|
|
|
|
|
|
|
|
Abstracts
A. Arico': Numerical solution of the nonlinear Schrodinger equation
Linear algebra techniques can be used to efficiently solve the
nonlinear Schrodinger equation
i qt = qxx + 2 q |q|2, where
q=q(x,t)
following the approach of the inverse scattering theory. We
present an overview of the procedure and discuss some numerical
experiments.
Z.J. Bai: Nonnegative Inverse Eigenvalue Problems with Partial Eigendata
(joint work with S. Serra Capizzano and Z. Zhao)
We consider the inverse
problem of constructing an n-by-n real
nonnegative matrix A from prescribed partial
eigendata. We first give the solvability
conditions for the inverse problem without
the nonnegative constraint and provide a
solution which is closest to a pre-estimated
analytical nonnegative matrix. To find a
nonnegative solution, we reformulate the
inverse problem as a monotone
complementarity problem and then propose a
nonsmooth Newton-type method for solving the
nonsmooth equation related to the monotone
complementarity problem. Under some very
mild assumptions, we show that our method
has simultaneously a global and quadratic
convergence. We also specialize our method
to the symmetric nonnegative inverse
problem, and to the cases of a prescribed
lower bound and of prescribed entries.
Numerical tests demonstrate the efficiency
of the proposed method and support our
theoretical findings.
D. Bertaccini: Adaptive updating preconditioners for image restoration with PDE models
We consider a semi implicit time step integration of a couple of nonlinear generalized
Alvarez–Lions–Morel-like partial differential equation models where the discretized equations
were solved by Krylov iterative solvers accelerated by an adaptive preconditioning paradigm
based on the update of incomplete factorizations which presents a global computational
cost slightly more than linear in the number of the image pixels, robust, able to manage
spatially variant blurring operators and competitive with multigrid techniques but with a
lower storage requirement and a greater flexibility. We demonstrated the efficiency of the
strategy by denoising and deblurring some images.
P. Brianzi: Ricostruzione di un segnale con formule di sovracampionamento
(in collaborazione con V. Del Prete)
Si affronta il problema della ricostruzione di un segnale a banda
limitata quando un numero finito di suoi campioni e delle sue
derivate sono persi.
Si studiano le matrici che intervengono nei sistemi lineari di
soluzione nella procedura di recupero e si affronta il particolare
il caso mal-condizionato in cui i campioni mancanti sono consecutivi
regolarizzando con il metodo di Tikhonov.
D. Fasino:
Ricostruzione di conduttività nella magneto-quasistatica con metodi level-set
Si presenta un metodo numerico per la risoluzione di un problema inverso nella
magneto-quasistatica (una approssimazione della magnetodinamica per basse frequenze).
Il problema consiste nella identificazione del coefficiente di conduttività
in un dominio tridimensionale omogeneo in cui possono essere presenti piccole
inclusioni, a partire da misure di distorsione di un campo magnetico esterno.
Computazionalmente, il problema richiede l'inversione regolarizzata di una
forward map fortemente non lineare; la soluzione cercata è una funzione costante
a tratti con un supporto piuttosto limitato, che può assumere solo un numero
molto piccolo di valori possibili, noti a priori. La regolarizzazione introdotta
può essere vista come un vincolo di sparsità sulla
discretizzazione del gradiente di una funzione parametrizzata mediante la
tecnica dei metodi level-set.
C. Estatico: Sviluppi recenti nella teoria della regolarizzazione,
con un’applicazione in imaging
(in collaborazione con P. Brianzi, F. Di Benedetto, M. Pastorino, A. Randazzo e G. Rodriguez)
Formally, any inverse problem can be modeled by an operator equation
A(x) = y,
where
A : X
® Y is the so-called “forward” operator between two functional
spaces
X and
Y, and
xÎX is the “cause” of some “effects”
yÎY . By means of
the knowledge of (some approximation of) the effects
y, the aim is to recover the
unknown cause
x. Many well known solving schemes approximate the solution x
by means of the minimization of a cost functional like
F(x) = ||A(x) − y||Y +
l||x||X
where
X
and
Y
are both Hilbert spaces. Iterative regularization algorithms for
the minimization of
F
give rise in general to over-smoothed solutions and the
discontinuities present in real solutions
x are not well restored.
More recently, it has been investigated the behavior of iterative methods based
on more general Banach spaces
X and
Y . The new “geometry” of the Banach
spaces can substantially reduce the over-smoothing of the iterative restoration
process.
In this talk, we discuss the regularization approach in Banach spaces. A
non-linear iterative method is applied to a non-linear inverse scattering problem
where the dielectric distributions
x
of a 2D domain must be recovered by means of
its scattered microwave field
y
outside the domain. We will show how the new
computational results well outperform
classical “Hilbertian regularization”.
M. Donatelli:
Matrici di regolarizzazione quadrate per problemi mal-posti
discreti di grandi dimensioni
(in collaborazione con L. Reichel)
Si considera la soluzione di problemi mal-posti discreti lineari di grandi dimensioni
mediante regolarizzazione con il metodo di Tikhonov.
Ricorrendo a metodi iterativi basati sul procedimento di Arnoldi si ha la necessita' di utilizzare
matrici di regolarizzazione quadrate. Tali matrici solitamente sono approssimazioni
alle differenze finite di operatori differenziali. In questo seminario si discutera'
come costruire matrici di regolarizzazione
quadrate per preservare particolari componenti della soluzione con un costo computazionale contenuto.
Una tecnica generale ed efficace per ottenere tali matrici e' l'imposizione di opportune condizioni al contorno.
F. Ferri:
Importance of 3D deconvolution in Confocal Microscopy of biopolymer gels
(in collaborazione con D. Magatti e M. Molteni)
Biopolymer gels are three‑dimensional (3D) networks
of fibers grown from the polymerization of biological macromolecules, usually proteins,
that polymerize under the action of some specific enzyme. The bio-functional
properties of the network is directly linked to its mechanical and physical structure,
which, in turn, depends on the physical parameters characterizing the gel, such
as fibers diameter and stiffness, distribution of fiber length and branching/cross-linking
points, mesh size, fractal morphology.
Among non invasive techniques
suitable for studying these systems, Confocal Microscopy (CM) is one of the
most prominent, having the distinctive advantages of allowing the direct
examination of the samples in real space and as a function of time. The quantitative
analysis of the 3D confocal images, however, requires a non-trivial 3D
deconvolution of the raw data, which might be quite critical because the Point
Spread Function (PSF) of the microscope is often larger than the fibers
diameter, it can be non isotropic and non uniform, and its experimental
determination is not straightforward as well. Assessing and testing the performances
of the deconvolution method used in the data analysis is, therefore, of crucial
importance.
In this talk we present a novel method for the generation of in
silico 3D-filamentous networks that exhibit the same structural and
morphological properties of fibrin gels (the gels involved in the process of
blood coagulation) and can be easily exploited for validating the deconvolution
method. The network is made of cylindrical straight fibers of known diameter
joined together at some randomly distributed nodal points. The resulting 3D
network exhibits a fractal morphology similar to that of the fibrin gel (mass
fractal dimension ~1.2-1.5) and, after convolution with the point spread
function of our confocal set-up, appears strikingly similar to the confocal
images taken on fibrin gels. Furthermore, based on these in silico gels, we have developed a simple statistical method for
the determination of some of the gel physical parameters mentioned above (fibers
diameter, gel mesh size, fractal dimension). This method of analysis, based on
the 3D correlation function of entire confocal stack, is rather simple, fast
and reliable, but if applied to the same in
silico gels convolved with the PSF of the microscope, produces highly inaccurate
results. Thus, in the case of real data, it is of crucial importance to
deconvolve them before any quantitative analysis.
N. Mastronardi: An algorithm for computing a new factorization of indefinite
symmetric matrices
(in collaborazione con P. Van Dooren)
Let
AÎRnxn
be a symmetric indefinite matrix and let
{I1,I2,I3}
be its inertia, i.e.,
I1,
I2
and
I3
are the number of eigenvalues of
A
greater, equal or less than zero. Without loss of generality, we assume
I1
³ I2+I3.
Then, the following factorization of
A
exists,
A = QLQT
where
Q is an orthogonal matrix and
L is an lower anti--triangular one,
i.e., a matrix with zero entries above the anti--diagonal, with a bulge
matrix of size
I1-I2-I3
in the main diagonal.
In this talk a stable and efficient algorithm for computing the latter
factorization is described.
Moreover, the computation of the inertia with the proposed algorithm is
straightforward.
Comparisons with algorithms available in the literature are shown.
B. Morini: Risoluzione di sistemi lineari in problemi ai minimi quadrati
vincolati ed applicazioni nella ricostruzione di immagini
Si considerano procedimenti iterativi per la risoluzione di
problemi ai minimi quadrati lineari, vincolati e di grandi
dimensioni. L'efficienza dei metodi di ottimizzazione considerati
dipende fortemente dalla formulazione dei sistemi lineari algebrici
che nascono ad ogni iterazione e dalla loro risoluzione iterativa.
Presentiamo due strategie per la risoluzione della fase di algebra
lineare ed risultati numerici ottenuti su problemi di ricostruzione
di immagini.
P. Novati: Preconditioning linear systems via matrix function evaluation
(in collaborazione con M. Redivo-Zaglia, M.
R. Russo)
For the solution of discrete ill-posed
problems, a novel preconditioned iterative
method based on the Arnoldi algorithm for
matrix functions is presented. The method is
also extended to work in connection with
Tikhonov regularization. Numerical
experiments arising from the solution of
integral equations and image restoration are
presented.
M. Piana: A linear inverse problem in X-ray solar astronomy
(in collaborazione con S. Allavena)
The scientific aim of the NASA mission Reuven Ramaty High Energy Solar
Spectroscopic Imager (RHESSI) is to collect X-ray observations with high
spatial and spectral resolution. The purest data provided by RHESSI are
Fourier components of the X-ray radiation sampled at discrete points of
the spatial frequency plane, named visibilities. This talk will describe
an interpolation/extrapolation method for the reconstruction of X-ray
images from RHESSI visibilities.
M. Semplice: Simulazioni numeriche con PDE paraboliche degeneri per la
conservazione ed il restauro monumentale
(in collaborazione con M. Donatelli, S. Serra-Capizzano)
Dopo un breve excursus su recenti modelli matematici di interesse nel
campo della conservazione dei monumenti e dei manufatti marmorei,
consideriamo la discretizzazione di equazioni paraboliche non lineari (e
degeneri) con la tecnica delle differenze finite e la loro applicazione
nelle simulazioni.
In particolare si descrivono la struttura dei sistemi non lineari da
risolvere ad ogni passo temporale e la scelta di opportuni
precondizionatori per il metodo di Newton-Krylov impiegato. Si
illustrano sia i risultati di convergenza ed ottimalità recentemente
ottenuti, che i limiti della loro applicabilità, illustrando i possibili
sviluppi futuri di questa ricerca.
S. Serra Capizzano: Canonical eigenvalue distribution of multilevel block
Toeplitz sequences with non-Hermitian symbols
(in collaborazione con M. Donatelli, N. Maya)
Let f be a
bounded symbol from the linear space of the
complex
s x s
matrices to
Ik=(-p,p)k
. We consider the sequence of multilevel
block Toeplitz matrices
{Tn(f)},
n=(n1, ..., nk).
When s=1, thanks to the work of Tilli,
if R(f) has empty interior and does
not disconnect the complex plane, then
{Tn(f)}
~l
(f, Ik). Here we generalize
the latter result for the case where the
role of R(f) is played by
Uj=1...s
R(lj(f)),
lj(f)
being the eigenvalues of the matrix-valued
symbol f. The result is extended to
the algebra generated by Toeplitz sequences
with bounded symbols.
D. Sesana: Successioni di matrici g-circolanti e g-Toeplitz: analisi spettrale,
precondizionatori regolarizzanti e applicazioni
(in collaborazione con C. Estatico, E. Ngondiep e S. Serra-Capizzano)
For a given nonnegative integer
g, a matrix
An
of size
n
is
called
g-Toeplitz if
its entries obey the rule
An=[ar-gs]r,s=0,...,n-1.
Analogously, a matrix
An
again of size
n
is called
g-circulant
if
An=[a(r-gs) mod n]r,s=0,...,n-1. In this talk we describe the asymptotic properties, in term of
spectral distribution, of both
g-circulant and
g-Toeplitz sequences in the case where
ak
can be
interpreted as the sequence of Fourier coefficients of an integrable
function
f
over the domain
(-p,p). Moreover, we consider the preconditioning problem which
is well understood and widely studied in the last three decades for
g=1, and we show that,
while a standard preconditioning cannot be achieved for
g³2, the result has a positive implication since there exist choices of
g-circulant sequences which can be used as basic regularizing preconditioning
sequences for the corresponding
g-Toeplitz structures. Few numerical experiments are presented
and discussed.
|
|
|
|
|
|
|
|
|