Solution of the Interval Equations of Dynamics by using Adaptive Taylor Series

Andrzej Pownuk Naveen Kumar Goud Ramunigari The University of Texas at El Paso, The University of Texas at El Paso, El Paso, Texas, USA El Paso, Texas, USA Department of Mathematical Sciences Department of Civil Engineering [Link] Email: [Link]@[Link] Email: andrzej@[Link] Iwona Skalna M.V. Rama Rao Vasavi College of Engineering University of Science and Technology AGH Department of Civil Engineering Department of Applied Computer Science Hyderabad, India Cracow, Poland E-mail: [Link]@[Link] E-mail: skalna@[Link]
Abstract In this paper we would like to discuss solution of the interval equations of dynamics by using adaptive Taylor series. In this approach in order to get upper (u) and lower bound (u) of the solution u = u(x, p) by expanding the function u using Taylor series with respect to the variable p. Then this approximation can be applied for calculation of u and u. Using special procedure for error estimation it is possible to control accuracy of the calculations.

I. S OLUTION SET OF THE INTERVAL EQUATIONS Let us consider some equation F (u, x, p) = Q(x, p) (1)

y(0) and p2 = dy(0) additionally derivative with respect to dt y y each uncertain parameter p1 , p2 have constant sign. Because of that it is possible to calculate the solution exactly by using appropriate endpoints of the parameters. 1 cos(t) + 1 sin (t) , t 0, 2 2 cos(t) + 1 sin (t) , t , 2 (7) y(t) = 2 cos(t) + 2 sin (t) , t , 3 2 3 1 cos(t) + 2 sin (t) , t 2 , 2 2 cos(t) + 2 sin (t) , t 0, 2 1 cos(t) + 2 sin (t) , t , 2 y (t) = 1 cos(t) + 1 sin (t) , t , 3 2 2 cos(t) + 1 sin (t) , t 3 , 2 2 (8)

(algebraic, differential, integral, etc.) where x Rn , p p Rm . Solution set of the equation (1) can be defined in the following way u(x) = {u(x, p) : F (u, x, p) = Q(x, p) , p p} u(x) = min u(x), u(x) = max u(x) (2) (3)

Solution is shown in the Fig. 1. It is interesting to note

Interval solution can be also defined as a solution of the following optimization problems min u(x, p) F (u, x, p) = Q(x, p) u(x) = (4) pp max u(x, p) F (u, x, p) = Q(x, p) u(x) = (5) pp II. T HE USE OF MONOTONICITY A. Free vibrations Let us consider the following differential equation d2 y + 2 y = 0 dt2 (6)

Fig. 1.

Interval solution of the equation 6

k where 2 = m . Lets assume that y(0) [1, 2], dy(0) [1, 2] dx and = 1. In this case we have two interval parameters p1 =

that the interval solution is continuous but non-differentiable. Interval solution can be created by using 4 different point solution (compare Fig. 2). Because the function y = y(t, p) is monotone (as a function of variable p), then the interval solution is exact. It is possible to verify the interval solution by using more simulations (compare Fig. 3). For each

Interval solution is shown on the Fig. 4. In order to get the

0.6t 2e 0.5t sin (t) , t 0, 2 2e sin (t) , t 2 , y (t) = 1e0.5t sin (t) , t , 3 2 0.6t 1e sin (t) , t 3 , 2 2

(13)

Fig. 2. Interval solution of the equation 6 can be calculated by using 4 solutions

Fig. 4.

Interval solution of the equation 10.

exact interval solution it is enough to find only 2 solutions (compare Fig. 5).

Fig. 3. Interval solution of the equation 6. 10 solutions which can be used as veryfication of our results.

timestep t it is possible to find such combination of parameters pmin (t), pmax (t) such that y(t) = y(t, pmin (t)), y(t) = y(t, pmax (t)) (9)
Fig. 5. Interval solution of the equation 10 can be calculated exactly by using point solutions.

Specific form of the formulas (9) is shown in the equations (7,8). Because the solution is monotone (with respect to the variables p, then the functions pmin (t),pmax (t) have finite number of values. If we know these functions we can calculate the interval solution (if appropriate method of solution exist). B. Damped vibrations Let us consider the following differential equation dy dy 2 2 + 0 y = 0 + 2 (10) dt dt2 General solution of that equation can be written in the following form y = Aet sin (t + ) (11) In calculations we assume that A A, A = [1, 2], [0.5, 0.6], = 0, = 1. In this case the solution is also monotone with respect to the interval parameters. Because of that it is possible to calculate the exact solution in the following way 0.5t 1e 0.6t sin (t) , t 0, 2 1e sin (t) , t 2 , (12) y(t) = 2e0.6t sin (t) , t , 3 2 0.5t 2e sin (t) , t 3 , 2 2

C. Free vibrations - non-monotone solution Let us consider the equation (6) with the following initial conditions y(0) = 0, dy(0) = = p [0.5, 0.6]. The solution dt is shown on the Fig. 6. In this case the sign of partial derivative

Fig. 6. Interval solution of the equation 10 can be calculated exactly by using point solutions.

is not constant. Because of that it is not possible to get the interval solution by using finite number of point solutions (compare Fig. 6). If we try to calculate the solution by using equations y(t) = y(t, pmin (t)), y(t) = y(t, pmax (t)), then the functions pmin (t), pmax (t) have infinitely many values, which is a big problem in calculations and future applications. In other words, it is not possible to get the exact interval solution [y(t), y(t)] by using finite number of point solutions (compare Fig. 6). From the point of view of applications it is important to find the interval solution [y(t), y(t)] as well as appropriate combinations of parameters pmin (t), pmax (t), which generate these solutions. III. N UMERICAL METHOD In previous section we discussed analytical methods for solution of differential equations. Unfortunately in applications usually it is not possible to get the analytical solution. In that situations we have to apply the numerical methods. Let us consider a system of first order parameter dependent differential equations dy dt = f (t, y, p) (14) x(0, p) = x0 (p) pp Partial derivative with respect to the uncertain parameter y vi = pi satisfy the following differential equation. dvi = f (t,y,p) + f (t,y,p) vi dt p y (15) vi (0, p) = v0i (p) = x0 (p) pi pp

y p

IV. H ERMITE APPROXIMATION METHOD Let us consider the following differential equation dy dt = p cos(pt) y(0) = 0 p [p, p]

(18)

Exact solution is equal to y = sin(pt) and we can use them in verification of the results. Solution is not monotone with respect to parameter p. In this approach we are going to solve the equation (18) for p = mid(p). After that it is necessary to find partial derivative y , which satisfy the p following differential equation d dt y = cos(pt) ptsin(pt) p y(0) (19) p = 0 p [p, p]

Lets find the solution of the equation (18) and equation (19) for the mid point p0 = mid(p). Using that solution and appropriate derivative it is possible to create interpolation polynomial. (20) y(t, p) H(t, p0 , p)

In order to increase the accuracy of the calculations it is possible use more solutions, which correspond to different values p0 , p1 , ..., pn . y(t, p) H(t, p0 , p1 , ..., pn , p) In this approach the following interpolation conditions can be applied y(t, pi ) = H(t, p0 , ..., pn , pi ), (22) y(t, pi ) H(t, p0 , ..., pn , pi ) = p p (23)

If the function y = y(t, y, p) is monotone (with respect to the variable p), then in order to find the interval solution it is necessary to find the solution of the equation (15) for the mid point p0 = mid(p) (i.e. y = y(t, p0 ) and vi = vi (t, p0 ). Then it is necessary to find all combinations of parameters, which correspond to the sign of the function of vi = vi (t, p0 ) = y(t,p0 ) pi . p(t)min = pi , p(t)max = pi , if p(t)min = pi , p(t)max = pi , if y(t, p0 ) pi 0

where H(t, p0 , p) is some polynomial with unknown coefficients. Now using classical optimization method it is possible to find approximate values of upper and lower bound of the solution y(t) min{H(t, p0 , ..., pn , p) : p p}, y(t) max{H(t, p0 , ..., pn , p) : p p}. (24) (25)

(16)

For p [0, 1] example solution is shown on the Fig. 18. It is clear that if we use more point solution then the accuracy of the approximation H(t, p0 , ..., pn , p) is better. ACKNOWLEDGMENT The authors would like to thank... R EFERENCES
A [1] H. Kopka and P. W. Daly, A Guide to LTEX, 3rd ed. Addison-Wesley, 1999.

y(t, p0 ) <0 pi

(17)

If the functions p(t)min , p(t)max has finite number of values, then it is possible to find the exact interval solution by using appropriate endpoints y(t) = y(t, pmin (t)), y(t) = y(t, pmax (t)). Solutions y(t, pmin (t)), y(t, pmax (t)) can be calculated by using any numerical method. Unfortunately usually the function y = y(t, p) is not monotone (with respect to the parameter p). Methods which can be applied in these situations will be presented in the next sections.

Harlow, England:

Fig. 7. Interval solution of the equation 18 can be calculated exactly by using point solutions.