We constructed three two-step semi-implicit hybrid
methods (SIHMs) for solving oscillatory second order ordinary differential equations (ODEs).
The first two methods are three-stage fourth-order and three-stage fifth-order with dispersion order six and zero dissipation.
The third is a four-stage fifth-order method with dispersion order eight and dissipation order five.
Numerical results show that SIHMs are more accurate as compared to the
existing hybrid methods, Runge-Kutta Nyström (RKN) and Runge-Kutta (RK)
methods of the same order and Diagonally Implicit Runge-Kutta Nyström
(DIRKN) method of the same stage. The intervals of absolute stability or periodicity of
SIHM for ODE are also presented.
1. Introduction
Second-order ordinary differential equations (ODEs) which are oscillatory in nature often arise in many scientific areas of engineering and applied sciences such as celestial mechanics, molecular dynamics, and quantum mechanics. Consider the numerical solution of the initial value problem (IVP) for second-order ODEs in the form
(1)y′′=f(t,y),y(t0)=y0,y′(t0)=y0′,
in which the first derivative does not appear explicitly. Apparently, some of the most common methods used for solving second-order ODEs numerically are Runge-Kutta Nyström (RKN) and Runge-Kutta methods for Runge-Kutta method the IVPs need to be reduced to a system of first-order ODEs twice the dimension. The IVP can be solved using a particular explicit hybrid algorithms which were developed by Franco [1] or a multistep method for special second-order ODEs as in Yap et al. [2]. Franco [3] proposed that (1) can be solved using a particular explicit hybrid algorithms or special multistep methods for second-order ODEs. Franco [3] constructed explicit two-step hybrid methods of order four up to order six for solving second-order IVPs by considering the local truncation error and order conditions developed by Coleman [4].
Most of the IVPs represented by (1) have solutions which are oscillatory in nature, making it difficult to get the accurate numerical results. To address the problem several authors [5–7] focused their research on developing methods with reduced phase lag and dissipation, where phase-lag or dispersion error is the difference of the angle for the computed solution and the exact solution and dissipation is the distance of the computed solution from the standard cyclic solution. The analysis of phase-lag or dispersion errors was first introduced by Brusa and Nigro [8]. Van der Houwen and Sommeijer [9] proposed explicit RKN methods of order four, five, and six with reduced phase-lag of order q=6, 8, 10, respectively. Senu et al. [7, 10] developed diagonally implicit RKN method with dispersion of higher order for solving oscillatory problems. In a related work Kosti et al. [11] constructed an optimized RKN method (OPRKN) based on the existing explicit four-stage fifth-order RKN method for the integration of oscillatory IVPs. In his derivation he used the phase-lag, amplification factor and the first derivative of the amplification factor by equating them to zero. Later, Kosti et al. [12] also developed an OPRKN method based on the same explicit RKN method, in which he used the phase lag, amplification factor, and the first derivative of the phase-lag properties instead of using only the first derivative of amplification factor in his previous work.
In this paper, we constructed three-stage fourth-order and three-stage fifth-order methods with dispersion order six and zero dissipation and also four-stage fifth-order method with dispersion order eight and dissipation order five. It is done by taking the dispersion relations for the semi-implicit hybrid methods and solving them together with the algebraic conditions of the methods. The intervals of stability of the methods are also presented. Finally, numerical tests on second-order differential equation for oscillatory problems are given.
2. Analysis Phase Lag of the Methods
An s-stage two-step hybrid method for the numerical integration of the IVP(1) is of the form
(2)Yi=(1+ci)yn-ciyn-1+h2∑j=1saijf(tn+cjh,Yj),222222222222i=1,…,s,yn+1=2yn-yn-1+h2∑i=1sbif(tn+cih,Yi),
where bi, ci, and aij can be represented in Butcher notation by the table of coefficients as follows:
(3)cAbT=c1a1,1⋯a1,s⋮⋮⋱⋮csas,1⋯as,sb1⋯bs.
The methods of the form (2) are defined by
(4)Y1=yn-1,Y2=yn,(5)Yi=(1+ci)yn-ciyn-1+h2∑j=1iaijf(tn+cjh,Yj),222222222222i=3,…,s,(6)yn+1=2yn-yn-1+h2[b1fn-1+b2fn+∑i=3sbif(tn+cih,Yi)],
where fn-1=f(tn-1,yn-1), fn=f(tn,yn), h=Δt=tn+1-tn and the first two nodes are c1=-1 and c2=0. The method only requires to evaluate f(tn,yn),f(tn+c3h,Y3),…,f(tn+csh,Ys) for each step after the starting procedure. This method is considered as a two-step hybrid method with s-1 stages per step and the semi-implicit hybrid with the diagonal elements being equal can be written in Butcher tableau as follows:
(7)-10c3a3,1a3,2γ⋮⋮⋮⋱γcsas,1as,2⋯as,s-1γb1b2⋯bs-1bs.
Phase analysis can be divided into two parts. First is the inhomogeneous part, in which the phase error is constant in time and second is the homogeneous one, in which the phase errors are accumulated as n increases. As proposed by Franco [3], the phase analysis is investigated using the second-order homogeneous linear test model
(8)y′′(t)=-λ2y(t)forλ>0,λ∈ℜ.
Alternatively when (2) are applied to (8), they can be written in vector form as
(9)Y=(e+c)yn-cyn-1-H2AY,(10)yn+1=2yn-yn-1-H2bTY,
whereH=λh, Y=(Y1,…,Ys)T, c=(c1,…,cs)T, and e=(1,…,1)T. From (9) we obtain
(11)Y=(I+H2A)-1(e+c)yn-(I+H2A)-1cyn-1.
Substituting (11) into (10), the following recursion relation is obtained:
(12)yn+1-S(H2)yn+P(H2)yn-1=0,
where
(13)S(H2)=2-H2bT(I+H2A)-1(e+c),P(H2)=1-H2bT(I+H2A)-1c.
Solving the difference system (12), the computed solution is given by
(14)yn=2|c||ρ|ncos(ω+nϕ),
where ρ is the amplification factor, ϕ is the phase, ω and c are real constants determined by y0 and y0′ and the hybrid parameters. The exact solution of (8) is given by
(15)y(tn)=2|σ|cos(χ+nH),
where n is the number of term, σ and χ are real constants determined by initial conditions. Equations (14) and (15) led to the following definition.
Definition 1 (phase lag).
Apply the hybrid method (2) to (8). Next we define the phase lag φ(H)=H-ϕ. If φ(H)=O(Hq+1), and then the hybrid method is said to have phase-lag order q. Additionally, the quantity d(H)=1-|ρ| is called amplification error. If d(H)=O(Hr+1), then the hybrid method is said to have dissipation order r.
From Definition 1, it follows that
(16)φ(H)=H-cos-1(S(H2)2P(H2)),d(H)=1-P(H2)
Let us denote S(H2) and P(H2) to be the following:
(17)S(H2)=2+∑i=1s-1αiH2i(1+γH2)s-2,P(H2)=1+∑i=1s-1βiH2i(1+γH2)s-2.
Based on the definition of phase lag, the dispersion relations are developed. For a zero dissipative method with three-stage (s=3), the dispersion relation of order six (q=6) is given by the following:
Order 6:
(18)α2=1γ(1360-γ2),
and the dispersion relations up to order eight for s=4 are given by
Order 6:
(19)β3-2β2γ-α3+γα2-12β2=-32γ2+1360-2γ3,
Order 8:
(20)14β22-(72γ2+γ+124)β2-γα3+γ2α2+(2γ+12)β3=120160-124γ2-2γ3-134γ4.
The following quantity is used to determine that the dissipation constant of the formula for s=3,4 is
for s=3(21)1-P(H2)=(-12β1+12γ)H2+(-12β2+14γβ1-38γ2+18β12)H4+(14γβ2-316γ2β1+516γ3+14β1β2222-116γβ12-116β13)H6+(5128β14-316γ2β2+532γ3β1222+364γ2β12+18β22-316β12β2+132β13γ222-18γβ1β2-35128γ4)H8+O(H10),
for s=4(22)1-P(H2)=(γ-12β1)H2222222222222+(12β1γ-γ2-12β2+18β12)H4222222222222+(-12β1γ2+12β2γ+γ3-12β3-18β12γ222222222222222+14β1β2-116β13)H6222222222222+(-γ4+12β1γ3-12β2γ2222222222222222+12β3γ-14β1β2γ+18β12γ2222222222222222+14β1β3+18β22+116β13γ222222222222222-316β12β2+5128β14)H8+O(H10).
From (12), the stability polynomial of hybrid method can be written as
(23)ξ2-S(H2)ξ+P(H2)=0.
The numerical solution defined by (12) should be periodic. The necessary conditions are
(24)P(H2)≡1,|S(H2)|<2,∀H2∈(0,Hp2),
and interval (0,Hp2) is known as the periodicity interval of the method. The method is called zero dissipative when d(H)=0, that is, if it satisfies conditions (16). Otherwise, as the method possesses a finite order of dissipation, the integration process is stable if the coefficients of polynomial in (23) satisfy the conditions
(25)P(H2)<1,|S(H2)|<1+P(H2),∀H2∈(0,Hs2),
and interval (0,Hs2) is known as the interval of absolute stability of the method.
3. Construction of the Methods
In this section, the fourth-and fifth-order SIHMs which require only three and four stages respectively are obtained. The derivations of the methods are based on the order conditions, dispersive and dissipative error, and minimization of the error constant Cp+1 of the method. The error constant is defined by
(26)Cp+1=∥(ep+1(t1)),…,(ep+1(tk))∥2,
where k is the number of trees of order p+2(p(ti)=p+2), for the pth-order method and (ep+1(ti)) is the local truncation error defined in Coleman [4].
The Order conditions of hybrid method given in [4] are
Order 2
(27)∑bi=1,
Order 3
(28)∑bici=0,
Order 4
(29)∑bici2=16,∑biaij=112,
Order 5
(30)∑bici3=0,∑biciaij=112,∑biaijcj=0,
Order 6
(31)∑bici4=0,∑bici2aij=130,∑biciaijcj=-160,∑biaijaik=7120,∑biaijcj2=1180,∑biaijajk=1360.
For the method, ci need to satisfy
(32)∑aij=(ci2+ci)2,(i=1,…,s).
3.1. SIHM with Three Stages
To derive the fourth-order SIHM method, we use the algebraic conditions up to order four (27)–(29), simplifying condition (32), zero dissipation conditions (β1=γ,β2=0), and dispersion relation of order six (q=6), (19). The resulting system of equations consists of five nonlinear equations with seven unknown variables to be solved. Therefore, we have two degrees of freedom. The coefficients of the methods are determined in terms of the arbitrary parameters c3 and a33 which are given by the expressions
(33)b1=16+6c3,b2=6c3-16c3,b3=16c3(1+c3),a31=-c3(30a33c3-c3-1)30,a32=715c32+a33c32+715c3-a33.
By minimizing the error constant from (26) we have c3=9/10 and a33=1/30. This method is denoted as SIHM3(4) which is given below:
(34)-1000091031001924130557222750513.
With this solution, the norms of the principal local truncation error coefficient for yn are given by
(35)∥τ(5)∥2=1.863×10-2,
where ∥τ(5)∥2 are the error equations for the fifth-order methods. This formula has dispersive order six and zero dissipation with a dispersion constant (13/302400)H7+O(H9). The interval of periodicity of the method is (0, 2.96).
Meanwhile, to derive the three-stage fifth-order SIHM method, the algebraic conditions (27) to (30) and equation (32) with dispersion relation of order six (q=6), equation (19), and zero dissipation conditions (β1=γ,β2=0) are solved simultaneously. This involves seven equations and seven unknowns need to be solved; hence it has a unique solution. This method is denoted as SIHM3(5) whose coefficients are given below:
(36)-100001130141513011256112.
With this solution, the norm of the principal local truncation error coefficient for yn is given by
(37)∥τ(6)∥2=1.147×10-1,
where ∥τ(6)∥2 are the error equations for the sixth-order method. This formula has dispersive order six and zero dissipation with a dispersion constant (13/302400)H7+O(H9). The interval of periodicity of the method is (0, 2.96).
3.2. SIHM Order Five with Four Stages
The SIHM method of order five is obtained by considering the order conditions up to order five which are (27) to (30) and (32) together with dispersion relations up to order eight, ((19) and (20)). Solving all the conditions simultaneously, and then the following family of solutions in terms of free parameters a41 and b3 is obtained:
(38)a31=360a41b3-30a41+1360b3,a32=-360a41b3+30a41-1+360b32+330b3360b3,a42=-360a41b3+30a41-29+360b32+330b330(12b3-1),a43=--1+20b320(12b3-1),a33=-b3+112,a44=-b3+112,2222b1=112,b2=56,b4=-b3+112,2222222222c3=1,c4=1.
By minimizing the error norm expression, we have a41=150617/771120 and b3=23/324.
With this solution, the norm of the principal local truncation error coefficient for yn is given by
(39)∥τ(6)∥2=9.772×10-2.
This fifth-order formula is dispersive order eight and dissipative order five with dispersion and dissipation constants are (241/881798400)H9+O(H11) and (277/44089920)H6+O(H8) respectively. This method is denoted as SIHM4(5), the coefficients are given below (see: The SIHM4(5) method) and the interval of absolute stability of the method is (0, 5.75). (40)-100001199385561403142818111506177711201858328560171201811125623324181.
Table 1 shows a comparison of the properties of the methods derived.
Summary of the properties of the SIHM3(4), SIHM3(5), and SIHM4(5) methods.
Method
q
r
∥τ(p+1)∥2
DPC
DSC
S.I/P.I
SIHM3(4)
6
∞
1.863×10-2
13/302400
—
(0, 2.96)
SIHM3(5)
6
∞
1.147×10-1
13/302400
—
(0, 2.96)
SIHM4(5)
8
5
9.772×10-2
241/881798400
277/44089920
(0, 5.75)
Note that DPC is dispersion constant, DSC is dissipation constant, P.I is periodicity interval, and S.I is stability interval.
4. Problems Tested and Numerical Results
In this section, SIHM3(4) is compared with five other fourth-order methods: DIRKN three-stage fourth-order derived by Senu et al. [10], DIRKN three-stage fourth-order derived by Sommeijer [13], Classical Runge-Kutta fourth-order given in Dormand [14], explicit three-stage fourth-order hybrid method derived by Franco [3], and Classical RKN fourth order given in Hairer et al. [15]. The fifth-order methods, SIHM3(5) and SIHM4(5) are compared with four other methods: DIRKN four-stage fourth-order derived by Senu et al. [7], Classical Runge-Kutta fifth order derived by Butcher [16], explicit four-stage fifth-order hybrid method derived by Franco [3], and Classical RKN fifth-order method given from Hairer et al. [15]. All the problems were executed for tend=104 except for Orbital problem tend=100. The test problems used are listed below.
Problem 1 (Chawla and Rao [17]).
We have
(41)y′′(t)=-100y(t),y(0)=1,y′(0)=-2.
Exact solution is y=-(1/5)sin(10t)+cos(10t)
Problem 2 (Attili et al. [18]).
We have
(42)y′′(t)=-64y(t),y(0)=14,y′(0)=-12.
Exact solution is y=(17/16)sin(8t+θ),θ=π-tan-1(4)
Problem 3 (Lambert and Watson [5]).
We have
(43)d2y1(t)dt2=-v2y1(t)+v2f(t)+f′′(t),d2y2(t)dt2=-v2y2(t)+v2f(t)+f′′(t),y1(0)=a+f(0),y1′(0)=f′(0),y2(0)=f(0),y2′(0)=va+f′(0).
Exact solution is y1(t)=acos(vt)+f(t),y2(t)=asin(vt)+f(t), and f(t)is chosen to be e-0.05t and parameters v and a are 20 and 0.1 respectively.
Problem 4 (an almost periodic orbit problem given in Stiefel and Bettis [19]).
We have
(44)y1′′(t)+y1(t)=0.001cos(t),y1(0)=1,y1′(0)=0,y2′′(t)+y2(t)=0.001sin(t),y2(0)=0,y2′(0)=0.9995.
Exact solution is y1=cos(t)+0.0005tsin(t), y2=sin(t)-0.0005tcos(t).
Problem 5 (inhomogeneous system studied by Franco [1]).
We have
(45)y′′(t)+(13-12-1213)y(t)=(g1(t)g2(t)),y(0)=(10),y′(0)=(-48),g1(t)=9cos(2t)-12sin(2t),g2(t)=-12cos(2t)+9sin(2t).
Exact solutions are
(46)y(t)=(sin(t)-sin(5t)+cos(2t)sin(t)+sin(5t)+sin(2t)).
Problem 6 (Allen and Wing [20]).
We have
(47)y′′(t)=-y(t)+t,y(0)=1,y′(0)=2.
Exact solution is y=sin(t)+cos(t)+t.
Problem 7 (inhomogeneous problem studied by Papadopoulos et al. [21]).
We have
(48)y′′(t)=-v2y(t)+(v2-1)sin(t),22222y(0)=1,y′(0)=v+1,
where v≫1.
Exact solution is y(t)=cos(vt)+sin(vt)+sin(t). Numerical result is for the case v=10.
Problem 8 (orbital problem studied by van der Houwen and Sommeijer [22]).
We have
(49)y1′′(t)=-4t2y1-2y2y12+y22,2y1(t0)=0,y′(t0)=-2π,(50)y2′′(t)=-4t2y2+2y1y12+y22,y2(t0)=1,y′(t0)=0,π2≤t≤tend.
Exact solution is y1(t)=cos(t2), y2(t)=sin(t2)
The following notations are used in Figures 1–16.
SIHM3(4): a semi-implicit hybrid method of order four with dispersive order six and zero dissipation.
SIHM3(5): a semi-implicit hybrid method of order five with dispersive order six and zero dissipation.
SIHM4(5): a semi-implicit hybrid method of order five with dispersive order eight and dissipative order five.
DIRKN(S1): a three-stage fourth-order dispersive order six method with “small” dissipation constant and principal local truncation errors derived by Senu et al. [7].
DIRKN(HS): a three-stage fourth-order derived by Sommeijer [13].
DIRKN(S2): a four-stage fourth-order dispersive order eight method with “small” dissipation constant derived by Senu et al. [10].
RKN4: a classical RKN method order four in [14].
RK4: a classical RK method order four in [14].
RKN5: a five-stage fifth-order RKN method derived by Butcher [16].
RK6: a seven-stage sixth-order RK method derived by Hairer et al. [15].
EXHBRD4: a explicit three-stage fourth-order hybrid method derived by Franco [3].
EXHBRD5: a explicit four-stage fifth-order hybrid method derived by Franco [3].
The efficiency curves for SIHM3(4) method for Chawla and Rao problem with h=0.125/2i,i=3,…,7.
The efficiency curves for SIHM3(4) method for Attili problem with h=0.5/2i,i=1,…,5.
The efficiency curves for SIHM3(4) method for Lambert and Watson problem with h=0.5/2i,i=1,…,5.
The efficiency curves for SIHM3(4) method for An almost Periodic Orbit problem with h=0.125/2i,i=1,…,5.
The efficiency curves for SIHM3(4) method for Inhomogeneous system with h=0.125/2i,i=1,…,5.
The efficiency curves for SIHM3(4) method for Allen and Wing problem with h=0.125/2i,i=2,…,6.
The efficiency curves for SIHM3(4) method for Inhomogeneous problem with h=0.125/2i,i=2,…,6.
The efficiency curves for SIHM3(5) and SIHM4(5) methods for Orbital problem with h=0.125/2i,i=4,…,8.
The efficiency curves for SIHM3(5) and SIHM4(5) methods for Chawla and Rao problem with h=0.125/2i,i=3,…,7.
The efficiency curves for SIHM3(5) and SIHM4(5) methods for Attili problem with h=0.125/2i,i=1,…,5.
The efficiency curves for SIHM3(5) and SIHM4(5) methods for Lambert and Watson problem with h=0.125/2i,i=1,…,5.
The efficiency curves for SIHM3(5) and SIHM4(5) methods for An almost Periodic Orbit problem with h=0.125/2i,i=1,…,5.
The efficiency curves for SIHM3(5) and SIHM4(5) methods for Inhomogeneous system with h=0.9/2i,i=1,…,5.
The efficiency curves for SIHM3(5) and SIHM4(5) methods for Allen and Wing problem with h=0.125/2i,i=1,…,5.
The efficiency curves for SIHM3(5) and SIHM4(5) methods for Inhomogeneous problem with h=0.125/2i,i=2,…,6.
The efficiency curves for SIHM3(5) and SIHM4(5) methods for Orbital problem with h=0.125/2i,i=4,…,8.
In order to evaluate the effectiveness of the semi-implicit hybrid methods, we solved several problems which have oscillatory solutions. To make a comparison of SIHM and other existing methods, one measure of the accuracy is examined using the absolute error which is defined by
(51)Absolute error=max{|y(tn)-yn|},
where y(tn) is the exact solution and yn is the computed solution.
For comparison purposes, in analyzing the numerical results, methods of the same order will be compared. The results are given in Figures 1–16. We present the efficiency curves where the common logarithm of the maximum global error along the integration versus the CPU time is taken. From Figures 1, 2, 3, 4, 5, 6, 7, and 8, we observed that the new SIHM3(4) is the most efficient for integrating second-order differential equations possessing oscillatory solutions, followed by diagonally implicit DIRKN(S1), original explicit hybrid method EXHBRD4, and other methods like DIRKN(HS), RKN4, and RK4.
From Figures 9, 10, 11, 12, 13, 14, 15, and 16, for the fifth-order methods we observed that SIHM4(5) is the most efficient, followed by SIHM3(5) and EXHBRD5 and the rest of the methods. Even though the new methods are semi-implicit and fairly expensive in terms of time consumed, they are still more efficient compared to the explicit counterpart.
5. Conclusion
In this paper three-stage semi-implicit hybrid methods of order four and five are developed and denoted by SIHM3(4) and SIHM3(5), respectively, they have dispersion order six and zero dissipation. We also developed method of four-stage and fifth order denoted by SIHM4(5) which has dispersion order eight and dissipation order five. All the three methods developed are suitable for solving second-order IVPs which are oscillatory in nature. From the efficiency curves shown in Figures 1–16, we can conclude that all the methods are very efficient compared to the well-known existing methods of the same order in the scientific literature.
FrancoJ. M.An explicit hybrid method of Numerov type for second-order periodic initial-value problems19955917990ZBL0844.6506110.1016/0377-0427(94)00011-OMR1343144YapL. K.IsmailF.SuleimanM.AminS. Md.Block methods based on Newton interpolations for solving special second order ordinary differential equations directly200843174180ZBL1186.6508810.3844/jmssp.2008.174.180MR2502059FrancoJ. M.A class of explicit two-step hybrid methods for second-order IVPs200618714157ZBL1082.6507110.1016/j.cam.2005.03.035MR2178096ColemanJ. P.Order conditions for a class of two-step methods for y″=f(x,y)2003232197220ZBL1022.6508010.1093/imanum/23.2.197MR1974223LambertJ. D.WatsonI. A.Symmetric multistep methods for periodic initial value problems1976182189202MR0431691Van de VyverH.A symplectic Runge-Kutta-Nyström method with minimal phase-lag20073671-21624ZBL1209.6507510.1016/j.physleta.2007.02.066MR2320869SenuN.SuleimanM.IsmailF.OthmanM.A singly diagonally implicit Runge-Kutta-Nyström method for solving oscillatory problems2011412155161ZBL1229.65124MR2841173BrusaL.NigroL.A one-step method for direct integration of structural dynamic equations19801556856992-s2.0-0019010803van der HouwenP. J.SommeijerB. P.Explicit Runge-Kutta (-Nyström) methods with reduced phase errors for computing oscillating solutions1987243595617ZBL0624.6505810.1137/0724041MR888752SenuN.SuleimanM.IsmailF.OthmanM.A fourth-order diagonally implicit Runge-Kutta- Nyström method with dispersion of high orderProceedings of the 4th International Conference on Applied Mathematics, Simulation, Modelling (ASM '10)July 201078822-s2.0-79958724693KostiA. A.AnastassiZ. A.SimosT. E.Construction of an optimized explicit Runge-Kutta-Nyström method for the numerical solution of oscillatory initial value problems2011611133813390ZBL1222.6506610.1016/j.camwa.2011.04.046MR2802004KostiA. A.AnastassiZ. A.SimosT. E.An optimized explicit Runge-Kutta-Nyström method for the numerical solution of orbital and related periodical initial value problems2012183347047910.1016/j.cpc.2011.11.002MR2875136SommeijerB. P.A note on a diagonally implicit Runge-Kutta-Nyström method198719339539910.1016/0377-0427(87)90208-1MR907809ZBL0637.65065DormandJ. R.1996Boca Raton, Fla, USACRC Pressxiv+368Library of Engineering MathematicsMR1383317HairerE.NørsettS. P.WannerG.2010Berlin, GermanySpringerButcherJ. C.20082ndChichester, UKJohn Wiley & Sonsxx+46310.1002/9780470753767MR2401398ChawlaM. M.RaoP. S.High-accuracy P-stable methods for y″=f(x,y)19855221522010.1093/imanum/5.2.215MR789886AttiliB. S.FuratiK.SyamM. I.An efficient implicit Runge-Kutta method for second order systems2006178222923810.1016/j.amc.2005.11.044MR2248483ZBL1100.65062StiefelE.BettisD. G.Stabilization of Cowell's method196913154175MR026325010.1007/BF02163234ZBL0219.65062Allen,R. C.Jr.WingG. M.An invariant imbedding algorithm for the solution of inhomogeneous linear two-point boundary value problems197414405810.1016/0021-9991(74)90004-7ZBL0273.65065MR0381323PapadopoulosD. F.AnastassiZ. A.SimosT. E.A phase-fitted Runge-Kutta-Nyström method for the numerical solution of initial value problems with oscillating solutions20091801018391846ZBL1197.6508610.1016/j.cpc.2009.05.014MR2678457van der HouwenP. J.SommeijerB. P.Diagonally implicit Runge-Kutta-Nyström methods for oscillatory problems1989262414429ZBL0676.6507210.1137/0726023MR987398