Seminar on Geometric Control Theory and its Applications
Academic year 2002 - 2003
- Wednesday May 21, 2003
Wen Shen (SISSA)
Small BV Solutions
of Hyperbolic Non-cooperative Differential Games
This is a joint work with Alberto
Bressan.
The talk will begin with a brief introduction to non-cooperative
differential game, which is actually an optimal control problem. I will
formally derive the Hamilton Jacobian equations
and the corresponding system of conservation laws.
The hyperbolicity of the system is essential,
and neccesary conditions are derived. In the positive case, the weak
solution of the system of conservation
laws determines an n-tuple of feedback strategies. These yield Nash
equilibrium solutions to the non-cooperative
differential game.
Some discussion about extensions and future work will be given at the
end of the talk, where we will also discuss partially cooperative
strategies which always yield a hyperbolic system.
- Wednesday May 7, 2003
Alexander Zuyev (ICTP / Inst. of Appl. Mathematics &
Mechanics, Donetsk)
On stabilizabily
conditions based on the critical Hamiltonians and symbols
Symmetric functions of the critical
Hamiltonians, called
symbols,
play an important role in such problems of nonlinear
control as characterization of the symmetries and the
feedback invariants.
We derive here a stabilizability condition in the class
of almost continuous feedback controls based on the
symbols.
The methodology proposed consists in defining a selector
of the multivalued covector field satisfying extra degree
condition on
a sphere close to the origin.
Then the above selector is extended to some
neighborhood in order
to get an exact differential form.
Integration of that form gives us a control Lyapunov function
candidate. The condition proposed is applied for analysis
of a planar control-affine system with polynomial homogeneous vector
fields.
Unlike known results in the literature, we consider the case
when each vector field vanishes at the origin and the
control is bounded. We give stabilizability conditions
explicitly in terms of the control system parameters.
This is joint work with Prof. Bronislaw Jakubczyk.
- Wednesday April 30, 2003 at 14:30:
Grigory Sklyar (University of Szczecin, Poland)
Representation
of affine control systems as series of nonlinear power moments and
optimality
We develop the approach to the
time-optimal problem based on the Markov moment problem method. The
initial point in our analysis is the representation of affine control
systems in a neighborhood of the equilibrium as the series of nonlinear
power moments. That allows, in particular, to examine analytic changes
of variables in the system as the corresponding analytic
transformations of its series. We consider the free algebra of
nonlinear power moments A an
d study the properties of finite-generated right ideals of this
algebra. In this way we obtain the following genera
lization of the remarkable R. Ree's theorem: Let J be an ideal
generated by finite number of Lee elements of A and J^\perp be its
orthogonal complement in A. Th
en J^\perp admits the orthogonal decomposition J^\perp = L \oplus
L^{sh},
where L is the orthogonal projection of the Lee subalgebra of A to
J^\perp and
L^{sh} = \{l1* l2* . . . *lk, k \geq2, l_j\in L\} , where * means the
shuffle product in A.
Using this theorem we obtain the result on the reduction (by an
analytic transformation) of the system series to th
e canonical form and define the (homogeneous) principle part of the
series. This, in turn, gives the asymptotic app
roximation of the time optimal problems for affine systems by problems
for the homogeneous systems whose series hav
e the same principle part.
Joint work with Svetlana Ignatovich (Kharkov National University).
- Thursday April 10, 2003 at 14:30:
A. Agrachev
On
Controllability of the Navier--Stokes Equation by Low Modes Forcing
We study controllability of the 2 and
3-dimensional Navier--Stokes equations
$$
\frac{\partial u}{\partial t}+(u\cdot\nabla)u+\nabla p-\nu\Delta
u=f, \quad \mbox{div} u=0 \eqno (1)
$$
with periodic data. Here $f$ is control which has the shape of a low
order (we stay with order two) trigonometric polynomial with respect to
the state variables:
$$
f(t,x)=\sum\limits_{|k|\le 2}e^{ik\cdot x}a_k(t),\quad
a_{-k}=\bar a_k,\quad a_k\cdot k=0,
$$
where $k=(k_1,k_2)$ or $k=(k_1,k_2,k_3),\ k_j\in\mathbb Z,\
|k|=\sum\limits_j|k_j|.
$$
The goal is to understand how the structure of the equations serves
for the energy propagation from low to higher frequencies.
e denote by $v^N$ the $N$-th order Fourier polynomial of the
function $v(t,x)$ with respect to the state variables.
Consider the finite dimensional Galerkin truncation
$$
\frac{\partial u^N}{\partial t}+\left((u^N\cdot\nabla)u^N\right)^N+
\nabla p^N-\nu\Delta u^N=f, \quad \mbox{div} u^N=0 \eqno (1^N)
$$
of the equation (1).
We prove that ($1^N$) is globally controllable for any $N$.
In the case of the 2-dimensional Navier--Stokes equation the same class
of controls provides the global approximate controllability of the
original (not truncated) system (1).
This is a joint work with Andrei Sarychev.
- Wednesday March 5, 2003 at 16:30:
Pierpaolo Soravia (Dip. Matematica - University of Padova)
Equazioni di
Hamilton-Jacobi a coefficienti discontinui e conseguenze
nella teoria dei controlli.
Discuteremo una classe di equazioni di
tipo Hamilton-Jacobi a
coefficienti discontinui. Per tali equazioni e' appropriato il concetto
di soluzione di viscosita', che garantisce l' esistenza di soluzioni.
In
generale pero' problemi al contorno hanno soluzioni multiple.
Presentero' alcune condizioni necessarie e sufficienti per l' unicita'
ed alcuni espliciti risultati di unicita' sotto ipotesi opportune, che
si applicano ad esempio al caso dell' equazione iconale dell' ottica
geometrica con indice di rifrazione discontinuo sotto la condizione
di Hormander.
- Wednesday February 26, 2003:
J. B. Pomet (INRIA, Nice, France)
On the (very)
formal integrability of the system
of PDEs for «flat outputs» of control systems
When studying flatness of control
systems, there appears some (overdetermined) systems of PDEs whose
order is not fixed, yet finite for any proper solution.
In order to analyse such equations, the main tool introduced here is a
valuation, that seems very natural to control system, and allows one to
compute formal power series solution although they potentially depend
on infinitely many variables.
(joint work with David Avanessoff et Laurent Baratchar)
- Wednesday December 4, 2002:
M. T. Nihtila (University of Kuopio, Finland)
Parametrization
of boundary control
Partial differential operators in $G
\times I$, where $G \subset \R^n$ is a bounded set and $I \subset \R$
an interval of
time variable, and their generalizations, pseudo-differential
operators, are studied here keeping in mind the goal of
modelling physical distributed-parameter phenomena. Control actions of
such systems usually take place on the
boundary $\partial G$. Symbolic calculus applied gives tools to form
e.g. compositions, formal adjoints, generalized right
and left inverses and so-called compatibility conditions. The operators
form an algebra $\mathcal D$ by using of which
typical boundary-value control problems can be formulated.
Parametrizability, which is a concept closely related to
flatness of ordinary controlled differential systems, means that for a
given control system $\L u=0$, where $\L$ is the
system operator, and the variable $u$ includes the actual control,
state and output variables, one can find an operator
$\S$ such that $\L u=0$ if and only if $u = \S f$ for some relevant
function $f$. It is required that the components $f_i$
of $f$ are $\mathcal D$-linearly independent. The pseudo-differential
operators and boundar-value operators are formed
by the matrix-like operators: r^+A+B K \L= T Q where the $KQ$-column
operates on the functions defined on $\partial
G \times I$, and the first column on functions in $\bar G \times I$.
The operator $r^+$ restricts the global operator $A$
to act in $\bar G \times I$. $T$ is the trace operator. In our
applications the computation rules in $\mathcal D$ give
explicitly the parametrization operator $\S$ subject to certain
existence assumptions of related pseudo-differential
operators or of one-sided inverse operators. This construction is based
on methods of homological algebra. Projective
freeness of a certain factor module (defined by the system equations)
implies parametrizability. Some examples of
partial differential control systems including boundary conditions are
presented to illustrate the parametrizability
concept and construction of the operator $\S$.
- Wednesday 27 November, 2002:
A. De Simone (Max Planck Institute - Leipzig)
Multiscale modelling of materials: two case
studies
- Wednesday 6 November, 2002 at 16:00:
T. Chambrion (SISSA)
Estimation
of the diameter of SU(n) under a single input left invariant control
system
Lie group formalism and Riemaniann
techniques can be used to get
an estimation of the diameter of SU(n) under a single input left
invariant
control system, that is to say a system with a drift and one unbounded
real control. The first step is to apply the technique of variation of
the
constant to approximate the reachable set at time T of the initial
system.
The second step consits in the replacement of the initial system by a
very
close Riemaniann one. Then, using the particular structure of the Lie
algebra su(n), it is possible to use well-known Riemaniannian results
to
get an upper bound for the diameter of the group.
- Wednesday 23 October, 2002:
Grégoire Charlot (SISSA)
Quasi-contact
S-R Metrics:
normal form in $R^ {2n}$, wave front and caustic in $R^ {4}$
This talk deals with sub-Riemannian
metrics in the quasi-contact case. First, in any even dimension, we
construct normal
coordinates, a normal form and invariants, which are the analogs of
normal coordinates, normal form and classical invariants
in Riemannian geometry. Second, in dimension 4, and thanks to this
"normal form", we study the local singularities of the
exponential mapping.
- Wednesday 9 October, 2002:
C. Altafini (SISSA)
Lie algebraic
aspects of the controllability of quantum systems
The controllability of the Schrodinger
equation and of the Markovian
master equation for N-level quantum mechanical systems is discussed
using Lie algebraic tools and notions from Lie semigroup theory.
The two level case is treated in detail.