Stokes' phenomenon in continuous limits of discrete Painlevé I
Christopher J. Lustri, John R. King
TL;DR
This work investigates how Stokes' phenomenon manifests in discrete versions of Painlevé I by comparing finite-difference discretisations and the integrable first discrete Painlevé equation through exponential asymptotics. By formulating infinite-order differential equations that interpolate the discrete problems, the authors identify singulants and Stokes lines, quantify exponential switchings, and study inner problems near poles. They demonstrate that finite-K discretisations exhibit increasingly rich Stokes structures that grow with $K$, but the integrable discrete Painlevé I limit eliminates Stokes switching, consistent with the Painlevé property and preventing moveable singularities. A general Delta analysis reveals parameter regimes where Stokes effects vanish (e.g., special $oldsymbol{ riangle}$ values), highlighting a deep link between discrete integrability and continuous markers of integrability, with implications for discretisation choices in preserving qualitative asymptotics.
Abstract
We use exponential asymptotic analysis to identify the relevance of Stokes' phenomenon to integrability in discrete systems. We study Stokes' phenomenon in two discrete problems with the same (leading-order) continuous limit, a finite-difference discretisation of the first continuous Painlevé equation and the first discrete Painlevé equation, as well as a family of differential equation associated with each discrete problem. This analysis reveals two important observations. Firstly, the orderly behaviour that characterises Stokes' phenomenon in discrete equations emerges naturally from corresponding continuous differential equations as the order of the latter increases, although this is not apparent at low orders. Secondly, Stokes' phenomenon vanishes in the continuum limit of the integrable discrete equation, but not the non-integrable discrete equation. This means that subdominant exponentials do not appear in the integrable equation, and therefore do not cause moveable singularities to form in the solution. The results are clarified further by consideration of one-parameter family of difference equations that interpolates between the two considered in detail.
