have a sampled vector field on a triangulation. Using interpolation of the
vector field inside the triangles, on can draw streamlines and even sketch the vector field
topology. It is clear that both the streamlines and the vector field topology depend on the
type of interpolation used, sometimes delivering wrong topologies:
Original vector field, linearly interpolated vector field, Powell-Sabin interpolated vector field, with the corre-
sponding vector field topologies [Scheuermann]

Streamlines, streamribbons
• The streamsurfaces are sometimes used to visualize the flow in 3D. They are formed of
the streamlines rooted at every point along a line segment or a curve.
• The streamribbons are generated from the tangent and normal unit vectors to the stream-
line. The twists of a streamribbon reflect the properties of the local flow.
Hedgehog versus streamlines [Amebarki], streamlines versus streamtapes [Chen]

The problem of seeding
There should not no be empty region, and no crowded
one!
Different strategies are used: seeds can be distributed
evenly, streamlines can be distributed evenly, seeds can
be distributed according to an entropy (reflecting the
areas carrying more “information”)...
Different seeding strategies: grid-seeded versus optimized [Turk]

Line integral convolution
A random noise pattern is chosen and a convolution is
made over every streamline.
Let v be a 2D-vector field, x a point,
2
x : I = [ 1, 1] ! R , a unit-speed parametriza-
tion of a streamline such that x(0) = x. The LIC at x
is defined as:
Z
Im(x) = ⌦(s) N ( x(s))ds.
I
where ⌦(s) is a Gaussian (for example) convolution
kernel on I, N (x) is the noise image in x.
The method is applicable to surfaces in higher dimensions, using multidimensional noise
fields.

Example: different strategies for the same vector field
[Scheuermann]

Vector Elds

Uploaded by

Copyright:

Available Formats

Vector Elds

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Vector Elds

Uploaded by

Copyright:

Available Formats

Vector fields

Chantal Oberson Ausoni

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 1

Vector fields: Outline

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 3

Vector fields: motivation, definition

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 4

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 5

Integral lines, singularities, vector field topology

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 6

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 7

Integral lines, singularities, vector field topology

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 8

Regular trajectory, two kind of stationary points, orbit [Frey]

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 9

Integral lines, singularities, vector field topology

u1(t) = (x + t, 0) and u2(t) = (x + t, t3).

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 10

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 11

Integral lines, singularities, vector field topology

cA( ) = det(A I3) = 0.

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 12

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 13

Integral lines, singularities, vector field topology

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 14

v(p) = Dv(p0)(p p0) + 0((p p 0 )2 )

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 15

Integral lines, singularities, vector field topology

The 2D hyperbolic singularities, and the non-hyperbolic center [Theisel]

Integral lines, singularities, vector field topology

The 3D hyperbolic singularities [Scheuermann]

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 18

• The point (0, 0) is a singularity. !

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 19

Integral lines, singularities, vector field topology

We can also represent all (normalized) vectors

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 21

Integral lines, singularities, vector field topology

In particular, if the Euler characteristic is non-zero, then

”You cannot comb a coconut. Can you comb a doughnut?”

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 22

Complex functions presenting singularities of index +1 and +2 [Scheuermann]

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 23

Integral lines, singularities, vector field topology

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 24

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 25

Integral lines, singularities, vector field topology

Topological skeleton of a vector field

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 26

Vector field topology: two sources, one saddle [Scheuermann]

Integral lines, singularities, vector field topology

Non-linear topology [Scheuermann]

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 28

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 29

Divergence, curl, Helmoltz-Hodge decomposition

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 30

Visualization of vector fields: small catalogue

ICS Summer school Roscoff - Visualization at the interfaces 28.7-8.8, 2014 32