Volume 11, 2007 15 THE ITERATION PROCESS FOR THE NONLINEAR TWO-DIMENSIONAL OSCILLATION PROBLEM Odisharia V.∗) , Peradze J.∗) , Peradze L.∗∗) ∗) I. Javakhishvili Tbilisi State University ∗∗) Georgian Technical University Abstract. The initial boundary value problem for an integro-differential Kirchhoff equation is considered in the case of a square domain. To find an approximate solution, step-by-step discretization is performed with respect to spatial variables and a time argument. The obtained cubic system is solved by the iteration method. The method error is estimated. Key words: Kirchhoff equation, Jacobi nonlinear iteration process, error estimate AMS subject classification 2000: 65M Let us consider the following initial boundary value problem Z ³ ´ 4 wtt − λ + 2 (wx2 + wy2 )dx dy (wxx + wyy ) = 0, π Ω (x, y) ∈ Ω, 0 < t ≤ T, w(x, y, 0) = w(0) (x, y), wt (x, y, 0) = w(1) (x, y), (x, y) ∈ Ω, w(x, y, t) = 0, (x, y) ∈ ∂Ω, 0 ≤ t ≤ T, (1) (2) (3) where Ω = {(x, y) | 0 < x < π, 0 < y < π}, ∂Ω is the boundary of the domain Ω, w(0) (x, y) and w(1) (x, y) are given functions, λ > 0 and T are given constants. Equation (1) is a two-dimensional analogue of the well-known Kirchhoff equation [1] Z ³ 2 π 2 ´ wtt − λ + w dx wxx = 0 (4) π 0 x for string vibration. The studies of many researchers are devoted to Kirchhoff type equations (for the bibliography see, e.g., [2], [3]). We present here one numerical method of the solution of problem (1)–(3). An approximate solution will be sought for as a finite sum wn (x, y, t) = n X wnij (t) sin ix sin jy, i,j=1 where the coefficients wnij (t) are calculated from the Galerkin system 00 (t) wnij h + λ+ n X 2 (p + l p,l=1 2 i 2 (t) )wnpl (i2 + j 2 )wnij (t) = 0, (5) 16 Bulletin of TICMI wnij (0) = 4 π2 4 0 wnij (0) = 2 π i, j = 1, 2, . . . , n, Z w(0) (x, y) sin ix sin jy dx dy, Ω (6) Z w(1) (x, y) sin ix sin jy dx dy, i, j = 1, 2, . . . , n. Ω We replace problem (5), (6) by the problem of finding functions w nij (t), i, j = 1, 2, . . . , n, where the relation between the functions wnij (t) and w nij (t) is such that 1 w nij (t). wnij (t) = p (7) i2 + j 2 We have to solve the following problem w00nij (t) h + λ+ n X i w2npl (t) (i2 + j 2 )w nij (t) = 0, (8) p,l=1 i, j = 1, 2, . . . , n, Z p 4 2 2 w(0) (x, y) sin ix sin jy dx dy, w nij (0) = 2 i + j π Ω Z 4p w0nij (0) = 2 i2 + j 2 w(1) (x, y) sin ix sin jy dx dy, π Ω i, j = 1, 2, . . . , n. (9) To solve the obtained Cauchy problem (8), (9) we will use a difference scheme of symmetrical type. To this end, we introduce the grid {tm | 0 = t0 < t1 < . . . < tM = T } with a constant step τ = tm − tm−1 . The approximate values w nij (tm ) are denoted by wm nij , i, j = 1, 2, . . . , n, m = 0, 1, . . . , M . The scheme has the form n m+u−1 2 i m−1 m−2 2 X (wm+u ) wm 1 X nh nij − 2w nij + w nij npl ) + (w npl + λ + 2 τ 2 u=−1,0 2 p,l=1 ×(i2 + j 2 ) m+u−1 o wm+u nij + w nij = 0, 2 i, j = 1, 2, . . . , n, m = 2, 3, . . . , M, (10) w0nij = wnij (0), τ w1nij = wnij (0) + w0nij (0) 2 Z i 2h 4 2τ (0) 2 (0) 2 ((wx ) + (wy ) )dx dy × + 2 λ+ 2 π π Ω Z (0) (0) ) sin ix sin jy dx dy, i, j = 1, 2, . . . , n. + wyy × (wxx Ω (11) Volume 11, 2007 17 From system (10) we write a subsystem for fixed m, 2 ≤ m ≤ M . Then n n o X 8 m−1 2 m m 2 w + 2λ + [(w ) + (w ) ] npl npl τ 2 (i2 + j 2 ) nij p,l=1 m−1 ×(wm nij + w nij ) = 8 f m , i, j = 1, 2, . . . , n, τ 2 (i2 + j 2 ) nij (12) where m−2 fm = 2wm−1 nij − w nij − nij ×(i2 + j 2 ) τ2 2 ³ λ+ m−2 wm−1 nij +w nij 2 Pn m−2 2 2 (wm−1 npl ) +(w npl ) p,l=1 2 ´ × , i, j = 1, 2, . . . , n. Let us agree that for the solution of problem (10), (11) we will use iteration m−1 layerwise, more exactly, knowing the approximate solutions wm−2 nij and w nij , m i, j = 1, 2, . . . , n, from (12) we find w nij , i, j = 1, 2, . . . , n, by iteration. As to the initial values, i.e. values at the zero and the first layer, formulas (11) make it possible to find w0nij and w1nij , i, j = 1, 2, . . . , n. m Denote the k-th iteration approximation wm nij by w nij,k , i, j = 1, 2, . . . , n, k = 0, 1, . . .. From our algorithmic approach it follows that (12) is a system of equations with respect to wm nij , i, j = 1, 2, . . . , n. To solve this system we use the nonlinear Jacobi iteration process. From (12) it follows that the process has the form w3nij,k+1 + aij w2nij,k+1 + bij w nij,k+1 + cij = 0, (13) where m−1 2 aij = wm−1 nij , bij = dij + (w nij ) + m−1 2 cij = (dij + (wm−1 nij ) )w nij − dij = 2λ + n X 8 , + j 2) τ 2 (i2 8 fm , + j 2 ) nij τ 2 (i2 (14) m−1 2 2 [(wm npl,k ) + (w npl ) ], k = 0, 1, . . . p, l = 1 p 6= i l 6= j In addition to (14) we need the values 2 8 2 , rij = dij + (wm−1 nij ) + 2 2 3 τ (i + j 2 ) 2³ 10 m−1 2 ´ m−1 sij = dij + (w nij ) w nij − 3 9 ³1 ´ 8 m m−1 w , i, j = 1, 2, . . . , n. − 2 2 + f nij τ (i + j 2 ) 3 nij (15) 18 Bulletin of TICMI Using the Cardano formulas [4], by (13) we write wm nij,k+1 in the explicit form wm (16) nij,k+1 = ψij,k , where aij ψij,k = − + σij,1 − σij,2 , i, j = 1, 2, . . . , n, (17) 3 3 ´1 i1 h ³ s2 rij 2 3 ij v sij σij,v = (−1) , v = 1, 2, i, j = 1, 2, . . . , n. (18) + + 2 4 27 To establish the convergence conditions for the iteration process used, we consider the Jacobi matrix n ∂ψ o ij,k J= , i, j = 1, 2, . . . , n, i1 , j1 = 1, 2, . . . , n. (19) ∂w ni1 j1 ,k Here ij and i1 j1 are respectively the numbers of a row and a column of matrix (19). By (14)–(18) on the principal diagonal of the matrix J we find zeros. As to nondiagonal elements, i 6= i1 , j 6= j1 , we find 2 X ∂ψij,k 1 m 1 h m−1 = − w ni1 j1 ,k 2w nij ∂wm 9 σ2 ni1 j1 ,k v=1 ij,v 3 ´− 1 i ³ 1 2 ´³ s2ij rij 2 +(−1)v wm−1 s + r + . ij ij nij 3 4 27 (20) From (18) we obtain the formulas σij,1 σij,2 = 3 ´1 3 3 ³ s2 rij σij,1 + σij,2 rij 2 ij 3 3 , σij,2 − σij,1 = sij , + = , 3 4 27 2 which together with (20) give ∂ψij,k (1) (2) = ψiji1 j1 + ψiji1 j1 , ∂wm ni1 j1 ,k (21) where ³ 2 ´ ´2−v m−1 v−1 − wm (s ) w ij 3 ni1 j1 ,k 3 nij ´−1 ³ ³ r ´v ij 2v 2v + σij,2 , v = 1, 2. × σij,1 + − 3 ³2 (v) ψiji1 j1 = (v) (22) Let us estimate |ψiji1 j1 |, v = 1, 2. To this end, we consider the functions ψ (u) (z) = (−r)u + 2 X 1 2u [z + (−1)v (z 2 + r3 ) 2 ] 3 , v=1 −∞ < z < ∞, r = const > 0, Volume 11, 2007 19 u = 1, 2. Each function ψ (u) (z), u = 1, 2, is even and increasing for any z ≥ 0 and therefore min −∞<z<∞ |ψ (u) (z)| = ψ (u) (0) = (2u − 1)ru , u = 1, 2. The latter relation, (14), (15) and (22) imply (u) |ψiji1 j1 | ≤ [τ 2 (i2 + j 2 )]u 2−u |sij |u−1 |wm−1 |wm ni1 j1 ,k |, u = 1, 2. nij | 26u − 20 From formulas (21), (23) and (14), (15) it follows that ¯ ∂ψ ¯ 1 n1 h ¯ ij,k ¯ m−1 2 2 2 m 2 2 2 ≤ | | λ+ τ (i + j )|w τ (i + j )|w ¯ ¯ ni1 j1 ,k nij ∂w ni1 j1 ,k 4 6 n ³ o X 1 m 5 m−1 2 ´i 2 m m−1 + (w ) + (w npl ) + |w nij | + |f nij | . 2 npl,k 9 p,l=1 (23) (24) Let us introduce the vectors n n n m m m = (wm−1 wm−1 n nij )i,j=1 , w n,k = (w nij,k )i,j=1 , f n = (f nij )i,j=1 . Besides, we also need vector and matrix norms. For the vector µ = (µs )N s=1 and N N P P |grs |. the matrix G = (grs )N |µs | and kGk1 = max r,s=1 we define kµk1 = Let us consider the sums n P s=1 1≤s≤N r=1 (i2 + j 2 )u , u = 1, 2. Taking into account i,j=1 that [5] n X l2u = l=1 n(n + 1)(2n + 1) ³ 3n2 + 3n − 1 ´u−1 , u = 1, 2, 6 5 we obtain n X n2 (n + 1)(2n + 1) ³ 6n2 + 6n − 2 ´u−1 (i2 + j 2 )u ≤ , u = 1, 2. 3 5 i,j=1 (25) Note that for u = 1 in (25) we have the equality. Let us estimate the norm of the Jacobi matrix (19). By virtue of (24) and (25) kJk1 τ 2 n2 (n + 1)(2n + 1) 12 n τ 2 (3n(n + 1) − 1) × max |wm | kwm−1 k1 × ni1 j1 ,k n i1 ,j1 15 n ³ o h X 1 m 2 5 m−1 2 ´i m +kwm−1 k +kf k , (26) × λ+ (w nij,k ) + (w nij ) 1 1 n n 2 9 i,j=1 ≤ i1 , j1 = 1, 2, . . . , n. 20 Bulletin of TICMI Applying the principle of compressed mapping [6], we assume that kJk1 ≤ q, 0 < q < 1, 1 (27) m m kwm kwm n,k − w n,0 k1 ≤ n,1 − w n,0 k1 , k = 1, 2, . . . , 1−q is fulfilled. From (26) it follows that for (27) to be hold, it is sufficient that the following biquadratic inequality α τ4 + β τ2 − γ ≤ 0 (28) be fulfilled with respect to the step τ . Here 1³ m 3n(n + 1) − 1 m−1 h α = kw n,0 k1 + kwn k1 λ + 15 2 n ´2 5 X i 1 m−1 2 m m + kw − w n,0 k1 + |w | , 1 − q n,1 9 i,j=1 nij β = kwm−1 k1 + kf m k, n n 1 ³ ´−1 12q 1 m m m γ = kw n,0 k1 + kw − w n,0 k1 . n2 (n + 1)(2n + 1) 1 − q n,1 Thus we come to a conclusion that if the parameter q and the step τ satisfy inequality (28), then system (12) has a unique solution with respect to m m n the unknowns wm nij , i, j = 1, 2, . . . , n. The vector w n = (w nij )i,j=1 consisting m of the components of solutions wm nij is the limit of a sequence of vectors w n,k , as k → ∞. Moreover, the estimate m kwm n,k − w n k1 ≤ qk kwm − wm n,0 k1 , k = 0, 1, . . . , 1 − q n,1 is true. Formula (7) enables us to construct with the aid of wm nij,k approximate solutions of the function wnij (t) at the grid points. The problem considered here for the one-dimensional equation (4) is studied in [7]. References [1] Kirchhoff, G., Vorlesungen über Mechanik, Teubner, Leipzig, 1883. [2] Medeiros, L., Limaco, I., Menezes, S., Vibrations of an elastic string, Mathematical Aspects, I and II., J. Comput. Anal. Appl., 4, 2(2002), 211–263. [3] Peradze, J., A numerical algorithm for the nonlinear Kirchhoff string equation, Numer. Math., 102, 2(2005), 311–342. [4] Kurosh, A., A course on higher algebra, Nauka, Moscow, 1975 (in Russian). [5] Gradstein, I., Rizhik, I., Tables of integrals, sums, series and products, Nauka, Moscow, 1971 (in Russian). [6] Trenogin, V. Functional analysis, Nauka, Moscow, 1980 (in Russian). [7] Peradze, J., The Jacobi nonlinear iteration method for a discrete Kirchhoff system, AMIM, I. Vekua Appl. Math. Inst. Tbilisi State Univ., 6, 1(2001), 81–89. Received April, 16, 2007; revised October, 15, 2007; accepted November, 7, 2007.