Fenichel-like Theory in Fast-Slow Dynamics

• Fenichel-like theory is an extension of the classical framework, establishing invariant slow manifolds across ODEs, PDEs, maps, and Banach-space systems.
• It refines traditional hyperbolicity by using spectral gaps, parameterized manifolds, and algebraic reductions to derive precise slow dynamics.
• The theory underpins explicit constructions and convergence proofs, advancing analysis in discontinuous, stochastic, and infinite-dimensional contexts.

Searching arXiv for recent and foundational papers on Fenichel-like theory across ODEs, PDEs, maps, and Banach-space settings. Fenichel-like theory denotes a family of extensions and analogues of the classical Tikhonov–Fenichel framework for fast-slow dynamics. In the classical finite-dimensional setting, one starts from a singularly perturbed system, passes to the limit $\varepsilon=0$ to obtain a critical manifold defined by an algebraic fast equilibrium relation, and then uses normal hyperbolicity to justify persistence of a nearby slow manifold together with convergence of the reduced dynamics. Recent work preserves this structural template in several nonclassical settings—discrete maps, parameterized critical manifolds, rate-based algebraic reductions, linear fast-reaction PDEs, Banach-space differential equations with slowly evolving parameters, regularized discontinuous foliations, and stochastic linear-noise reductions—while modifying the notions of hyperbolicity, invariance, and reduction to fit the ambient category (Jelbart et al., 2022, Kuehn et al., 2024, Doorakkers et al., 3 Oct 2025).

1. Classical template and the meaning of the analogy

The common reference point is the standard fast-slow ODE

$$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$

with singular limit

$0=F(u^0,v^0,0),\qquad \partial_t v^0=G(u^0,v^0,0).$

In the formulation recalled in the infinite-dimensional survey, Tikhonov gives convergence to the reduced problem, while Fenichel proves persistence of a compact normally hyperbolic critical manifold $S_0$ to a nearby locally invariant slow manifold $S_\varepsilon$, together with convergence of the restricted flow (Kuehn et al., 2024).

The later literature broadens this picture in two main directions. One direction preserves the dynamical conclusions—slow manifold persistence, attraction, and reduced-flow convergence—but changes the ambient phase space or the time-evolution object, as in maps, PDEs, and Banach-space ODEs. The other direction preserves the reduced slow dynamics while changing the computational formalism, for example by working with parameterized critical manifolds or with rate separations instead of a priori slow and fast variables (Feliu et al., 2019).

A useful intrinsic formulation, emphasized in coordinate-free reduction theory, starts from

$$\dot x = h^{(0)}(x)+\varepsilon h^{(1)}(x)+\varepsilon^2\cdots$$

and assumes near $a\in Z:=V(h^{(0)})$ that

$$\operatorname{rank}Dh^{(0)}(x)=r,\qquad \mathbb R^n=\ker Dh^{(0)}(x)\oplus \operatorname{im}Dh^{(0)}(x),$$

with every nonzero eigenvalue of $Dh^{(0)}(x)$ having negative real part on $Z$. Under these hypotheses, the system is locally equivalent to Tikhonov standard form, and the reduced vector field is the projection of $$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$0 onto $$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$1 along $$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$2 (Feliu et al., 2019).

2. Recurring structures across the literature

Despite major differences in setting, the same few ingredients recur: a singular object at $$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$3, a substitute for normal hyperbolicity, a nearby invariant slow structure for $$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$4, and a reduced dynamics obtained by restriction or projection.

Setting	Critical object / hyperbolicity test	Fenichel-like conclusion
Discrete fast-slow maps (Jelbart et al., 2022)	$$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$5; nontrivial multipliers satisfy $$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$6	Persistence of $$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$7, $$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$8, and invariant fiber foliations
Linear fast-reaction PDE (Kuehn et al., 2024)	$$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$9; modewise fast spectral separation after Fourier transform	Exact invariant attracting slow manifold $0=F(u^0,v^0,0),\qquad \partial_t v^0=G(u^0,v^0,0).$0, solution convergence, and $0=F(u^0,v^0,0),\qquad \partial_t v^0=G(u^0,v^0,0).$1
Banach-space fast-slow ODE (Doorakkers et al., 3 Oct 2025)	$0=F(u^0,v^0,0),\qquad \partial_t v^0=G(u^0,v^0,0).$2; uniform exponential stability of a two-parameter process	$0=F(u^0,v^0,0),\qquad \partial_t v^0=G(u^0,v^0,0).$3 invariant graphs $0=F(u^0,v^0,0),\qquad \partial_t v^0=G(u^0,v^0,0).$4, reduction map, exponential tracking, stable foliation
Parameterized critical manifold (Feliu et al., 2019)	$0=F(u^0,v^0,0),\qquad \partial_t v^0=G(u^0,v^0,0).$5 with known parameterization $0=F(u^0,v^0,0),\qquad \partial_t v^0=G(u^0,v^0,0).$6	Explicit reduced equation in parameter variables $0=F(u^0,v^0,0),\qquad \partial_t v^0=G(u^0,v^0,0).$7
Rate-based algebraic reduction (Apelt et al., 21 Jan 2025)	$0=F(u^0,v^0,0),\qquad \partial_t v^0=G(u^0,v^0,0).$8 at a Tikhonov-Fenichel parameter value	Direct reduced vector field in original coordinates; algorithmic detection of admissible reductions
Regularized discontinuous foliation (Panazzolo et al., 2017)	$0=F(u^0,v^0,0),\qquad \partial_t v^0=G(u^0,v^0,0).$9; normal hyperbolicity via $S_0$0 after blow-up	Sliding region as limit of invariant manifolds of regularized smooth systems
Stochastic LNA reduction (Eilertsen et al., 2021)	Deterministic critical manifold and an enlarged tangent fluctuation manifold $S_0$1	Slow scale linear noise approximation derived by geometric projection

This comparison shows that the adjective “Fenichel-like” usually signals structural analogy rather than literal identity with the classical theorem.

3. Coordinate-free, discrete, and algebraic reformulations

A major strand of the literature seeks to preserve the geometric content of Fenichel theory while changing coordinates, objects, or computational entry points. In discrete geometric singular perturbation theory, the basic map is written as

$S_0$2

with critical fixed-point manifold

$S_0$3

Here normal hyperbolicity is not an imaginary-axis condition but a unit-circle condition: if $S_0$4, then a point $S_0$5 is normally hyperbolic iff the nontrivial multipliers satisfy $S_0$6. The relevant projection is

$S_0$7

and the reduced map is

$S_0$8

The resulting theorems give persistence of $S_0$9, local stable and unstable manifolds, and invariant foliations by fibers. A fundamental difference from flows is that the reduced map degenerates to the identity at $S_\varepsilon$0, because there is no exact discrete analogue of slow-time rescaling (Jelbart et al., 2022).

For parameterized critical manifolds, the problem is different. The issue is not persistence but computation of the reduced vector field when the slow manifold is known as a parameterized set

$S_\varepsilon$1

If

$S_\varepsilon$2

is the intrinsic reduced equation, then the parameter dynamics is

$S_\varepsilon$3

where $S_\varepsilon$4 is characterized by

$S_\varepsilon$5

With any full-rank annihilator $S_\varepsilon$6 satisfying $S_\varepsilon$7, one obtains the explicit formula

$S_\varepsilon$8

This does not create a new slow manifold theorem; it gives a constructive pullback of the usual Tikhonov–Fenichel reduction to manifold parameters, especially useful in chemical reaction networks (Feliu et al., 2019).

A closely related but more algorithmic development replaces separation of variables by separation of rates. One starts from

$S_\varepsilon$9

with slow-fast rate scaling

$$\dot x = h^{(0)}(x)+\varepsilon h^{(1)}(x)+\varepsilon^2\cdots$$0

At a Tikhonov-Fenichel parameter value, one factors the fast field as

$$\dot x = h^{(0)}(x)+\varepsilon h^{(1)}(x)+\varepsilon^2\cdots$$1

and computes the reduction directly in original coordinates: $$\dot x = h^{(0)}(x)+\varepsilon h^{(1)}(x)+\varepsilon^2\cdots$$2 Necessary and sufficient algebraic conditions are formulated through Jacobian minors, characteristic polynomial coefficients, elimination ideals, and primary decomposition. The emphasis is explicit: this framework is an algebraic and algorithmic reduction theory, not a new general persistence theorem (Apelt et al., 21 Jan 2025).

4. Infinite-dimensional and functional-analytic extensions

The most direct infinite-dimensional prototype in the recent literature is a linear fast-reaction PDE on

$$\dot x = h^{(0)}(x)+\varepsilon h^{(1)}(x)+\varepsilon^2\cdots$$3

with

$$\dot x = h^{(0)}(x)+\varepsilon h^{(1)}(x)+\varepsilon^2\cdots$$4

and governing system

$$\dot x = h^{(0)}(x)+\varepsilon h^{(1)}(x)+\varepsilon^2\cdots$$5

where $$\dot x = h^{(0)}(x)+\varepsilon h^{(1)}(x)+\varepsilon^2\cdots$$6. Setting $$\dot x = h^{(0)}(x)+\varepsilon h^{(1)}(x)+\varepsilon^2\cdots$$7 yields the algebraic constraint

$$\dot x = h^{(0)}(x)+\varepsilon h^{(1)}(x)+\varepsilon^2\cdots$$8

and critical manifold

$$\dot x = h^{(0)}(x)+\varepsilon h^{(1)}(x)+\varepsilon^2\cdots$$9

After Fourier transform, the PDE decouples into a family of $a\in Z:=V(h^{(0)})$0 fast-slow ODEs indexed by $a\in Z:=V(h^{(0)})$1. This permits an explicit slow manifold,

$a\in Z:=V(h^{(0)})$2

which is invariant, attracting when $a\in Z:=V(h^{(0)})$3, and infinite-dimensional. The paper proves strong $a\in Z:=V(h^{(0)})$4 convergence of solutions, convergence of restricted semiflows, and

$a\in Z:=V(h^{(0)})$5

Because the construction is modewise and linear, no nonlinear graph transform or Lyapunov–Perron method is needed, and no spectral-gap condition is required for manifold existence (Kuehn et al., 2024).

A different infinite-dimensional extension treats

$a\in Z:=V(h^{(0)})$6

with $a\in Z:=V(h^{(0)})$7 in a Banach space $a\in Z:=V(h^{(0)})$8 and $a\in Z:=V(h^{(0)})$9. The critical manifold is

$$\operatorname{rank}Dh^{(0)}(x)=r,\qquad \mathbb R^n=\ker Dh^{(0)}(x)\oplus \operatorname{im}Dh^{(0)}(x),$$0

and the main hyperbolicity input is a functional-analytic replacement for finite-dimensional normal hyperbolicity: uniform exponential stability of the nonautonomous linear process

$$\operatorname{rank}Dh^{(0)}(x)=r,\qquad \mathbb R^n=\ker Dh^{(0)}(x)\oplus \operatorname{im}Dh^{(0)}(x),$$1

Under assumptions on this process, on nonlinear Lipschitz bounds, and on slow drift, the theory yields $$\operatorname{rank}Dh^{(0)}(x)=r,\qquad \mathbb R^n=\ker Dh^{(0)}(x)\oplus \operatorname{im}Dh^{(0)}(x),$$2 slow manifolds

$$\operatorname{rank}Dh^{(0)}(x)=r,\qquad \mathbb R^n=\ker Dh^{(0)}(x)\oplus \operatorname{im}Dh^{(0)}(x),$$3

with $$\operatorname{rank}Dh^{(0)}(x)=r,\qquad \mathbb R^n=\ker Dh^{(0)}(x)\oplus \operatorname{im}Dh^{(0)}(x),$$4 uniformly as $$\operatorname{rank}Dh^{(0)}(x)=r,\qquad \mathbb R^n=\ker Dh^{(0)}(x)\oplus \operatorname{im}Dh^{(0)}(x),$$5. It also yields a reduction map

$$\operatorname{rank}Dh^{(0)}(x)=r,\qquad \mathbb R^n=\ker Dh^{(0)}(x)\oplus \operatorname{im}Dh^{(0)}(x),$$6

and exponential tracking

$$\operatorname{rank}Dh^{(0)}(x)=r,\qquad \mathbb R^n=\ker Dh^{(0)}(x)\oplus \operatorname{im}Dh^{(0)}(x),$$7

The construction is Lyapunov–Perron based and treats attracting manifolds completely; finite-dimensional unstable directions are only sketched in an extension (Doorakkers et al., 3 Oct 2025).

These two infinite-dimensional theories are complementary. The linear PDE model gives an explicit exact prototype. The Banach-space theory gives an abstract persistence-and-reduction framework, but only for finite-dimensional slow variables and, in the main results, only in the attracting case. This suggests that “Fenichel-like” in infinite dimensions ranges from exact solvable linear models to fully functional-analytic invariant-manifold theory.

5. Regularization, stochastic reduction, and loss of normal hyperbolicity

In discontinuous dynamics, Fenichel-like theory appears after regularization and blow-up. For a discontinuous oriented $$\operatorname{rank}Dh^{(0)}(x)=r,\qquad \mathbb R^n=\ker Dh^{(0)}(x)\oplus \operatorname{im}Dh^{(0)}(x),$$8-foliation with smooth codimension-one discontinuity set $$\operatorname{rank}Dh^{(0)}(x)=r,\qquad \mathbb R^n=\ker Dh^{(0)}(x)\oplus \operatorname{im}Dh^{(0)}(x),$$9, a transition regularization has local generator

$Dh^{(0)}(x)$0

After the directional blow-up $Dh^{(0)}(x)$1, $Dh^{(0)}(x)$2, the desingularized vector field takes slow-fast form

$Dh^{(0)}(x)$3

The critical manifold on the exceptional divisor is defined by the height function

$Dh^{(0)}(x)$4

namely

$Dh^{(0)}(x)$5

and normal hyperbolicity is the transversality condition

$Dh^{(0)}(x)$6

Exactly here Fenichel theory is used: normally hyperbolic branches persist as invariant manifolds of the blown-up smooth system, and their blow-downs define sliding regions of the regularized discontinuous dynamics. The resulting criterion is

$Dh^{(0)}(x)$7

A notable consequence is that sliding can depend on the chosen transition function $Dh^{(0)}(x)$8, not only on the discontinuous vector field (Panazzolo et al., 2017).

Stochastic reduction near the thermodynamic limit exhibits a different adaptation. For the linear noise approximation,

$Dh^{(0)}(x)$9

the reduction is performed not on the deterministic critical manifold alone but on an enlarged manifold of states and tangent fluctuations. If the deterministic singular drift factors as $Z$0, the corresponding projection is

$Z$1

and the slow scale linear noise approximation is obtained by projecting both the perturbation and the noise onto the tangent bundle of the enlarged critical manifold $Z$2. In standard form, with $Z$3, the reduced fluctuation equation becomes

$Z$4

The paper argues that this GSPT-based derivation explains why existing ssLNAs differ: one must reduce using the singular geometry, not the full perturbed Jacobian. It also shows that loss of normal hyperbolicity in reverse QSSA forces branchwise reduction rather than a single global stochastic QSSA (Eilertsen et al., 2021).

The biochemical literature makes the boundary of the classical theory explicit. For the intermolecular autocatalytic zymogen activation reaction, small $Z$5 yields a genuine Fenichel regime with critical manifold $Z$6 and reduced equation

$Z$7

Small $Z$8 gives another Fenichel regime with critical manifold $Z$9. By contrast, the singular limit

$$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$00

produces a dynamic transcritical bifurcation: the critical set is the union of two lines, normal hyperbolicity is lost at their intersection, and tracking errors scale like $$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$01 rather than $$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$02. The same reduced equation may therefore arise either from Fenichel theory or from center manifold theory, depending on the path in parameter space. The paper further interprets chemical reversibility as a dynamic imperfection that destroys the transcritical structure (Eilertsen et al., 2021).

6. Stronger asymptotics, misconceptions, and scope

An important strengthening of the classical picture appears in multiscale malaria models. There the host variables are slow, vector variables are fast, and the critical manifold is given by the quasi-steady-state relation $$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$03. Standard Tikhonov–Fenichel theory gives finite-time approximation, but the paper adds asymptotic stability of the reduced flow and proves a uniform-in-time result on $$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$04. In its abstract form, the Chapman–Enskog approximation errors satisfy

$$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$05

after inclusion of bulk and initial-layer corrections. The first-order bulk correction to the slow manifold,

$$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$06

is identified explicitly with the first-order approximation of the Fenichel invariant manifold obtained from the invariance equation. The paper is equally explicit that classical Fenichel theory does not itself provide the initial-layer terms or the large-time estimates (Banasiak et al., 2023).

Several recurrent misconceptions are corrected by the modern literature. First, “Fenichel-like theory” is not synonymous with a single general persistence theorem. The parameterized-manifold and rate-based algebraic papers provide constructive reduction formulas or algorithmic detection of reductions, while relying on already available Tikhonov–Fenichel hypotheses for invariant-manifold existence (Feliu et al., 2019, Apelt et al., 21 Jan 2025). Second, normal hyperbolicity is category-dependent: it is an annular spectral gap about the unit circle for maps, uniform exponential stability of a two-parameter process for Banach-space systems, a modewise fast spectral condition in the linear PDE model, and the nonvanishing derivative $$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$07 for regularized discontinuous foliations (Jelbart et al., 2022, Doorakkers et al., 3 Oct 2025, Kuehn et al., 2024, Panazzolo et al., 2017). Third, reduced slow dynamics need not be obtained in the same way across settings: in discrete time the reduced map necessarily degenerates to the identity at $$\varepsilon \partial_t u^\varepsilon = F(u^\varepsilon,v^\varepsilon,\varepsilon),\qquad \partial_t v^\varepsilon = G(u^\varepsilon,v^\varepsilon,\varepsilon),$$08, and in stochastic LNA theory the reduced noise must be derived from the singular geometry rather than by ad hoc elimination (Jelbart et al., 2022, Eilertsen et al., 2021).

The scope limits are equally sharp. The explicit PDE prototype is essentially linear, and the paper identifies nonlinear Fourier-mode mixing as the major obstruction to a broader infinite-dimensional GSPT (Kuehn et al., 2024). The Banach-space theory treats attracting critical manifolds in full detail but leaves unstable directions largely outside the main theorem, and it always keeps the slow variables finite-dimensional (Doorakkers et al., 3 Oct 2025). In nonhyperbolic problems such as dynamic transcritical bifurcations, classical Fenichel persistence fails, so only branchwise or alternative reductions are available (Eilertsen et al., 2021). Uniform-in-time approximation is exceptional rather than generic and requires stability of the reduced dynamics in addition to fast attraction (Banasiak et al., 2023).

Taken together, these developments show that Fenichel-like theory is best understood as a transferable structural paradigm: identify a singular critical object, formulate the appropriate hyperbolicity condition in the ambient category, construct or justify the perturbed slow structure, and relate the full dynamics to a reduced flow by restriction, projection, or asymptotic matching. This suggests that the enduring content of Fenichel’s theory is less the specific finite-dimensional theorem than the geometric architecture it provides for singular perturbation problems across ODEs, maps, PDEs, discontinuous systems, and stochastic approximations.