Preprint
Short Note

This version is not peer-reviewed.

Predefined-Time Feasibility Under Bounded Actuation: Radial Bounds and Sharp Rotational Geometry

Submitted:

17 November 2026

Posted:

18 November 2026

You are already at the latest version

Abstract
A prescribed stabilization time under hard actuator limits is meaningful only when the initial state is reachable within that horizon. We first establish controller-independent radial arrival-time lower bounds for nonlinear control-affine systems and an explicit obstruction for globally Lipschitz dynamics. For linear systems with componentwise bounded inputs, classical support functions yield exact finite-horizon feasibility tests. The main quantitative result concerns planar rotational dynamics: we characterize precisely the time horizons for which the backward reachable set is a Euclidean disk and derive the sharp horizon-uniform Hausdorff bound for its deviation from the averaged disk, with an explicit optimal constant. An analytical example shows that a directional test can exclude an initial state that passes the simpler radial test. The results concern plant-level feasibility, not the construction of a stabilizing feedback.
Keywords: 
;  ;  ;  ;  ;  

MSC:  93B03; 93C10; 93D15

1. Introduction

A predefined-time stabilization claim specifies a uniform convergence-time bound T c > 0 as part of the design; it is stronger than convergence with an unspecified state-dependent settling time [1]. Under a hard bound on actuator authority, however, convergence-time requirements are meaningful only for initial conditions that can physically be steered to the target in time T c . The dependence of admissible stabilization domains on input constraints and prescribed convergence times has already been studied in the fixed- and prescribed-time literature [2,3,4]. Classical null-controllable regions and support-function methods [5,6], together with the standard convex-geometric framework of support functions and Minkowski addition [7], provide the mathematical background. In the linear setting below, these tools describe the time-integrated action of rotated actuator constraint sets.
This note connects these viewpoints through controller-independent feasibility tests. We derive a radial minimum-time barrier for nonlinear plants and a closed-form Lipschitz consequence, then use classical support functions to retain directional information. For planar rotational dynamics, our main quantitative results characterize exactly the horizons producing a disk and give the smallest horizon-uniform Hausdorff error constant relative to the averaged disk. These explicit specializations of classical convex reachability can be applied before controller design; they do not claim priority for the general method.
Throughout, ∥ · ∥ denotes the Euclidean norm for vectors and its induced operator norm for matrices, B 2 is the closed unit Euclidean ball, and an admissible control is a measurable function taking values in a prescribed compact set almost everywhere. The main object is the backward reachable set of the origin,
R T − ( 0 ) : = x 0 : some admissible control yields x ( T ; x 0 ) = 0 .
When the origin is an equilibrium under an admissible zero control, membership in (1) is equivalent to the ability to reach the origin by time T: after an earlier arrival, the zero control holds the state there. In particular, every predefined-time stabilization assertion on an initial domain Ω necessarily requires Ω ⊆ R T c − ( 0 ) . This inclusion is a feasibility requirement, not a sufficient condition for constructing a stabilizing feedback.

2. A Radial Minimum-Time Obstruction

Consider a control-affine system
x ˙ = f ( x ) + G ( x ) w , x ∈ R n , w ( t ) ∈ U ⊂ R m ,
where f and G are continuous and U is nonempty and compact. For r > 0 define
c ( r ) : = max ∥ x ∥ = r w ∈ U − x ⊤ ∥ x ∥ f ( x ) + G ( x ) w .
The maximum exists; continuity of f , G and compactness of the unit sphere and U also make c continuous on ( 0 , ∞ ) . It measures the largest inward radial velocity available anywhere on the sphere of radius r. Because it is a spherical maximum, it can overestimate what can be attained along any single trajectory; the resulting bound is therefore necessary rather than sufficient.
Proposition 1. 
Let an admissible absolutely continuous trajectory of (2) start at x 0 ≠ 0 and reach the origin at some finite time T. Set R = ∥ x 0 ∥ and suppose c ( r ) > 0 for 0 < r ≤ R . Then
T ≥ ∫ 0 R d r c ( r ) ,
where the integral is improper and may equal + ∞ . If c ( r ) > 0 for all r > 0 and ∫ 0 ∞ d r / c ( r ) = ∞ , no admissible controller can give a finite settling-time bound uniform over R n .
Proof. 
Write r ( t ) = ∥ x ( t ) ∥ . For almost every time at which r ( t ) > 0 , the chain rule and (3) give r ˙ ( t ) ≥ − c ( r ( t ) ) . Fix 0 < ε < R and let T ε be the first time r ( T ε ) = ε . By continuity of r, the set of times in [ 0 , T ε ] at which r ( t ) = R is nonempty and compact. Denote its last element by s ε . On [ s ε , T ε ] the radius stays in [ ε , R ] : an excursion above R would force another crossing of R. Thus
F ε ( ρ ) : = ∫ ε ρ d σ c ( σ )
is continuously differentiable along this part of the trajectory, and
d d t F ε ( r ( t ) ) = r ˙ ( t ) c ( r ( t ) ) ≥ − 1 a . e .
Integration from s ε to T ε yields T ≥ T ε − s ε ≥ ∫ ε R d ρ / c ( ρ ) . Letting ε ↓ 0 proves (4), including the divergent case. The final assertion follows by taking initial radii arbitrarily large. □
The radial integral is useful when a global control-authority envelope is available. A particularly simple alternative, requiring no computation of the spherical maximum, is available for globally Lipschitz plants.
Corollary 1. 
Suppose f ( 0 ) = 0 and, for every x , z ∈ R n ,
∥ f ( x ) − f ( z ) ∥ ≤ L f ∥ x − z ∥ , ∥ G ( x ) − G ( z ) ∥ ≤ L G ∥ x − z ∥ ,
while ∥ w ∥ ≤ U for every w ∈ U . Put a = L f + U L G and b = U ∥ G ( 0 ) ∥ . If a > 0 and b > 0 , every admissible trajectory that reaches the origin from x 0 ≠ 0 by time T satisfies
T ≥ 1 a log 1 + a ∥ x 0 ∥ b .
For a = 0 < b the bound is T ≥ ∥ x 0 ∥ / b . If b = 0 , no nonzero initial state reaches the origin in finite time. In particular, none of these globally Lipschitz plants admits a finite uniform settling-time bound over all of R n under the stated hard input bound.
Proof. 
For r ( t ) = ∥ x ( t ) ∥ > 0 ,
r ˙ ( t ) ≥ − ∥ f ( x ( t ) ) ∥ − ∥ G ( x ( t ) ) ∥ U ≥ − a r ( t ) − b
almost everywhere. Scalar comparison gives r ( t ) ≥ ( r ( 0 ) + b / a ) e − a t − b / a if a > 0 ; evaluating at a hitting time gives (5). For a = 0 , integrate r ˙ ≥ − b instead. When b = 0 , comparison yields r ( t ) ≥ r ( 0 ) e − a t > 0 for every finite t (with the evident constant-radius interpretation if a = 0 ). The lower bounds diverge with ∥ x 0 ∥ whenever b > 0 . □
For example, when a , b > 0 , an initial ball Ω = B 2 ( 0 , R ) can be stabilized by a prescribed time T c only if
R ≤ b a e a T c − 1 ,
with R ≤ b T c for a = 0 . This is an outer radius: it does not certify that every state within the ball is feasible. The obstruction relies on a global growth restriction. Hard saturation alone does not preclude a uniform-time bound if the autonomous drift supplies sufficiently strong inward motion at large radii.

3. Exact Finite-Horizon Geometry for Linear Plants

For a linear system
x ˙ = A x + w , w ( t ) ∈ [ − u ¯ , u ¯ ] n ,
the radial estimate can be improved by retaining the directional information discarded by Cauchy–Schwarz. Let R T − ( 0 ) denote the set in (1) for (7), and let h T ( p ) : = sup x ∈ R T − ( 0 ) p ⊤ x be its support function. The following formulas are standard consequences of linear variation of constants and convex reachability [5,6]; we record them to extract explicit prescribed-time tests.
Proposition 2. 
For every T > 0 , R T − ( 0 ) is compact, convex, centrally symmetric, and
R T − ( 0 ) = ∫ 0 T e − A s v ( s ) d s : v ( s ) ∈ [ − u ¯ , u ¯ ] n ,
h T ( p ) = u ¯ ∫ 0 T ∥ e − A ⊤ s p ∥ 1 d s .
Consequently, x 0 ∈ R T − ( 0 ) if and only if
p ⊤ x 0 ≤ h T ( p ) for all p ∈ R n .
For each fixed nonzero p, a support-maximizing boundary point is realized by a bang–bang measurable input.
Proof. 
Variation of constants and x ( T ) = 0 give x 0 = − ∫ 0 T e − A s w ( s ) d s . Symmetry of the input box lets us replace − w by v, proving (8). The set is a compact convex Aumann integral of compact convex-valued continuous multifunctions. For each p, maximizing the linear functional pointwise gives
max ∥ v ∥ ∞ ≤ u ¯ p ⊤ e − A s v = u ¯ ∥ e − A ⊤ s p ∥ 1 .
Integration proves (9); the maximizer has components v i ( s ) = u ¯ sgn ( ( e − A ⊤ s p ) i ) , with arbitrary values at zeros. Finally, a closed convex set is the intersection of its supporting half-spaces, which proves (10). □
The exact test separates feasibility from any particular controller synthesis. In contrast to (6), it resolves directional restrictions and can certify either inclusion or exclusion of a proposed initial state. The expression itself is classical; the useful point here is its explicit finite-time specialization below.

4. Rotational Dynamics and Time-Averaged Actuator Authority

Let n = 2 in (7), with
A = ω 0 − 1 1 0 , ω ≠ 0 .
Because A ⊤ = − A , the matrix e A s is a rotation. The Euclidean radius of a free trajectory is constant; all inward radial motion is supplied by the bounded input. It follows directly that every admissible path satisfies r ˙ ≥ − 2 u ¯ whenever r > 0 , and hence
R T − ( 0 ) ⊆ 2 u ¯ T B 2 .
The exact support function reveals considerably more.
Theorem 1. 
Let u ¯ > 0 and T > 0 . For (11) with the componentwise input constraint in (7), one has
u ¯ T B 2 ⊆ R T − ( 0 ) ⊆ 2 u ¯ T B 2 .
Moreover, the backward reachable set is a Euclidean diskif and only if T = k π / ( 2 | ω | ) for some integer k ≥ 1 . At precisely these horizons,
R T − ( 0 ) = 4 u ¯ T π B 2 .
At every horizon its deviation from the averaged disk satisfies
error R T − ( 0 ) , 4 u ¯ T π B 2 ≤ C * u ¯ | ω | ,
where theoptimal uniform constantis
C * : = 2 2 − 16 π 2 − 4 π arccos 4 π 2 ≈ 0.08435278035 .
The bound is sharp uniformly over T > 0 : no smaller constant can replace C * in (15) for all horizons.
Proof. 
Since e A s is orthogonal, ∥ p ∥ ≤ ∥ e A s p ∥ 1 ≤ 2 ∥ p ∥ . The support-function criterion for disk inclusions and (9) then give (13). For p = ∥ p ∥ ( cos θ , sin θ ) ⊤ , rotational symmetry yields
h T ( p ) = u ¯ ∥ p ∥ | ω | ∫ 0 | ω | T g ( θ + s ) d s , g ( z ) = | cos z | + | sin z | .
For ω < 0 , replace θ by − θ in this formula; the set of angular directions, and all the conclusions, are unchanged. The even function g has period P = π / 2 and mean m = P − 1 ∫ 0 P g ( s ) d s = 4 / π . If | ω | T = k P , its integral over k periods is m | ω | T for every θ , so (14) follows.
To establish necessity, let L = | ω | T and suppose R T − ( 0 ) is a disk. Its support function must then be independent of θ . But the first nonconstant cosine coefficient of g is
a 1 = 4 π ∫ 0 π / 2 ( cos s + sin s ) cos ( 4 s ) d s = − 8 15 π ≠ 0 .
The corresponding complex Fourier coefficient of θ ↦ ∫ 0 L g ( θ + s ) d s is a nonzero multiple of ∫ 0 L e 4 i s d s = ( e 4 i L − 1 ) / ( 4 i ) . Angular independence therefore forces e 4 i L = 1 , or L ∈ P Z ; since T > 0 , necessarily L = k P , k ≥ 1 . This proves the claimed equivalence.
For the optimal Hausdorff bound, define the P-periodic primitive
Q ( t ) : = ∫ 0 t g ( s ) − m d s .
Its periodicity follows from ∫ 0 P ( g − m ) = 0 . By (17),
h T ( p ) − 4 u ¯ T π ∥ p ∥ = u ¯ ∥ p ∥ | ω | Q ( θ + L ) − Q ( θ ) .
On [ 0 , P ] one has Q ( t ) = sin t − cos t + 1 − m t . Set α = π / 4 − arccos ( m / 2 ) , so 0 < α < P / 2 . Since Q ′ ( t ) = sin t + cos t − m , the minimum of Q is attained at α and its maximum at P − α . Direct evaluation gives
max Q − min Q = 2 2 − m 2 − m arccos ( m / 2 ) = C * .
Thus | Q ( θ + L ) − Q ( θ ) | ≤ C * for every θ , L . For compact convex sets, the Hausdorff distance equals the supremum of the difference of their support functions over the unit sphere; (18) proves (15). Sharpness is attained by choosing θ = α and L = k P + ( P − 2 α ) for any integer k ≥ 0 : then Q ( θ + L ) − Q ( θ ) = Q ( P − α ) − Q ( α ) = C * . The corresponding positive horizon T = L / | ω | achieves equality in (15), so the constant cannot be reduced. □
Remark 1. 
The exact disk identity (14) characterizes reachability, not a feedback that achieves convergence at the prescribed time. Its if-and-only-if condition singles out complete quarter-turn sweeps of componentwise actuator authority. The absolute error in (15) is independent of the horizon, while the reference disk has radius 4 u ¯ T / π . More precisely,
error R T − ( 0 ) , ( 4 u ¯ T / π ) B 2 4 u ¯ T / π ≤ π C * 4 | ω | T ,
so therelativeshape discrepancy, normalized by this radius, is O ( T − 1 ) for fixed ω ≠ 0 . Furthermore, (18) provides a quadrature-free formula for any T: with L = | ω | T and an angular variable ϑ = θ for ω > 0 (respectively − θ for ω < 0 ), h T ( p ) = ( u ¯ ∥ p ∥ / | ω | ) [ m L + Q ( ϑ + L ) − Q ( ϑ ) ] . The periodic function Q is evaluated explicitly by Q ( t ) = sin s − cos s + 1 − 4 s / π , where s = t − ( π / 2 ) ⌊ 2 t / π ⌋ ∈ [ 0 , π / 2 ) . Thus the exact directional test requires no numerical integration.

5. A Planar Rotational Feasibility Example

Consider (11) with ω = 2.8 , u ¯ = 0.5 , and a desired horizon T c = 5.8097 . The same parameters occur in the two-state spacecraft illustration of [8]: the inertia data L 1 = L 2 = 5 , L 3 = 19 and k 3 = 1 yield ω = ( L 3 − L 1 ) k 3 / L 1 = ( 19 − 5 ) / 5 = 2.8 . Its published, input-normalized equations are x ˙ 1 = − 2.8 x 2 + v 1 , x ˙ 2 = 2.8 x 1 + v 2 , where v i = sat ( u i ) denotes the applied actuator input, bounded by u ¯ = 0.5 . Here this model serves solely as a linear reachability benchmark. The simple radial bound (12) excludes states with ∥ x 0 ∥ > 2 u ¯ T c ≈ 4.10808 , but gives no answer for many states of smaller norm.
For the coordinate direction p = e 1 = ( 1 , 0 ) ⊤ , formula (17) can be evaluated without numerical integration. Write | ω | T c = k π / 2 + δ , where k = 10 and δ ≈ 0.559197 . Because g ( s ) = cos s + sin s on 0 ≤ s ≤ π / 2 ,
h T c ( e 1 ) = u ¯ | ω | 2 k + sin δ + 1 − cos δ ≈ 3.69336 .
Hence x 0 = ( 4 , 0 ) ⊤ violates the exact necessary-and-sufficient condition (10), because e 1 ⊤ x 0 = 4 > h T c ( e 1 ) . It cannot be steered to the origin by T c under any input respecting the saturation bound, even though ∥ x 0 ∥ = 4 < 4.10808 is not excluded by the radial test. This strict refinement is a useful reason to preserve directional information when a linear model is available.
Conversely, the inner inclusion in (13) gives a positive feasibility certificate: every state in the disk of radius u ¯ T c = 2.90485 can be driven to the origin by some admissible control. In particular, the initial condition x 0 = ( 0.5 , − 0.5 ) ⊤ has norm about 0.7071 and is physically feasible within the horizon. This conclusion is about the existence of an admissible control, not the stability or optimality of any specific adaptive or neural-network implementation.

6. Concluding Discussion

Hard actuator bounds impose a reachability test before controller synthesis. The radial envelope (3) yields a general arrival-time obstruction and, under global Lipschitz assumptions, the explicit no-go bound (5). For input-constrained linear plants, classical support functions give the exact directional test (10). Its rotational specialization determines precisely the disk-shaped horizons and the sharp uniform estimate (15)–(16); the spacecraft benchmark demonstrates a strict improvement over the radial test without simulations.
The results concern necessary reachability conditions and exact plant-level feasibility sets, not feedback stabilization. Constructing a robust feedback with prescribed settling time on a feasible initial domain remains a separate problem.

Author Contributions

Robert Vrabel: Conceptualization, formal analysis, writing–original draft preparation, and writing–review and editing.

Funding

This research received no grant funding.

Institutional Review Board Statement

Not applicable.

Data Availability Statement

No datasets were generated or analyzed in this study. All reported numerical values follow from the displayed analytical formulas.

Conflicts of Interest

The author declares no conflicts of interest.

References

  1. Jiménez-Rodríguez, E.; Sánchez-Torres, J. D.; Loukianov, A. G. On optimal predefined-time stabilization. Internat. J. Robust. Nonlinear Control 2017, 27, 3620–3642. [Google Scholar] [CrossRef]
  2. Garg, K.; Arabi, E.; Panagou, D. Prescribed-time convergence with input constraints: A control Lyapunov function based approach. 2020 American Control Conference (ACC), 2020; pp. 962–967. [Google Scholar] [CrossRef]
  3. Garg, K.; Panagou, D. Characterization of domain of fixed-time stability under control input constraints. 2021 American Control Conference (ACC), 2021; pp. 2272–2277. [Google Scholar] [CrossRef]
  4. Garg, K.; Arabi, E.; Panagou, D. Fixed-time control under spatiotemporal and input constraints: A quadratic programming based approach. Automatica 2022, 141, 110314. [Google Scholar] [CrossRef]
  5. Hu, T.; Lin, Z.; Qiu, L. An explicit description of null controllable regions of linear systems with saturating actuators. Syst. Control Lett. 2002, 47, 65–78. [Google Scholar] [CrossRef]
  6. Le Guernic, C.; Girard, A. Reachability analysis of linear systems using support functions. Nonlinear Anal. Hybrid. Syst. 2010, 4, 250–262. [Google Scholar] [CrossRef]
  7. Schneider, R. Convex Bodies: The Brunn–Minkowski Theory, 2nd expanded ed.; Cambridge University Press: Cambridge, 2014. [Google Scholar] [CrossRef]
  8. Cai, N.; Wu, W.; Tong, S. Predefined-time adaptive neural network optimal output feedback control for nonlinear systems. Sci. China Inf. Sci. 2026, 69, 209201. [Google Scholar] [CrossRef]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.