Professional Documents
Culture Documents
U.
mjimenez@gmail.com
Jimenez Urrea, Liliana.
Universidade Estadual de Campinas, BRASIL.
lilijimenezu@hotmail.com
Kalise, Dante.
Universidad Tecnica Federico Santa Mara, CHILE.
dante.kalise@usm.cl
Lopez Erazo, Tulio Emiro.
Universidad del Cauca, COLOMBIA .
tele2607@hotmail.com
Mancinelli, Elina.
Universidad Nacional de Rosario, ARGENTINA.
elina@fceia.unr.edu.ar
Mena, Hermann.
Escuela Politecnica Nacional, ECUADOR.
hmena@math.epn.edu.ec
Merino, Pedro.
Escuela Politecnica Nacional, ECUADOR.
merino@math.epn.edu.ec
Monroy Perez, Felipe.
UAM Azcapotzalco, M
EXICO.
fmp@correo.azc.uam.mx
Moreno, Joali.
Universidad Central de Venezuela, VENEZUELA.
jmoreno@kuaimare.ciens.ucv.ve
14
Parente, Lisandro.
Universidad Nacional de Rosario, ARGENTINA.
lparente@fceia.unr.edu.ar
Recalde, Diego.
Escuela Politecnica Nacional ECUADOR.
recalde@math.epn.edu.ec
Salazar, Mara Fernanda.
Escuela Politecnica Nacional, ECUADOR.
msalazar@math.epn.edu.ec
Semet, Frederic.
Universite de Valenciennes, FRANCE.
frederic.semet@univ-valenciennes.fr
Sosa, Wilfredo.
Instituto de Matematica y Ciencias Anes, PER
U.
sosa@uni.edu.pe
Soto, Mara.
Banco Central del Ecuador, ECUADOR.
marlima5@yahoo.com.mx
Sukhomlin, Nikolay.
Universidad Autonoma de Santo Domingo, REP. DOMINICANA.
w17971865@gmail.com
Thera, Michel.
Universite de Limoges, FRANCE
michel.thera@unilim.fr
Torres, Luis Miguel.
Universitat Magdeburg, GERMANY.
Escuela Politecnica Nacional, ECUADOR.
ltorres@math.epn.edu.ec
Torres, Ramiro.
Escuela Politecnica Nacional, ECUADOR.
rtorres@math.epn.edu.ec
15
16
Abstracts
QVI systems related to some production optimization problems
In this paper we study the Quasi Variational Inequalities (QVI) Systems which
appear in the optimization of some problems related to a multi-item single
machine, where at any time the machine is either idle or producing any of m
dierent items.
Our aim is to obtain an optimum production schedule for each of the cases
we propose, therefore we dene a policy control as a pair (production state,
time) which states at each instant of time which item must be produce. The
cost to be optimized takes into account a running cost and the production
switching costs, as the integral cost functional has an innite horizon a discount
factor is used to guarantee that the integral converge.
Using the dynamical programming methods we arrive to Hamilton-Jacobi-
Bellman equations associated to these problems which are given by QVI sys-
tems. Our central result is the characterization of the optimal cost function
as the unique solution in the viscosity sense of this QVI systems.
We focus our attention on two particular cases: the ergodic case and the
piecewise deterministic one. In Gonzalez R.L.V., Rofman E., Sur des solutions
non bornees de lequation de Bellman associee aux probl`emes de commutation
optimale avec constraints sur letat. Comptes Rendus Acad. Sc. Paris, (1993),
a deterministic problem akin to these ones was studied.
In the ergodic case the objective is to nd an optimal production schedule
that minimizes the average cost for an innite horizon. By using dynamic pro-
gramming techniques and taking into account the switching cost, it is possible
to nd an optimal feedback policy, in terms of any solution in the viscosity
sense of a rst order QVI system. This system is obtained considering a se-
quence of optimization problems with non zero discount rate. We prove the
existence of solution of this QVI system and the uniqueness of the optimal
average cost.
Secondly we deal with the case where the demand varies randomly accord-
ing to a piecewise deterministic process. The demand changes are described
by a Poisson processes and the demand value takes a nite number of values.
This type of problems belong to the class of optimal control problems of jump-
ing processes with state space constraints. We describe the optimal scheduling
problem of a multi-item single machine production system where, besides the
inner production, external purchases are allowed to cope with the demands.
We analyze the HJB equation associated to this problem, considering both
the integral and the dierential expression of this equation. We present the
existence, uniqueness and regularity of the optimal cost function.
17
Aragone, Laura
Universidad Nacional de Rosario
Rosario, ARGENTINA
laura@fceia.unr.edu.ar
Mancinelli, Elina
Universidad Nacional de Rosario
Rosario, ARGENTINA
elina@fceia.unr.edu.ar
Condiciones de existencia en optimizacion fuzzy
En un problema de optimizacion usualmente consideramos que todos los datos
e informaciones correspondientes al problema estan perfectamente determina-
dos. Sin embargo en la practica este no es siempre el caso y puede ser necesario
considerar incertidumbre en uno o varios aspectos del problema, bien sea por
informacion insuciente, porque la informacion actual no es objetiva o simple-
mente por errores en la medicion de los datos.
El analisis de sensibilidad es un aspecto fundamental en un problema de
optimizacion y en cierta forma ataca un aspecto relacionado con la incertidum-
bre en los datos. Otra forma muy frecuente para modelar situaciones de opti-
mizacion bajo incertidumbre es ofrecida por la optimizacion estocastica pero
no siempre las situaciones de incertidumbre que presenta un problema son de
tipo probabilstico. Varias herramientas diferentes al analisis estocastico han
sido presentadas para adaptarse mejor seg un diferentes contextos especcos,
podemos nombrar por ejemplo el analisis intervalar y la teora de conjuntos
fuzzy.
Desde la aparicion del concepto de conjuntos fuzzy (fuzzy sets) introducido
por Lofti Zadeh en 1965 han aparecido numerosas aplicaciones de este concepto
y de los aspectos teoricos y practicos desarrollados a partir de el, algunos
componentes de esta teora fuzzy son la logica fuzzy, aritmetica fuzzy, teora
de la posibilidad, ecuaciones e inclusiones diferenciales fuzzy entre otros.
La optimizacion fuzzy trata justamente del estudio y analisis de problemas
de optimizacion donde uno o varios aspectos del problema estan especica-
dos por alg un elemento fuzzy. Con cierta profundidad se han estudiado hasta
ahora problemas de programacion matematica en los cuales las restricciones
son consideradas como conjuntos fuzzy de R
n
o incluso problemas en donde
algunos de los coecientes del problema son considerados n umeros fuzzy. Nu-
merosos artculos y aplicaciones con este enfoque de optimizacion fuzzy han
aparecido en los ultimos 30 anos.
Otro tipo problema interesante, poco estudiado hasta ahora, pero que esta
comenzando a recibir mas atencion es el problema de optimizacion sobre es-
pacios fuzzy, esto es, problemas de optimizacion donde las variables de de-
cision son tambien de tipo fuzzy. Este problema presenta m ultiples aspectos
matematicos interesantes y consideramos que tiene un amplio potencial de
aplicacion.
18
A pesar de que el espacio de decision pueda ser dotado con operaciones de
suma y multiplicacion por escalar razonables, estas no forman en general un
espacio vectorial. Por otro lado, las posibles metricas de que puede dotarse este
espacio le proporcionan cada una diferentes propiedades topologicas. Parece
natural considerar que dado que los conjuntos fuzzy son funciones de pertenen-
cia sobre un conjunto universal, un problema de optimizaci on en espacios fuzzy
pueda verse como un problema de optimizacion sobre un espacio funcional, de
forma analoga a lo que sucede en un problema de control optimo. Esta idea
debe considerarse con cuidado por que la estructura algebraica y topologica
del espacio fuzzy no constituye en general un espacio vectorial topologico o un
espacio vectorial normado.
En este trabajo pretendemos mostrar una introduccion a la optimizacion
fuzzy, con un enfasis especial sobre la optimizacion en espacios fuzzy. Quere-
mos mostrar resultados recientes sobre las condiciones de existencia del optimo
para este tipo problema y discutir las posibilidades de aplicacion a futuro.
Baez Sanchez, Andres David
Universidad Estadual de Campinas
Campinas, BRASIL
adavidbaez@gmail.com
Isomorsmo de grafos e variancias do PQA
Uma instancia simetrica de ordem n do Problema Quadratico de Aloca cao
(PQA) possui duas outras instancias a ela associadas: uma instancia relaxada,
de ordem n(n 1)/2, cuja solu cao e polinomial, caracterizada pela matriz
Q = FD
T
, onde F e D sao vetores que contem os triangulos superiores das
matrizes de dados da instancia; e uma instancia-padrao, de ordem n, na qual os
valores dos vetores F e D sao ordenados em ordens opostas. As variancias dos
valores das solu coes dessas instancias podem ser obtidas em tempo polinomial,
[GW70], [ABQG02].
O problema do isomorsmo de grafos pode ser modelado como um PQA,
cujas matrizes de dados sejam descritivas dos dois grafos em questao. Dois
grafos serao isomorfos se e somente se o valor maximo da fun cao objetivo do
PQA formado por suas matrizes de adjacencia for igual ao n umero comum m
de arestas dos dois grafos. Esta verica cao exige, portanto, a resolu cao exata
de uma instancia do PQA, que e NP-hard.
Em [ABQG02] e denido o isomorsmo de instancias do PQA e se mostra
que duas instancias isomorfas possuem a mesma variancia. Para utilizar este
resultado na caracteriza cao do isomorsmo de um par de grafos, comparase
a instancia a ele correspondente, com instancias construidas com duas copias
do primeiro, e com duas copias do segundo grafo. Se os dois grafos forem
isomorfos, as variancias associadas a essas tres instancias serao iguais.
Esta igualdade de variancias, porem, nao e suciente para caracterizar o iso-
morsmo: dois grafos podem ser nao isomorfos e apresentar variancias iguais
neste teste. Foram utilizadas, entao, valora coes invariantes das arestas, ou
seja, valora coes dependentes das propriedades do grafo e nao de sua rotula cao.
19
Testes realizados com a valora cao dada pela contagem de percursos de com-
primento 2 entre vertices ligados por arestas que correspondem ao n umero de
ciclos de comprimento 3 de que cada aresta participa mostraram ser a mesma
capaz de distinguir entre grafos nao isomorfos de ate 500 vertices, com apenas
uma troca de aresta dentre um maximo de 30.000 arestas. O custo computa-
cional desta valora cao e O(n2), assim como o custo da variancia relaxada,
enquanto as outras duas sao O(n4).
A um par de grafos valorado como acima descrito se pode, ainda, aplicar
uma sobrevalora cao correspondente `a soma dos valores das arestas adjacentes,
em O(m). Outras possibilidades podem ser consideradas, tais como a soma de
graus, em grafos nao regulares, ou os valores dos afastamentos dos vertices que
denem as arestas; enm, a contagem de percursos pode ser levada a compri-
mentos maiores, embora a custo mais elevado.
[GW70] Graves, G.W. e Whinston, A.B. (1970). An algorithm for the qua-
dratic assignment problem. Management Science 17, 453-471.
[ABQG02] Abreu, N.M.M., Boaventura-Netto, P.O., Querido, T.M. e Gouvea,
E.F.. (2002). Classes of quadratic assignment problem instances: isomorphism
and diculty measures using a statistical approach. Discrete Applied Mathe-
matics 124, 113-116.
Boaventura Netto, Paulo Oswaldo
Universidade Federal do Rio de Janeiro
Rio de Janeiro, BRAZIL
boaventu@pep.ufrj.br
Control optimo de problemas elpticos con restricciones puntuales
sobre el gradiente del estado
Algunos autores han abordado recientemente el analisis numerico de problemas
de control optimo gobernados por ecuaciones en derivadas parciales elpticas
con restricciones puntuales sobre las derivadas del estado. Estos trabajos se
han apoyado en los resultados teoricos probados por Casas y Fernandez (Ap-
plied Mathematics and Optimimization 27, 35-56, 1993). Aquel estudio se
centro en la demostracion de la existencia de un multiplicador de Lagrange
asociado a las restricciones sobre el estado y las correspondientes condiciones
de optimalidad de primer orden. En esta conferencia volvemos sobre el mismo
problema y aportamos nuevos resultados que pueden conducir a mejoras sig-
nicativas en las estimaciones del error de las aproximaciones numericas. El
problema se formula en los siguientes terminos.
(P)
Min J(u) =
L(x, y
u
(x))dx +
N
2
u2(x)dx
u(x) a.e en , [y
u
(x)[ x K
20
donde y
u
es la solucion del problema de Dirichlet
Ay + a
0
(x, y) = u en
y = 0 sobre ,
siendo un abierto acotado de R
n
, n = 2 o 3, es la frontera de , A un
operador elptico de segundo orden en , K
es un conunto compacto,
< < < + y > 0.
Para este problema deducimos las condiciones sucientes de optimalidad de se-
gundo orden y un resultado de regularidad inesperado para el control optimo.
Casas, Eduardo
Universidad de Cantabria
Santander, ESPA
NA.
eduardo.casas@unican.es
Plan maestro de abastecimiento de alimentos para Guayaquil
En el presente trabajo se plantea la utilizacion de un modelo conceptual para
el Abastecimiento de Alimentos en la Ciudad de Guayaquil. Este modelo esta
basado en la creacion de entidades logsticas (puntos de transferencia, nodos
logsticos y plazas logsticas) cuyo proposito fundamental es mejorar la organi-
zacion en terminos de distribucion y comercializacion de los diversos productos
que demanda la poblacion de la ciudad. En base a estas entidades, se realizan
los analisis respectivos para determinar la conguracion optima del sistema.
Para ello, se emplea el concepto de la programacion matematica, dise nando un
modelo que permite minimizar los diferentes costos que intervienen en todo el
Sistema de Abastecimiento planteado: costos de construccion, de operacion,
de transporte, y de produccion. El modelo fue dise nado sobre las distintas
cadenas agroindustriales de alimentos elegidas para el estudio, teniendo sub-
modelos para cada una de las cadenas homologadas con cinco eslabones co-
munes: Produccion, Transporte, Transformacion, Distribucion al consumidor
y Flujos internos en la ciudad. Se lo ha construido de tal forma que genera los
ujos y vol umenes de las cadenas de acuerdo con una demanda previamente
sectorizada dentro de la ciudad, la misma que es estimada en base a una
poblacion proyectada y a una canasta basica de consumo de alimentos al cabo
de un horizonte de tiempo preestablecido. Finalmente, seg un los escenarios
considerados, se plantea la utilizacion de esta herramienta para la localizacion
optima de las plazas de mercado existentes actualmente en la ciudad y las
propuestas en el estudio.
Cepeda de la Torre, Carlos Alberto
Escuela Politecnica del Litoral
Guayaquil, ECUADOR
ccepeda@espol.edu.ec
21
Numerical and optimization methods for active vibration control
and model updating in vibrating structures: linking control to in-
dustry
The use of active feedback control strategy is an eective way to stabilize and
control dangerous vibrations, such as resonance and other forms of instability,
in vibrating structures, including bridges, highways, buildings, and space and
air crafts. These structures are distributed-parameter systems. However, be-
cause of practical considerations such systems are very often discretized into
systems of matrix second-order dierential equations using nite-element tech-
niques; control is then designed and implemented on this discretized system
and, nally applied to a real-life structure. Unfortunately, the existing vibra-
tions control techniques, even for these simplied models, have some severe
numerical diculties and engineering limitations. The vibration industries
have approached vibration control problems in an ad hoc way and many of
the current engineering practices are not supported by strong mathematical
foundations. In the last few years, the speaker and his collaborators have
developed a practical computational approach for feedback control in nite
element structures. The distinctive features of this approach are (i) control
can be designed directly on the nite model without requiring transformation
to a standard rst-order state-space form, (ii) the algorithms require knowl-
edge of only a small number of eigenvalues and eigenvectors of the associated
quadratic matrix pencil, and (iii) above all, no a priori reduction of the order
of the model or controller is required, irrespective of the size of the systems.
Exploiting the parametric nature of the approach, optimization-based numeri-
cally robust algorithms have also been developed to design feedback controllers
which remain as insensitive as possible with respect to small perturbations in
the data. This approach has also been successfully applied to another related
industrial problem, namely, the nite-element model updating problem which
concerns updating an analytical nite element model with experimental data
so that the updated model can be used with condence for future design and
constructions. The minimal computational and engineering requirements of
this new approach make it readily applicable to feedback control design and
model updating in even very large practical-life structures. In this talk, these
recent advances will be reviewed and a brief discussion will be presented on
future directions of research in this area.
Datta, Biswa
Northern Illinois University
Illinois, USA
dattab@math.niu.edu
Optimal control of partial dierential equations with ane control
constraints
The numerical solution of PDE optimal control problems involving ane point-
wise control constraints is investigated. Optimality conditions are derived and
22
a semi-smooth Newton method is presented. Global and local superlinear con-
vergence of the method are obtained for linear problems. Dierently from box
constraints, in the case of general ane constraints a proper weighting of the
control costs is essential for superlinear convergence of semi-smooth Newton
methods. This is also demonstrated numerically by controlling the two dimen-
sional Stokes equations with dierent kinds of ane constraints.
De Los Reyes, Juan Carlos
Escuela Politecnica Nacional
Quito, ECUADOR
jcdelosreyes@math.epn.edu.ec
Neural network control of materials sinterization system
This report describes the development of control strategies based on neural
network for a complex thermal system on range of 20C to 1400C. We use an
adaptive neural network identier with parallel structure and neurocontroller
on line with adaptive learning. To control the temperature of the oven, a
5-3-1 neural network structure, was formed by process control signal. The
system dynamic was identied with a neural 6-1 structure neural network and
We used an estimate systems output for including in controller. Both tan-
gent sigmoidal activation function. All system is proved with a typical curve
for materials sinterization in physical analysis. We applied adaptive feedback
back-propagation algorithm, the input to the controller was selected before test
estimate and measure signal output and some past systems input. Inverse Di-
rect strategy control with system identication was used on development. This
work was made in Universidad de Nari no, San Juan de Pasto Colombia by In-
telligent Systems and Instrumentation Research Group.
Fajardo Fajardo, Daro Fernando
Universidad de Nari no
San Juan de Pasto, COLOMBIA
dario@udenar.edu.co
Ruiz Rosero, Juan Pablo
Universidad de Nari no
San Juan de Pasto, COLOMBIA
jpruiz84@gmail.com
Martnez Benavides, John Jairo
Universidad de Nari no
San Juan de Pasto, COLOMBIA
john jairo25@hotmail.com
23
A semismooth Newton algorithm for Bingham uids in channels
In this work, we are concerned with the boundary value-problem describing
the ow of a Bingham uid in a bounded channel R
d
, d 1, 2, consid-
ering multiple boundary conditions. We analyze the variational setting of this
problem given by a variational inequality of the second kind. We prove that
this inequality is a necessary optimality condition of a minimization problem
in a convex set of H
1
(). Then, by using Fenchels duality theory, we deeply
analyze the problem and obtain a primal-dual optimality system relating the
solutions for the primal and dual problems. Unfortunately, this system is ill-
conditioned since the dual problem has not unique solution. Thus, we propose
a Tikhonov regularization, which is a penalty-type smoothing, to overcome
this issue. Penalty-type smoothings imply a local regularization of the original
problem, and, dierently from global smoothings, stay as close as possible to
the original problem formulation and reduce in this manner the ill-conditioning.
Since semi-smooth Newton methods have been successfully applied to in-
nite dimensional complementarity problems like the Signorini or contact prob-
lem and to simulation and control of uids, we propose a second order algo-
rithm, based on these methodology, to numerically solve the problem. We
deeply analyzed the constructed algorithm. Particulary, we prove that it is
locally superlinear convergent. Finally, we approximate the problem by using
a conforming nite element discretization, and we numerically test the behav-
ior of the algorithm when used to study two problems: the ow of a Bingham
uid in a reservoir with a forcing term and the ow of a Bingham uid in a
given geometry of R
2
, considering homogenous Dirichlet boundary conditions,
non-homogeneous Dirichlet boundary conditions and stress free boundary con-
ditions.
Gonzalez, Sergio
Escuela Politecnica Nacional
Quito, ECUADOR
sgonzalez@math.epn.edu.ec
De los Reyes, Juan Carlos
Escuela Politecnica Nacional
Quito, ECUADOR
jcdelosreyes@math.epn.edu.ec
A Kantorovich-type result for vector optimization
For an unconstrained multiobjective problem, we propose a Newton-like method
(together with a globalization strategy). The method neither scalarizes the
original problem nor uses ordering information. Under the assumptions of
strict convexity and twice continuous dierentiability of all criteria, as in the
real valued case, the method locally superlinearly converges to an optimum
(a Pareto point). Also as in the scalar case, whenever the Hessians of all ob-
jectives are Lipschitz continuous, the order of convergence is quadratic. As
24
a by product, local existence of ecient point is obtained under semi-local
assumptions.
We also consider an extension of the method for the vector optimization
case, i.e., when the partial order is induced by a closed convex cone. Assum-
ing that the objective function is twice continuously dierentiable and locally
strongly convex with respect to the ordering cone, we show, by means of a
Kantorovich-like technique, that the method locally converges to an ecient
point with quadratic order.
Gra na Drummond, Luis Mauricio
Universidade Federal do Rio de Janeiro
Rio de Janeiro, BRAZIL
lmgd@facc.ufrj.br
Semismooth Newton methods for portfolio optimization
We are considering a continuous-time market model of one bank account and
one stock. Trading the stock incurs transaction costs proportional to the trad-
ing volume. Given the interest rate r, the stock trend parameter , the trans-
action costs and the volatility , the values X
0
and X
1
of the bank account
and the stock, respectively, evolve according to
dX
0
(t) = rX
0
(t) dt (1 +) dL(t) + (1 ) dM(t) (1a)
dX
1
(t) = X
1
(t) dt + X
1
(t) dW(t) + dL(t) dM(t). (1b)
L(t) and M(t) denote the cumulative purchases and sales of the stock and thus
determine the trading strategy. The objective is to maximize the expected util-
ity of the total wealth at a given terminal time T. Admissible trading strategies
ensure positive total wealth at all times. The associated value function satises
a Hamilton-Jacobi-Bellman equation of the type
(2) maxV
t
+LV, L
B
V, L
S
V = 0 on (
1
,
1
) (0, T),
subject to terminal and boundary conditions. Here, L
B
, L
S
and L are rst and
second order linear dierential operators. The boundaries of the sets where
the maximum is attained by either one of the terms determine the optimal
trading policy. In the presentation, we address semismooth Newton methods
which allow for the ecient numerical solution of (2) and give some examples.
Griesse, Roland.
Technische Universitaet Chemnitz
Chemitz, GERMANY
roland.griesse@mathematik.tu-chemnitz.de
25
Compact convex sets of constant width (in 2D) or of constant thick-
ness (in 3D) : new stu from old ones.
When Minkowskis theory of convexity appeared, some mathematicians
said that he discovered a nice mathematical joy which, unfortunately, is quite
useless. About a century passed, and now the theory of convex sets is a very
important applied branch of mathematics.
V. Boltyanski, in Geometric methods and optimization problems (1999).
The geometrical convexity (the one treating of the structure of convex bod-
ies, their smoothness, volume,...) and the functional convexity (the one
treating of properties of convex functions, their use in optimization,...) are
two areas of mathematics, well delineated but dierent, with, for each of them,
their research and working communities, mathematical questions, journals, etc.
There however is a contact point which allows to link the two elds, that of
variational problems concerning convex bodies. These questions lie at the con-
uence of three domains of mathematics: Variational analysis and calculus,
Optimal control, Shape optimization. In this communication, essentially ped-
agogical and of synthesis, we present some properties, mainly of a variational
type, of convex bodies of constant width (in 2D) or of constant thickness (in
3D), laying particular emphasis on the fundamental dierences between 2D and
3D. We thus arrive at the front of some open problems (some are long-standing
ones), for which results and techniques from Variational calculus, Optimal con-
trol, or Shape optimization, have failed to provide answers, up to now. We
shall present in detail the conjecture on convex bodies of constant thickness in
3D of minimal volume. The most conclusive results on that subject, in the 3D
context, have been obtained recently by T. Bayen, Th. Lachand-Robert, E.
Oudet (two papers published in 2007).
La convexite dans le plan et dans lespace presente un sujet passionnant,
la convexite, `a la fois par sa simplicite, sa naturalite et sa puissance, pour au
moins trois raisons :
- au niveau des questions que lon peut se poser naturellement, geometriques,
visibles ;
- du fait que la convexite est une notion qui se rencontre dans de nom-
breuses branches des mathematiques ;
- du fait de son utilite, de sa force, dans de nombreuses applications.
M. Berger in Convexite dans le plan et au-del`a, 2 volumes, collection
Opuscules, editions Ellipses (2006).
Hiriart-Urruty, Jean-Baptiste
Institut de Mathematiques, Universite Paul Sabatier
Toulouse, FRANCE
jbhu@cict.fr
26
Feedback Solutions to Quantum Control Problems
Control of quantum systems described by Schroedinger equation is considered.
Feedback control laws are developed for the orbit tracking via a controlled
Hamiltonian. Asymptotic tracking properties of the feedback laws are ana-
lyzed. Numerical integrations via time-splitting are also analyzed and used to
demonstrate the feasibility of the proposed feedback laws. Also, the receding
horizon feedback law is analyzed and it improves the performance of the pro-
posed feedback laws.
Ito, Kazufumi
North Carolina State University
USA
kito@math.ncsu.edu
An Augmented Lagrangian method for equilibrium problems
We propose an Augmented Lagrangian method for equilibrium problems, whose
feasible sets are given by a nite set of convex inequalities. In each step, the
primal variables are updated by solving an unconstrained equilibrium prob-
lem, while the dual variables are updated through a closed formula. In the
particular case of optimization problems, we recover the proximal Augmented
Lagrangian method proposed by Rockafellar in 1976. Under monotonicity as-
sumptions on the bifunction which denes the equilibrium problem, we estab-
lish global convergence of the generated sequence to a solution of the problem.
Iusem, Alfredo
Instituto de Matematica Pura e Aplicada
Rio de Janeiro, BRAZIL
iusp@impa.br
Um metodo de regiao de conanca para minimizacao irrestrita sem
derivadas
Em otimiza cao, um problema se diz irrestrito se seus parametros podem as-
sumir quaisquer valores. Para o caso de nosso interesse neste trabalho, min-
imizamos uma fun cao objetivo de varias variaveis, sobre todo o espa co R
n
.
Nossa motiva cao para estudar possveis algoritmos para resolver estes proble-
mas, e a grande demanda de prossionais de varias areas por tais ferramen-
tas. Nas aplica coes apresentadas em Powell [2003], calcular o valor da fun cao
F(x), dado o vetor x, e muito caro, e os valores das derivadas de F em x
nao sao facies de encontrar, ou porque F(x) resulta de algum fenomeno fsico
ou qumico, ou, mais comumente, porque este e o resultado de uma complexa
simula cao de um computador. O objetivo deste trabalho e estudar a estrutura
do modelo quadratico Q, construdo por interpola cao por um n umero denido
27
de pontos que sao determinados de uma forma particular, considerando val-
ores da fun cao F. Metodos de regiao de conan ca sao utilizados, no entanto,
quando uma aproxima cao para F(x); x R
n
, e construda a partir dos val-
ores da fun cao objetivo. Entao o seguinte vetor de variaveis e gerado usual-
mente procurando o mnimo da aproxima cao Q, em uma conveniente regiao
de R
n
. A aproxima cao e chamada fun cao modelo contida no espa co linear
M das fun coes de R
n
a R com dimensao m =
1
2
(n +1)(n +2). Cada modelo e
denido por condi coes de interpola cao da forma Q(x
i
) = F(xi), i = 1, 2, . . . , m,
onde x
i
sao os pontos de interpola cao que estao em posi coes que garantem a
nao singularidade do sistema formado por as condi coes, e m esta no intervalo
[n + 1, m]. O metodo utilizado trabalha com posi coes inicias dos pontos de
interpola cao, o ajuste do raio da regiao de conan ca, o calculo do vetor ext, e
a sele cao de um ponto para ser substitudo para avalia c ao das singularidades e
erros das equa coes de interpola cao do modelo, utilizando fun coes de Lagrange.
Em Powell [2003], e provado que a cota de erro pode controlar o ajuste do
raio de uma regiao de conan ca, de forma a se obter excelentes resultados
de convergencia num algoritmo para problemas de minimiza cao irrestrita. O
algoritmo proposto em Powell [2002, 2003], sera testado em alguns problemas
da cole cao Hock-Schittkowski [1981], onde observa-se que a escolha de M e
importante para identicar o trabalho em cada itera cao. A compara cao dos
resultados numericos do metodo e feita com um metodo de busca padrao de
Lewis-Torczon [2000], o qual dene em cada itera cao um conjunto padrao de
dire coes de busca no ponto atual na procura de melhores valores para F.
Jimenez Urrea, Liliana
Universidade de Campinas
Campinas, BRAZIL
lilijimenezu@hotmail.com
da Rocha Lopes, Vera Lucia
Universidade de Campinas
Campinas, BRAZIL
vlopes@ime.unicamp.br
Numerical approximation of the LQR problem in a highly damped
wave equation
The aim of this work is to obtain optimal order error estimates for the LQR
(Linear-quadratic regulator) problem in a highly damped 1-D wave equation.
We consider a nite element discretization of the system dynamics and a con-
trol law constant in the spatial dimension. To solve the LQR problem, we
seek a feedback control which depends on the solution of an algebraic Riccati
equation. Optimal error estimates are proved in the framework of the approx-
imation theory for control of innite-dimensional systems. Finally, numerical
results are reported to illustrate that the optimal rates of convergence are
achieved.
28
Kalise, Dante
Universidad Tecnica Federico Santa Mara
Valparaso, CHILE
dante.kalise@usm.cl
Hernandez, Erwin
Universidad Tecnica Federico Santa Mara
Valparaso, CHILE
erwin.hernandez@usm.cl
Otarola, Enrique
Universidad Tecnica Federico Santa Mara
Valparaso, CHILE
enrique.otarola@usm.cl
Un metodo que no usa derivadas para resolver un problema de pro-
gramacion no lineal con restricciones lineales
En el a no 2002, Marco Sciandrone, Stefano Lucidi y Paul Tseng publican el
artculo titulado: Objetive-Derivative-Free methods for constrained optimiza-
tion [2], en el que presentan dos metodos de descenso suciente para resolver
un problema de optimizacion no lineal:
minimizar f(x) (3)
sujeta a: g
i
(x) 0, i I, (4)
donde f : R
n
R (funcion objetivo) y g
i
: R
n
R (funcion restriccion
para cada i), son continuamente diferenciables en R
n
para todo i I, con
I = 1, . . . , m N. Los metodos propuestos por estos autores para resolver
(1)-(2), no hacen uso del gradiente de la funcion f, ni de aproximaciones al
mismo. Estos metodos se enfocan alrededor de tres aspectos fundamentales: la
b usqueda de una matriz de rango completo tal que sus columnas corresponden
a los gradientes de las restricciones casi activas en el punto actual, la b usqueda
de un conjunto de direcciones que satisfacen alguna condicion, las cuales gen-
eran positivamente el cono polihedral convexo en el punto actual y el calculo
de un paso a ser dado a lo largo de una direccion de descenso.
Los autores, entre otros comentarios, mencionan lo simple que puede re-
sultar intentar resolver con estos metodos, problemas de la forma (1)-(2) que
involucren restricciones de igualdad. El trabajo que desarrollamos consiste en
adaptar el primero de los metodos propuestos por Lucidi, Sciandrone y Tseng
para resolver un problema de programacion no lineal con restricciones lineales
de desigualdad y de igualdad. Formalmente consideramos el problema:
minimizar f(x) (5)
sujeta a: g
i
(x) 0, i I (6)
h
i
(x) 0, l E, (7)
dondeI = 1, . . . , m N, E = m+1, . . . , m+p N y f : R
n
R, (n 1)
es una funcion continuamente diferenciable en R
n
. Ademas, suponemos que
29
las funciones g
i
: R
n
R y h
i
: R
n
R son lineales para todo i I y todo
l E.
El algoritmo nal lo implementamos en MATLAB 6.5 y lo usamos para
resolver algunos problemas de la coleccion Hock-Schittkowski [1, 4]. Ademas,
comparamos los resultados obtenidos en la solucion de dichos problemas con
esta adaptacion, con el metodo de B usqueda patron, propuesto por Robert
Michael Lewis y Virginia Torczon [3].
[1] Hock, W. e Schittkowski, K., 1981. Test examples for nonlinear program-
ming codes, Springer, Berlin.
[2] Lucidi, S., Siandrone, M., e Tseng, P., 2002. .Objetive-derivative- free
methods for constrained optimization, Mathematical Programming, pp. 37-
59.
[3] Lewis, R. M. e Torczon, V., 2000. Pattern search methods for linearly
constrained minimization, SIAM Journal of optimization, vol. 10, No. 3, pp.
917-941.
[4] Schittkowski, K. 1987. More test examples for nonlinear program- ming
codes, Springer, Berlin.
Lopez Erazo, Tulio Emiro
Universidad del Cauca
Popayan, COLOMBIA
tele2607@hotmail.com
Ecient solution of large scale matrix equations arising in LQR/LQG
design for parabolic PDEs
Feedback control of systems govern by partial dierential equations has been
an engineers desire for many decades now. Linear Quadratic Regulator (LQR)
design and Linear Quadratic Gaussian (LQG) design have been investigated
in the literature since the pathbreaking work of Lions, showing that these
methods are valuable tools in achieving this task.
The solution of large scale Lyapunov and Riccati equations is a major
task in applying the above techniques to semi-discretized partial dierential
equations constraint control problems. In this context, the same task appears
when Balanced truncation model order reduction are applied. The software
LyaPack has shown to be a valuable tool in the task of solving these equations
since its introduction in 2000.
Here we want to discuss recent improvements and extensions of the un-
derlying algorithms and their implementation in the successor library M.E.S.S
(matrix equation sparse solver).
Mena, Hermann
Escuela Politecnica Nacional
30
Quito, ECUADOR
hmena@math.epn.edu.ec
Benner, Peter
Technische Universitat Chemnitz
Chemnitz, GERMANY
benner@mathematik.tu-chemnitz.de
Saak, Jens
Technische Universitat Chemnitz
Chemnitz, GERMANY
jens.saak@mathematik.tu-chemnitz.de
Error estimates for the FEM approximation of a semilinear elliptic
control problem with state constraints and nite dimensional control
space
We derive error estimates for a nite element based approximation of a semi-
linear elliptic optimal control problem with a nite-dimensional control space
set along with equality and inequality constraints on the state, which are re-
quired only in nitely many points of the spatial domain. If pointwise state
constraints are given in this case, then they will often be active only in nitely
many points. Therefore, nitely many point constraints seem to be inter-
esting for numerical computations, if the location of active points has been
found approximately. Our problem is equivalent to a nonlinear programming
problem in a nite-dimensional space. We refer Merino et.al. [2007] where
optimality conditions for semilinear control problems with nite-dimensional
control space and pointwise state constraints in the whole spatial domain are
discussed. Computations indicate that the numerical approximation of the
optimal controls and states is of the order h
2
, where h > 0 is the mesh size of
the nite element scheme. In fact, we can show that order the error is equal to
the error made by the nite-element error for the state. Here, no interpolation
error of the optimal control occurs that in the case of control functions which
limits the order of the order of the approximation. If pointwise state con-
straints on the whole domain are given, then there are only a few results. We
refer to Casas [2002] who studied the semilinear problem for functional con-
trols and obtain an orther of h
2n/2
, with n be the dimension of the domain.
In Deckelnick and Hinze [2000] linear-quadratic problems with pointwise state
constraints are considered and obtain the order h
2n/2
. Since the Lagrange
multipliers associated with the (nitely many) pointwise state constraints are
regular Borel measures, which appear in the adjoint equation of the control
system, it is quite surprising the error near to h
2
. It is the nite-dimensional
nature of our problem that explains the high order of the error. Numerical
experiments are presented.
[1] Casas, E. (2002). Error estimates for the numerical approximation of semi-
linear elliptic control problems with nitely many state contraints. ESAIM:
31
Control, Optimization and Calculus of Variations 8, 345-374.
[2] J. C. de los Reyes, P. Merino, J. Rehberg, and F. Troeltzsch. Optimality
conditions for state-constrained PDE control problems with nite-dimensional
control space]. submitted, 2006.
[3] M. Deckelnick and M. Hinze. Convergence of a nite element approximation
to a stateconstrained elliptic control problem. SIAM J. Numerical Analysis,
to appear, 2006. Cantabria, 2000.
[4] C. Meyer. Error estimates for the nite-element approximation of an el-
liptic control problem with pointwise state and control constraints. WIAS,
Preprint 1159, 2006.
Merino, Pedro
Escuela Politecnica Nacional
Quito, ECUADOR
merino@math.epn.edu.ec
Troltzsch, Fredi
Technische Universitat Berlin
Berlin, GERMANY
troeltzsch@math.tu-berlin.de
Vexler, Boris
Technische Universitat M unchen
M unchen, GERMANY
vexler@ma.tum.de
Optimal control on nilpotent Lie groups
Let M be a smooth manifold of dimension n, and let be a rank k < n distri-
bution of smooth vector elds satisfying the bracket generating condition. In
this lecture we present the optimal control problem determined by , together
with the energy functional of -horizontal curves, for the case when M is a
step-2 nilpotent Lie group and is a nite collection of left invariant vector
elds.
The iteration of the Lie bracket of vector elds in yields the ag of
modules of vector elds
1
2
l
TM, where 1 = and
i+1
=
i
+[,
i
]. The distribution is said to be bracket generating, if for each
p mboxM, there exist a positive integer m for which
m
p
= T
p
M.
An absolutely continuous curve t p(t), is said to be -horizontal, if
p(t) (p(t)), almost everywhere. A sub-Riemannian metric is dened by
a smooth varying inner product p , )
p
in (p). For horizontal curves
t p(t), the energy functional is dened as usual,
c(p) =
1
2
p, p).
32
We approach the problem as a sub-Riemannian geodesic problem, that is,
the one of minimizing functional c, in the class of -horizontal curves. We use
the Pontryagin Maximum Principle and the associated Hamiltonian formalism
to derive necessary conditions for geodesics. We integrate some cases of the
Hamiltonian equations and derive some geometric properties of the geodesics
and small radiispheres.
Monroy Perez, Felipe
UAM-Azcapotzalco
Mexico D.F., M
EXICO
fmp@correo.azc.uam.mx
A symmetry preserving alternating projection method for matrix
model updating
The Matrix Model Updating Problem (MMUP), considered in this paper, con-
cerns updating a symmetric second-order nite element model so that the up-
dated model reproduces a given set of desired eigenvalues and eigenvectors
by replacing the corresponding ones from the original model, and preserves
the symmetry of the original model. In an optimization setting, this is a
constrained nonlinear optimization problem. Taking advantage of the spe-
cial structure of the constraint sets, it is rst shown that the MMUP can be
formulated as an optimization problem over the intersection of some special
subspaces and linear varieties on the space of matrices. Using this formulation,
an alternating projection method is then proposed and analyzed. The projec-
tions onto the involved subspaces and linear varieties are characterized. To
the best of our knowledge, an alternating projection method for MMUP has
not been proposed in the literature earlier. A distinct practical feature of the
proposed method is that it is implementable using only a few measured eigen-
values and eigenvectors. No knowledge of the eigenvalues and eigenvectors of
the associated quadratic matrix pencil is required. The results of our numer-
ical experiments on both illustrative and benchmark problems show that the
algorithm works well. The paper concludes with some future research prob-
lems.
Moreno, Joali
Universidad Central de Venezuela
Caracas, VENEZUELA
jmoreno@kuaimare.ciens.ucv.ve
Variable metric proximal decomposition methods
In this work, we extend the general decomposition scheme known as Hybrid
Proximal Decomposition Method (HPDM). This extended method deals with
two kinds of inexactness in both stages of the decomposition scheme and allows
33
for the use of a variable metric in the subproblems. We prove global conver-
gence and we characterize the local linear convergence rate. Also, we show that
the Splitting Method for Composite Mappings proposed by Pennanen and the
Proximal Alternating Directions Method proposed by He et al. can be put in
this scheme as special cases.
Parente, Lisandro
Universidad Nacional de Rosario
Rosario, ARGENTINA
lparente@fceia.unr.edu.ar
Lotito, Pablo
Universidad Nacional del Centro de Buenos Aires
Buenos Aires, ARGENTINA
plotito@exa.unicen.edu.ar
Solodov, Mikhail
Instituto de Matematica Pura e Aplicada
Rio de Janeiro, BRAZIL
solodov@impa.br
Polyhedral considerations for a ower production and packing prob-
lem.
An Ecuadorian company cultivates about ninety varieties of owers which are
then combined according to certain rules (called recipes) into several kinds of
products sold to the customers. We have considered an optimization model
for assigning owers to products in such a way that the global revenue is
maximized. Our model is based on the idea of considering each product as an
array of slots that have to be lled with owers. For this purpose, each slot
has a list of admissible varieties associated to it. There are many possibilities
to assemble a product by selecting owers for a xed recipe. For example,
several varieties of owers of the same color are feasible in a recipe. For
mixed products which contain several colors, the number of feasible ower
combinations for a single product turns out to be exponential. So, we faced a
large scale integer programming problem.
In this work we present a theoretical analysis of the polyhedron associated
to this problem for which we proved it is NP-hard. We demonstrated, under
certain hypotheses, the existence of some families of facets. Computational
experiments, using the SCIP v1.00 solver, showed that our families of facets
improved a twenty percent (in average) the solving time of the integer pro-
grams for some specic set of instances.
Recalde, Diego
Escuela Politecnica Nacional
Quito, ECUADOR
recalde@math.epn.edu.ec
34
An integer programming approach for the single source minimum
cost unsplittable ow problem
In the single source unsplittable min-cost ow problem commodities must be
routed simultaneously from a common source node to certain sink nodes in a
given digraph. The demand of each commodity must be routed along a single
path respecting arc capacities and the total cost must be minimized. Sev-
eral approximation algorithms have been developed in order to solve dierent
versions of this problem. Dinitz, Garg, and Goemans proved that any given
splittable ow satisfying certain demands can be turned into an unsplittable
ow such that the ow value on any arc exceeds the ow value on that arc,
in the given ow, by no more than the maximum demand. Goemans conjec-
tured that this result even holds when we include arc costs, then it is required
that the cost of the unsplittable ow must not exceed the cost of the given
splittable ow. We consider here an integer programming based approach and
formulate a model for nding the minimum cost unsplittable ow of conges-
tion bounded by a xed given parameter. We present a solution algorithm
based on the branch-and-bound framework. As a rst step, two alternatives
for the computation of lower bounds are given , namely, Linear relaxation and
Lagrangian relaxation. Then, a branch and bound scheme for the original
problem is designed by taking advantage of the inner structure of the integer
program. Preliminary computational results regarding to Lower bounds as well
as to the whole branch-and-bound algorithm for the integer model are shown
and compared with solutions given for integer program solvers. We have used
the same input data considered in previous investigations. Such tests reveal
that most of the time our algorithm achieves the optimal solution or nishes
with a good approximation in a reasonable running time. Moreover, when-
ever our algorithm obtains the optimal solution, we compare it with the cost
of the min-cost (splittable) ow and verify that the Goemans conjecture holds.
Salazar, Mara Fernanda
Escuela Politecnica Nacional
Quito, ECUADOR
msalazar@math.epn.edu.ec
Recent advances for location-routing problems
Location-routing problem (LRP) is a generic name for a relatively new class of
NP-hard combinatorial optimization problems at the intersection of location
and transportation analysis. Generally speaking the LRP consists of deter-
mining locations for facilities from which customers are served on routes with
the objective of minimizing the overall cost. More precisely the LRP involves
three types of decisions related to: i) facility location; ii) customer allocation
to facilities; iii) vehicle routing. The overall cost includes the costs for using
and operating the facilities and the transportation costs. Many variants of the
35
LRP are addressed in the literature. They mainly dier by the side constraints
imposed and by the number of intermediate facilities considered. In this talk
we focus on some of variants for which we describe mathematical programming
models and recent solution methods. In particular, we consider: i) the capac-
itated location-routing problem in which capacity restrictions are imposed on
the depots and on the vehicles; ii) the two-echelon location-routing problem
in which the customer demands are transported from the depots to customers
through intermediate facilities.
Semet, Frederic
Universite de Valenciennes
Valenciennes, FRANCE
frederic.semet@univ-valenciennes.fr
Generalized Nash games and equilibrium problems
We reformulate the generalized Nash equilibrium problem as an equilibrium
problem so that solving the former problem is reduced to solving the latter
problem. We use Ky Fans Lemma to obtain a new existence result for equilib-
rium problems, consequently for the generalized Nash equilibrium problems,
which does not invoke monotonicity and convexity of the objective function.
Sosa, Wilfredo
Instituto de Matematica y Ciencias Anes
Lima, PER
U
sosa@uni.edu.pe
Programacion estocastica y algoritmos en-lnea para el PGSC
En este artculo se aborda el problema de la gestion de fondos que la banca
privada mantiene en el Banco Central del Ecuador. El problema es formulado
como un programa lineal estocastico multietapa, donde el parametro sujeto a
incertidumbre es el ujo de caja diario. La tecnica de programacion con re-
curso resulta inaplicable al momento de resolver instancias reales, debido a que
el n umero de escenarios crece explosivamente conforme aumenta el n umero de
das dentro del horizonte de estudio. Como alternativa, se proponen algoritmos
en-lnea para determinar el monto de inversiones y retiros a realizarse diaria-
mente sobre una cuenta en el extranjero, con el n de maximizar la utilidad
esperada dentro de un cierto horizonte de tiempo, mientras el riesgo de que se
presente una situacion de falta de liquidez se mantiene acotado. El desempe no
de estos algoritmos se estudia mediante simulaciones computacionales sobre
instancias extradas del registro historico del BCE.
Soto Lima, Mara Consuelo
Banco Central del Ecuador
Quito, ECUADOR
36
marlima5@yahoo.com.mx
Nuevo metodo de resolucion exacta del problema inverso de modelos
de tipo de Black-Scholes-Merton con volatilidades determinsticas
El resultado principal de este trabajo es un nuevo metodo de resolucion ex-
acta del problema inverso para un gran n umero de modelos nancieros de
tipo de Black-Scholes-Merton. Construimos una ecuacion algebraica para la
volatilidad con coecientes expresados en funcion de las variables observables
en el mercado y una variable que puede ser facilmente calculada usando los
datos del mercado. Resolvemos el problema inverso para algunos modelos con
volatilidades variables.
Nuestro metodo se puede aplicar a numerosos modelos de tipo Black-
Scholes de valuacion del precio de opciones, el Merton-Black-Scholes-Cox mod-
elo estructural de defaultable bonos, varios modelos del riesgo de credito,
Sharpe ratios, derivados del mercado cambiario, modelo de Black Scholes ajus-
tado por la asimetra y curtosis, Cox-Ingersoll-Ross modelo, modelo de Vasicek,
etc.
De hecho, nuestro metodo esta basado en la existencia de una simetra
especca. Para todos los modelos de este tipo construimos una ley de conser-
vacion que llamamos Ley de conservacion de strike price que es intrnseca
a backward y forward problemas. La resolucion del problema inverso de
dichos modelos es posible si el sistema dinamico tiene la simetra denida por
un especial backward operador que se puede interpretar como el Operador
de strike price. Este operador nos ayuda a construir la expresion explcita
del backward propagador de Derman. Dicho tipo de leyes de conservacion
tienen el misma papel que las integrales de movimiento en mecanica.
Tambien comprobamos que el problema de la solucion de la ecuacion para-
bolica de Black Scholes, junto con la Ley de conservacion de strike price es
equivalente al problema de la solucion de la ecuacion de Black Scholes junto
con la conocida condicion estandar para el momento de maduracion. Este
hecho signica que el conjunto de esta ley de la conservacion y de la ecuacion
Black Scholes constituye una forma implcita de la soluci on clasica de Black
Scholes.
El concepto de leyes de conservacion fue originalmente desarrollado por
los fsicos, pero los economistas se han visto considerablemente interesados en
el. Argumentamos que la fenomenologica Sticky Strike Rule de Derman y
nuestra exacta Ley de conservacion de strike price tienen el misma origen.
Particularmente, establecemos el papel importante de la elasticidad en todos
estos modelos.
Ademas, nuestro metodo permitio descubrir la estructura compleja del Es-
pacio de riesgos cuyo concepto introducimos. Revelamos que varios modelos
actualmente usados en las nanzas cuantitativas incluyen mas de una volatil-
idad para un subyacente. Descubrimos la base teorica de la bien conocida
asimetra entre los casos de in-the-money y out-of-the-money.
37
Sukhomlin, Nicolay
Universidad Autonoma de Santo Domingo, Pontifcia Universidad Catolica
Madre y Maestra
Santo Domingo, REP
UBLICA DOMINICANA
w17971865@gmail.com
A sensitivity analysis of a class of variational inequalities
Let us consider a general variational inequality: (1I) nd u K Dom
such that
Au f, v u) + (v) (u) 0, v K.
Here K is a nonempty closed convex set of X, f is a xed element in the
topological dual X
.
The question of giving sucient conditions for the existence of a solution of
problem (1I) is a central problem in the study of variational inequalities. This
question has been at the origin of many contributions during the last years.
Several theoretical existence results for variational inequalities in general re-
exive Banach spaces and governed by a general operator A (not necessarily
linear) are well known when a coerciveness condition hold for the operator
A. We can cite for instance the contributions of J.L. Lions, Brezis, Browder.
etc. However, the variational formulation of many engineering problems leads
generally to non-coercive variational inequalities (e.g. problems in mechanics
which admits nontrivial virtual rigid body displacement). These problems are
formulated by semi-coercive variational inequalities and were studied rst by
Fichera and Lions & Stampacchia, Duvaut & Lions (for problems with fric-
tional type functionals). Recently many mathematicians and engineers has
focused their attention on non-coercive unilateral problems, using several dif-
ferent approaches such as the critical point theory, the Leray-Schauder degree
theory, the recession analysis or the regularization method by approximating
non-coercive problems by coercive ones (see e.g. Adly - Goeleven & Thera,
Ang - Schmidt & Vy, Baiocchi - Gastaldi & Tomarelli). The main concern
of these contributions is to obtain necessary or sucient conditions for the
solvability of such problems in a general setting by imposing some compactnes
conditions and some compatibility conditions on the term f. More recently,
Adly - Ernst & Thera has considered situations in which the existence of the
solution is stable with respect to small uniform perturbations of the data of
the problem. This result should be of great interest for problems in nance
and engineering where the data are known only with a certain precision and it
is desired that further renement of the data should not cause the emptiness
of the set of solutions.
In this presentation, we will review, old and new results on existence result
for the solvability of variational inequalities. We will also focus on a converse
of the famous Lions & Stampacchia Theorem of 1967.
38
Thera, Michel
Universite de Limoges
Limoges, FRANCE
michel.thera@unilim.fr
Line planning in Quito
Line planning is an important step in the phase of strategic planning in a
public transportation system. In the present work, we present an optimization
model aiming at the reduction of the global operational costs while ensuring
a certain level of quality of service, in terms of available transport capacity.
The computational complexity of this model is discussed for some transporta-
tion network topologies arising in the context of the Troleb us System, which
is by now the largest public transportation system in Quito, carrying around
250, 000 passengers daily. Moreover, we have considered how several other
factors aect the computational complexity of the model. If either xed cost
is allowed or the transportation modes are greater than two, then a reduction
from the Minimization Knapsack Problem and 3-Dimensional Matching Prob-
lem can be used to show that our Line Planning Model is NP-hard. Finally,
results of computational tests based on real data are reported.
Torres, Ramiro
Escuela Politecnica Nacional
Quito, ECUADOR
rtorres@math.epn.edu.ec
Set packing and set covering problems appearing in biological net-
works
Given a ground set X, a family T of subsets from X, and a positive weight
function c : T R
+
, the set packing problem asks for a subfamily T
1
T of
mutually non intersecting sets such that the sum of their weights is maximized.
Conversely, the set covering problem asks for a mimimum weight subfamily
T
2
T with the property that every element x X is contained at least in
one set of T
2
. Considering the hypergraph H where the nodes are the elements
from X and the hyperedges the sets in T, these problems can be formulated as
nding either a maximum weight matching (set packing) or a minimum weight
hyperedge cover (set covering) in H.
Set packing and set covering are fundamental problems in combinatorial
optimization which are related to applications in many areas such as vehi-
cle routing, line planning in public transportation, frequency assignment in
telecommunications, bandwidth packing, scheduling of airline crews, to cite a
few.
The analysis of metabolite networks has received increasing attention in
the recent past. A typical approach for metabolic pathway analysis consists in
39
considering the solutions to the system:
Sv = 0, v
i
0, i I,
where S R
mn
is the stoichiometric matrix of the network, which repre-
sents a mapping of n reaction rate vectors on m metabolites into a space of
concentration time derivatives, and I R := 1, . . . , n is the set of thermo-
dynamically irreversible reactions. This system is obtained from the metabolic
balance equation at the so-called quasi-steady state. A solution v R
n
to this
system is called a ux vector. Moreover, the set of all feasible solutions denes
a polyhedral cone C, which is called the ux cone.
In this talk I will explore, from a polyhedral viewpoint, the structure of
certain set packing and set covering problems that arise in the context of bi-
ological networks; for instance, to create non-conicting activation rules for
reactions involved in a regulatory structure, or to determine minimal inter-
vention sets. This is ongoin research that I am currently starting as a part
of a postdoctoral stay at the Otto-von-Guericke Universitat in Magdeburg,
Germany.
Torres, Luis Miguel
Universitat Magdeburg and Escuela Politecnica Nacional
Magdeburg, GERMANY
ltorres@math.epn.edu.ec
40
Information and tips
Talks. The LAWOC opening ceremony and monday presentations will
take place at the Paraninfo of the Universidad Andina Simon Bolvar. The rest
of the talks will take place in Aula 21 at Universidad Andina Simon Bolvar.
Coee breaks. Coe will be served in the Sala social.
Power Supply. Please note that the voltage in Ecuador is 110 V and
American Plugs are needed.
Useful Phone Numbers.
Organization 098122984
Emergency 911 Police 101
Firemen 102 Airport 2 440 080
International Phone Calls and Internet. Computers with internet ac-
cess are available at the computer laboratory of the Universidad Andina Simon
Bolvar. A limited amount of wireless internet accounts will also be available.
Wireless internet access and computers are also available at the 6th oor of
the Administration building at EPN Quito.
International calls can be made either from ANDINATEL (see item 29 in
the map) or from internet cafes. Some internet cafes are listed in the map
reference.
Transportation in Quito. Quito has two main bus systems: Trole and
Ecova. Both run on sole set lanes across the city north to south, with extra
bus lines running to suburbs from the northern and southern stations. The
Trole has 2 routes running mainly along Av. 10 de Agosto, and Ecova cross
along Av. 6 de Diciembre. The fare for both systems is US $0.25.
Taxis are a good transportation alternative. You can get to almost any
point in town for about $3. Legal taxis (yellow cars) have to be certied by
local government. That certication placard is prominently displayed on both
sides of and in the upper right windscreen of both taxis and buses. It is not
recommended to use taxis lacking this local government seal of approval. They
will all pick you up from wherever youre calling. Here you will nd a list of
Radio Taxi companies to use:
Taxi Amigo Tel. 2222222 / 2333333 / 2222220
City Taxi Tel. 2633333 / 2630220
Central Radio Taxis Tel. 2406806 / 2811111 / 2267900
RTR Tel. 2533878 / 2535600
JJ Taxi Tel. 2639639 / 2639999 / 2628400.
41
Cash dispensers. Most of the restaurants and services accept credit cards.
However, it is recommended to have extra cash. You can get it at places nearby
to the workshop building (see the map references).
Money exchange. If you need to exchange money, there are some places
near the LAWOC.
Vazcorp (outside the map). Travellers checks in other currencies.
Address: Amazonas and Roca (Hotel Alameda Real). Tel. 2225442.
Multicambios (Item 28 in the reference map).
Address: Colon 919 and Reina Victoria, Tel. 2561747; Roca 720, Tel.
2567344 and at the airport.
Food and drinks. A practical choice is to take a taxi and go to Plaza
Foch. There you can nd a lot of restaurants and bars. You can nd several
possible options at dierent prices.
Restaurants. Specic restaurant choices are given in the following list.
El Esmeraldas. Ecuadorian seafood.
Address: Isabel La Catolica y Luis Cordero.
Naranjilla Mecanica. Fusion food.
Address: Tamayo y General Veintimilla
Al forno. Italian Pizza.
Address: Baquerizo Moreono E7-86 y Almagro.
Pims Grilled food.
Address: Isabel La Catolica 915 y Luis Cordero.
Tibidabo. Catalunian food.
Address: Francisco Salazar 934 e Isabela Catolica.
Antojera Chilena. Chilenian food.
Address: Isabel La Catolica 1129 y Coru na.
Time Out. Grilled food and sports bar.
Address: Isabel La Carolica y Luis Cordero. Swisshohel.
El Pobre Diablo. International food and Bar.
Address: Isabel La Catolica E12-06 y Galavis esq. La Floresta.
Swisshotel. International food.
Address: Isabel La Catolica E12-06 y Galavis esq.
La Briciola. Italian food.
Address: Toledo 1255 y Luis Cordero
La Vi na. International food.
Address: Isabel La Catolica y Luis Cordero.
La Choza. Ecuadorian food.
Address: 12 de Octubre N24-551 y Cordero
Happy Panda. Chinese food.
Address: Isabel La Catolica N94-464
Quito Deli. Cafeteria.
Address: Isabel La Catolica 915 y Luis Cordero.
Mediterraneum. Seafood.
Address:Andaluca 376 y Cordero
Enten ca. Mediterranean food.
Address: Coru na e Isabela Catolica.
42
Tratoria. Italian food.
Whimper N21-29
Barlovento. Seafood and Ecuadorian food.
Address: 12 de Octubre 2511 y Orellana.
Hunters. Grilled food.
Address: 12 de Octubre 2517
Hotel Quito. International food.
Address: Gonzales Suarez N27-142
Ceviches de la Rumi nahui. Ecuadorian seafood.
Address: 12 de Octubre y Colon.
Gran Casa China. Chinese food.
Address: Cordero E9-242 y Tamayo.
Corrientes 348 Argentinean food.
Address: Edmundo Carvajal OE4-226 y el Condor.
Mi Cocina Ecuadorian food.
Address: Centro Comercial Megamaxi, 6 de Diciembre y Calle Aleman.
106-110
Amarilo Grilled food and Bar.
Address: Juan Gonzales N35-113.
Rincon Italiano Italian food.
Address: Av. Los Shyris41-61 e Isla Floreana.
Rincon la Ronda Ecuadorian and International food.
Address: Bello Horizonte E48-45 y Av. Almagro.
43
44
Touristic information
Quito is the capital of Ecuador and a recommended touristic city. With
attractions ranging from parks to historical centers and museums, there are
many possibilities for sightseeing and visits. Outside the city, there are green
mountains and snow-capped volcanoes to enjoy.
Parque Itchimba. The Itchimba Park is located at the summit and
slope of the Itchimbia hill, at the oriental limit of Quitos Historic Center at
2 900 meters. It is considered as a natural lookout point because of the great
view that you get of Quitos Historic Center in either way. The total area
of the park is 54 hectares; in it you can nd around 400 varieties of owers
and 40 species of birds. One of its main attractions is the Itchimba Cultural
Center or Crystal House, which was rebuilt from the old metallic structure of
the Santa Clara Market, built in Hamburg, Germany, in 1889 and brought to
Ecuador in pieces.
Hours Mon-Sun 10 a.m. to 6 p.m.
Location El Itchimba, east of Old Town
Prices Musseum tiket $1 pp.
Teleferiqo. One of the citys most popular attractions is El Teleferiqo,
where six-person cable cars transport you up to the side of Volcan Pichincha
to 4,050m (13,280 ft.). At the top, you will have a magnicent view of the city
and surrounding mountain peaks and of the Pichincha volcano. The air is thin
up there. The ambitious and very modern complex includes an oxygen bar to
replenish the weary traveler. Youll also nd souvenir stands and shops, and
a couple of restaurants.
Hours 10 a.m. to 6 p.m.
Location El Teleferiqo. Calle Arnulfo Araujo y Av. Occidental
Prices Cable car ride to the top $4 pp.
El Panecillo. From a distance, the hill that hosts a huge statue of the
winged virgin does indeed look like a panecillo (small bread roll). Since its
directly south of the city, this hill was an ideal place to construct the 45m-
high (148-ft.) statue. The signicance of the Panecillo Hill dates back to Inca
times, when it was known as Shungoloma (Hill of the Heart) and used as a
place to worship the sun. Later the virgins statue was constructed in the
Colonial time. These days, most people come up here for the 360
views of
Quito. Try to get here in the morning (around 10am), before the clouds settle
in around the nearby mountains. On a clear day, you can see Cotopaxi in the
distance. This is a relatively quick ride from Old Town, and a taxi should only
cost about $3.
45
Hours Mon-Fri 9am-6pm; Sat-Sun 9am-5pm.
Location El Panecillo, south of Old Town.
Prices Admission to climb to the top of the monument $1 pp.
Getting to the Equator 0
o
. The Mitad del Mundo is around 15 kilo-
meters from Quito. There is a small town, that owes its name from the fact
that its fount in the latitude 0