A model for communication in a multi-segment market
Igor BYKADOROV*, Andrea ELLERO** and Elena MORETTI**
*Sobolev Institute of Mathematics, Siberian Branch of the Russian Academy of Sciences **Department of Management, Università Ca'Foscari Venezia
Abstract. We consider a linear optimal control model for the marketing of seasonal products which are produced by the same firm and sold by retailers in different market segments The horizon is divided in two consecutive non-intersecting intervals, called production and selling periods, respectively. The production period state variables are the inventory levels and two kinds of goodwills (consumers' and retailers' goodwill, respectively) while the selling period state variables are the sales levels and the two kinds of goodwills. In the production interval there are three kinds of controls: on production, quality and advertising, while in the selling one the controls are on communication via advertising, promotion ad-dressed to consumers and incentives given to retailers. We consider the case of several kinds of commu-nications. The optimal control problem is transformed into an equivalent nonlinear programming prob-lem.
Keywords. Optimal control, advertising, seasonal products, segmentation. .
1. Introduction
In this paper we consider a linear optimal control model for the marketing of seasonal products which are produced by the same firm and sold by retailers in differ-ent market segmdiffer-ents. More precisely, the firm produces and sells a seasonal product with different attributes in two consecutive time intervals, the first one devoted to production and the second one to the sale of the product. In the production period the firm can control production, product quality and advertise the product both to retailers and to the different market segments. In the selling period the firm has a wider communication activity via promo-tion to consumers, incentives to retailers and advertising to both. We assume that the communication efforts of the firm act on the goodwill that consumers and retailers as-sign to the product. The motion equations of the model refer explicitly to the consumers' and retailers' goodwills as state variables. The firm seeks the maximum profit but has also to keep a minimum level of goodwills at the end of the selling period. This way we generalize the linear models proposed in [1]-[3] (see also [4]-[6]).
2. A multi-segment linear model
We consider a model withn
market segments,r
retailers and
d
kinds of communications. As in [2], we assume that production and sales take place in two dis-tinct intervals[
t
0,
t
1]
(production period) and[
t
1,
t
2]
selling period). We will use the index sets
}.
,
1
{
},
,
1
{
},
,
,
1
{
n
J
r
K
d
I
2.1. State variables and controls
Through the production period we consider
n
state variables
)
(t
m
i inventory level in thei
-th segment,i
I
, at timet
[
t
0,
t
1]
,and
n
1
controls
(t)
u
i production expenditure rate in thei
-th seg-ment,i
I
, at timet
[
t
0,
t
1]
,
q
expenditure rate for improving quality; as in the one-segment linear model [2],q
will be consid-ered constant and nonnegative.Through the selling period the state variables are
(t)
x
i sales level in thei
-th segment,i
I
, at time]
,
[
t
1t
2t
.Moreover, through both production and selling periods we consider the state variables
(t)
C
i consumer goodwill in thei
-th segment,i
I
, at timet
[
t
0,
t
2]
,
(t)
R
j goodwill of the j-th retailer,j
J
, at time]
,
[
t
0t
2t
, andd
controls
)
(t
a
k expenditure rate in thek
-th type ofcommu-nication,
k
K
, at timet
[
t
0,
t
2]
.The consumer goodwills
C
i(t)
explain how easily the consumers in thei
-th segment choose to buy the product, while the goodwillsR
j(t)
take into account howfa-vorably the
j
-th retailer keeps the product in stock (see [2]). For what concerns controlsa
k(t
)
, we consider that during the production period the communication co-4, Acad. Koptyug avenue, Novosibirsk, 630090, Russia,Phone +7 383-363-46-06, Fax: +7 383-333-25-98, e-mail: bykad@math.nsc.ru
incides with advertising, while in the selling period ad-vertising, promotion and incentives are all available.
2.2. Dynamics
The dynamics of the model is rather different in the pro-duction period
[
t
0,
t
1]
and in the selling period[
t
1,
t
2]
. The dynamics in the production period[
t
0,
t
1]
is de-scribed by2n
r
motion equations:n
equations for the inventory levelsm
i(t
)
,n
equations describing thegoodwills
C
i(t)
of the consumers' segments andr
equations for the goodwills
R
j(t)
of the retailers.For the
i
-th segment,i
I
, the inventory level dynamics is the following),
(
)
(
t
u
t
m
i
i i where
i
marginal productivity of production expenditure,1
i
.Further, in the
i
-th segment,i
I
, the goodwill level dynamics is given by the motion equations
K k k p k C i C it
C
t
a
t
C
i i(
)
(
),
)
(
( )
where
i C
decay rate of goodwillC
i,
0
i C
,
) ( p k Ci
marginal productivity of expenditure incommu-nication
a
k, via advertising, in terms of goodwilli
C
in the production period, C(pk)
0
i
.
Remark that C(pk)
0
i
means that, in the productionpe-riod, communication expenditure
a
k for advertisingin-creases the goodwill
C
i.Finally, for the
j
-th retailer,j
J
, the goodwill level dynamics is given by the motion equations),
(
)
(
)
(
t
R
t
( )a
t
R
k K k p k R j R j j
j
where
j R
decay rate of goodwillR
j,
0
j R
,
) ( p k Rj
marginal productivity of expenditure incommu-nication
a
kvia advertising, in terms of goodwillR
j in the production period, R(pk)
0
j
.The dynamics in the selling period
[
t
1,
t
2]
is also de-scribed by2n
r
motion equations:n
equations for sales levelsx
i(t)
ini
-th segment and, as in the produc-tion period,n
equations for goodwillC
i ini
-thseg-ment and
r
equations for goodwillR
jofj
-th retailer.For the
i
-th segment,i
I
, the inventory level dynam-ics is the following
J j j R x i C i i it
x
t
C
t
R
t
x
j i i(
)
(
)
)
(
)
(
,
)
(
1 ) (q
l
t
a
i i x K k k s k x
where
i
saturation aversion parameter,
i
0
,
i
C
goodwill productivity in terms of sales originated by goodwillC
i,
0
i C
,
j iR x
goodwill productivity in terms of sales ini
-th segment originated by thej
-th retailer goodwillj
R
,
0
j iR x
,
) (s k xi
marginal productivity of expenditure incommu-nication
a
k via promotion in terms of sales levelx
i during the selling period, (xsk)
0
i
,
i
x
l
marginal sale productivity of the expenditure rateq
for improving quality,
0
i x
l
,
1 0 1q
t
t
q
the total expenditure for improvingquality in the production period.
Further, for
i
-th segment,i
I
, the goodwill level dynamics is given by the motion equation
K k C k s k C i C i i it
x
t
C
t
a
t
l
q
C
i i i(
)
(
)
,
)
(
)
(
( ) 1
where
i
word-of-mouth in terms of goodwillC
i,
i
0
,
i
C
decay rate of goodwillC
i,
Ci
0
,
) (s
k Ci
marginal productivity of expenditure incommu-nication
a
k via advertising in terms of goodwill iC
in the sales period, C(s)k
0
i
,
i
C
l
marginal productivity of quality expenditure rateq
for goodwillC
i,
0
i
C
l
.
Finally, for the
j
-th retailer,j
J
, the goodwill level dynamics is given by the motion equation
K k R k s k R j R jt
R
t
a
t
l
q
R
j j j(
)
(
)
,
)
(
( ) 1
where
j R
decay rate of goodwillR
j,
Rj
0
,
) ( s
k Rj
marginal productivity of expenditure incommu-nication
a ,
k via advertising and incentives, in terms of goodwillR
j in the sales period,0
) (s
k Rj
j
R
l
marginal productivity ofq
for goodwillR
j,
0
j Rl
.
2.3. Objective functionThe total profit of the firm is determined as the difference between the total income and the total expenditures (for production, product quality and communication) and is given by the following objective function:
dt
q
t
u
t
m
c
t
x
p
t t i I i I i i i I i i i 1 0)
(
)
(
)
(
2
2 0,
)
(
t t k K kt
dt
a
where
ip
product sale price ini
-th segment,p
i
0
,
ic
inventory marginal cost per time unit ini
-th seg-ment,c
i
0
.2.4. Boundary conditions
The natural boundary conditions for the inventory and sales levels are
.
),
(
)
(
,
0
)
(
)
(
t
0x
t
1x
t
2m
t
1i
I
m
i
i
i
i
The boundary conditions for the consumer goodwills are
,
,
)
(
,
)
(
t
0C
0C
t
2C
2i
I
C
i
i i
i
where
0 iC
initial value of goodwillC ,
i
2 i
C
lower bound of goodwillC
i at final timet
2.The boundary conditions for the retailers' goodwills are
,
,
)
(
,
)
(
t
0R
0R
t
2R
2j
J
R
j
j j
j
where
0 jR
initial value of goodwillR
j,
2 j
R
lower bound of goodwillR
jat final timet
2.The boundary conditions for the control variables are
],
,
[
,
],
,
0
[
)
(
t
u
i
I
t
t
0t
1u
i
i
],
,
[
,
],
,
0
[
)
(
t
a
k
K
t
t
0t
2a
k
k
].
,
0
[ q
q
where
iu
upper bound foru
i(t
)
,
u
i
0
,
ka
upper bound fora
k(t
)
,
a
k
0
,
q
upper bound forq
,q
0
.
2.5. Formulation of the model
In order to formulate the linear model we consider
n
-dimensional state variables vectorsm
(
t
),
x
(
t
),
C
(
t
)
with elements
m
i(
t
),
x
i(
t
),
C
i(
t
),
i
I
, respectively;
r
-dimensional state variables vectorR
(t
)
with elementsR
j(
t
),
j
J
;n
-dimensional control vector)
(t
u
with elementsu
i(
t
),
i
I
;
r
-dimensional control vectora
(t
)
with elementsK
k
t
a
k(
),
;
n
-dimensional constant vectorsp
,
c
,
C
0,
l
x,
l
C,
u
C ,
2 with elementsp
ic
iC
il
xl
CC
iu
ii
I
i i,
,
,
,
,
,
,
0 2 , respectively;
r
-dimensional constant vectorsR
0,
l
R,
R
2 with elementsR
j,
l
Rj,
R
j,
j
J
2 0
, respectively;
d
-dimensional constant vectora
with elementsK
k
a
k,
;diagonal constant matrices
,
C,
,
C,
of ordern
with diagonal elements
i,
Ci,
,
Ci,
i,
I
i
, respectively;diagonal constant matrix
R of orderr
withdiagonal elements
j
J
j
R
,
;constant
n
d
matrices
C(p),
x(s),
C(s) with elements Cpk xsk Cski
I
k
K
i i i,
,
,
,
) ( ) ( ) (
, respectively;constant
r
d
matrices
R(p),
R(s) with elementsK
k
J
j
s k R p k Rj,
j,
,
) ( ) (
, respectively;constant
n
r
matrix
R with elementsJ
j
I
i
j iR x,
,
.Moreover, let us define
],
,
[
,
)
(
)
(
)
(
],
,
[
,
)
(
)
(
)
(
0 2t
t
1t
2t
A
t
x
t
X
t
t
t
t
R
t
C
t
A
,
,
2 2 2 0 0 0
R
C
A
R
C
A
, , 0 0 ) ( ) ( ) ( p R p C p R rn nr C E
,
,
,
0
0
0
) ( ) ( ) ( ) (
R C x s R s C s x s R rn rn nr C R Cl
l
l
L
E
Q
where, as usual,
0
nr and0
rn are
n
r
and,respec-tively,
r
n
zero matrices.Moreover, let
e
n ande
d ben
-dimensional and,respectively,
d
-dimensional unit vectors while0
n and dThis way the linear marketing model with
n
seg-ments,r
retailers andd
kinds of communications isProblem
P
: maximize
1 0 2 0,
)
(
)
(
)
(
)
(
2 t t t t T d T n T Tdt
t
a
e
dt
q
t
u
e
t
m
c
t
x
p
subject to],
[
),
(
)
(
)
(
),
(
)
(
t
u
t
A
t
A
t
E
( )a
t
t
t
0,t
1m
p
,
)
(
,
0
)
(
t
0A
t
0A
0m
n
],
,
[
,
)
(
)
(
)
(
1 1 2 ) (t
t
t
L
q
t
a
E
t
QX
t
X
s
,
)
(
),
(
)
(
,
0
)
(
2 2 1 2 1x
t
m
t
A
t
A
t
x
n
].
,
0
[
,
)
(
0
,
)
(
0
n
u
t
u
d
a
t
a
q
q
Remark that it is possible to replace the vector inequality
)
(
)
(
t
2m
t
1x
with an equality. In fact, if bycontradic-tion in an optimal solucontradic-tion of
P
x
(
t
2)
m
(
t
1)
, it would be possible to decrease production and, at the same time, increase the value of the objective function, due to the special form of the motion equations.The model proposed is rather general and allows different specifications. One of them is the following. Assume the aim of the firm is to determine the optimal expenditure rate for each segment and each retailer. Then
r
n
d
. Moreover, let each kind of communicationact only on a single retailer or on a single segment of the market. It means that for every
i
I
andk
K
\ i
{
}
one has C(pk)
x(sk)
C(sk)
0
i i i
and that for everyJ
j
andk
K
\
{
n
j
}
one has R(pk)
R(s)k
0
j j
.Another possible interpretation of the model will be given in Conclusions.
3. Decomposition into parametric
subproblems
To discuss an approach for solving problem
P
, let us define
n
-dimensional vectorm
~
, whose elementsm
~
i,
I
i
, are parameters;
(
n
r
)
-dimensional vectorsA,
~
A
, whose elementsJ
j
A
A
I
i
A
A
i i
nj,
nj,
~
,
,
,
~
, respectively, are pa-rameters too .
As in [3] problem
P
can be solved by first solving the optimal control problems, depending on the parame-tersm
~
m
(
t
1),
A
~
A
(
t
1),
A
A
2 and then solving a nonlinear programming problem in which the parameters are the decision variables. Denote.
~
,
~
0
~
A
m
X
A
X
nThe parametric subproblems
P
1
P
(
m
~
1)
,P
2
P
2(
A
~
)
and
P
3
P
3(
m
,
q
,
A
~
,
A
)
can be written this way ProblemP
1: maximize
1 0,
)
(
)
(
t t T n Tdt
t
u
e
t
m
c
subject to),
(
)
(
t
u
t
m
.
)
(
0
,
~
)
(
,
0
)
(
t
0m
t
1m
u
t
u
m
n
n
ProblemP
2: maximize
1 0,
)
(
t t T da
t
dt
e
subject to),
(
)
(
)
(
t
A
t
E
( )a
t
A
p.
)
(
0
,
~
)
(
,
)
(
t
0A
0A
t
1A
a
t
a
A
d
ProblemP
3: maximize
2 1,
)
(
t t T da
t
dt
e
subject to,
)
(
)
(
)
(
1 ) (L
q
t
a
E
t
QX
t
X
s
.
)
(
0
,
)
(
,
~
)
(
t
1X
X
t
2X
a
t
a
X
d
Moreover, problem
P
1 , due to the special form of its motion equations, can be replaced byn
simpler prob-lemsP
i
P
(i)(
m
~
i),
i
I
1 ) ( 1 , this way Problem 1() iP
: maximize
1 0,
)
(
)
(
t t i i im
t
u
t
dt
c
subject to),
(
)
(
t
u
t
m
i
i i].
,
0
[
)
(
,
~
)
(
,
0
)
(
0 i 1 i i i it
m
t
m
u
t
u
m
LetF
~
1(i)
F
~
1(i)(
~
i),
i
I
,
F
~
2
F
~
2(
A
~
)
andF
~
3
)
,
~
,
,
(
~
3m
q
A
A
F
be the optimal values of problems,
,
) (1
i
I
P
i
P
2andP
3 respectively. Then problemP
is equivalent to a nonlinear programming problem where the following objective function
p
m
~
t
1t
0
q
F
~
1()F
~
2F
~
3 I i i T
(1) must be maximized. In the following we will assume that the general position condition (GPC) ([7], p.166) holds.The optimal solutions of problems
P
1(),
i
I
,
i
coincide with the ones for the case
n
1
(see [3]), i.e. for everyi
I
the optimal solution of problemP
1(i) is,
)
,
(
,
)
,
(
,
0
)
(
1 0 *
t
t
t
u
t
t
t
t
u
i i u i u i,
)
,
(
),
(
)
,
(
,
0
)
(
1 0 *
t
t
t
t
t
u
t
t
t
t
m
i i i u u i i u i
where i i i uu
m
t
t
i
~
1
. The optimum value is
,
~
2
)
~
(
~
() 2 1 i i i i i i i im
m
u
c
m
F
(2) wherem
~
i
[
0
,
iu
i(
t
1
t
0)].
(3) In problemP
2, due to GPC, the number of switches in the optimal controla
*k(
t
)
k
K
cannotbe more than
n
r
([7], p.166). Denote these switching times by
1(k),
n(k)r
k
K
. Lett
0
1(k)
n(k)r
n(k)r1,
k
K
,
(4) and denote,
)
(
,
~
0 ( ) 1 ( ) ) ( 1 0 0 p k t p k t t pE
e
a
t
D
A
e
A
e
G
where
E
k( p) isk
-th column of matrixE
(p). One has
,
),
1
(
,
)
(
) ( ) ( ) (odd
is
r
n
r
n
d
even
is
r
n
r
n
d
G
K k p k K k p k p (5) where}.
1
,
{
,
)
(
)
1
(
)
(
1 ) ( ) ( ) (
r
n
r
n
l
D
l
d
l j k j p k j p k
Moreover,
(
n
r
)
-dimensional vectorv
exists such that( )
1
,
,
{
1
,
},
) (r
n
j
K
k
E
e
v
T kp k j
(6) and the optimum value of the objective function is
,
),
1
(
,
),
(
~
) ( ) ( 2odd
is
r
n
r
n
T
even
is
r
n
r
n
T
F
K k p k K k p k (7) where}.
1
,
{
,
)
1
(
)
(
1 ) ( ) (
r
n
r
n
l
a
l
T
l j k j j k p k
Now consider problem
P
3. For sake of simplicity we willconsider only the case where all eigenvalues of matrix
Q
are distinct. If some of them coincide, similar calculations can be made, as in [2]. Let
S
be the matrix of eigenvec-tors of matrixQ
. Let us denote.
}
,
,
,
{
1 2 2S
1QS
diag
n r
Due to GPC, since the eigenvalues of the motion
equa-tions matrix are real, then
k
K
the number ofswitches in the optimal control
a
*k(
t
)
cannot be more than2
n
r
. Let us denote the switching times by) ( 2 ) ( 1
,
,
k r n k
. Let,
,
2 ) ( 1 2 ) ( 2 ) ( 1 1t
k
K
t
kn r k r n k
(8) and denote,
)
(
~
1 1 1 1 ) (e
2S
X
e
1S
X
e
1e
2S
L
G
s
t
t
t
t
,
,
)
(
1 1 ( ) ) (K
k
E
S
e
a
t
D
ks
k t
ks
where
E
k( s) isk
-th column of matrixE
(s).Let
M
I
be the number of negative eigenval-ues of matrixQ
.Let
M
be even. Then
,
,
)
1
2
(
,
),
2
(
) ( ) ( ) (odd
is
r
r
n
d
even
is
r
r
n
d
G
K k s k K k s k s (9) where}.
1
2
,
2
{
,
)
(
)
1
(
)
(
1 ) ( ) ( ) (
r
n
r
n
l
D
l
d
l j k j s k j s k
Further, there exists a
(
2
n
r
)
-dimensional vectorw
such that},
2
,
1
{
,
,
1
) ( 1 ) (r
n
j
K
k
E
S
e
w
T ks k j
(10) and the optimum value of the objective function is
,
),
1
2
(
,
),
2
(
~
) ( ) ( 3odd
is
r
r
n
T
even
is
r
r
n
T
F
K k s k K k s k (11) where}.
1
2
,
2
{
,
)
1
(
)
(
1 ) ( ) (
r
n
r
n
l
a
l
T
l j k j j k s k
Let
M
be odd. In this case
,
),
2
(
,
),
1
2
(
) ( ) ( ) (odd
is
r
r
n
d
even
is
r
r
n
d
G
K k s k K k s k s (12) and there exists a(
2
n
r
)
-dimensional vectorw
such that (10) holds, and the optimum value of the objective function is
.
),
2
(
,
),
1
2
(
~
) ( ) ( 3odd
is
r
r
n
T
even
is
r
r
n
T
F
K k s k K k s k (13)4. The parametric optimization problem
Now we can formulate the parametric optimization prob-lem. Recall that the following conditions hold:If
M
is even, then the parametric optimization problem which has to be solved to obtain the solution of problemP
is the following
P
:
maximize (1) subject to(3),(14),(4)-(6),(8)-(10), where
F
~
1(i) is (2)
i
I
,F
~
2 is (7),F
~
3 is (11).If
M
is odd, then the parametric optimization prob-lem to be considered becomes
P
:
maximize (1) subject to (3),(14),(4)-(6),(8),(12),(10), where
F
~
1(i)
i
I
is (2),F
~
2 is (7),F
~
3 is (13).Both
P
andP
are nonlinear programming prob-lems with3
n
r
2
1
parametersm
~
,
q
,
A
~
,
A
, withd
r
n
)
2
3
(
switching time variables 1( ),
,
n(k)rk
,,
,
,
,
2( ) ) ( 1k
K
k r n k
with(
n
r
)
-dimensionalvar-iable vector
v
and(
2
n
r
)
-dimensional variable vec-torw
, with(
3
n
2
r
)
(
d
1
)
nonlinear equality con-straints and some box concon-straints. The objective functions of the parametric problems are rather simple, but the equality constraints are nonlinear and cannot be solved explicitly. Remark that the constraints of the parametric optimization problem are necessary and sufficient condi-tions for the feasibility of problemP
.Conclusions
We have proposed and investigated an optimal control model with several kinds of communications in order to determine the optimal expenditure rate for each segment and each retailer in a multi-segment market. Remark that if there are only one retailer and one market segment (i.e.
1
r
n
) and if we consider the whole communicationexpenditure rate a priori subdivided into two parts, one for the retailer and one for consumers, we obtain the model presented in [1]. Moreover, for this case we have seen in particular that the sign of eigenvalues of the ma-trix of motion equations determines the type of optimal alternate communication. This implies that if saturation aversion and goodwill decay, which can be considered as negative factors for the firm, are “stronger” than word-of-mouth and goodwill productivity, which are positive for the firm, than it is more convenient to advertise only in the middle of the selling period. Otherwise it is conven-ient to advertise at the beginning of the selling period and then to refresh the goodwill of the firm at the end of the period. In the model considered here, the type of optimal alternate communication in the selling period depends from the evenness/oddness of the number of negative eigenvalues of the matrix of motion equations.
We hope that a similar kind of interpretation (the relation between positive and negative factors for the firm, the retailers and the market segments) can be found for this general case of the model. At this aim the investi-gation of the parametric optimization problem
P
seems to be particularly relevant.References
[1] Bykadorov, I., Ellero, A. and Moretti, E.: “Minimi-zation of communication expenditure for seasonal Products”, RAIRO Operations Research, Vol.36, p. 109-127, 2002.
[2] Bykadorov, I., Ellero, A. and Moretti, E.: “A model for the marketing of a seasonal product with differ-ent goodwills for consumer and retailer”, Journal of Statistics & Management Systems, Vol. 6, p. 115-133, 2003.
[3] Favaretto, D. and Viscolani, B.: “A single season production and advertising control problem with bounded final goodwill”, Journal of Information and Optimization Sciences, Vol. 21, p. 337-357, 2000. [4] Hernández-Castaneda S., Zapato-Lillo P.: “Normal
form of some infinite extensive games and continui-ty properties”, to appear in International Journal of Biomedical Soft Computing and Human Sciences. [5] Hernández-Castaneda S., Zapato-Lillo P.:
“Existance of Nash equilibrium in infinite extensive games”, to appear in International Journal of Bio-medical Soft Computing and Human Sciences. [6] Kaku I., Li Zh. and Xu Ch.: “An Effective
Heuris-tic Algorithm based on Segmentation for Solving a Multilevel Lot-sizing Problem”, International Jour-nal of Biomedical Soft Computing and Human Sci-ences. V. 15, No. 1, Special Issue on Total Opera-tions Management, p. 53 - 59, 2010.
[7] Seierstad, A. and Sydsæter, K.: “Optimal Control Theory with Economic Applications”, North--Holland, Amsterdam, 1987.
Igor BYKADOROV He received the
Degree in Mathematics from Novosibirsk State University in 1983, Ph.D. degree in Mathematics at Sobolev Institute of Math-ematics, Novosibirsk. His present research interests include mathematical models for economic applications, game theory, pric-ing.
Andrea ELLERO He received the Degree
in Mathematics from the University of Padova, Ph.D. degree in Mathematics applied to Economics from University of Trieste. He currently investigates the ap-plication of mathematical models to mar-keting and revenue optimization serving as an associate professor.
Elena MORETTI She received the
Degree in Physics at the University of Padova. Here main research interests are optimal control and mathematical pro-gramming applied to management and