Ìàòåìàòèêà / 4. Ïðèêëàäíàÿ ìàòåìàòèêà
 
Àipanov Azamat Sh.
Süleyman Demirel
University, Faculty of Economics, 
Almaty, Kazakhstan
 
GLOBAL
ASYMPTOTIC STABILITY OF OSCILLATORY-ROTARY MOTION OF A MATHEMATICAL PENDULUM
 
1. Introduction. The problem
covered is a stability of the mathematical pendulum moving under the force of
gravity, viscous friction and external force with constant moment. It is
assumed that initial deviation and velocity of the pendulum may be very large,
so that it is capable of exhibiting not only oscillatory, but also rotational
motion. When the damping coefficient  is sufficiently large
the pendulum's motion becomes stable due to the dissipation of energy by the
friction force. The pendulum eventually comes to a damped oscillations mode and
converges to the equilibrium point. When
 is sufficiently large
the pendulum's motion becomes stable due to the dissipation of energy by the
friction force. The pendulum eventually comes to a damped oscillations mode and
converges to the equilibrium point. When  is small there is a
possibility that the pendulum will start to rotate without stopping if the
external force moment acting on it prevails. Thus there is a threshold value
 is small there is a
possibility that the pendulum will start to rotate without stopping if the
external force moment acting on it prevails. Thus there is a threshold value  such that when the
damping coefficient
 such that when the
damping coefficient  , the motion of the pendulum tends to its equilibrium
position with any initial conditions.
, the motion of the pendulum tends to its equilibrium
position with any initial conditions.
Several attempts to determine the value of  made by
F. Tricomi, L. Amerio, G. Seifert, W. Hayes,
C. Böhm and other scientists in 30th – 50th years of the XX century
were not successful. Only lower and upper estimates in the forms of
 made by
F. Tricomi, L. Amerio, G. Seifert, W. Hayes,
C. Böhm and other scientists in 30th – 50th years of the XX century
were not successful. Only lower and upper estimates in the forms of  and
 and  were found. It was not
until 1954 when Japanese mathematician M. Urabe from the University of
Hiroshima proposed a numerical algorithm [1, 2] for calculating
 were found. It was not
until 1954 when Japanese mathematician M. Urabe from the University of
Hiroshima proposed a numerical algorithm [1, 2] for calculating  . However, it turned out in some cases that the algorithm
doesn't converge. As for the analytical formula for finding
. However, it turned out in some cases that the algorithm
doesn't converge. As for the analytical formula for finding  , the problem has not been solved yet.
, the problem has not been solved yet.
The modification of Urabe's method proposed by the
author [3] has an increased rate of convergence. The results obtained here can
be used for the analysis of stability of systems with angular coordinates
(phase systems) where oscillation and rotation processes take place. These are,
for example, rotor systems in mechanics, synchronous generators in electric
power industry, phase-locked loop (PLL) systems in radio technology and others.
2. The problem
statement. The mathematical pendulum represents a rod of length  , with a mass
, with a mass  on its end. There is
the force of gravity
 on its end. There is
the force of gravity  and an external force
with constant moment
 and an external force
with constant moment  acting on the
pendulum. It moves in a viscous ambient where the force of friction is equal
 acting on the
pendulum. It moves in a viscous ambient where the force of friction is equal  and is directed
opposite to the velocity vector (
 and is directed
opposite to the velocity vector ( is a friction coefficient which depends on ambient
viscosity,
 is a friction coefficient which depends on ambient
viscosity,  is a velocity of the
pendulum). It is assumed that the inequality
 is a velocity of the
pendulum). It is assumed that the inequality  is satisfied.
 is satisfied.
The motion of the mathematical pendulum considered
here is described by the system of differential equations [4]
 (1)
                                                   (1)
which can be reduced to the form
 (2)
                                                (2)
where  ;
;   is an angle of the
pendulum's deviation from the vertical;
 is an angle of the
pendulum's deviation from the vertical;  ;
;   is an angular velocity;  a parameter
 is an angular velocity;  a parameter  is a damping coefficient;
 is a damping coefficient;   is a
 is a  -periodic function;
-periodic function;   .
.
On the set  the function
 the function  turns to zero at two points:
at
 turns to zero at two points:
at  and at
 and at  . We designate
. We designate  and point out that the following
relationships are true:
 and point out that the following
relationships are true:
 ,
,   ;
;   ,
,   ;
;   ,
,   .
.
The stationary set  of the system (1) on
the plane
 of the system (1) on
the plane  consists of the stable
equilibrium points (
 consists of the stable
equilibrium points ( ,
,  ) and the unstable equilibrium points (
) and the unstable equilibrium points ( ,
,  ), where
), where  .
 .
Let's denote the solution of the system (1) with the
initial conditions  ,
,  as the functions
 as the functions  ,
,  , where
, where  .
.
D e f i n i t i o n . The system (1) is called globally asymptotically stable, if its
solution ( ,
,  ) for any initial conditions
) for any initial conditions  ,
,  goes with
 goes with  to some (stable or unstable)
equilibrium state from the set
 to some (stable or unstable)
equilibrium state from the set  .
.
We consider on the plane  two special phase
trajectories:
 two special phase
trajectories:  which starts from the
unstable equilibrium point (
 which starts from the
unstable equilibrium point ( ,
,  ) and
) and  which tends to the unstable
equilibrium point (
 which tends to the unstable
equilibrium point ( ,
,  ). The images of the phase trajectories
). The images of the phase trajectories  ,
,  corresponding to the cases  a)
 corresponding to the cases  a)  ,  b)
,  b)  and  c)
 and  c)  are presented on the Figure 1.
 are presented on the Figure 1.
 


 
T h e o r e m   1 .
(E.A. Barbashin and V.A. Tabuyeva [4, p. 70]). There exists a unique value of  such that the following relationships hold
true
(Figure 1):
 such that the following relationships hold
true
(Figure 1):
       a)  when
 when  ,                                                                     
      (3)
,                                                                     
      (3)
b)  when
 when  ,                                                                                (4)
,                                                                                (4)
c)  when
 when  .                                                                                 (5)
.                                                                                 (5)
Thereby, as follows from Theorem 1, the two phase trajectories  and
 and  merge together:
 merge together:  ,
,  , when
 , when  (Figure 1 (b)). The
criterion of the global asymptotic stability of the system (1) has a form of an
inequality
 (Figure 1 (b)). The
criterion of the global asymptotic stability of the system (1) has a form of an
inequality  , setting the problem of determining the value of
, setting the problem of determining the value of  .
.
3. Modification of
the Urabe's algorithm. The Urabe's original algorithm [1, 2] is based
on assumption of analyticity of the function  on the segment
 on the segment  . But as the Figure 1 (c) suggests, when
. But as the Figure 1 (c) suggests, when  the phase trajectory
 the phase trajectory  cannot be represented as power
series on the segment
 cannot be represented as power
series on the segment  . This is the main disadvantage of the Urabe's algorithm. The
modification of this method is rather based on Theorem 1 by E.A. Barbashin
and V.A. Tabuyeva and the expansions of the functions
. This is the main disadvantage of the Urabe's algorithm. The
modification of this method is rather based on Theorem 1 by E.A. Barbashin
and V.A. Tabuyeva and the expansions of the functions  ,
,  into Taylor series on the segments
 into Taylor series on the segments  ,
,  respectively.
 respectively.
Step 1. Expanding the
function  on the segment
 on the segment  into the Taylor power series
we get
 into the Taylor power series
we get
 (6)
                             (6)
where  . Repeat the same procedure with
. Repeat the same procedure with  on the segment
 on the segment  
 
 (7)
                              (7)
where  . Whereas
. Whereas  and
 and  is a
 is a  -periodic function the following equalities hold true:
-periodic function the following equalities hold true:  ,
,  , moreover
, moreover  ,
,   .
.
Step 2. The sought-for
value of  lies on the segment
 lies on the segment  , where
, where  can be taken as any of the lower
estimations
 can be taken as any of the lower
estimations  and
 and  can be taken as any of the
upper estimations
 can be taken as any of the
upper estimations  . Let us take the first approximation equal to
. Let us take the first approximation equal to  .
.
Step 3. We seek the
function  which satisfies the equation
 which satisfies the equation
 (8)
                                 (8)
on the segment  in the form of power series
 in the form of power series
 (9)
                                               (9)
where  ,
,   
 
Step 4. Plugging the
series (6) and (9) into the formula (8) we get the equation

which can be written as  . Equating the coefficients
. Equating the coefficients  ,
,  to zero we get the system of
equations
 to zero we get the system of
equations

From this we find the values of  :
:
 
   

Here  is a positive root of the
quadratic equation
 is a positive root of the
quadratic equation  because
 because  ,
,   .
.
Step 5. Plugging the found
coefficients  into the formula (9) we
calculate the value
 into the formula (9) we
calculate the value  .
.
Step 6. We are searching
for the function  that satisfies the equation
 that satisfies the equation
 (10)
                                       (10)
on the segment  in the form of power series
 in the form of power series
 (11)
                                            (11)
where  ,
,  
Step 7. Plugging the series
(7) and (11) into the formula (10) we get the equation

which can be reduced to  . Equating the coefficients
. Equating the coefficients  ,
,   to zero, we get the system
of equations
 to zero, we get the system
of equations

From this we find the values of  :
:
 
    
                                                   

Here  is a negative root of the
quadratic equation
 is a negative root of the
quadratic equation  since
 since  ,
,   .
. 
Step 8. Plugging the found
coefficients  into the formula (11) we
calculate the value of
 into the formula (11) we
calculate the value of  .
.
Step 9 (dichotomy
method). Let's consider the following three cases:
9a) If  , then from (3) follows that $
, then from (3) follows that $ , i.e. the value of
, i.e. the value of  lies in the segment
 lies in the segment  . We take the next approximation equal to
. We take the next approximation equal to  .
.
9b) If  , then from (4) follows that
, then from (4) follows that  , i.e. we found
, i.e. we found  – the sought-for value of
the damping coefficient. The algorithm stops.
 – the sought-for value of
the damping coefficient. The algorithm stops.
9c) If  , then from (5) follows that
, then from (5) follows that  , i.e. the value of
, i.e. the value of  lies in the segment
 lies in the segment  . We take the next approximation equal to
. We take the next approximation equal to  .
.
During the computer implementation of this algorithm
the fulfillment of the condition  serves as a criterion of
stoppage of the algorithm on the step 9b, where
 serves as a criterion of
stoppage of the algorithm on the step 9b, where  is a sufficiently small
number which specifies the accuracy of calculations. If this inequality
doesn't hold true, on the step 9a or 9c we determine the segment
 is a sufficiently small
number which specifies the accuracy of calculations. If this inequality
doesn't hold true, on the step 9a or 9c we determine the segment  which contains
 which contains  and the length of this
segment is going to be half the length of the initial segment
 and the length of this
segment is going to be half the length of the initial segment  .
.
Next we repeat the steps 3-9 taking the value of  instead of
 instead of  ; determine the segment
; determine the segment  which contains
 which contains  and select the middle point
of this segment as a next approximation
 and select the middle point
of this segment as a next approximation  and so forth. Thus, on the
 and so forth. Thus, on the  th iteration the length of
th iteration the length of  is going to be
 is going to be  times smaller than the
length of
 times smaller than the
length of  , which means that it is possible to reach very high accuracy in a
reasonable amount of iterations. Furthermore, one should consider as many
summands in the expansions (6), (7), (9) and (11) as needed to achieve the
given accuracy of calculations.
, which means that it is possible to reach very high accuracy in a
reasonable amount of iterations. Furthermore, one should consider as many
summands in the expansions (6), (7), (9) and (11) as needed to achieve the
given accuracy of calculations.
4. Numerical
example. Let us look at the numerical example illustrating the efficiency of the
proposed modification of the Urabe's algorithm. We consider the equations of
the mathematical pendulum motion (1) with following values of parameters
 ,
,   ,
,   .                                             (12)
.                                             (12)
Using the formulas for lower and upper estimates of  we can determine the
interval
 we can determine the
interval  in which
 in which  lies. Notably, there are
following lower estimations [4] in the form of inequality
 lies. Notably, there are
following lower estimations [4] in the form of inequality  :
:
(a) F. Tricomi estimate:
 ;                                          (13)
;                                          (13)
(b) W. Hayes estimate:
 ;                                       
(14)
;                                       
(14)
(c) C. Böhm estimate:
 .                                       
(15)
.                                       
(15)
There are also the upper estimates [4] in the form of  :
:
(d) G. Seifert estimate:
 ;                                 
(16)
;                                 
(16)
(e) F. Tricomi estimate:
 ;                                                 (17)
;                                                 (17)
(f) L. Amerio estimate:
 .                              (18)
.                              (18)
For the example considered here the following estimates were gained from
the formulas (13)-(18):
(a)  ;    (b)
;    (b)  ;     (c)
;     (c)  ;
;
(d)  ;    (e)
;    (e)  ;    (f)
;    (f)  .
.
Thus, we have the  estimates for the value of
 estimates for the value of  . In the examined example
. In the examined example  , so the conditions
, so the conditions  are also satisfied.
Meanwhile it is impossible to say anything about the mutual position of
 are also satisfied.
Meanwhile it is impossible to say anything about the mutual position of  and
 and  (what of the next
inequalities holds true:
 (what of the next
inequalities holds true:  or
 or  ?). So we are not able to say whether the system is stable or unstable
based on the estimates (13)-(18).
?). So we are not able to say whether the system is stable or unstable
based on the estimates (13)-(18).
Computer calculations using the modified algorithm of
M. Urabe showed that  . From this we can see that the criterion
. From this we can see that the criterion  is satisfied, which means
that the system (1) with the given values of parameters (12) is globally
asymptotically stable.
 is satisfied, which means
that the system (1) with the given values of parameters (12) is globally
asymptotically stable.
The calculations based on the original algorithm
carried out for the different values   ,
,   ,
,   ,  where
,  where  is the number of
terms in the Taylor series expansion, don't converge. On the other hand, the
iteration process in the modified algorithm rapidly converges to the value
 is the number of
terms in the Taylor series expansion, don't converge. On the other hand, the
iteration process in the modified algorithm rapidly converges to the value  , the condition
, the condition  , where
, where  , which signifies the end of the process, is fulfilled on the 24th
iteration for
, which signifies the end of the process, is fulfilled on the 24th
iteration for  .
.
5. Conclusion. The paper contains
the results related to the stability analysis of the mathematical pendulum: the
global asymptotic stability of the system under the condition  , the theorem by E.A. Barbashin and V.A. Tabuyeva in existence and
uniqueness of the threshold value
, the theorem by E.A. Barbashin and V.A. Tabuyeva in existence and
uniqueness of the threshold value  , the upper and lower estimates of
, the upper and lower estimates of  by F. Tricomi, L. Amerio,
G. Seifert, W. Hayes, C. Böhm, and the modification of the
Urabe's method for numerical calculation
 by F. Tricomi, L. Amerio,
G. Seifert, W. Hayes, C. Böhm, and the modification of the
Urabe's method for numerical calculation  . The Urabe's original algorithm has a following drawback. When
. The Urabe's original algorithm has a following drawback. When  the function
 the function  has a positive value at the
point
 has a positive value at the
point  , while it is undefined at the point
, while it is undefined at the point  when
 when  . That is why the modified algorithm uses the expansion of the function
. That is why the modified algorithm uses the expansion of the function  on the segment
 on the segment  and the expansion of the
function
 and the expansion of the
function  on the segment
 on the segment  so that there is no need to calculate
 so that there is no need to calculate
 . The fact that the functions are expanded into Taylor series on the
segments of smaller length also leads to the increase in accuracy of the
algorithm.
. The fact that the functions are expanded into Taylor series on the
segments of smaller length also leads to the increase in accuracy of the
algorithm.
The values of  on the segment
 on the segment  and
 and  on the segment
 on the segment  (as the solutions of the
differential equations (8) and (11) respectively) can be also calculated
numerically using Runge – Kutta method.
 (as the solutions of the
differential equations (8) and (11) respectively) can be also calculated
numerically using Runge – Kutta method.
 
References:
1. M. Urabe. Infinitesimal deformation of the periodic
solution of the second kind and its application to the equation of a pendulum
// J. Sci. Hiroshima Univ., Ser. A, Vol. 18, 1954. – P. 183-219.
2. M. Urabe. The least upper bound of a damping
coefficient ensuring the existence of a periodic motion of a pendulum under
constant torque // J. Sci. Hiroshima Univ., Ser. A, Vol. 18, 1955. – P. 379-389.
3. A.Sh. Aipanov. Modification of the Urabe's method //
Theses of Int. Sci. Students Conf. "VII Kolmogorov Readings", Sect.
Math., Moscow, 2007. – P. 13-14 (in Russian).
4. E.A. Barbashin, V.A. Tabuyeva. Dynamic systems with
cylindrical phase space. – Moscow: Nauka, 1969 (in Russian).