Professional Documents
Culture Documents
JMSS-2013 2
JMSS-2013 2
JMSS-2013 2
David Publishing
D DAVID PUBLISHING
Contents
Applied Mathematics
55 Interpolation in 2 with Minimum Value of the Uniform Norm of the Laplace Operator
Sergey I. Novikov
83 Oscillation Theorems for Second Order Differential Equations and Their Applications
Xhevair Beqiri, Elizabeta Koci
System Science
110 Calculation of Infinite Series Through Polygamma Functions’ Algorithms and Laplace Transforms
Héctor Luna-García, Luz María García Cruz and R. Mares
Journal of Mathematics and System Science 3 (2013) 55-61 D DAVID PUBLISHING
Sergey I. Novikov
Institute of Mathematics and Mechanics, Ural Branch of RAS, Yekaterinburg 620990, Russia
Received: September 11, 2012 / Accepted: December 11, 2013 / Published: February 25, 2013.
Abstract: Problems, which are studied in the paper, concern to theoretical aspects of interpolation theory. As is known, interpolation is
one of the methods for approximate representation or recovery of functions on the basis of their given values at points of a grid.
Interpolating functions can be chosen by many various ways. In the paper the authors are interested in interpolating functions, for which
the Laplace operator, applied to them, has a minimal norm. The authors interpolate infinite bounded sequences at the knots of the
square grid in Euclidian space. The considered problem is formulated as an extremal one. The main result of the paper is the theorem, in
which certain estimates for the uniform norm of the Laplace operator applied to smooth interpolating functions of two real variables are
established for the class of all bounded (in the corresponding discrete norm) interpolated sequences. Also connections of the considered
interpolation problem with other problems and with embeddings of the Sobolev classes into the space of continuous functions are
discussed. In the final part of the main section of the paper, the authors formulate some open problems in this area and sketch possible
approaches to the search of solutions. In order to prove the main results, the authors use methods of classical mathematical analysis and
the theory of polynomial splines of one variable with equidistant knots.
ϕ2 k −1 ( x) sin kx=
= , ϕ2 k ( x) cos
= kx (k 1,..., m) ,
process of constructing another function
then the authors arrive at the trigonometric
L( x) ≡ L( x; f ) such that L( x ) = f ( x ( k ) ) ,
(k )
interpolation. The choice of the system of functions
k ∈ A. Function L( x) is called the interpolating
{ϕν } is determined by the properties of the function
function or the interpolant. Here A is the set of all
class for which an interpolation process is studied and
non-negative integers or a finite subset of it. Values of
(or) an interpolation algorithm is constructed. The
f can be not given at the points lying between the
interpolation by algebraic and trigonometric
knots of the grid. In that case, the authors deal with
polynomials, splines and many other systems of
interpolation of discrete data { f ( x )}
(k )
k∈A
. Usually functions of one variable was investigated in sufficient
detail.
For many applications, the interpolants that change
Corresponding author: Sergey I. Novikov, senior
researcher, research field: approximation theory. E-mail: regularly upon transition from one knot of a grid to
Sergey.Novikov@imm.uran.ru.
56
Interpolation in
2
with Minimum Value of the Uniform Norm of the Laplace Operator
other knots are most important. This leads to problems modifications of the extremal interpolation problem
of choosing of interpolants on the basis of were researched (see Refs. [11, 12], and many other
minimization of a norm of the derivative of order papers). For the history of these problems and
m (m ≥ 1) for fixed interpolated data. Originally, contributions of many authors, the authors refer to the
such a minimization problem was considered by reader to Refs. [13, 14], and references therein.
J.Favard [1] for functions of one variable in the case of Interpolation by functions of several variables was
the uniform norm. Actually, this problem as well as studied much less. Here the problem of finding of
some more general problems of interpolation (so-called optimal interpolants has certain specific details. Since
on ( n ≥ 2) there are partial derivatives of the
n
multiple interpolation) was studied in Ref. [1]. Later
Boor [2], Karlin [3], Tikhomirov and Bojanov [4], same order with respect to the different variables, the
Fisher and Jerome [5], and many others found solutions authors have to minimize either the Sobolev semi-norm
of the Favard type interpolation problems for various of order m or a norm of a linear differential operator
interpolation conditions and spaces of interpolating (of order m ) applied to interpolants. Also
functions. The number that is equal to the norm of spline-similar functions arose here ([7, 15], and other
papers). In ( n ≥ 2) , the extremal interpolation
n
m -th derivative of the “minimal” interpolant is called
the value of the Favard interpolation problem. problems were considered for operators of the mixed
derivatives ∂ / ∂x1 1 ∂x2 2 ...∂xn n , m1 + m2 + … +
m m m m
A key role in studying of the minimization problems
with interpolation restrictions played the splines, i.e. mn = m [11], and for linear differential operators that
the smooth piecewise-polynomial functions. In 1957, were represented as a superposition of n
J.Holladay [6] found an interesting property of the one-dimensional linear differential operators with
interpolating cubic splines. It turned out that they constant real coefficients [16]. The corresponding
provide the least L2 -norm of the second derivative difference operators were constructed by a
over the corresponding set of interpolating functions. superposition of n appropriate one-dimensional
This is so-called the minimum norm property. It was linear difference operators, and classes of interpolated
transferred on interpolating splines of an arbitrary odd data were defined by these operators. Every
degree and L-splines [7]. The minimum norm property one-dimensional difference operator was chosen so
served as a starting point for many subsequent that it vanish the grid values of all functions from the
generalizations (see Refs. [8, 9] and other papers). null space of the differential operator. Solutions of
Often it is required to interpolate not one set of these problems were obtained in the above quoted
numerical data, but some collection. On this way, the papers. However, the most of the partial linear
authors arrive at problems of extremal interpolation. differential operators cannot be represented as a
The class of interpolated data whose m -th finite superposition of one-dimensional ones. One of such
differences are bounded, is chosen. The problem of differential operators is the well-known Laplace
operator ∆ =∂ / ∂x1 + ... + ∂ / ∂xn . In the case
2 2 2 2
extremal interpolation is to find a minimum value for
L∞ -norm of m -th derivative of interpolants under when the Laplace operator in n , n ≥ 2 is
interpolation of the class of data by smooth functions of considered, there is no appropriate statement of the
one variable at the equidistant knots. This means that extremal interpolation problem, since there is no finite
the authors take the supremum of the values of the difference operator that annihilates all harmonic
Favard interpolation problems over the chosen class of functions on a rectangular lattice. More information
interpolated data. First, this problem was solved by about this and other results in this area can be found in
Subbotin in the paper [10]. Later, various the paper [14].
57
Interpolation in
2
with Minimum Value of the Uniform Norm of the Laplace Operator
Now the authors pass to statement of the problem This value is the norm of the Laplace operator applied
considered in the present paper. to the "best" function from the class Y∞ ( z ) when the
=
Let z {z j ,ν : ( j ,ν ) ∈ 2 } be an arbitrary real "worst" sequence from set M ∞ is interpolated.
The problem || ∆u ||∞ → u∈inf Y ( z ) is the Favard type
=
sequence, {
|| z ||l∞ sup | z j ,ν |: ( j ,ν ) ∈ 2 } . The
∞
= (1) ≤ 2 f ∞
∆f ∞
(2)
z∈M ∞ u∈Y∞ ( z ) ∞ ∂x ∞
58
Interpolation in
2
with Minimum Value of the Uniform Norm of the Laplace Operator
∂f otherwise. Since z
j ,ν ∈ M ∞ the authors have
( )
1/ 2
≤ 2 f ∆f
A∞ ( 2 ) ≥ inf || ∆u ||∞
∞ ∞
∂y ∞ (3)
u∈Y∞ ( z )
The same inequalities for functions defined on
Let u = u ( x, y ) be an arbitrary function from the
n , n ≥ 3 were established in Ref. [22].
class Y ( z) .
Another tool is the parabolic B-spline of one variable ∞
Now the authors decompose 2 into closed
with equidistant knots. B-splines are well known in
squares of the unit measure
mathematics. Detailed information about B-splines and
2 = Q j ,ν ,
their applications can be found, for example, in Refs.
where
[23-25], and references therein. The authors shall give
an explicit representation and list some properties of
Q=
j ,ν {( x, y ) ∈ 2
: j ≤ x ≤ j + 1, ν ≤ y ≤ ν + 1 }
and j , ν run over all integers. It is evident that the
the parabolic B-spline with equidistant knots.
neighboring squares are intersected in sets of zero
Let a ∈ be an arbitrary fixed point, h > 0 be
plane measure. In every square Q j ,ν the authors
any number, and δ =− {a h / 2, a − h / 6, a + h / 6, a + h / 2}
represent the differentiable function u by Taylor’s
be a partition of the interval [ a − h / 2, a + h / 2] .
formula (see Ref. [26])
Define the following function: B (t ) = 0 if
=
u ( x , y ) u ( x0 , y 0 ) +
t < a − h / 2 and t ≥ a + h / 2 ,
∂u ( x0 + t ( x − x0 ), y0 + t ( y − y0 ))
1
9 h + ( x − x0 ) ∫ dt + (4)
2
B (t=
) t − a + ∂x
2 2 0
∂u ( x0 + t ( x − x0 ), y0 + t ( y − y0 ))
1
if a − h / 2 ≤ t < a − h / 6 , + ( y − y0 ) ∫ ∂y
dt
h 2
0
B ( t= − 3(t − a )
2
) 3 where ( x0 , y0 ) is an arbitrary fixed point from Q .
4 j ,ν
9 h
2 y0 =ν if ν ≠ 0 and y0= ν + 1 if ν = 0 , the authors
B (t=
) t − a −
2 2 have u ( x0 , y0 ) = 0 . From Eq. (4) it follows the
if a + h / 6 ≤ t < a + h / 2 . inequality
∂u ( x, y )
Function B (t ) is the parabolic B-spline with knot | u ( x, y ) | ≤ | x − x0 | max +
( x , y )∈Q ν ∂x
set δ and has the following properties:
j,
Now the authors take the least upper bound over all
so that z
j ,ν
= 1 if ( j ,ν ) = (0, 0) and z
j ,ν = 0 points ( x, y ) ∈ and get the inequality
2
59
Interpolation in
2
with Minimum Value of the Uniform Norm of the Laplace Operator
for every point ( j ,ν ) ∈ 2 . Now the authors put inequalities Eq. (2) are known, and the products of the
G j ,ν ( x, y ) = Fj ,ν ( x, y ) if the point ( x, y ) belongs to parabolic B-splines can be written for any their number
in corresponding to dimension of the space.
the fixed square V j ,ν , and G j ,ν ( x, y ) = 0 otherwise.
The problem considered in this paper can be
The authors define a function F = F ( x, y ) by
generalized by means of replacement of the uniform
F ( x, y ) = ∑G j ,ν
( x, y ) . Function F lies in the norms l∞ and L∞ by the norms l p and
j ,ν ∈
Lp (1 ≤ p < ∞) respectively. In that case, the set of
class Y∞ ( z ) and
interpolated data is defined as follows
∆F = ∑ ∆G ν ( x, y ) ≤
1/ p
∑ z j
∞ j, p
ν
j , ∈ ∞
=
M p = z : z {z j } j∈n , ≤ 1 , the
j∈n
≤ ∑
j ,ν ∈
∆G j ,ν ( x, y ) ≤ 48
class of interpolating functions as
∞
The proved theorem gives only an interval, in which the value Eq. (1) is
1/ p
Eq. (1) lies. A sharp value of Eq. (1) is unknown. It is
possible the upper estimate can be decreased by means = Ap n ( )
sup inf ∆u ( x) dx .
z∈M p u∈Yp ( z ) n ∫
p
the norm of the Laplace operator applied to the smooth [14] S.I. Novikov, Problems of extremal functional
interpolation, in: Proceedings of the International Summer
bounded interpolating functions, on the class of the
Mathematical Stechkin Workshop on Function Theory,
interpolated bounded sequences. Aleksin Russia, 2007, pp.100-109. (in Russian)
[15] J. Duchon, Splines minimizing rotation-invariant semi
Acknowledgments norms in Sobolev spaces, Lecture Notes in Mathematics
571 (1977) 85-100.
This research was supported by the Russian [16] S.I. Novikov, V.T. Shevaldin, A problem of extremal
Foundation of Basic Researches (project no.11-01-00347) interpolation for multivariate functions, Proceedings of the
and by the Program of Integration Projects between the Steklov Institute of Mathematics (Supplementary issue) 1
(2001) 150-166.
Ural and Siberian Branches of RAS. [17] S.I. Novikov, Interpolation with minimal value of the
norm of the Laplace operator in a ball, Proceedings of
References Institute of Mathematics of NAS of Ukraine 5 (1) (2008)
[1] J. Favard, Sur l’interpolation, Journal de Mathematiques 248-262. (in Russian)
Pures et Appliquèes 19 (9) (1940) 281-306. [18] S.I. Novikov, Interpolation in a ball with minimum value
[2] C. de Boor, On the best interpolation, Journal of of L p -norm of the Laplace operator, Proceedings of the
Approximation Theory 16 (1) (1976) 28-42. Steklov Institute of Mathematics (Supplementary issue)
[3] S. Karlin, Interpolation properties of generalized perfect 277 (1) (2012) 152-160.
splines and solutions of certain extremal problems, [19] S.I. Novikov, Interpolation in a square with minimum
Transactions of the American Mathe-matical Society 206 value of the norm of the Laplace operator, in: Abstracts of
(1975) 25-66. International Conference “Theory of Approximation of
[4] V.M. Tikhomirov, B.D. Bojanov, Some convex problems functions and its Applications” dedicated to the 70th
of approximation theory, Serdica 5 (1) (1979) 83-96. (in anniversary of prof. A.I. Stepanets. Kyev, Institute of
Russian) Mathematics of NAS Ukraine, 2012, p.78. (in Russian)
[5] S. Fisher, J. Jerome, Minimum norm extremals in function [20] S.I. Novikov, Interpolation in a square with minimum
spaces, Lecture Notes in Mathematics 479 (1975) 1-209. value of the uniform norm of the Laplace operator,
[6] J.G. Holladay, A smoothest curve approximation, Proceedings of the Steklov Institute of Mathematics 18 (4)
Mathematical Tables and other Aids to Computation 11 249-257. (in Russian)
(1957) 233-243. [21] G. Timofeev, Kolmogorov’s type inequality with the Laplace
[7] J.H. Ahlberg, E.N. Nilson, J.L. Walsh, The Theory of operator, in: Theory of Functions and Approximations, 1983,
Splines and Their Applications, Academic Press, New Saratov, Saratov State University, pp.84-92. (in Russian)
York and London, 1967. [22] V.G. Timofeev, An inequality of Landau type for
[8] M. Atteia, Generalisation de la definition et des proprietes functions of several variables, Mathematical Notes 37 (5-6)
des “spline functions”, Comptes Rendus des Seances de (1985) 369-377.
L’Académie des Sciences 260 (13) (1965) 3550-3553. [23] I.J. Schoenberg, Cardinal interpolation and spline functions,
[9] P.-J. Laurent, Approximation et optimisation, Hermann, Journal of Approximation Theory 2 (1969) 167-206.
Paris, 1972. [24] C. de Boor, Splines as linear combinations of B-splines, in:
[10] Yu.N. Subbotin, On the relationship between finite Approximation Theory II, Proceedings of International
differences and corresponding derivatives, Proceedings of Symposium, Austin, Texas, USA, 1976, pp.1-47.
the Steklov Institute of Mathematics 78 (1965) 24-42. (in [25] S.B. Stechkin, Yu.N. Subbotin, Splines in numerical
Russian) Mathematics, Nauka, Moscow, 1976 (in Russian).
[11] Yu.N. Subbotin, Functional interpolation in the mean with [26] V.I. Burenkov, Sobolev spaces on domains, Teubner Texts
the least n-th derivative, Proceedings of the Steklov in Mathe-matics, Vol. 137, G.Teubner Verlag GmbH,
Institute of Mathematics 88 (1967) 30-60. (in Russian) Stuttgart, 1998.
[12] V.T. Shevaldin, A problem of extremal interpolation, [27] C. de Boor, K. Höllig, S. Riemenschneider, Box Splines,
Mathematical Notes 29 (4) (1981) 310-320. Applied Mathematical Science, Vol. 98, Springer Verlag,
[13] Yu.N. Subbotin, Variations on a spline theme, New York and Berlin etc., 1993.
Fundamental and Applied Mathematics 3 (4) (1997) [28] R. Adams, Sobolev spaces, Academic Press, New York,
1043-1058. (in Russian) San Francisco and London, 1975.
Journal of Mathematics and System Science 3 (2013) 62-67 D DAVID PUBLISHING
Irene Duarte Gandica1, Lina Marcela Ocampo1, Edwin Fernando Duque Marín1, Maria Mercedes González2
1. Escuela de Investigación en Biomatemática,Universidaddel Quindío, Armenia, Quindío 630001, Colombia
2. Grupo GYMOL, Universidaddel Quindío, Armenia, Quindío 630001, Colombia
Received: January 8, 2012 / Accepted: February 11, 2013 / Published: February 25, 2013.
Abstract: The poliomyelitis is an acute infection produced by the polio virus that affects the human central nervous system. It is
transmitted by fecal-oral and respiratory contact.There are two types of vaccine, OPV (live attenuated virus) and IPV (inactivated
polio virus). Currently, there is a plan of vaccination until the age of 5 with OPV. The children vaccinated expel a virus (derived from
the vaccine) to the environment, and some of the people that have oral contact with them, get vaccinated by the herd behavior.
Nevertheless, taking into account the lately observed facts about the reversion to virulence of the oral polio vaccine during its
circulation in the environment, a change in the current vaccination schedule is being contemplated, where the oral polio vaccine can
be replaced by the inactivated vaccine. Nowadays, In Colombia the inactivated oral polio vaccine is recommended for children
presenting immune deficiency who are vaccinated with IPV. These children do not expel poliovirus to the environment. This work
presents a mathematical model that describes the dynamics of the infection in a population where the two types of vaccination are
carried out. The population is divided into two groups of age and Michaelis-Menten interactions. Different strategies of vaccination
are simulated and analyzed.
Key words: Poliomyelitis, vaccine OPV, vaccine IPV, vaccination, mathematical model.
mothers with antibodies are protected in natural form highly incident times, that is, it mostly affects the
against the paralytic disease during some weeks. The poorest non-immunized populations and preserves the
immunity is acquired after an infection by the wild same epidemiological characteristics. On the other
virus or by vaccination. The immunity acquired by the hand, it is important to mention that the appearance of
natural infection (that includes mild and subclinical outbreaks caused by vaccine-derived virus is a
infections) or by the complete series of the live phenomenon that has raised the need to reach the world
poliovirus vaccine orally administered, produces an eradication.
antibody response as well as a local response in the In 1994 the ICCPE (International Certification
intestinal cells. It is estimated that this is a life-long Commission for poliomyelitis Eradication) concluded
immunity and that it can block the infection by that the natural circulation of the wild poliovirus in the
subsequent wild viruses, interrupting the chain of Americas had been interrupted. One of the risks to be
transmission. The application of the IPV confers faced in the poliomyelitis post-eradication stage is the
humoral immunity. But the mucosal immunity is appearance of polio cases due to an oral
relatively smaller. Consequently, the IPV does not vaccine-derived virus [3-6]. Once the world
offer resistance to the carrying and spread of the wild certification of polio eradication was achieved, the
virus in the community. It is believed that there is little World Health Organization recommends to interrupt
or no cross-protective immunity among different types the use of the OPV, leaving the IPV as the only
of poliovirus. alternative. Before presenting the scenarios in this
The poliovirus is an enterovirus that is divided into stage, all the risks representing the interruption of the
three antigenic types: 1, 2 and 3. All 3 types can cause vaccination with VPO should be analyzed, since the
paralysis. Type 1 more frequently, type 3 in a smaller costs generated when having the IPV as the only option
measure, and type 2 rarely causes it. Most of the are high and many countries can refuse to use it.
epidemics are caused by the poliovirus type 1. Vaccine This work presents a model in differential equations
related cases, containing the three types of virus, are that describes the dynamics of the infection in a
generally produced by type 2 or 3. The poliovirus population where the two types of vaccination are
derived from the Sabin vaccine, which has caused carried out. It intends to contribute, through the
outbreaks in Egypt, Philippines, Haiti, Madagascar and simulation of different hypothetical scenarios, to the
the Dominican Republic, is a virus that has mutated in design of a plan to minimize and/or avoid the risks of
more than 1% regarding the VDPV (original Sabin the replacement of the OPV by the IPV.
strain) and that has reverted to the neurovirulence. Two
2. Mathematical Model
types of vaccine-derived poliovirus are known: The
iVDPV (i of immunodeficiency), isolated in immune The model considers the following assumptions:
deficient people, and the cVDPV (c of circulating), that - A closed human population is considered (Global
has been isolated in outbreaks and has shown to have death rate = birth rate).
the same biological and epidemiological characteristics - No migrations are taken into account.
of the wild virus. The initiative of polio world - It is considered that an individual has acquired the
eradication has significantly diminished the quantity of infection, only, if he develops a flaccid paralysis and it
cases in the world: From around 350,000 estimated is demonstrated that it was caused by the poliovirus.
cases in 1988 to just 784 notified in 2003. Otherwise, - From a year, the circulating vaccine-derived
the disease keeps behaving, from an epidemiological poliovirus can recover virulence.
point of view, in the same way it used to happen in - The people who have never been vaccinated and
64 Model for the Vaccination Against the Poliomyelitis
have contact with the vaccine virus (that has mutated variables is:
between a 1 and 15%), can develop infection. 1
1 1 1 1 1 1
- A proportion of the people having oral contact
1
with the Sabin virus (that has mutated less than 1%)
2
circulating in the environment, develop immunity. 2 2 2 2 1
Acknowledgment
We would like to thank Departamento
Administrativo de Ciencia, Tecnología e Innovación,
COLCIENCIAS, and Universidad del Quindío.
References
[1] Restrepo A, Robledo J, Leiderman E, Restrepo M, Botero
D, Bedoya V., Enfermedades Infecciosas.
Corporaciónpara Investigaciones Biológicas, Medellín,
2003.
[2] A. Roy, M. Robert, Infectious Diseases of Humans,
Oxford University Press Inc., New York, 1991.
[3] M. Eichner, K. Dietz, Eradication of poliomyelitis: When
Fig. 4 Human populations and virus behavior when the
can one be sure that polio virus transmission has been
initial populations are: . , . , . ,
terminated?, American Journal of Epidemiology 143 (8)
. , , , and the vaccination coverage
(1996) 16-22.
is: , . .
[4] P. Fine, G. Oblapenko, R. Sutter, Polio control after
total population. Comparing the results of the certification: Major issues outstanding, Bulletin of the
World Health Organization 82 (2004) 47-52.
simulations, it is concluded that a single unit of Sabin
[5] Organización Panamericana de la Salud, Erradicación de
virus in the environment leads to infection, and given la Poliomielitis, Guía Práctica,Publicación Científica y
the strong resistance of the virus, it is necessary to be Técnica No. 607, 2005.
very attentive to the circulating virus quantity in the [6] N. Alvis, F. De la Hoz, J. Narváez, Impacto económico
de la introducción de la vacuna inactivada inyectable
environment, since this represents one of the most
contra la poliomielitis en Colombia,Revista
impacting factors on the increase of the infected Panamericana de Salud Pública 27(5) (2010) 352–359.
population. [7] R.J.T. Radboud, M.A. Pallansch, O.M. Kew,V. M.
It is also shown that the increase of the initial Cáceres, R.W. Sutter, K.M. Thompson, A dynamic model
of poliomyelitis outbreaks: Learning from the past to help
susceptible population causes an increase of the Inform the future, American Journal of Epidemiology
infected population. 162 (2005) 358-372.
[8] T.J. Bailey, The Mathematical Theory of Infectious
4. Conclusions Diseases, Hafner Press, New York, 1957.
[9] F. Brauer, C. Castillo-Chavez, Mathematical Models in
This leads to conclude that before suspending the use Population Biology and Epidemiology, Springer-Verlag,
of the oral vaccine, the percentage of the vaccinated New York, 2001.
people with OPV should be very close to 100% of the [10] K.L. Sutton, H.T. Banks, C. Castillo-Chavez, Estimation
of invasive pneumococcal disease dynamics parameters
population, since the probability of infection is
and the impact of conjugate vaccination in Australia,
considerably low. Mathematical Biosciences and Engineering 5 (1) (2008)
Finally, this work constitutes a first attempt of 175-204.
mathematical modeling of the poliomyelitis in the [11] C. Castillo-Chavez, A. Rivas, N. Hengartner, J. Hyman, G.
Chowell, Critical response to post-outbreak vaccination
post-eradication stage. The simulations have permitted against foot-and-mouth disease, Contemporary
to evaluate the human population and virus behavior Mathematics, American Mathematical Society, 2005.
corresponding to different vaccination scenarios. This [12] J.D. Murray, Mathematical Biology, Springer–Verlag,
New York, 1993.
allows to an approximation to real settings that lead to
[13] B. Carnahan, H.A. Luther, J.O. Wilkes, Applied
the evaluation of the risks involving the replacement of Numerical Methods, John Wiley and Sons, New York,
the OPV vaccine by the IPV vaccine. 1969.
Model for the Vaccination Against the Poliomyelitis 67
[14] R.J. Burden, J.D. Faires, Numerical Analysis, Thomson the World Health Organization 60 (3) (1982) 405-422.
Brooks, USA, 2005. [17] W. Dowdle, C. Wolff, Post-eradication poliovirus
[15] C. Johnson, Numerical Solution of Partial Differential facility-associated community risks, Biologicals 34 (2006)
Equations by the Finite Element Method, Cambridge 127-132.
University Press, New York, 1987. [18] T. Hovi, L. Shulman, H. Van der Avoort, J. Deshpande,
[16] B. Cvjetanovic, B. Grab, H. Dixon, Epidemiological M. Roivainen, E. Gourville, Role on environmental
models of poliomielitis and measles and their application poliovirus surveillance in global polio eradication and
in the planning of immunization programmes, Bulletin of beyond, Epidemiology and Infection 140 (1) (2011) 1-13.
Journal of Mathematics and System Science 3 (2013) 68-74 D DAVID PUBLISHING
Received: January 22, 2012 / Accepted: February 12, 2013 / Published: February 25, 2013.
Abstract: The purpose of variable selection is to identify important predictors for response variables. Although there are many
varieties of variable selection methods, almost all of them have a problem of not accounting for the relationship between predictors.
Therefore it would well happen that the selected subset of identified predictors leads to hard-to-interpret model consisted of only
interaction terms. In design of experiments, the analysis is driven by the effect heredity principle which governs the relationship
between an interaction and its corresponding main effects. In this paper, the authors extend the variable selection method the Lasso
with effect heredity principle to its Bayesian version. In the example, the authors analyze the data obtained from typical screening
design Plackett-Burman design and compare the result from the ordinary Bayesian Lasso and proposed method.
principle [3] and analyzed the experiments with where p is the number of main effects.
complex aliasing patterns [5]. Yuan et al. extended the The Lasso is proposed for variable selection and
least angle regression [6] to follow the principle [7]. parameter estimation and the estimates can be
Choi et al. and Noguchi et al. addressed the problem interpreted as constrained version of ordinary least
applying the Lasso [8] with effect heredity principle squares. It fits a linear model by solving the
[9-10]. optimization problem
p p
| | |
When the authors take into account not only main
min(Y Xβ ) (Y Xβ ) (
T
i ij
|) (3)
effects but interactions as candidate factors in the
i 1 i j j 2
analysis of designed experiment, the number of factors where > 0 is the tuning parameter controlling the
can be larger than observations. In this paper, the amount of shrinkage for the fitted coefficients.
authors have use of the variable selection method the Sufficiently large value of produces shrunken
Bayesian Lasso [11] which is Bayesian version of the estimates of regression coefficients to be exactly zero.
Lasso because of their merit of being able to apply It may improve prediction accuracy and interpretation.
even when the number of variables exceeds the In screening experiments, this characteristic
number of observations. And the authors extend the corresponds to the effect sparsity principle.
Bayesian Lasso to follow the effect heredity principle. Tibshirani suggests that estimates of the Lasso can
Extension to Bayesian approach let us obtain credible be interpreted as a Bayesian posterior mode using
intervals of estimates. normal likelihood and an independent Laplace prior to
The rest of paper is organized as follows. In Section each regression coefficient [8]. The first explicit
2, the authors explain the brief overview of the Lasso application of the Bayesian Lasso was proposed by
and its Bayesian version. In Section 3, the authors Park and Casella [11].
introduce our models and shows parameter structure 3. Models
in the hierarchical model. The authors demonstrate the
proposed method through two examples in Section 4, 3.1 Strong and Weak Heredity Model
and conclude with a discussion in Section 5. In this paper, the authors extend the Bayesian Lasso
for selecting subset following effect heredity principle.
2. Lasso and Bayesian Lasso
Choi et al. [9] considered the strong heredity model
Usually, experimental data can be modeled in the with following reparameterization of the coefficient
form of the following general linear model for the interaction terms xixj.
Y Xβ ε (1) p p
f s ( x) i xi ij i j xi x j (4)
where the authors have n observations on a response Y i 1 i j j 2
70 Bayesian Lasso for Screening Experiments
do not obey the heredity principle. In Bayesian Table 1 Plackett-Burman 12-run design with response data.
Design Response Y
variable selection, the heredity principle is adopted by No. A B C D E F G H I J K M1 M2
setting conditional probability for an interaction given 1 1.058 6.058
2 1.004 4.733
its corresponding main effects [3]. For example, if the 3 -5.200 4.625
4 5.320 5.899
authors consider two factors A and B and their 5 1.022 7.000
6 -2.471 5.752
interaction AB, the conditional probability for the 7 2.809 5.682
8 -1.272 6.607
interaction being active is as follows 9 -0.955 5.818
p 0 if ( A , B ) (0, 0) 10 0.644 5.917
11 -5.025 5.863
Pr( AB 1) p1 if one of A , B 1 12 3.060 4.809
p if ( A , B ) (1, 1)
2 Bayesian Lasso
1.5
whether the effect is active or not. That is, A can take
1.0
0.5
two values 0 or 1. When A = 0, main effect A is
0.0
inactive and A = 1 means A has active effect. By
-0.5
introducing the above indicator parameter into the
-1.0
consider the model selected from every possible Strong Heredity Model
combination of factors.
2
1
4. Examples
0
and Eq. (5) to the Bayesian Lasso and analyze the data
-2
the true model Y = A + 2AB + 2AC + where ~ N(0, Fig. 1 Boxplots for samples from posterior distributions in
the simulated screening experiment M1.
=0.25), that is, factor A has an active effect and
interactions AB and AC are active while the remaining which have strong effect, posterior median estimates
factors D-K are inactive. The authors consider a total are compared. Note that the median estimates won’t
of 66 candidate effects consisted of 11 main effects be exactly zero therefore the authors need to specify
and 55 two-factor interactions. Through this example, threshold value to identify effects. Add to using
the authors confirm our proposed model can identify turning parameter, specifying the amount of a
the active effects and compare the results of the usual threshold value also can be interpreted as effect
Bayesian Lasso. sparsity principle.
Fig. 1 shows boxplots of 50,000 samples from each If the authors ignore the factors whose absolute
66 factor’s posterior distribution obtained by the usual value of median estimate under 0.25, identified factors
Lasso and proposed methods. To identify the factors are only interactions AB and AC. Active main effect A
72 Bayesian Lasso for Screening Experiments
can’t be identified with the ordinary application of the Table 2 Factors and levels for cast fatigue experiment.
Bayesian Lasso and this is the same result of the Level
Lasso and least angle regression. The strong heredity Factor - +
A :Initial structure As received treat
model identify the subset including main effects A, B,
B :Bead size Small Large
C, E, H, I, K and the interaction IK. The selected C :Presure treat None HIP
subset does not conform to the true model but follows D :Heat treat Anneal Solution treat/age
strong heredity principle. When the authors apply E :Cooling rate Slow Rapid
F :Polish Chemical Mechanical
weak heredity model, active main effect A and
G :Final treat None Peen
interactions AB and AC can be identified. The authors
Bayesian Lasso
0.6
Hamada and Wu could not identify any of the
0.4
0.2
important effects [4]. It can be said our proposed
0.0
method have the advantages.
-0.2
-0.4
4.2 Cast Fatigue Experiment
-0.6 A B C D E F G AC AE AG BD BF CD CF DE DG EG
Hunter et al. [14] used the 12-run design to study Strong Heredity Model
Hamada and Wu [4] and the main effect F and Weak Heredity Model
means changing chemical to mechanical on polish and Fig. 2 Boxplots for samples from posterior distributions in
the simulated screening experiment M2
no peening probably is able to make fatigue lifetimes
longer. Therefore, the goal of this experiment would interactions AE and FG are identified by applying
be to identify the subset (F, FG) which has adjR2 of the threshold value 0.1. This result is same as the
0.871. Note that the procedure of the stepwise type Lasso. The interaction AE is identified but its
variable selection method ignores stochastic errors corresponding main effects A and E are not. And the
inherited in the stages of variable selection. It makes subset consisted of these factors has adjR2 of 0.553.
their theoretical properties rather hard to understand. When the authors apply the strong heredity model and
The authors applied our proposed models and the set the threshold value 0.3 for absolute median
ordinary Bayesian Lasso to the data M2 in Table 1. Fig. estimates, only one main effect F is identified and the
2 shows the results. From the result of the ordinary subset has adjR2 of 0.395. On the other hand, the
application of the Bayesian Lasso, main effect F and authors can identify the important factors F and FG
Bayesian Lasso for Screening Experiments 73
Bayesian Lasso
with the weak heredity model and the combination of
3
the factors is following the weak heredity principle.
2
Here, the authors can see the variance of samples from
1
posterior distributions for inactive main effects is
0
smaller than inactive interactions when proposed
-1
models are applied. This might be interpreted that the
-2
A C E G I K AD AG AJ BD BG BJ CE CH CK DG DJ EG EJ FH FK GJ HJ IK
model can consider effect hierarchy principle.
Strong Heredity Model
However, in Fig. 1, there are not so large difference of
5
variance of the samples between main and interaction
4
3
effects. Therefore, it might depend on the number and
2
size of effects.
1
0
5. Discussion
-1
-2
A C E G I K AD AG AJ BD BG BJ CE CH CK DG DJ EG EJ FH FK GJ HJ IK
In the examples, the Bayesian Lasso identified same Fig. 3 Boxplots for samples from posterior distributions in
factors as the Lasso and the Bayesian Lasso with the the example from Hamada and Wu [4].
effect heredity principle could select the subset limitation of the design that may require additional
following the principle. The weak heredity model experimentation to resolve. The authors have two
works well where there are a few active interactions choices of the model, strong and weak heredity, based
that are partially aliased with the main effects. Here, on the Bayesian Lasso. However, in practice, the
the authors illustrate some limitations with the brief authors don't know which version to use. Thus, the
example. The authors used the proposed methods on authors should try both of them and consider the result
Hamada and Wu’s [4] constructed example, the same with physical background. Limitation for variable
design as Table 1 with true model Y = 2A 4C 2BC selection with the methods might depend on the effect
2CD , where ~ N(0, =0.5). Fig. 3 shows the size and number of considered factors. Choosing
result. When the authors set threshold value 0.25 for adequate design might be helpful to identify active
absolute value of posterior samples, the Bayesian factors in screening experiment.
Lasso selects the subset (C, AK, AF, DG, BH) and its
Acknowledgments
adjR2 is 0.870. The strong heredity model selects the
subset (C, G, D, J, H) which has adjR2 of 0.913. The This work was supported by grants from the Tokyo
weak heredity model also can't identify active factors University of Science of Japan.
even true model is following the weak heredity References
principle. The selected subset is (C, G, BH, DG, GH)
[1] C.F.J. Wu, M. Hamada, Experiment: Planning, Analysis,
with adjR2 of 0.872. The interaction BH is contained and Parameter Design Optimization, 2nd ed., New York,
without its main effects B and H. This illustrates a Wiley, 2000, pp. 172-173.
74 Bayesian Lasso for Screening Experiments
Received: January 22, 2012 / Accepted: February 12, 2013 / Published: February 25, 2013.
Abstract: Schrödinger’s equation is one the equations that mark the beginnings of the systematic quantum physics. It was shown that
it follows from the Dirac’s equation and the relationship with classical physics, i.e. with classical field theory was established. The
subject of this work is the relationship between classical relativistic physics and the quantum physics. Investigation carried out in this
work, shows that the free electromagnetic field, spinor Dirac’s field without mass, spinor Dirac’s field with mass, and some other
fields are described by the same vibrational formulation. The conditions that a field be described by Maxwell’s equations of motion
are given in this work, and some solutions of these conditions are also given. Non-relativistic approximation of the equations of the
non-quantified field are the Schrödinger’s equations. Dirac’s equation as a special case, contains Maxwell’s equations and the
Schrödinger’s equation.
1
F C C A A
After separating the real and imaginary parts of
2 each four components of the Dirac’s field function,
G A , F C
(11) the authors have
Taking F = G = 0, these equations become:
F g G 2 A 0
F
0, F
0 (12)
F g F 2C 0 (18)
1
i.e., equations of the free Maxwell’s field. The authors
see that equations of the free Maxwell’s field are, in
F A A C C
2
1
F C C A A
fact, a special case of the canonical Eq. (11). If the
authors chose C 0, then we obtain equations of 2
the standard electromagnetic theory of the field, which G A , F C
are the equations of the conventional theory of the free For F G 0, 0 The authors obtain
electromagnetic field. Therefore, it follows that the canonical equation for the real Proca’s field,
same dynamic equations describe free Dirac’s field if A , C are four-vectors,
and the electromagnetic field [2]. In the classical rotB 0 E 2 A
theory of the electromagnetic field, Lagrange’s
divB 2 f (19)
variables are the potentials A of the electromagnetic
field, whereas the tensor of the electromagnetic field rotE 0 B 2C
forms the four-vector and the potential A . The divE 2
authors now discuss the field with the mass term. In For the spinor field, introducing linear combination
that case the density of the Lagrangean is of spinor and
K , i
2 (13) I
1 1
i , II i (20)
where is a scalar constant. In order that this 2 2
1 1
must be a
Lagrangean be a relativistic scalar, or I II , I II
scalar as well. Using the Euler-Lagrange’s equations, 2 i 2
the authors obtain Lagrange’s equations of motion Eq. (19) become,
2 0, 2 0 (14)
i
I 0, i
II 0 (21)
According to the definition of the conjugated These are Dirac equations with eight -
momentum, the authors have: component functions and with positive and negative
iK , iK (15) mass terms. According to this, the authors conclude
Now, the density of the Hamiltonian, is that we come to the Dirac’s field described by the
H 0 0 Lagrangean Eq. (13) and both Eq. (21), if we remain
in the scope of the classical relativistic theory [2].
1
0 i i i i K 2
(16)
K 3. Conditions of the Maxwell’s Field
Canonical equations in covariant notation follow:
In the contemporary physics of the electromagnetic
K 2 0
fields, Maxwell’s operator plays a very important role.
1
0 (17) The question arises, which physical objects are
K formulated by the same or analogous equations of
1 0
0 physics. We shall discuss this question in this section.
K Maxwell’s system of equations of the free field, we
0 K 2
0
can write in the matrix form [3]
78 Relationship Between Classical and Quantum Physics
D 0 (22) only field that satisfies the Maxwell’s Eqs. (22) and
D (23) (23). The authors are interested in whether any other
where field exists that can be described by these equations.
D (24) For this purpose, the authors will seek for the
is called Maxwell’s operator, and
are matrices conditions of the relativistic invariance of the field
that satisfy conditions of the antisymmetry equations. The authors start from the Lorentz’s
, 2g (25)
transform given by Ref. [3]
ch -sh 0 0
1 0 0 ai
0
, i
x a x , a -sh ch
(26) 0 0 (30)
0 1 88 ai 0
0 0 1 0
0 0 0 1 0 0 1 0
0 0 0 1
0 0 1 0 0 0 0 1
a 1
, a 2
,
0 1 0 0 1 0 0 0 Corresponding field transform, the authors will
1 0 0 0 0 1 0 0 write in the form:
0 1 0 0 R , S , 1, R 1, S 1 (31)
1 0 0 0 (27) where R and S are operators of the transform for
a3
0 0 0 1 and , respectively.
0 0 1 0 Requirement for the relativistic invariance of the
Where and Φ are given in components: Lagrangean,
Ex x K R 0 R K
K (32)
E -A y gives R R
0 0
y (33)
Ez -A z whereas relativistic invariance of the Eq. (23) gives
F f R 1a S
,
0
, (28) (34)
Bx Cx
Therefore, the conditions of the relativistic
By Cy invariance of the “Maxwellian” Eqs. (22) and (23) are
B Cz given by Eqs. (33) and (3). One of the solutions of the
z
G - set of Eqs. (33) and (34) is
and F and G are scalars. We operate with the operator a b c d
D on the field functions , and obtain
R , S (35)
b a 88 d c 88
0 E rotB G 0 where
0 B rotE F 0 (29) ch 0 0 0
divE 0G 0 ch 0 0
a
divB 0 F 0 0 ch 0
Chosing scalars F, G such that F = G = 0, the 0 0 0 ch
authors obtain Maxwell’s equations for the free field. 0 0 0 sh
Thus, the authors have shown that we can describe
0 0 sh 0
electromagnetic field by Eq. (22) and Eq. (23). These b
0 sh 0 0
equations are called “Maxwell’s equations”. We
make the question: Is the electromagnetic field the sh 0 0 0
Relationship Between Classical and Quantum Physics 79
ch 0 0 0 1 0 0 0 0 0 0 0
0 ch 0 0 0 ch 0 0 0 0 sh 0
c
0 0 ch 0 0 0 ch 0 0 sh 0 0
0 0 0 ch 0 0 0 1 0 0 0 0
R
0 0 0 sh 0 0 0 0 1 0 0 0
0 0 sh 0 0 0 sh 0 0 ch 0 0
d (36)
0 sh 0 0 0 sh 0 0 0 0 ch 0
sh 0 0 0 0 0 0 0 0 0 0 1
we obtain
ε is a parameter confined by the condition:
R→1, S→1 for Ф→0 i.e. , 0 0 Ex Ex , Bx Bx
E y E y ch Bzsh, By By ch Ezsh
4. Some of the Solutions and Their
Ez Ezch By sh, Bz Bzch E y sh
Interpretation
F F , G G ' (38)
The authors will seek for the list of solutions of the In this case is consisted from the antisymmetric
Eqs. (33) and (34) for the values of the parameters tensor F and the scalars G and F . That means
1 1 that for the choice 1, , we have two-vector
, 1, , 0, , 1.
fiel A , C , or tensor-scalar field F , G , F .
2 2
If we chose λ = 1, , operating with the For F G 0, C 0 this field is reduced into the
operator S on classic standard electromagnetic field.
ch 0 sh 1
0 0 0 0 0 If the authors chose , , than R S .
2 2
0 0 0 0 0 0 0 0
Therefore, and are of the same magnitude
0 0 0 0 0 0 0 0 in relation to the Lorentz’s transform. Operating with
0 0 0 ch sh 0 0 0 operator R on , the authors obtain:
S
0 0 0 sh ch 0 0 0
Ex Ex ch Gsh
0 0 0 0 0 0 0 0 2 2
0 0 0 0 0 0 0 0
Ez Ezch By sh
sh 0 0 0 0 0 0 ch 2 2
G Gch Exsh
the authors obtain 2 2
Ax Ax ch sh
Bz By ch Ezsh (39)
C x C x ch f sh 2 2
E y E y ch Bzsh
Ay Ay , C y C y (37) 2 2
Az Az , C z Cz F F ch Bxsh
2 2
ch Axsh, f f ch C xsh
Bz Bzch E y sh
In this case is consisted by four-vectors 2 2
( , A) and ( f , C ).
Operating with the operator R on ' ,
Bx Bx ch F sh
2 2
80 Relationship Between Classical and Quantum Physics
F F ch Bxsh, G G 'ch Exsh and the components j 1, 2,3 of the Maxwell’s
equations will be
Ax Ax , C x Cx (40)
1 B j
jkl El 0 (46)
Ay Ay ch C zsh xk c t
C y C y ch Azsh 1 E j
jkl Bl 0 (47)
Az Azch C y sh xk c t
B j
C z Czch Ay sh 0 (48)
x j
f f ,
E j
In this case A and C form antisymmetric tensor 0 (49)
x j
of the second rank, and f are scalars, and
E x , E y, Ez,G and B x , By , Bz, F form Substitute the Crammers’ complex vector
four-vectors analogous electrodynamics where j E j iB j in these equations, multiplying Eqs.
E,B
form antisymmetric tensor of the (48) and (49) by i jkl and thereafter, adding Eq. (46)
electromagnetic field. Dirac’s equation with Eq. (47), and Eq. (48) with Eq. (49), the authors
0, without mass term is 0 ,
iD obtain:
where is a field function with components l i j
jkl 0 (50)
Ex , E y , Ez , F , Bx , By , Bz , G xk c t
j
Operating with the operator D on the field function 0 (51)
Eq. (28), we obtain Eq. (29). For the values x j
F G 0 Eq. (29) are the Maxwell’s equations for Differentiate the Eq. (50) according to x j
the free field.
2 l i j
jkl 0 (52)
5. Description of the Maxwell’s Equations in x j xk c t x j
the Form of the Dirac’s Equation We will now try to bring the expression
We will show that Maxwell’s equations in the l i j
jkl 0 (53)
vaccum can be written in the form of the Dirac’s xk c t
equation for massless particles [4] into the form of the Dirac’s equation for leptons (43)
Relationship Between Classical and Quantum Physics 81
j iD I 0, iD II 0 (62)
i c i jkl i l (54)
t xk The authors will pass to the non-relativistic
The authors use the matrix Schrödinger’s field.
S k jl i jkl (55) The authors begin with II 0, I 0, properties
to obtain of the matrices Eqs. (25), (26) and (27), and the
authors will take that [5, 6]
j
i c Sk jl Pk l (56)
t I eikx 0
(63)
i.e., Maxwell’s equations in the form of the Dirac’s
equation. Matrix S k jl is 3 3, k 1, 2,3 Then the authors will have:
matrix i 0 i j a j ikx0
e 0 (64)
111 112 113 i j a j i 0
S1 jl i j1l i 211 212 213 i 0 i j a j 0 (65)
311 312 313 2 i 0 i j a j 0 (66)
0 0 0 0 0 0 In the non-relativistic limit:
S x i 0 0 1 0 0 i i 0 2 (67)
0 1 0 0 i 0 that
From the equation (66), we take for
0 0 i i
j a j (68)
S2 jl S y 0 0 0 2
i 0 0 Substitute in Eq. (33) and obtain
0 i 0 i
i 0 i j a j j a j
S3 jl S z i 0 0 (57) 2
0 0 0 1 1
2
j i a j ai ai a j
2
These matrices have the properties
(69)
Si , S j i ijk S k (58)
From the assymmetrical condition Eq. (25) the
S 2 S12 S22 S32 2 (59) authors have:
so that the authors have: 0 aj 0 ai 0 ai 0 aj
2 g ji
S 2 S S 1 2 (60) a j
0 a i
0 a i
0 a j
0
j a a a a
j i i j
2 ij
(70)
i.e. S 1 i c S k jl Pk l (61)
t
a j a i a i a j 2 ij
and the Eq. (61) describes the particle with a spin
S 1. But, this particle is the quant of the Using Eq. (70), the Eq. (69) takes the form
electromagnetic field―a photon. 1
i 0 (71)
2
6. Non-Relativistic Boundary and the
which is the Schrödinger’s equation for the free
Schrödinger’s Equation
particle. Now we take the case when the particle is
From the Dirac’s equations with the positive and under the influence of the external electromagnetic
negative mass term field. The motion of a free relativistic particle of mass
82 Relationship Between Classical and Quantum Physics
m and charge q is described by the Dirac’s equation where E E mc 2 , i , j 2i ijk k and
[7-9] 2
p p p I .
i ic mc 2 (72)
t
where the 3 matrices i and the matrix are given 7. Conclusions
by
Investigation carried out in this work shows that the
0 i I 0
i i and (73) free electromagnetic field, spinor Dirac’s field without
0 0 I mass, spinor Dirac’s field with mass, and some other
are the Pauli spin matrices, and I is the unit 2 2
i,
fields are described by the same variational
matrix. Describe how the Dirac’s equation should be formulation. The conditions that a field be described
modified if particle comes under the influence of an by Maxwell’s equations of motion are given in this
external electromagnetic field described by the work, and some solutions of these conditions are also
four-potential A A , A . Hence deduce that the
0
given. Non-relativistic approximation of the equations
Dirac equation for a particle of charge q in the presence of the non-quantified field are the Schrödinger’s
of an electrostatic central potential V r qA , with
0
equations. Dirac’s equation as a special case, contains
A 0, may be written as Maxwell’s equations and the Schrödinger’s equation.
Hˆ where Hˆ ic mc V r (74)
2
i The authors conclude that the basis of the quantum
t
By seeking energy eigenstates of the form physics lie on the classical physics, including special
iEt theory of relativity.
r , t r exp (75)
References
and writing the Dirac wave function r in terms
of two-component spinors [1] K. Ljolje, S. Vobornik, Canonical formulations of fields
r
satisfying the D'Alambert's equation, Fizika 11 (4) 171
r (76) (1979) 171-180.
r [2] K. Ljolje, Some remarks on variational formulations of
show that the Dirac equation may be written in the physical fields, Fortschr. Phys. 36 (1) 1988 9-32.
[3] S. Ahmetaj, Physical Fields Described by Maxwell`s
form
E V mc r c p r 0
Equations © 2007: American Institute of Physics.
2
[4] D. Bjorken, S.D. Drell, Relativistic Quantum Field,
E V mc r c p r 0 (77)
McGraw-Hill, New York, 1965.
2
[5] E.D Hoker, D. Gieser, Quantum Field Theory, Los
Angeles, 2004
Show that for a weak potential in the [6] P.J. Mulders, Introduction to Quantum Field Theory,
non-relativistic limit, these equations are reduced to an Amsterdam, 2004
energy eigenvalue equation of Schrodinger form [7] R. Casalbuoni, Advanced Quantum Field Theory, Firenze,
1999.
ˆ 2
Received: January 14, 2012 / Accepted: February 12, 2013 / Published: February 25, 2013.
Abstract: The authors present several oscillation theorems for differential equation of second order
(r (t ) g ( ( x(t )) x' (t ))' q (t ) f ( x(t )) 0 (1)
and for differential equation with damping term
Mx' ' (t ) p (t ) x' (t ) q (t ) x(t ) 0 (2)
where M > 0, r(t) is positive continuous function. The conclusion is based also on building function where coefficients are involved in
the equation and positive functions used by Philo H (t , s ) and averaging techniques. Our results generalized, extend to some already
known oscillation criteria in the literature. Also, here we give some applications of oscillation solution of . (1) and (2), where p(t) and
q(t) are positive. The original purposes of differential equation are the mathematical formulation of the vibration frequency and the
amplitude profile of a vibrating string with friction which the mass may have to encounter air resistance in its motion and in electric
circuit containing an ac voltage source, an indicator, a capacitor, and a resistor in series is analyzed mathematically, the equation that
results is a second order linear differential equation with oscillatory solution.
Denote
2. Main Result
1
(r (t1 ) g ( ( x(t1 )) x' (t1 )) 2
c
The nature of solutions of second order differential 2
equations is studied by many authors. Here, the authors Now by (7) it follows that:
give some conditions for coefficients where the (2) t
r (t )q(t ) F ( x(t ) c (r ( s )q( s ))' F ( x( s ))ds
have bounded solutions. Taking in account the result
t1
the authors have obtained from previous researches , t
(r ( s )q ( s ))'
here the authors present more generalized criteria that c r ( s )q( s )F ( x( s ))ds (8)
t1
r ( s)q( s)
define oscillation solution of (1) and (2) that can be
applied in dynamics , electricity etc. Hence, by Granwalls-it inequality, the authors get
t
( r ( s ) q ( s ))'
Before the authors prove the theorems, the authors r (s)q(s)
ds
H (t , s )
f (l ) g (l )dl , (6) and
s
h1 (t , s ) H (t , s ) .
then the solution x(t ) of (1) such that The authors define certain conditions for the
x(t1 ) 0, t1 t 0 for some t1 [t 0 , ) is bounded. coefficients of (1) using positive function of Philo,
Proof Let x(t ) be a arbitrary solution of (1) such
that x(t1 ) 0, t1 t 0 . when the solution is oscillatory.
Oscillation Theorems for Second Order Differential Equations and Their Applications 85
Theorem 2.3 Let exist constant k, m, M, such that: exists and c (a, b) , such that
f ' , ' k 0 , 0 m g (t ) M (11) w' (t ) A0 (t ) A1 (t ) w 2 (t ) (16)
and then
t
1 h2 (t , s ) r ( s ) M
2 H 1 ( s, a )h1 ( s, a )
c 2
lim sup ( q ( s ) H (t , s ) ) ds = (12) 1
( H (s, a) A0 (s)
)ds
H (c , a ) a
2
t H (t , t ) 0 t0
4k 4 A1 ( s )
2 H 1 (b, s )h2 (b, s )
c 2
then (1) is oscillatory . 1 (17)
H (b, c ) a
( H (b, s ) A0 ( s )
4 A1 ( s )
)ds 0
Proof Assumed that x(t ) is a nonoscillatory
solution of (1). Then exist t 0 , such that x(t ) 0 , for Proof If multiplying (16) with H (t , s) 0 and
t t0 . integrate from t to c, the authors have
c
H
The authors define ( s, t ) w' ( s )ds
r (t ) g ( ( x(t )) x' (t ) t
w(t ) , for t t 0 (13)
f ( ( x(t ))
c c
2 H 1 ( s, t )h1 ( s, t )
c 2
t 2 2
w (t , s )k H (t , s ) ds
ds , 4 A1 ( s )
t
rM t
0
q( s) H (t , s)ds H (t , t0 )w(t0 ) 2 H 1 ( s, a ) h1 ( s, a )
t0 c
1
H
( s, a ) A0 ( s ) ds w(c ) (18)
2 H
(c, a ) 4
wk H (t , s ) h2 (t , s ) rM
t a A1 ( s )
ds Similar to this, if (16) multiplies with H (t , s ) 0
0
4k 2 hence
t
thus by (15) it follows that
H
t 2
( s, t ) A0 ( s )ds w(c) H (t , c)
1 h (t , s )r ( s ) M
lim sup
t H (t , t 0 ) t
(q( s ) H (t , s ) 2
4k 2
)ds
c
1
2
0 t H 2 (t , s ) h (t , s )
H (t , s) A1 (s) w(s) 2
is bounded , that is the contradiction with condition 2
ds
c A1 ( s )
given in the theorem.
Lemma 2.4 Let A0 , A1 C ([t 0 , ), ) , where c 2 1
H (t , s )h2 (t , s )
2
ds
A1 0 and w C 1 ([t 0 , ), ) . If (a, b) [t 0 , ) t
4 A1 ( s )
86 Oscillation Theorems for Second Order Differential Equations and Their Applications
p 2 (t ) p ' (t )
if t b and dividing with H , the authors get w' (t ) w 2 (t ) Kq(t ) 0
4M 2M
b 1 the rest of the proof is similar.
H (b, c) A ( s ) H (b, c)h1 (b, c) ds w(c)
2
1
H (b, c) c
0
4 A1 ( s ) (19) Application For instance, the mass may have to
If (18) and (19) are added, the authors obtain (17). encounter air resistance in its motion. A typical
Based on the relation of Hile, for (2) can be defined physical assumption is that the frictional force points
oscillation nature of solutions like as follows: opposite to the direction of motion and has magnitude
f ( x) proportional to the speed. If there is no external forces,
Theorem 2.5 Let q(t ) 0 , and K 0 , for
x this leads to the differential equation
x 0 , where K is constant. If for large t , mu ' 'u ' ku 0 (21)
p (t ) O(1) and where u ' is magnitude of frictional force, with
characteristic equation
K p 2 ( s) p' ( s )
[ M
q( s )
4M 2
2M
]ds 0
t
(20)
mr 2 r k 0
t
[7] J. Ohriska, A. Zulova, Oscillation critweria for second [9] Ravi P. Agarval, Said R. Grace, Donal O’Regan,
order non-linear differential equation, Institute of Oscillation Theory for second order linear, half-linear
mathematics, Safarik University, Jesenna, 10 (2004) 1-10. superlinear and sublinear dynamic equation (2002) Florida
[8] R. Feynman, The Feynman Lectures on Physics, Pearson, institute of technology.
California, 1989.
Journal of Mathematics and System Science 3 (2013) 89-95 D DAVID PUBLISHING
Received: December 19, 2012 / Accepted: January 11, 2013 / Published: February 25, 2013.
Abstract: In this article, the true model of size p is considered. The importance of the independent variables will be studied. The model
by dropping independent variable one at a time is called a reduced model, and the size of the model is p-1. The non-centrality quantity
is employed to measure the strength of influence on the true model for the independent variable dropped. The larger influence is defined
to be more important in the true model. In this article, the goal is ranking the importance of the independent variables to study the
structure of the models. It is important in the practical application. The confidence interval approach is used to rank the degrees of the
influence of the independent variables in linear models and achieve the goal. An illustrative example is given, and the modeling
procedure is studied to check the assumptions step by step in this example to make sure the assumptions satisfied in the true model. As
a result, the proposed method can be used efficiently.
SS p
matrix of independent variables of rank p, model assumption, it is shown that and
2
0 ,.1,..., p 1 ' is a p 1 SS p1,m
n
2
p and n ( p 1),
vector of are distributed as '2
,
2 p 1, m
denotes the n n identity matrix. Model Eq. (1) is chi-square distribution on degrees of freedom with
non-centrality ( =0). Thus, E SS p ( n p )
referred as the true model of size p. From this model, 2
independent variables.
SS 1 2 2 2
E p 1, m 1 p 1, m . It Lemma 1.1 see Ref. [1]: Let Y have the
n p n p n p non-central F distribution with u, v degrees of freedom,
shows that the noncentrality p 1, m will affect the and non-centrality , denoted by Fu ,v , . Then
prediction error in reduced model. PY y [1 e 1.794148 x ]1 (2)
Note that Q (m ) , m = 1, 2, …, p-1, are idempotent where
and symmetric, thus they are positive semi-definite. 1
2(u 2 ) 2 uy
3
p 1, m can be interpreted as the contribution of X m
in making significant given that other variables are
9(u ) 2
9v u
already in the model. the maximum absolute approximation error is 0.05
By Taylor’s expansion, for u 1, v 5 and 0 .
SS p 1,m E SS p 1,m (n p 1) 2 2 2 p 1,m Now, consider the problem of the confidence
interval estimates of p 1 , p 1 p 1,1 ,..., p 1, p 1 .
E
SS p E SS p (n p) 2
Let p 1,[1] p 1,[ 2 ] ... p 1,[ p 1] denote the
(n p 1) 2 p 1,m
ordered p 1, m s . Require that, for each p 1, m with
'
(n p)
the confidence coefficient 1 , 0 1
thus
satisfying:
n p SS p 1, m n p 1 inf P c2 ,m ˆ p 1, m p 1, m c1, m , m 1,2,., , , p 1 1 0 (3)
E p 1, m
2 SS p 2
Where 0 are given.
p 1
n p SS p 1,m n p 1
Use ˆ p 1,m to Theorem 1.2 For each p 1, m with the confidence
2 SS p 2 coefficient 1 , 0 1 , satisfying
estimate the parameter p 1, m .
inf P ˆp 1, m c1m p 1, m ˆp 1, m c2 m 1
Rewrite ˆ p 1,m as follows:
where
16C 2 2 4C 2
3
1 SS SS p 1 1 1 2A A 4C (1 A) 1
2
ˆp 1, m p 1, m ˆm 1 u H
9u 9u
1
2 SS p 2 2 2 2 u
c1m max
2( A2 C 2 (1 A))
(n p)
SS p 1.m SS p 1 u H 1
3
where ˆ m .
2 u
1
A
SS p
(n p) 3
1 E E2 4DF where
p 1 p 1,1 , p 1, 2 ,..., p 1, p 1 T . The
c2m 1
Let 2 2D
statistic ̂m has the non-central F distribution on 1 and 2 2
u 1 , n p , 1 A , 1 B
n-p degrees of freedom with non-centrality parameter 9 9u
p 1, m , denoted by F1, n p , p 1, m . Note that the ̂ m is ,
C
1
ln D A2 C 2 1 A
the statistic used to evaluate the importance of the 1.794148 2
Ranking Independent Variables in Linear Models 91
2(u ) 2 p 1, m 2 2 m
inf P ˆp 1, m c1m p 1, m ˆp 1, m c2 m 1 (4)
9(u )
2
9v u
Let
P ˆp 1, m p 1, m c1m 1
2
(5) or
2 p 1 , m 1 2 c 2 m 3 1
2 1
2
and 1
9v 9u
P ˆp 1, m p 1, m c2 m
(6)
1
(10)
9 u 9 v 2 p 1, m 1 2 c 2 m 3
2 2 2 2
2
From Eqs. (5) and (6), it will be satisfied Eq. (4) as
1
follows: ln
1 . 794148 2
inf P c ˆ 2m, c p 1, m p 1 m 1m C2
Let u 1 , n p , where
inf P ˆ p 1, m p 1 m , c ,1 m 36
1
sup P ˆ p 1, m , c2 m
p 1 m C ln
1.794148 2
F B 2 C 2 1 B 0 1
P ˆ p 1, m 2 p 1, m c1m
2
Thus E 4 DF 0 .
2
1
From Eq. (11) P p1,m H ˆ p 1, m 2 c1m
2
E E 2 4 DF E E 2 4 DF
Y or Y 1.794148 x (p1)1 , m 1
2D 2D [1 e ]
1 1
And AYB0 implies A 2 p1,m 1 2c2m 3 B
2 u 1 2c1m 2(u 2 H )
3
The following results are obtained. 1 1 2
9v u H 9(u H )
3
1 E E 2 4 DF 1
p 1, m 1 2c2 m 2
2(u 2 H ) 2 u 1 2c1m
2
2 2D
3
3
9(u H ) 2 9v u H
or 1 E E 2 4 DF
1 2 c2m
p 1, m
2 2D 1
2 u 1 2c1m 3
1B 1
3
1 1
and
p 1, m c2 m 9v uH
2 A 2 1
Let 2
2(u 2 H ) 2 u 1 2c1m
2
3
3
1 E E 2 4DF 1 B 1
3
c2m max 1 , c2 m
2
2 D
2 A 2
(13) 9(u H ) 2 9v u H
3
1 E E 2 4DF 1
1
2 2D 2 u 1 2c1m 3
1 1
be guaranteed the positive of p 1, m . Since 9v u H
1
E E 2 4 DF
2 2
4 2 u 1 2c1m
3
2D
E E 2 4 DF
9u 9v u H
2 A2 1 2
ln (15)
E E 2
1.794148
2 A2
Let 1 A, , C
2 1
2E E 2B B ln
2 9v 1.794148 2
2 A2 A A A
Thus u(1 2c1m )
Let X
E E 4 DF 2
B uH
0 From Eq. (15)
2D A
Thus 1
2 u 1 2c1m 3
3
1 1
1 E
p 1, m
E 2 4 DF 1 2 c 2 m (14) 9v u H 1 2
ln
2
2D
1
1. 794148
2 2
4 2 u 1 2c1m
3
H 9u 9v u H
From Eq. (7)
Ranking Independent Variables in Linear Models 93
1
AX 3 1 1 u H 1
3
C and c1m 1
2 u
1
4
2 2 A
1 A X
3
9u Thus
3
1 1 16C 2 2 4C 2 (16)
2A A 4C (1 A) 1
1
2
1 u H
Since C 0 and AX 1 0 , thus X 9u
3 3 9u
. 1
2 u 2( A2 C 2 (1 A))
A c1m max
1
4 1 u H
3
1 2 2 1
AX 1 C 1 A X 3 2 u 1
3
A
9u
From Eqs. (13) and (16), the confidence interval of
2 1
4 2
p 1,m is as follows:
A2 X 3 2 AX 3 1 C 2 1 AX 3 the parameter
2
9u
1
inf P ˆp 1, m c1, m p 1, m ˆp 1, m c2 m 1
p 1,m
A
2
4C
2
C 2 (1 A) X 2 AX 1
3 3
0 as an interval estimate of p 1,m with confidence
9u
coefficient 1 .
4C 2
1
2 A 4 A 2
4 A 2
C 2
1 A 1
9
2. Example
u
X3 2 2
2 A C 1 A In the following example, the data were originally
3 reported in Ref. [2] for assessing the effects of air
2
2 A 4 A2 4A2 C 2 1 A1 4C pollution on the median housing value (medv).
9u
The dependent variable is Median value of owner
X
2A2 C 2 1 A homes. The considered independent variables are as
follows:
3 Crim: Per capita crime rate;
2
2 A 16C A2 4C 2 1 A1 4C
2
Zn: Proportion of residential land zoned 25K+;
9u 9u
Indus: Proportion of non-retail bus acres per town;
2A2 C 2 1 A Chas: Charles River connection dummy;
nox: Nitric oxides concentration pp 10M;
1 3 rm: Average # of rooms per dwelling;
and X ( ) . age: Proportion of owner-occ dwellings before
A
1940;
1
71.794748
dis: Wtd dist to five Boston employment ctrs;
when 0 1 1 e 3 2
2
and n p C
2 36 rad: Access index to radial hwys;
tax: Full-value prop-tax rate per $10K;
then A C 1 A 0
2 2
ptratio: Pupil-teacher ratio by town;
16C 2 2 4C 2 b: Proportion of blacks per town;
then A 4C 2 1 A1 0
9u 9u lstat: % lower status of the population.
There are 506 observations. The data was used by
3
2A 16C 2 2 4C 2 the classification and regression trees (CART) methods
A 4C 2 (1 A)1
1u H 9u 9u to integrate the interaction into the linear regression
c1m 1
2 u 2( A C (1 A))
2 2
models. There are eight branches in the CART tree.
The correspondence eight levels are defined as follows:
94 Ranking Independent Variables in Linear Models
Level 1: rm ≤ 6.941, lstat ≤ 14.4, dis ≤ 1.38; (4) Testing the equality of the variances of the
Level 2: rm ≤ 6.941, lstat ≤ 14.4, dis > 1.38 residuals.
Level 3: rm ≤ 6.941, lstat > 14.4, crim ≤ 6.99 H 0 : 12 22 ...... k2 2
Level 4: rm ≤ 6.941, lstat > 14.4, crim > 6.99 The systematic sampling method is used to separate
Level 5: rm > 6.941, rm ≤ 7.437, nox ≤ 0.66 residuals into five groups. Here, the size of these
Level 6: rm > 6.941, rm ≤ 7.437, nox > 0.66 groups is n1 102, n2 n3 n4 n5 101.
Level 7: rm > 6.941, rm > 7.437, nox ≤ 0.68 Let X i 1 , X i 2 , , X in , i 1, 2, , k , be k
i
Level 8: rm > 6.941, rm > 7.437, nox > 0.68 independent random samples from k normal;
The modeling procedures are as follows: populations
N i , i2 , i 1, 2, , k ,
(1) Transform the dependent variable 2
j 1
distribution under each fixed combination of ,
between the variance 2
i and all other 2j s , and
independent variables. Let y medv 0.134912 . Use
si2
CART (the classification and regression trees) to find ˆi k 1
as an estimator of i , where
the interaction between the explanatory variables. Each ( s ) 2
j
k
1 ni 1 ni
of sub-sample, class 5 and class 6 are merged into one
class. Class 7 and class 8 are also merged into one class.
si2
ni 1 j 1
( X ij X i ) 2 and X i
ni
X
j 1
ij ,
hypotheses is accepted.
( 12 , , k2 ) [ k ]
2
Ranking Independent Variables in Linear Models 95
Received: January 3, 2013 / Accepted: February 11, 2013 / Published: February 25, 2013.
Abstract: The authors consider relational databases organized over an ordered domain with some additional relations – a typical
example is the ordered domain of rational numbers together with the operation of addition. In the focus of our study are the FO
(first-order) queries that are invariant under order-preserving permutations–such queries are called order-generic. It was discovered that
for some domains order-generic FO queries fail to express more than pure order queries. The collapse result theorem was proved for
locally generic queries over a linearly ordered domain with the Pseudo finite Homogeneity Property (or / and the Isolation Property) by
Belegradek et al.. Here the authors consider a circularly ordered domain and prove the collapse result theorem over a quasi circularly
minimal domain.
non-empty set with the mapping assigning to each Then an interpretation of this unary relation in such a
relational symbol in L the relation of the same arity system is called a pseudo finite set.
over this set. Let M be an infinite structure of Definition 2.4 The authors say that a complete
signature L. Here the authors consider circularly theory Т has the Isolation Property if there is a cardinal
ordered structures. This means that L includes a ternary such that for any pseudo-finite set A and for any
relational symbol K of which the interpretation in M element a of a model of the theory Т there exists A 0
satisfies to axioms of the circular order. The authors fix A such that |A 0 | < and tp ( a / A 0) isolates tp ( a /
the circuit of database SC and enter the following A).
notations: Definition 2.5 Let A M where M is a circularly
L 0 = {K}, L’ = L 0 SC, L’’ = L SC ordered structure. The set A is convex if for any
A query of a database can be formally determined as elements a, b A the following holds: for any
a mapping which is accepted by the state of a database element c M such that K(a, c, b) the authors have
and it makes a new relation of fixed arity over M . that c A or for any element c M such that
The authors consider two languages for querying. K(b, c, a) the authors have c A.
Queries of the first language are formulas of signature Definition 2.6 Let A, B be convex disjoint subsets of
L’–the authors name them by limited. Queries of the a circularly ordered structure M . The authors say that
second language are formulas of signature L’’ – the A is the immediate predecessor of the set B (denote this
authors name them by expanded. In Ref. [7] the authors by A = IP(B)) or В is the immediate successor of the set
have proved reducibility of expanded queries to limited А (denote this by В = IS(А)) if for any a A, b B and
ones over an 0-categorical quite o-minimal domain. c M such that K(a, c, b) the authors have that c A
Here the authors prove reducibility of expanded B.
queries to limited ones over a circularly ordered Notation 2.7 (1) Let M = M, =, <, … be a linearly
domain. ordered structure. The authors denote by c(M) the
Recall the necessary definitions. structure M* = M, =, K3, … in which the linear order
Definition 2.2 A k-аry query is locally generic < is replaced by the ternary relation K3 as follows: for
over finite states if a if and only if ( a ) every a, b, c M K(a, b, c) a b c b c a
((s)) for every partial < -isomorphism : X M, cab
where X M, for any finite state s over X and for any (2) Let M1, M2, …, Mm be linear orderings. If the
k-tuple a in X. authors replace < by K3 as above, the authors denote
A state s expands an universe M of the signature by c(M1, …, Mm) the circularly ordered sum M1, …,
L to a L -structure which the authors denote by Mm such that for every 1 i m–1 Mi = IP(Mi + 1) and
( M , s ). Mm = IP(M1).
Definition 2.3 A -state for a L -structure W (3) K0(x, y, z):= K(x, y, z) y x y z x z .
is pseudo finite in W if (W , s ) is a model of the Definition 2.8 Let I M where M is a circularly
first order L -theory of all (W , r ) where r is a ordered structure. The set I is called an open interval if
finite state over W . I = {c M: M K0(a, c, b)} for some elements a, b
A pseudo finite set is a special case of a pseudo M.
finite state. A signature consisting of one unary If I is an open interval, the authors will sometimes
relation and some other relations, and such systems of write I = (a, b) if the authors wish to show the
this signature in which all closed formulas true in all endpoints of I. Similarly, the authors can define closed,
finite systems of the signature hold are considered. semi-open-semi-closed and etc. intervals in M . By an
98 On the Isolation Property Over a Database Domain
interval in M the authors will understand any of the M, =, K3, … is called weakly circularly minimal if
above called types of intervals in M . Obviously, both every definable (with parameters) subset of the
an interval and a singleton are convex sets. structure M is an union of finitely many convex sets.
The following notion has been originally introduced Weakly circularly minimal structures have been
by Dugald Macpherson and Charles Steinhorn in Ref. originally studied in Ref. [14].
[8]. They have described circularly ordered groups Example 2.13 Let M = c(Q1, Q2, Q3, Q4), =, K3,
that are circularly minimal. P1, where Qi is a copy of the ordering of rational
Definition 2.9 [8] A circularly ordered structure M numbers for each i 4, and P(M) = Q1 Q3.
= M, =, K3, … is called circularly minimal if any It can be proved that M is a weakly circularly
definable (with parameters) subset of the structure minimal structure. It is easy to see that P(M) is an union
M is an union of finitely many intervals and points of two convex sets that are neither intervals nor points.
in M . Therefore, M is not circularly minimal. The authors
The notion of circular minimality is a variant of can also observe that P(M) consists of two convex sets
o-minimality. O-minimality has been developed in each of which is not -definable. This contrasts with
many works [9-13]. the weakly o-minimal case, where every parametrically
Example 2.10 Let G = {z C | |z| = 1}, where C is definable set is a union of finitely many definable with
the set of complex numbers. Consider the structure G the same parameters convex sets.
= G, =, *, K3, where * is a binary operation being
3. Results
the multiplication on complex numbers.
Obviously, G, * is a group. It isn’t difficult to Generalizing the notion of circular minimality, the
prove that G is circularly minimal. authors call a circularly ordered structure M quasi
Example 2.11 Let M = c( + *, Q, + *, Q), circularly minimal if every definable subset of the
=, K3, where is the order of natural numbers, * is structure is a finite union of points, intervals and
the inverse order on natural numbers, and Q is the -definable sets. A complete theory T is said to be
order of rational numbers. quasi circularly minimal if every its model is quasi
Obviously that M is a circularly minimal circularly minimal.
structure. Denote the first elements of these two There are quasi circularly minimal theories that are
orderings that are equivalent to + * by a and c, not circularly minimal. The simplest example is the
and the last elements of these orderings by b and d. theory T of densely ordered sets without endpoints
Consider the following formula: with a distinguished subset that is dense in the universe,
(x) := y z [K0(y, x, z) t1 t2 (K0(y, t1, x) and its complement is also dense in the universe. It can
K0(x, t2, z))] be showed that T is the theory of the structure
Obviously that (x) (K0(b, x, c) K0(d, x, a)), i.e. R, , K 3 , Q is quasi circularly minimal. It is not
(M) is an union of two open intervals, but the authors circularly minimal since the distinguished subset is not
see that the endpoints of these intervals are not a finite union of intervals and points.
definable over : a, b, c, d acl() \ dcl(). This is The authors prove the following theorem:
in contrast with the o-minimal case where acl(A) = Theorem 3.1 Any quasi circularly minimal theory
dcl(A) for all A M. has the Isolation Property.
The following notion has been introduced jointly Proof Let A be a pseudo finite set in a model M
with D. Macpherson: of a quasi circularly minimal theory T , and a M .
Definition 2.12 A circularly ordered structure M = Show that there exists A0 A such that
On the Isolation Property Over a Database Domain 99
Mohamed Abdelwahed
Department of Mathematics, College of Science, King Saud University, Riyadh 11451, Kingdom of Saudi Arabia
Received: January 21, 2013 / Accepted: February 11, 2013 / Published: February 25, 2013.
Abstract: The authors propose a numerical algorithm for the two-dimensional Navier-Stokes equations written in stream
function-vorticity formulation. The total time derivative term is treated with a first order characteristics method. The space
approximation is based on a piecewise continuous finite element method. The proposed algorithm is used to simulate the mechanical
aeration process in lakes. Such process is used to combat the degradation of the water quality due to the eutrophication phenomena. For
this application high computing facilities and capacities are required. In order to optimize the computing time and make possible the
simulation of real applications, the authors propose a parallel implementation of the numerical algorithm. The parallelization technique
is performed using the Message Passing Interface. The efficiency of the proposed numerical algorithm is illustrated by some numerical
results.
Key words: Navier-Stokes equations, stream function, vorticity, high performance computing.
wind
formulation has two main advantages. The first one is
related to the automatic satisfaction of the divergence
free condition. The second one concerns the reduction
air bubbles of the equations number. Using Eq. (2), from Eq. (1)
injector
the authors obtain the following system:
Fig. 1 Aeration process.
Γs
t u v f in [0,T ]
= 0 in [0,T ] (3)
Ω = d on [0,T ]
Γinj
g on [0,T ]
Γf
n
Fig. 2 Domain. Here the boundary data ψd and g are calculated using
ud and the relations Eq. (2).
boundary). Where uwind is the wind velocity and uinj is The second alternative concerns the reduction of the
the injector velocity. computing time through the parallel implementation of
For such application high computing facilities and the numerical algorithm using MPI (Message Passing
capacities are required [7]. In fact, let us consider, for Interface) [11].
example, only a real two dimensional simulation. The The paper is organized as follows. In Section 2, the
domain is a two dimensional cross section of a lake authors introduce the numerical analysis of the used
whose average dimensions can reach 500 m × 20 m, model. The time discretization scheme is presented in
the injector length is about 10m with more than 100 Section 2.1. The used finite element method for the
holes of 1 cm for its diameter. In order to detect the space approximation is described in Section 2.2. In
bubbles effect, it is necessary to make a mesh around Section 3, the authors illustrate the parallel algorithm
the injector area with elements having a size of the and the authors present some numerical results
order of bubbles volume: this should generate a mesh showing the efficiency of the proposed method. In
of the order of 10 million nodes. To have an idea of the Section 4 the authors deal with a real application. The
difficulty, a simulation for T = 10 mn with a mesh of authors treat the aeration process using the obtained
5.105 nodes, requires approximately two days on a parallelized finite element algorithm. The paper ends
good computer. with some concluding comments in Section 5.
In order to reduce the computing time and make
possible the numerical simulation of the dynamic 2. Numerical Analysis
aeration process with more fine meshes, the authors The authors present in this section a numerical
propose two alternatives. analysis of the system Eq. (4). The time discretization
The first one consists in reducing the number of scheme is based on the characteristics method [12, 13].
equations by using the the steam function―vorticity The space approximation is performed using the
formulation [8-10]. The stream function ψ and the regularization-stabilization technique introduced in
vorticity ω are linked to the velocity field u by the Refs. [14, 15] for the biharmonic problem.
relations:
2.1 Time Discretization
u and u (2)
The formulation is more suitable compared Let ψn+1 and ωn+1 being the approximations of ψ and
with primitive variables (velocity and pressure) for ω at time tn+1 = (n + 1) t , where t is the time step.
representing the kinematic nature of the flow. Such a The time discretization of the problem Eq. (4) leads to
A Parallel Finite Element Algorithm for Navier-Stokes Equations 103
Ch, there exist two triangles K and K’ of Th such that K represents the jump of the normal derivative on each
edge T of Ch, where vK = v|K and n vK is the normal
K
≠ K’ and T = ∂K ∩ ∂K’.
The authors consider the discrete spaces: derivative of vK on ∂K.
Yh { C 0 (); K Th , |K P1 ( K )} The authors obtain the following discrete
formulation:
Yh0 Yh H 01 ()
a (h , h ) b( h , h ) g h d
Where P1(K) is the space of piecewise linear functions
defined on K . ( h ) Yh
The authors denote by ωh, ψh and Fh respectively the
approximations of ω, ψ and F in Yh. The classical b(h ,h ) d ( h ,h ) Fh h d (13)
h Yh
0
discrete variational formulation of the problem Eq. (10)
is written as:
h d Yh , h Yh
0
d ( h ,h ) h h d ( h ,h ) Yh Yh0 results concerning this technique are described in Ref.
[15] for the bi-harmonic problem and in Ref. [16] for
The authors remark that the compatibility constant
0 the quasi-Stokes equations.
between spaces Yh and Yh is independent on h but the
associated bilinear form a(.,.) to Yh is elliptic for the 2.3. Linear System
norm of L2(Ω) and consequently the coercivity Let n the number of mesh nodes, (θi) i = 1, n be a base
constant depends on h. To make up for such 0
of Yh and (ηi) i = 1, n a base of Yh . The system Eq. (13)
inconvenient, the authors use an approximation method is written for all j = 1, …, n:
based on a technique of stabilization described in Refs. n n
F d
Ah ( h , h ) mes(T ) [ h h ][ h h ]dT h j
T
T Ch which is equivalent to the following linear system :
( h , h ) Yh Yh MX = N (15)
with where
*The matrix A = (ai, j)1 ≤ i, j ≤n with ai, j = a (θi, θj) is to distribute the lines of the matrix between the
computed from the bilinear form a(.,.). processors. Each processor contains the total vector X
*The matrix S = (si, j)1 ≤ i, j ≤n with si, j = Ah (θi, θj) and a block Mi of lines of the matrix (Fig. 3).
represents the bilinear form Ah (.,.). Each processor carries out locally the product of a
* The matrix B = (bi, j)1 ≤ i, j ≤n block Mi (compound of a some number of lines of the
with bi, j = b(ηi, ηj). matrix M) and of the vector X. The result is a vector
* D = -λB. (MX)i of dimension the number of lines of the block Mi.
* G and F are associated to the second member with After that, a synchronous communication is used for
the update of the total vectors after each product
G j g j d and Fj Fh j d .
matrix-vector. No communications are needed in the
For the storage of these matrices, the authors used local computing of each processor.
the Morse compact method keeping only non-zero 3.1.2 Scalar Product
terms. For the resolution, the authors used the Each processor computes a scalar partial product of
BICGSTAB (bi-conjugate gradient stabilized) two corresponding local vectors by multiplying these
algorithm [17. two vectors and by adding the components of the
partial result. Then each processor sends to all
3. Parallelization
processors its partial result and receives from each the
The goal of the parallel implementation is to make partial resulting product which is added to its own
possible the simulation of real application cases, result. Consequently, a total communication is
needing high performance computing, in reasonable necessary in this scalar product.
computing time. The authors can ideally hope to divide 3.1.3 Linear Operations
the real time (Elapsed time) by the number of used Each processor contains one part of each total vector
processors. The authors present in the following a considered as a local vector. It is possible to make a
parallelization of the solver. The implementation was linear combination of two local vectors; the obtained
done with Fortran and MPI library for communications vector result is a local vector. Consequently, no
[11]. communication is needed for this operation.
Some carried out numerical tests on one-processor The most important part of overhead is generally
machine showed that the resolution part of the linear resulting from the exchanges of messages between
system Eq. (15) represents the major part of the total processors (send/receive). For that, one starts with the
CPU (computing) time of the code. Consequently the distribution of all the necessary data to be used by
authors focused on the parallel treatment of this part of M X MX
the solver which uses the BICGSTAB algorithm. In M1 MX 1
fact at each iteration step, the following operations are
M2 MX 2
down.
3.1.1 Matrix-Vector Product M3 MX 3
minimize computing time. The strategy used consists Fig. 3 Matrix-vector product.
106 A Parallel Finite Element Algorithm for Navier-Stokes Equations
architecture. 0
1 2 3 4
Speed−up
two processors. Concerning speed-up gotten with four 8
authors increase the number of unknowns (mesh 4), the Fig. 5 Speed-up.
authors can reach speed-up near to its optimal value on 100
00
11 00
11 00
11
00
11
00
11
50
00
11
00
11 00
11 00
11
vary the mesh, the efficiency raises. Indeed, when the 40
00
11 00
11 00
11
00
11 00
11 00
11
00
11
authors raise the problem size, the speed-up improves 11
00 00
11 00
11 00
11
00
11
30
00
11 00
11 00
11 00
11
00
11
00
11 00
11 00
11 00
11
whatever is the number of processors, what entails an 00
11
20
00
11 00
11 00
11 00
11
00
11 00
11 00
11 00
11
00
11
00
11
10
improvement of the efficiency. But, when the authors 00
11
00
11 00
11 00
11 00
11
0 00
11 00
11 00
11
fix the mesh (for example the mesh 1) and continue
1 2 3 4
Mesh
Fig. 6 Efficiency.
increasing the number of processors, the efficiency is
reduced. Indeed, if the authors increase the number of section of the lake containing one injector. The
processors, the communication cost rises. Therefore, computational domain is a two dimensional domain
the speed-up and efficiency decrease. Thus, the with length 150 m and maximum height 25 m
increase of the number of processors requires an containing one injector with 5m length. The used mesh
increase of the problem size in order to guarantee the is about 497,635 nodes. The authors used the following
efficiency of the parallelization, which is called boundary conditions: the no slip condition on the
scalability. bottom, uwind = (0.01m/s, 0,0), and uinj = (0,0,0.1
m/s).The numerical simulation is done using 16
4. Numerical Simulation of the Aeration
processors for T = 10 min. The execution time is about
Process
4.26 h. The authors remark that the execution time
This section is devoted to a real application. The using one processor is about two days.
authors give a numerical simulation of the aeration Fig. 8 shows a zoom near the injector (see Fig. 7) of
process using the obtained parallelized stabilized finite the velocity field due to the wind effect. The authors
element algorithm. remark that this effect concerns only the top of the
The authors present in this section a numerical domain which is not sufficient to aerate the lake
simulation of the dynamic created using the parallel domain. The authors present in Fig. 9 a zoom near the
solver. The authors consider a two dimensional cross injector of the created dynamic by the mechanical
108 A Parallel Finite Element Algorithm for Navier-Stokes Equations
After 1 min
5. Conclusions
In this work, the authors have proposed a parallel
After 4 min
algorithm for the two dimensional Navier-Stokes
equations written in stream function-vorticity
formulation. In order to optimize the computing time,
the authors have proposed a parallel implementation
for the obtained solver. The parallelization process is
performed using MPI. The obtained solver is used to
simulate the mechanical aeration process in an
eutrophic lake. The proposed approach will be
extended in a forthcoming work to a three dimensional
model taking into account the two-phases air bubbles
and water. The authors denote finally that the proposed
algorithm is general and can be adapted for many other
important engineering problems related to the
Navier-Stokes equations. After 7 min
A Parallel Finite Element Algorithm for Navier-Stokes Equations 109
Received: January 22, 2012 / Accepted: February 12, 2013 / Published: February 25, 2013.
Abstract: Here is introduced some novel algorithms which made use of polygamma functions to get the exact limits of a broad class of
infinite series. Moreover, Laplace transform is used to find the sum of many convergent infinite series. These exact limits are found in
different branches of physics for some special cases series and are in complete agreement with the values found by other authors.
Moreover, the methods presented here are generalized and applied to other wide variety of sums, including alternating series. Finally,
these methods are simple and quite powerful to calculate the limits of many convergent series as you can see from the examples
included.
0
1 1 z c ln z
A 0
e Ax dx (3) dz
n c 1 z
2
n 1 1
1 2
Considering now N , the result is
n 1 n 1
2
2
1
6
1
S c, d
e dx
e cx e x
dx (6) This first series appears in the black body theory [4].
0 c d 1 e x 2.3 Generalization of Infinite Series Limits
x
Changing z e variables and using the Eulerian
To generalize the infinite series so far treated, let us
functions properties [3], the Eq. (6) finally is expressed
1
as: consider the infinite series , and apply
n a
1
m
1 z z d c n 1
S c, d dz
c d 0 1 z the most general Feynman identity
d 1 c 1 1 1
m 1! 0
x m 1e Ax dx (10)
c d A m
QN 1 n c0 c1n ... cN 1n N 1 (14) commute. Now, let us present two applications of the
the infinite series is expressed inpartial fractions as Laplace transform for the evaluations of some infinite
k Aj series.
S a0 ,..., ak ; c0 ,..., cN 1 (15) For the first example, the following expression is
n a
mi
n 1 j 1
j considered
where A j are the coefficients in the fractions. s ae at be bt
L
And applying the Feynman identity done in Eq. (10), s a s b a b
the series can be written as: Then, the infinite series associated with this
S a0 ,..., ak ; c0 ,..., cN 1 expression is given by
nx ae
ax
k Aj (16)
n be bx
x
m j 1 n a j x
e dx n a b
e dx
n 1 j 1 m j 1 0 n 1 n a n b 0
Following similar procedures to those already made, which becomes
1
n 1 at a bt b
it finally reaches the value of the infinite series in terms
n a n b a b 1 t
dt
of polygamma function. n 1 0
e nt t k 1e at dt
The Feynman identities given by Eqs. (3), (8) and n a
k
n 1 n 1 0
(10) correspond to the Laplace transforms of 1, x, xm-1,
x a ln x
1 k 1
f s is the Laplace 1
k 1
respectively. In general, if
transform of g x , one has
0
1 x
dx (21)
where x e t , variable change is used.
f s e sx g x dx (18)
Finally, the limit of infinite series is
k
0
Where the sum f n can be written as 1 a 1
k k 1
(22)
n a
k
n n 1
f n g x e nx dx
n
(19) To finish the algorithms exposure, it looks important
to give some information on an alternating series
0
considering that the addition and integration operations example, which have not yet been considered in this
Calculation of Infinite Series Through Polygamma Functions’ Algorithms and Laplace Transforms 113
work. Due to these series are important in physics, they applications. The algorithms developed are also useful
will be extensively treated in another article applying for solving the alternating series; another important
our general algorithms used here. As a small example, convergent series class.
the following alternating series is solved by the Section
Acknowledgments
2 methods.
Considering the series given by Leibnitz expression Thanks to E. Poulain for fruitful discussions. One of
1
n
2n 1 .
n 1
the authors (HLG) thanks CONACYT for his
fellowship as member of SNI for financial support.
Using standard methods of Section 2,
1 References
n
e 1 e 2 nx dx (23)
x n
n 1 2n 1
[1] P.Ramond, Field Theory: A Modern Primer, Frontiers in
0 n 1
Physics, Vol 74, 2nd ed.,Westview Press, United States,
1 e 2 nx can be
n
and since the infinite series 2001.
n 1 [2] G.B. Arfken, H.J. Weber, F.E. Harris, Mathematical
also evaluated, finally one has Methods for Physicists: A Comprehensive Guide, 7th ed.,
1 e x
n 1
dt
arctan t 0 Elsevier Academic Press, USA, 2012, pp. 321-344
1
2 x
dx
n 1 2 n 1 1 e 1 t 2
0 0
4 (Chapter 5).
[3] M. Abramowitz, I.A. Stegun, Handbook of Mathematical
5. Conclusions Functions: With Formulas, Graphs, and Mathematical
The limits of a broad class of convergent infinite Tables, Dover Publications, USA, 1964.
series were obtained using some novel algorithms. [4] L.I. Schiff, Quantum Mechanics, 3rd ed., McGraw Hill,
New York, 1968.
These algorithms are very flexible and can be used to
[5] I.S. Gradshteyn, I.M. Ryzhik, A. Jeffrey, D. Zwillinger,
evaluate many series that are important in different Table of Integrals, Series and Products, 6th ed., Academic
physics disciplines, as well as in other relevant sciences Press, USA, 2000, p. 11.