On alternating-conjugate splitting methods
J. Bernier, S. Blanes, F. Casas, A. Escorihuela-Tomàs · 2025
Read On alternating-conjugate splitting methods by J. Bernier, S. Blanes, F. Casas, A. Escorihuela-Tomàs (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-08453.txt.
Introduction
Bernier, Blanes, Casas, and Escorihuela-Tomàs present a 2025 class of splitting methods for stepping equations through time. Splitting breaks a hard flow into easy parts taken in turn. Complex step sizes can buy accuracy but often wreck long runs. The new class pairs a complex recipe with its mirror twin, built from the flipped signs of the same steps. The pair cancels the worst long-run drift.
On light-like and energy-keeping systems the paired schemes held their shape far better than past complex rivals. The team check which traits the pair keeps exactly. New members up to sixth order beat the best real-step schemes on set problem kinds. Sim folk gain a sharper tool for long quantum and orbit runs.
Catalog record
| CREATOR | J. Bernier, S. Blanes, F. Casas, A. Escorihuela-Tomàs |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 80 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-08453
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
On alternating-conjugate splitting methods J. Bernier, S. Blanes, F. Casas, A. Escorihuela-Tomàs Abstract The new class of alternating-conjugate splitting methods is presented and analyzed. They are obtained by concatenating a given composition involving complex coefficients with the same composition but with the complex conjugate coefficients. We show that schemes of this type exhibit good long time behavior when applied to linear unitary and linear Hamiltonian systems, in contrast with other methods based on complex coefficients, and study in detail their preservation properties. We also present new schemes within this class up to order 6 that exhibit better efficiency than state-of-the-art splitting methods with real coefficients for some classes of problems. 1 Introduction Splitting methods are widely used for the numerical integration in time of initial value problems defined by ordinary and partial differential equations that can be subdivided into different parts easier to solve than the original system [ 17 ] . Splitting methods are typically easy to implement, present favorable properties concerning error propagation and preserve in addition a wide variety of structural properties inherent to the continuous system they are approximating. In this respect, they constitute paradigmatic examples of geometric numerical integrators [ 4 , 14 ] . Relevant examples are Hamiltonian systems separated into kinetic and potential energy in classical mechanics and the time-dependent Schrödinger equation in quantum mechanics, where high-order splitting schemes have shown their effectiveness (see [ 9 ] for a recent review). Advantageous as they are, splitting methods have also some drawbacks. One of them is related with an essential feature of this class of schemes: they necessarily involve negative coefficients when their order is three or higher. This makes them unsuitable, in particular, for non-reversible systems such as reaction-diffusion equations, since the presence of negative coefficients induce severe instabilities in the numerical solution. Even in the context of ordinary differential equations, having negative coefficients usually leads to large truncation errors, and so it is recommended to include additional (free) parameters to reduce these errors. Splitting methods of order higher than two can still be used in equations evolved by semigroups, such as parabolic equations of evolution, if they involve instead complex coefficients having positive real part , as first shown in [ 15 , 11 ] . In that way not only the stability is recovered but in addition splitting methods of higher orders typically improve the efficiency with respect to lower orders over a wide range of accuracies [ 6 ] . In fact, methods of this class were already considered in the first explorations of splitting methods in the early 1990s [ 2 , 20 , 21 , 22 ] and later in the context of symplectic integration of Hamiltonian systems [ 12 ] and quantum mechanics [ 1 , 19 ] . This was somehow natural, since it was soon realized that the order conditions, even of low order, admit complex solutions in addition to the real ones. More recently, a systematic study of splitting methods with complex coefficients has been carried out, not only by designing new high order schemes but also by analyzing their preservation properties, for both ordinary and partial differential equations [ 7 , 5 , 3 , 8 ] . As is well known, when considering splitting methods with real coefficients, left-right palindromic (or symmetric) compositions are usually preferred, since their construction is easier and in addition they provide time-symmetric approximations to the exact solution. The situation is different, however, with complex coefficients, at least in the context of linear equations evolving in the unitary group [ 5 , 3 ] and also for parabolic problems [ 8 ] . In fact, when considering complex coefficients, other patterns are more suitable. In particular, symmetric-conjugate schemes (i.e., methods with complex coefficients which are symmetric in the real part and antisymmetric in the imaginary part) present good preservation properties in both cases. This is true even when dealing with real ODEs and the numerical approximation is projected after each time step to its real part [ 5 ] . Although the results presented in the above mentioned references concerning symmetric-conjugate splitting methods are promising for their practical application to integrate linear unitary problems and evolution equations of parabolic type, it is fair to say that the analysis of splitting methods involving complex coefficients is far from complete. In particular several questions naturally arise: are there other types of schemes also possessing good preservation properties when applied to this class of problems? In the case of real Hamiltonian systems, is the energy almost preserved for long time integrations, as for splitting methods with real coefficients? What happens when one applies these methods to nonlinear evolution problems, such as nonlinear Schrödinger or complex Ginzburg–Landau equations? In this work we address some of these issues, and in particular we present and analyze a rather general class of splitting methods with complex coefficients obtained by concatenating a given composition with the same composition but with complex conjugate coefficients. We show that schemes of this type (which can be properly called alternating-conjugate methods) exhibit a good long time behavior when they are applied to linear unitary and Hermitian flows with a sufficiently small time step, and also for linear Hamiltonian systems, due to their special spectral properties. This is in contrast with symmetric and symmetric-conjugate schemes and account for some observations previously reported in the literature. The plan of the paper is as follows. In section 2 we carry out some numerical experiments illustrating the behavior exhibited by different splitting methods of order 3 and 4, and in particular the good preservation properties of alternating-conjugate schemes. Section 3 is devoted to analyze the qualitative properties of these methods applied to linear unitary and Hermitian flows, whereas section 4 deals with linear real Hamiltonian systems. Section 5 presents an strategy to construct alternating-conjugate composition schemes based on the study of the modified vector field. The new methods are then illustrated on several numerical examples. Finally, section 6 contains some concluding remarks. 2 Some numerical observations In this section we consider the simple problem of approximating U(t)=\mathrm{e}^{itH} , where H is a N\times N Hermitian matrix. As is well known, U(t) is a unitary matrix, U^{*}(t)=U(t)^{-1}\in\mathbb{C}^{N}\times\mathbb{C}^{N} , and can be seen as the solution of the differential equation \frac{dU}{dt}=iHU,\qquad U(0)=I. (2.1) Suppose H can be decomposed as H=A+B so that computing \mathrm{e}^{itA} and \mathrm{e}^{itB} is easier (or less costly) than solving ( 2.1 ). Then it makes sense to construct splitting methods of the form \Psi_{h}^{[p]}=\mathrm{e}^{ia_{s+1}hA}\,\mathrm{e}^{ib_{s}hB}\,\mathrm{e}^{ia_{s}hA}\,\cdots\,\mathrm{e}^{ia_{2}hA}\,\mathrm{e}^{ib_{1}hB}\,\mathrm{e}^{ia_{1}hA} (2.2) with suitable coefficients a_{j} , b_{j} so that U(h)=\Psi_{h}^{[p]}+\mathcal{O}(h^{p+1}),\qquad h\rightarrow 0 for a time step t=h . Simple examples are the Lie–Trotter scheme \Psi_{h}^{[1]}=\mathrm{e}^{ihA}\,\mathrm{e}^{ihB} and the Strang splitting \Psi_{h}^{[2]}\equiv\mathrm{e}^{\frac{1}{2}ihA}\,\mathrm{e}^{ihB}\,\mathrm{e}^{\frac{1}{2}ihA}, (2.3) of order p=1 and 2, respectively. Methods with complex coefficients already exist at order 3 [ 2 ] , \Psi_{h}^{[3]}=\mathrm{e}^{ia_{1}hA}\,\mathrm{e}^{ib_{1}hB}\,\mathrm{e}^{ia_{2}hA}\,\mathrm{e}^{i\overline{b}_{1}hB}\,\mathrm{e}^{i\overline{a}_{1}hA},\qquad a_{1}=\frac{\gamma}{2},\;\;b_{1}=\gamma=\frac{1}{2}\pm i\frac{\sqrt{3}}{6},\;\;a_{2}=\frac{1}{2}, (2.4) which corresponds to the composition of the Strang splitting \Psi_{h}^{[3]}=\Psi_{\gamma h}^{[2]}\,\,\Psi_{\overline{\gamma}h}^{[2]} (2.5) and order 4, namely \displaystyle\Psi_{h,p}^{[4]}=\mathrm{e}^{ia_{1}hA}\,\mathrm{e}^{ib_{1}hB}\,\mathrm{e}^{ia_{2}hA}\,\mathrm{e}^{ib_{2}hB}\,\mathrm{e}^{ia_{2}hA}\,\mathrm{e}^{ib_{1}hB}\,\mathrm{e}^{ia_{1}hA} (2.6) \displaystyle\Psi_{h,sc}^{[4]}=\mathrm{e}^{ia_{1}hA}\,\mathrm{e}^{ib_{1}hB}\,\mathrm{e}^{ia_{2}hA}\,\mathrm{e}^{ib_{2}hB}\,\mathrm{e}^{i\overline{a}_{2}hA}\,\mathrm{e}^{i\overline{b}_{1}hB}\,\mathrm{e}^{i\overline{a}_{1}hA}. The coefficents a_{j},b_{j} in ( 2.6 ) are obtained by inserting the Strang splitting ( 2.3 ) into the triple-jump composition [ 13 , 20 , 24 ] \Psi_{\gamma_{1}h}^{[2]}\,\,\Psi_{\gamma_{2}h}^{[2]}\,\,\Psi_{\gamma_{3}h}^{[2]} (2.7) with \displaystyle\Psi_{h,p}^{[4]}:\qquad\gamma_{1}=\gamma_{3}=\frac{1}{2-2^{1/3}\mathrm{e}^{2i\ell\pi/3}},\qquad\gamma_{2}=1-2\gamma_{1},\qquad\ell=1,2 \displaystyle\Psi_{h,sc}^{[4]}:\qquad\gamma_{1}=\overline{\gamma}_{3}=\frac{1}{4}\pm i\,\frac{1}{4}\sqrt{\frac{5}{3}},\qquad\gamma_{2}=\frac{1}{2}, respectively. Notice the different pattern in the distribution of the coefficients: