Study of Systems with Variable Length using Processes Without Collisions
C. S. Sousa∗, A. D. Ramos∗∗ and A. Toom∗
Introduction
We consider a class of particle processes with a finite number of types of particles, which we call Processes Without Collisions or PWC for short. As the discrete time goes on, any particle may die or transform itself into one or several particles of any types with certain probabilities, but there are no collisions, that is every transformation applies to only one particle and
probabilities of its transformations do not depend on other particles.
We pay special attention to the limit, when the number of particles tends to infinity and the quantities of particles of all types may be treated as real rather than integer numbers. In fact, this approximation is often used in various sciences, sometimes without a proper foundation. In
[
1],
we prove that under mild conditions the resulting dynamical systems have at least one fixed point and under additional conditions tend to it when time tends to infinity.Study of Systems with Variable Length using Processes Without Collisions
C. S. Sousa∗, A. D. Ramos∗∗ and A. Toom∗
Definitions and Description of Process
Particles of type
k = 1, . . . , n
will be calledk
-particles. A generic state of our process is a vectorq =
q
1 .. .q
n
∈ Ω = ZZ
n+,
where
q
k∈ ZZ
+ denotes the number ofk
-particles. For any vectorc ∈ C
n we denote thenorm of
c
bykck =
nX
i=1|c
i|.
Distance inIR
n:∀v, v
0∈ IR
n:
dist(v, v
0) = kv − v
0k.
Study of Systems with Variable Length using Processes Without Collisions
C. S. Sousa∗, A. D. Ramos∗∗ and A. Toom∗
We call a
n
-dimensional vectorv =
v
1 .. .v
n
∈ IR
n positive ifv
1≥ 0, . . . , v
n≥ 0
andv
1+ . . . + v
n> 0.
For any vector
v ∈ IR
n we callv
normalized ifkvk = 1
. We denoteD =
v
1 .. .v
n
: v
1≥ 0, . . . , v
n≥ 0, v
1+ . . . + v
n= 1
,
D
+=
v
1 .. .v
n
: v
1> 0, . . . , v
n> 0, v
1+ . . . + v
n= 1
.
Study of Systems with Variable Length using Processes Without Collisions
C. S. Sousa∗, A. D. Ramos∗∗ and A. Toom∗
The Probability Operator
We denote the operator of norming by
N orm(v) = v/kvk,
for any non-null vectorv.
We denote by
Π
the set of probability distributions onΩ = ZZ
n+.
Definition 1 Let
P ∈ Π.
We denote byP
q the value ofP
at vectorq ∈ Ω.
We say that thedistribution
P
is in fact finite, if the set{q ∈ Ω :
P
q> 0}
is finite.We denote by
Π
0 the set of in fact finite probability distributions onΩ.
Π
0 is the space where we shall work.Π
andΠ
0 are convex.At every step of discrete time all particles decide independently into what they are going to transform themselves. Every
k
-particle turns into a vectorq
with a probabilityθ
k,q.Type of particles Transitional probabilities Resulting vector
Study of Systems with Variable Length using Processes Without Collisions
C. S. Sousa∗, A. D. Ramos∗∗ and A. Toom∗
We denote by
δ
k∈ Ω
the vector having one at thek
-th place and zeros at all the otherplaces and by
δ
0∈ Ω
the vector having zeros at all the places, that is,δ
k=
0
.. .1
.. .0
∈ Ω
andδ
0=
0
.. .0
∈ Ω.
Type of particles Resulting vector Transformation
k
δ
0 Deathk
δ
k no changek
2 · δ
k MitosisStudy of Systems with Variable Length using Processes Without Collisions
C. S. Sousa∗, A. D. Ramos∗∗ and A. Toom∗
Of course,
∀k :
X
q∈Ω
θ
k,q= 1.
We assume that there is a constant
C
such that(q
1+ . . . + q
n> C) =⇒ (θ
k,q= 0).
that is, no particle may transform into more than
C
particles at once. Now, we define an operatorM : Π → Π
in two steps.Study of Systems with Variable Length using Processes Without Collisions
C. S. Sousa∗, A. D. Ramos∗∗ and A. Toom∗
Step 1:
For every
q ∈ Ω
we denote by∆
q∈ Π
0
the distribution concentrated in one vector
q ∈ Ω.
We define
M ∆
q∈ Π
0
as the result of application of
M
to∆
q for every q:M ∆
q def=
nX
k=1 qkX
j=1V
kj.
Here
V
kj are jointly independent random vectors. Every random vectorV
kj equalsp
with a probabilityθ
k,p for allp ∈ Ω.
Hereq
k is thek
-th component ofq.
Thus we have defined howM
acts on∆
q.Step 2:
Now we define how
M
acts on any probability distributionP ∈ Π
0,
assuming linearity:M P =
X
q∈Ω
(M ∆
q) · P
q,
where
P
q is the value of distributionP
on vectorq.
Thus the operatorM : Π
0
→ Π
0 is defined.Study of Systems with Variable Length using Processes Without Collisions
C. S. Sousa∗, A. D. Ramos∗∗ and A. Toom∗
The Deterministic Operator
In practice we deal with macro-amounts, that is large amounts of particles, so large that we use continuous macro-units instead of discrete micro-units. For example, talking of a chemical or nuclear reaction, we measure the amounts of substances in macro-units, for example
grams or kilograms rather than in micro-units like the number of molecules or atoms.
In this connection, we consider the non-integer analog of our operator, in which instead of non-negative integer numbers of particles of each type, we have non-negative real numbers, called densities
d
1, . . . , d
n.
We are interested only in proportion of different particles rather than in the total amount. For this reason, we use the operator of normalization which acts on any probability distribution
P
Study of Systems with Variable Length using Processes Without Collisions
C. S. Sousa∗, A. D. Ramos∗∗ and A. Toom∗
Definition 2 For all
d ∈ D
we define the operatorM : D −→ D
˜
by the formula˜
M d = N orm (M
0d) ,
whereM
0=
X
r∈Ωr
1θ
1,r· · ·
X
r∈Ωr
1θ
n,r .. . ...X
r∈Ωr
nθ
1,r· · ·
X
r∈Ωr
nθ
n,r
.
Properties ofM :
˜
•
The mapM : D → D
˜
has at least one fixed point.Study of Systems with Variable Length using Processes Without Collisions
C. S. Sousa∗, A. D. Ramos∗∗ and A. Toom∗
Relation between
M, M
0and
M
˜
Now let us describe the relation between operators
M, M
0 andM .
˜
For any real number
x ≥ 0
we denote byround(x)
the smallest integer number which is not less thanx
and call this operator rounding. For any positive vectorv
we denote byround(v)
the vector, whose components are rounded components ofv
:round(v) =
round(v
1)
.. .round(v
n)
.
Study of Systems with Variable Length using Processes Without Collisions
C. S. Sousa∗, A. D. Ramos∗∗ and A. Toom∗
•
First, givend ∈ D
andL > 0,
we transform it into a micro-stateround(L · d).
•
Then, sinceround(L · d) ∈ Ω,
we can applyM
to the distribution∆round
(L·d),
which is concentrated in
round(L · d),
and obtainM ∆round
(L·d),
which is a random vector distributed inΩ.
Normalizing
M ∆round
(L·d),
we obtain a random vectorN orm
³
M ∆round
(L·d)´
distributed inD.
(1)Now we can say how
M, M
0 andM
˜
are related:•
the distributionM ∆round
(L·d)L
concentrates toM
0· d,
whenL −→ ∞;
and•
the distribution (1) concentrates toM d,
˜
whenL −→ ∞.
Study of Systems with Variable Length using Processes Without Collisions
C. S. Sousa∗, A. D. Ramos∗∗ and A. Toom∗
Theorem 1
∀d ∈ D, ∀ε > 0,
P rob
Ã
dist
à M ∆round
(L·d)L
, M
0· d
!
> ε
!
L→∞−→ 0.
Theorem 2∀d ∈ D, ∀ε > 0,
P rob
dist
µ
N orm
³
M ∆round
(L·d)´
,
M d
˜
¶
> ε
L→∞−→ 0.
Study of Systems with Variable Length using Processes Without Collisions
C. S. Sousa∗, A. D. Ramos∗∗ and A. Toom∗
Definition 3 Let
M
be a square matrix of ordern.
We callM
reducible if there is a permutation of indices which reduces the matrixM
to
A
0
B
C
,
where
A
andC
are square matrix and0
is null matrix. Otherwise we call the matrixM
irreducible.
Definition 4 Let
M
be a non-negative and irreducible square matrix of ordern
and letλ
dombe its dominant eigenvalue. Let
h
be the number of eigenvalues of the matrixM,
whose moduli are equal toλ
dom.
Ifh = 1,
we callM
primitive. Ifh > 1,
the matrix is calledimprimitive, and
h
is called the index of imprimitivity.Theorem 3 Let the matrix
M
0 be primitive. Then the sequence of vectorsM
˜
td
has a limitStudy of Systems with Variable Length using Processes Without Collisions
C. S. Sousa∗, A. D. Ramos∗∗ and A. Toom∗
Application: Systems with Variable Length
Systems with Variable Length or SVL is the name of a new class of 1-D particle systems,
whose components can appear and disappear in the process of interaction. We consider SVL with discrete time, with only two types of particles
1
and2.
This process includes three operators death, flip and mitosis, denoted by Mortα
,
Flipβ andMitγ respectively.
•
Mortα:
one component in the state1
disappears with probabilityα
independently from the other components.•
Flipβ:
changes the state of one component from1
to2
with probabilityβ
independentlyfrom the other components.
•
Mitγ:
one component in the state1
duplicates itself with probabilityγ
independently from the other components.Study of Systems with Variable Length using Processes Without Collisions
C. S. Sousa∗, A. D. Ramos∗∗ and A. Toom∗
Our process is the sequence of measures
µ
t= µQ
t,
(2)where
Q =
MortαFlipβMitγ.
At every time step first operates the operator death, then the operator flip and finally the operator mitosis. Here we write operators after measures (as in the references
[
2]
e[
3]
). Let us represent the process (2) in our terms. We user =
1
0
, s =
0
1
, u =
0
2
, v =
0
0
and the following transitional probabilities
θ
1,r= 1, θ
2,r= (1 − α) β, θ
2,u= (1 − α) (1 − β) γ, θ
2,v= α
Study of Systems with Variable Length using Processes Without Collisions
Study of Systems with Variable Length using Processes Without Collisions
C. S. Sousa∗, A. D. Ramos∗∗ and A. Toom∗
In this case,
M
0=
1
(1 − α) β
0
1 − (1 − α) β + (1 − α) (1 − β) γ − α
.
(3)M
0 is reducible, and therefore not primitive. The eigenvalues of the matrixM
0 are:λ
1= 1
andλ
2= 1 − (1 − α) β + (1 − α) (1 − β) γ − α.
The eigenvectors associated with eigenvalue
λ
1 are
v
10
.
The eigenvectors associatedwith eigenvalue
λ
2 are
(1 − α) β
− (1 − α) β + (1 − α) (1 − β) γ − α
v
2v
2
.
Study of Systems with Variable Length using Processes Without Collisions
C. S. Sousa∗, A. D. Ramos∗∗ and A. Toom∗
Results:
•
Letγ >
β (1 − α) + α
(1 − β) (1 − α)
.
Then– the stochastic process (2) has at least two different invariant measures;
– its deterministic approximation with matrix
M
0 (3) has at least two different fixed points.•
Letγ ≤
β (1 − α) + α
(1 − β) (1 − α)
.
Then– the stochastic process (2) has only one invariant measure;
Study of Systems with Variable Length using Processes Without Collisions
C. S. Sousa∗, A. D. Ramos∗∗ and A. Toom∗
Acknowledgements
C. S. Sousa acknowledge financial support from Facepe/Brazil. A. D. Ramos and A. Toom were supported by CNPq.
References
[1] C.S. Sousa. Processos de Part´ıculas sem Colis˜oes. Tese de Doutorado, Universidade Federal de Pernambuco, Pernambuco, (2007).
Veja http://www.de.ufpe.br/˜ toom/ensino/doutorado/alunos/caliteia/CaliTese.pdf. [2] A. Toom. Non-ergodicity in a 1-D particle process with variable length. Journal of
Statistical Physics, v. 115, n. 3/4, (2004), pp. 895-924.
[3] A.D. Ramos. Processos de Part´ıculas com Comprimento Vari ´avel. Tese de Doutorado, Universidade Federal de Pernambuco, Pernambuco, (2007).