Moving-Parameter Asymptotics

• Moving-Parameter Asymptotics is defined as an asymptotic regime where a varying control parameter interacts with the limiting process, altering standard asymptotic behaviors.
• The methodology involves techniques such as free-boundary transformations, matched asymptotics, and Riemann–Hilbert analysis to capture phenomena like depinning, phase transitions, and scaling changes.
• Applications span ergodic depinning, orthogonal polynomial limits, and Lévy process boundary problems, providing insights into nonuniform regimes and parameter-dependent transitions.

Moving-parameter asymptotics denotes an asymptotic regime in which the quantity of interest is governed by a parameter that is itself varied, generated, or coupled to the limiting process. In the cited literature, this includes depinning laws with $\mu\to0^+$ in inhomogeneous media, free boundaries $s(t)$ whose evolution is part of the solution, moving barriers $1\pm t^\gamma$ in first-passage problems, large inhomogeneity parameters in Painlevé equations that force an $\alpha$- or $n$-dependent rescaling of the independent variable, and weight parameters $\alpha,\beta,\lambda\to\infty$ in norm asymptotics for hypergeometric orthogonal polynomials (Scheel et al., 2016, Aiki et al., 2013, Aurzada et al., 2013, Schmidt et al., 2024, Sobrino et al., 2022). Across these settings, the leading law is determined not only by the limit itself, but by the way the parameter approaches a threshold, boundary, large-value regime, or geometric constraint.

1. Forms of the moving parameter

A first class consists of threshold problems in which a control parameter approaches a critical value. In depinning of fronts in ergodic media, the moving parameter is $\mu=a-a_{\mathrm c}\to0^+$, and the central observable is the average speed $\bar s(\mu)$ near the depinning threshold. In the homogeneous case one has heuristically $s\sim \mu^1$, while in ergodic media the asymptotic exponent depends on the local dimension $\kappa$ of the ergodic measure near the critical medium state (Scheel et al., 2016). In transmission problems with Robin-type coupling, the parameter is the permeability $s(t)$0, and the limits $s(t)$1 and $s(t)$2 correspond respectively to complete decoupling and full unification of the problem (Fukao, 10 Nov 2025).

A second class is formed by problems in which the “parameter” is part of the solution. In the carbonation free-boundary model, the front location $s(t)$3 is itself the moving parameter, with interface law $s(t)$4, and the large-time asymptotic statement is the two-sided diffusive scaling estimate $s(t)$5 (Aiki et al., 2013). In fast-reaction asymptotics for concrete carbonation, the small parameter $s(t)$6 collapses a diffuse reaction layer onto a moving interface $s(t)$7, while the form of the resulting sharp-interface model depends on how the transport parameter $s(t)$8 scales relative to $s(t)$9 (Evans et al., 2011).

A third class uses a time-dependent boundary or barrier. For asymptotically $1\pm t^\gamma$0-stable Lévy processes, the moving boundary is $1\pm t^\gamma$1, and the asymptotic problem asks whether the survival probability retains the same exponent as in the constant-boundary case (Aurzada et al., 2013). For random walk bridges, both the observation time $1\pm t^\gamma$2 and the bridge length $1\pm t^\gamma$3 move, and the asymptotics of $1\pm t^\gamma$4 depend on how $1\pm t^\gamma$5 approaches $1\pm t^\gamma$6 and on the ratio $1\pm t^\gamma$7 (Sloothaak et al., 2017).

A fourth class is large-parameter asymptotics in which the parameter reorganizes the geometry of the independent variable. For generalized Hastings–McLeod solutions of the inhomogeneous Painlevé-II equation, the large parameter is $1\pm t^\gamma$8, and the relevant scaled variable is $1\pm t^\gamma$9; the scaled plane splits into a pole-free region and a pole region, with algebraic asymptotics in the first and elliptic/$\alpha$0-functional asymptotics in the second (Schmidt et al., 2024). For rational Painlevé-III solutions $\alpha$1, the asymptotics are formulated in the scaled coordinate $\alpha$2, and inside the eye-shaped domain one needs the finer decomposition $\alpha$3 to resolve the local pole-zero lattice (Bothner et al., 2018).

A fifth class appears in orthogonal-polynomial theory when the weight parameter moves. The papers on Jacobi, Laguerre, and Gegenbauer polynomials treat fixed degree $\alpha$4 and fixed $\alpha$5, while $\alpha$6, $\alpha$7, or $\alpha$8 tend to $\alpha$9, and derive asymptotics for $n$0-norms, Shannon-type integrals, and complexity-related quantities (Sobrino et al., 2021, Sobrino et al., 2022). Related two-parameter behavior occurs in quasilinear parabolic equations with small viscosity $n$1 and steep initial-layer parameter $n$2, where the structure of the asymptotic expansion depends on the ratio $n$3 or $n$4 (Zakharov, 2015).

2. Mechanisms that determine the leading law

A recurrent mechanism is reduction to a simpler effective dynamics followed by identification of the singular contribution. In depinning, the reduced front-position equation $n$5 yields the passage-time identity

$n$6

and near the critical state one has $n$7. The singularity of the integral is then controlled by the local dimension $n$8 through $n$9, which produces the trichotomy $\alpha,\beta,\lambda\to\infty$0, $\alpha,\beta,\lambda\to\infty$1, or $\alpha,\beta,\lambda\to\infty$2 (Scheel et al., 2016).

In free-boundary carbonation, the decisive object is not a singular integral but an integrated moment identity containing the term $\alpha,\beta,\lambda\to\infty$3. This is why the natural growth law is diffusive. The proof of the lower bound combines positivity of the incoming concentration $\alpha,\beta,\lambda\to\infty$4, the energy inequality, the interface law $\alpha,\beta,\lambda\to\infty$5, and the moment relation; the upper bound follows by dropping nonnegative terms and using boundedness of $\alpha,\beta,\lambda\to\infty$6 and $\alpha,\beta,\lambda\to\infty$7 (Aiki et al., 2013).

In moving-boundary first-passage for Lévy processes, the central mechanism is a decomposition

$\alpha,\beta,\lambda\to\infty$8

where $\alpha,\beta,\lambda\to\infty$9 is a one-sided subordinator extracted from the relevant Lévy tail. The subordinator tracks the moving boundary $\mu=a-a_{\mathrm c}\to0^+$0, while the remainder process $\mu=a-a_{\mathrm c}\to0^+$1 retains the constant-boundary persistence exponent. This yields

$\mu=a-a_{\mathrm c}\to0^+$2

whenever $\mu=a-a_{\mathrm c}\to0^+$3 and the relevant tail of the Lévy measure is regularly varying with index $\mu=a-a_{\mathrm c}\to0^+$4 (Aurzada et al., 2013).

In orthogonal-polynomial parameter asymptotics, the main mechanism is degeneration of the polynomial family under parameter growth. For Jacobi,

$\mu=a-a_{\mathrm c}\to0^+$5

while for Gegenbauer,

$\mu=a-a_{\mathrm c}\to0^+$6

For Laguerre, a shifted large-$\mu=a-a_{\mathrm c}\to0^+$7 limit yields a Hermite profile,

$\mu=a-a_{\mathrm c}\to0^+$8

so the moving parameter changes not only the size but the effective local polynomial model (Sobrino et al., 2022).

A further mechanism is constraint formation by penalization. In the transmission problem, the energy

$\mu=a-a_{\mathrm c}\to0^+$9

has the same bulk part for all $\bar s(\mu)$0, but the term $\bar s(\mu)$1 disappears as $\bar s(\mu)$2 and enforces $\bar s(\mu)$3 on $\bar s(\mu)$4 as $\bar s(\mu)$5. This suggests that the parameter acts as a soft-to-hard transmission constraint (Fukao, 10 Nov 2025).

3. Analytical frameworks

The literature uses several distinct analytical frameworks, each adapted to a different kind of moving parameter. In translation-equivariant front propagation, the decisive reduction is to an invariant manifold $\bar s(\mu)$6 carrying the skew-product dynamics

$\bar s(\mu)$7

The asymptotic theorem is then an ergodic statement about the reciprocal speed (Scheel et al., 2016).

Free-boundary reaction models instead rely on fixed-domain transforms and energy methods. In the carbonation problem, the rescaling $\bar s(\mu)$8 rewrites the PDE on the fixed cylinder $\bar s(\mu)$9, and weak-solution theory is built by truncation of the nonlinear exchange term, positivity and $s\sim \mu^1$0 bounds, energy estimates, and passage to the limit (Aiki et al., 2013). In the fast-reaction sharp-interface problem, matched asymptotics separates outer regions from one or several inner reaction layers, and the resulting interface conditions $s\sim \mu^1$1, $s\sim \mu^1$2 depend on the chosen scaling regime (Evans et al., 2011).

Integrable large-parameter problems are treated by Deift–Zhou nonlinear steepest descent for Riemann–Hilbert problems. For generalized Hastings–McLeod functions, the large-$s\sim \mu^1$3 analysis uses a genus-zero $s\sim \mu^1$4-function in the pole-free region and a genus-one $s\sim \mu^1$5-function in the pole region; the leading approximation is algebraic in one regime and $s\sim \mu^1$6-functional in the other, with errors $s\sim \mu^1$7 away from the exceptional set (Schmidt et al., 2024). Rational Painlevé-III solutions are likewise analyzed through a Riemann–Hilbert representation, but with a macroscopic/microscopic split $s\sim \mu^1$8, an eye-shaped genus transition in the $s\sim \mu^1$9-plane, and a different asymptotic structure at half-integer values of the secondary parameter $\kappa$0 (Bothner et al., 2018).

Moving-boundary persistence problems use fluctuation theory rather than PDE reduction. For Lévy processes, the method combines a $\kappa$1-dependent decomposition into a remainder plus a subordinator, Laplace-transform estimates, renewal functions for ladder processes, and local limit estimates (Aurzada et al., 2013). For random walk bridges, the near-endpoint regime requires an asymptotic density for $\kappa$2 under survival, Doney-type local estimates for killed random walks, and a final integration against the Gaussian bridge kernel (Sloothaak et al., 2017).

Parameter asymptotics of orthogonal-polynomial norms are dominated by explicit integral analysis. The $\kappa$3 regime for weighted norms is handled by Laplace’s method around the maximizer of $\kappa$4; endpoint-dominant Jacobi unweighted norms use an endpoint Laplace expansion; and the large-parameter Laguerre case uses asymptotic expansions for generalized integrals together with family limits to Hermite (Sobrino et al., 2022). The Jacobi complexity paper follows a related route: fixed-degree endpoint limits, hypergeometric evaluations, and gamma/digamma asymptotics produce formulas such as $\kappa$5 and $\kappa$6 (Sobrino et al., 2021).

Variational formulations supply another general framework. Brownian-motion probabilities over globally subanalytic sets are reduced to one-variable radial Gaussian integrals with constructible amplitude, leading to Puiseux-log expansions as $\kappa$7 and constructible asymptotics as $\kappa$8 (Kaiser et al., 2017). In the transmission problem, the singular limits $\kappa$9 and $s(t)$00 are encoded by Mosco convergence of the convex energies $s(t)$01 to $s(t)$02 and $s(t)$03, respectively (Fukao, 10 Nov 2025).

4. Regime changes and phase transitions

Many moving-parameter problems exhibit genuine phase transitions, meaning that the leading asymptotic form changes qualitatively at a critical scaling. In ergodic depinning, the critical quantity is the local dimension $s(t)$04. The speed law is “soft depinning” for $s(t)$05, “critical logarithmic depinning” for $s(t)$06, and “hard depinning” for $s(t)$07: $s(t)$08 This is the sharpest statement in the paper and is directly tied to how often the ergodic medium samples the most pinning state (Scheel et al., 2016).

In random walk bridges, the phase transition occurs when the observation time approaches the terminal conditioning time. If $s(t)$09, the bridge probability retains a regularly varying $s(t)$10-type behavior with slowly varying factor $s(t)$11. If $s(t)$12, the asymptotic depends on the comparison scale $s(t)$13, with three regimes: $s(t)$14 The critical regime is described by the scaling function

$s(t)$15

This is a direct phase transition in the moving ratio $s(t)$16 (Sloothaak et al., 2017).

In integrable systems, the regime change is geometric. For generalized Hastings–McLeod functions, the scaled plane divides into a pole-free region and a pole region; the first is governed by the cubic algebraic equation $s(t)$17, while the second requires a genus-one spectral curve and a $s(t)$18-functional formula (Schmidt et al., 2024). For rational Painlevé-III solutions, the eye-shaped domain $s(t)$19 confines poles and zeros in the scaled $s(t)$20-plane, the interior carries a locally uniform lattice when $s(t)$21, and the half-integer cases replace the two-dimensional pole field by accumulation along “eyebrows” (Bothner et al., 2018). The cited paper explicitly states that the limits $s(t)$22 and $s(t)$23 do not commute.

Fast-reaction interface models show a different kind of regime change: the limiting free-boundary problem depends on how transport coefficients scale with the fast-reaction parameter. When $s(t)$24, one obtains one-phase sharp-interface limits; when $s(t)$25, two-phase Stefan-type limits remain; and rapidly varying diffusivities can produce a nonstandard two-scale moving-boundary problem in which the macroscopic interface law is closed by a microscale boundary-value problem (Evans et al., 2011). The two-parameter parabolic Cauchy problem displays the same principle in a different form: the structure of the formal expansion depends on whether one studies $s(t)$26 or $s(t)$27, and the relevant inner variables change accordingly (Zakharov, 2015).

Transmission problems provide a particularly clean soft-to-hard transition. As $s(t)$28, the interface Robin term vanishes and the limit is two independent Neumann problems; as $s(t)$29, the trace mismatch satisfies $s(t)$30 on $s(t)$31, and the limit is a unified regime with continuity of state and continuity of flux across the interface (Fukao, 10 Nov 2025).

5. Representative asymptotic laws

Representative leading laws from the literature can be organized as follows.

Setting	Moving regime	Leading asymptotic
Front depinning in ergodic media	$s(t)$32	$s(t)$33, $s(t)$34, or $s(t)$35
Carbonation free boundary	$s(t)$36	$s(t)$37
Lévy moving boundary	$s(t)$38, $s(t)$39	$s(t)$40
Random walk bridge	$s(t)$41	three regimes by $s(t)$42
Jacobi Shannon spreading length	$s(t)$43, fixed $s(t)$44	$s(t)$45
Laguerre weighted norm	$s(t)$46, fixed $s(t)$47	$s(t)$48
Transmission problem	$s(t)$49 or $s(t)$50	$s(t)$51 and $s(t)$52 solution rates

These laws are not interchangeable. In depinning, the law is dictated by local geometric sampling of a bottleneck state (Scheel et al., 2016). In carbonation, the diffusive $s(t)$53 scale comes from a moment identity featuring $s(t)$54 (Aiki et al., 2013). In Lévy persistence, the exponent $s(t)$55 is stable under boundary motion only below the threshold $s(t)$56 (Aurzada et al., 2013). In fixed-degree Jacobi asymptotics, the large-parameter law $s(t)$57 is accompanied by $s(t)$58, $s(t)$59, and limiting constants for the Cramér–Rao, Fisher–Shannon, and LMC complexities (Sobrino et al., 2021). In Laguerre parameter asymptotics, more than one large-$s(t)$60 regime appears, including a Hermite-controlled asymptotic after the shift $s(t)$61 (Sobrino et al., 2022).

A plausible implication is that “moving-parameter asymptotics” is less a single technique than a family of singular-limit problems in which the decisive object may be a residence-time integral, a free-boundary moment identity, a ladder-process renewal function, a variational penalty, or a spectral-curve transition.

6. Scope, limitations, and boundary cases of the concept

The cited literature also marks the boundaries of the subject. One paper on generalized linear mixed models states explicitly that it is not a moving-parameter asymptotics paper in the usual LAN/local-alternative/triangular-array sense. Its asymptotic content is instead a parameter-structured “precise asymptotics” with two sample-size indices $s(t)$62, block-dependent rates, and asymptotic mutual independence between fixed effects, random-effects covariance, and dispersion parameters (Bhaskaran et al., 2022). This is still relevant because it exhibits parameter-block-dependent normalization, but it does not fit the local-to-boundary or singular-geometry pattern of the other examples.

Several papers identify open problems where fixed-parameter asymptotics cease to be uniform. The Painlevé-III study is explicit that the fixed-$s(t)$63 theory breaks down near half-integers and that a double-scaling limit in which $s(t)$64 approaches $s(t)$65 as $s(t)$66 is needed to describe the “closing of the eye” (Bothner et al., 2018). The Lévy moving-boundary paper proves persistence stability only in the subcritical regime $s(t)$67 and does not establish a theorem at the critical threshold $s(t)$68 (Aurzada et al., 2013). The Jacobi complexity paper treats $s(t)$69 with fixed $s(t)$70, but states that the regime $s(t)$71 with fixed $s(t)$72 is open for its LMC/Fisher–Shannon analysis and identifies varying Jacobi polynomials as a separate open problem (Sobrino et al., 2021).

Some results are explicitly formal rather than fully justified. The two-parameter parabolic paper constructs formal asymptotic expansions and emphasizes that the structure depends on the relation between $s(t)$73 and $s(t)$74, but it does not present a global composite error theory (Zakharov, 2015). The fast-reaction carbonation paper derives one-phase, two-phase, and micro-macro moving-boundary limits by matched asymptotics, while noting that well-posedness of the two-scale free-boundary problems, convergence proofs, and corrector estimates remain open (Evans et al., 2011). The Brownian-motion paper proves full Puiseux-log asymptotic expansions for fixed globally subanalytic sets, but its strongest parameter-uniform definability statement is confined to the univariate spatial case (Kaiser et al., 2017).

These limitations are substantive rather than peripheral. They indicate that moving-parameter asymptotics is often controlled by nonuniformity: approaching a threshold, coalescing roots, changing genus, forcing a hard interface constraint, or coupling multiple scales can invalidate a fixed-parameter expansion and require a new asymptotic geometry.