betaFeedback

WorksEngineering

Control Invariance on Manifolds through Tangent Bundle Splitting

Abstract

Control invariance for interconnected non-linear systems on manifolds requires conditions under which subsystem safety constraints can be combined to form an invariant set. This paper develops local product type control invariant sets on manifolds using tangent bundle splitting, adapted product coordinates, and subsystem tangency conditions satisfied simultaneously by one admissible input at each state. The resulting conditions establish control invariance of product sets without requiring independent subsystem dynamics, with separate control choices possible when inputs are independently constrained and each acts only on its subsystem. This provides a way to verify subsystem safety without discarding dynamic coupling and is illustrated by a torus example in which adapted coordinates give a curved constraint region a product representation.

Introduction

The state of a network of NN interconnected systems can be described in Euclidean space by the partition x=(x1,⋯ ,xN)∈Rn=Rn1×⋯×RnNx=(x_1,\cdots,x_N)\in\mathbb{R}^n=\mathbb{R}^{n_1}\times\cdots\times\mathbb{R}^{n_N}, with each xix_i being the state of the ithi^{\mathrm{th}} subsystem. In this paper, we assume each subsystem ii can be controlled independently, using ui:R→Uiu_i:\mathbb{R}\to\mathcal{U}_i with Ui⊂Rmi\mathcal{U}_i\subset\mathbb{R}^{m_i} compact, and the dynamics evolve by

x˙i=F(xi,x−i,ui),xi∈Rni,ui∈Ui,\dot{x}_i=F(x_i,x_{-i},u_i),\qquad x_i\in\mathbb{R}^{n_i},\qquad u_i\in\mathcal{U}_i,

where x−ix_{-i} collects the states of the other subsystems on ii. Given this partition, a natural question is whether there exist sets Ωi\Omega_i such that an admissible control keeps xi(t)∈Ωix_i(t)\in\Omega_i for every subsystem ii and all t≥0t\geq0, whenever xi(0)∈Ωix_i(0)\in\Omega_i for all ii. The main tools for answering this question lie in set invariance theory [1] and viability theory [2]. For linear systems, [3] gives conditions for the existence of such sets based on the coupling strength, i.e., {Ωi(∥x−i∥)}i=1N\{\Omega_i(\lVert x_{-i}\rVert)\}_{i=1}^N, and [4] shows the maximum coupling that allows these sets to exist. These ideas were exploited in distributed control for linear networks in [5] and [6] by allowing a time-varying state partition. As a result, partitioning the state is key to obtaining distributed guarantees from a network.

Set invariance is closely linked to safety, as trajectories remain “trapped” within the set. For example, consider a scalar function, known as a Control Barrier Function (CBF), h:Rn→Rh:\mathbb{R}^n\to\mathbb{R}; its zero superlevel set Ω={x:h(x)≥0}\Omega=\{x:h(x)\geq0\} serves as the safety certificate. The authors in [7] and [8] employ quadratic programmes and Lie derivatives to ensure this level set encloses the system trajectories. The stochastic setting was handled by [9], and probabilistic results were obtained in [10]. An alternative view of safety certification can be achieved via reachability analysis using the Hamilton-Jacobi-Bellman equation [11]. For networks, existing work, see [10] and [12], examines how subsystem constraints can be combined into an invariant set while accounting for dynamic coupling. Recent work in [13] proposes global certification via only local certificates. This implies that the certificates have a product-like structure.

Geometric control describes dynamics intrinsically on manifolds [14], [15], allowing system structure to be expressed independently of coordinates.Safety filters based on CBFs on manifolds have been explored using backstepping techniques [16]. Ibuki et al. Ibuki et al. [17] formulated CBFs on S2\mathbb{S}^2 for collision avoidance. Networks evolving on manifolds have also been an active research area, as shown in power converters [18] and attitude tracking [19]. For interconnected systems on manifolds, a global state partition may not exist, coordinates are no longer globally defined, and identifying coordinates and subsystems is no longer straightforward. In recent work, Abdelgadil et al. Abdelgadil et al. [20] show how to model multi-agent systems on a manifold by splitting the tangent bundle via distributions. This type of splitting identifies subsystem velocity directions, but does not by itself provide compatible coordinates or a safety guarantee. In this paper, we construct invariant sets with a local product structure for interconnected systems evolving on manifolds. Following [20], subsystem directions are represented by distributions {Di}iN\{\mathcal{D}_i\}_{i}^{N} that are direct sums of the tangent bundle of the ambient manifold MM. These distributions generalise coordinates in curved spaces and can be assigned to each subsystem; we show that the distributions we build are involutive. This does not, by itself, provide compatible subsystem state coordinates. On these distributions, we introduce the notion of “adapted chart” Φ:M→Rn\Phi:M\to\mathbb{R}^n which satisfies Ω=Φ−1(Ω1×⋯×ΩN)\Omega=\Phi^{-1}(\Omega_1\times\cdots\times\Omega_N). As such, each subsystem distribution corresponds to its own coordinate block and, within these coordinates, one admissible input must satisfy all subsystem tangency conditions simultaneously. Furthermore, we give conditions for which a splitting supports local product representations of constraints and allows their control invariance to be verified through subsystem conditions.

For the problem formulated in Section II, this paper:

  1. The paper establishes necessary and sufficient conditions for the existence of coordinates adapted to the given subsystem distributions in Proposition 1, and Section III. These coordinates allow subsystem constraints to be represented as a product without assuming the manifold is, globally, a Cartesian product.

  2. Establishes the control invariance of compact product sets through subsystem tangency conditions satisfied by

a common input in Theorem 5, and describes the set and conditions intrinsically on MM in Section IV.

iii) Separates feasible subsystem input choices under Cartesian product input constraints and input alignment, without requiring independent subsystem dynamics in Corollary 1.

The existence of adapted charts and their associated distributions are conditions guaranteeing the existence of an agent interpretation of the non-linear dynamics and pave the way for decentralised and distributed safety certificates.

Notation: We use standard geometric notation from [21], [22]. Let MM be a smooth nn-dimensional manifold without boundary, with tangent spaces TxMT_xM and tangent bundle TM=⨆x∈MTxMTM = \bigsqcup_{x\in M} T_xM. The differential of a smooth map H:M→NH:M\to N at xx is dHx:TxM→TH(x)N\mathrm{d}H_x:T_xM\to T_{H(x)}N, and Γ(TM)\Gamma(TM) is the set of smooth vector fields. For X,Y∈Γ(TM)X,Y\in\Gamma(TM), the Lie bracket is [X,Y]h=X(Yh)−Y(Xh)[X,Y]h=X(Yh)-Y(Xh) for h∈C∞(M)h\in C^\infty(M), where Xh:=dh(X)Xh:=\mathrm{d}h(X). A smooth distribution D⊂TM\mathcal{D}\subset TM of constant rank rr is a smooth vector subbundle whose fibres D(x)⊆TxM\mathcal{D}(x)\subseteq T_xM are rr-dimensional. It is involutive if its smooth local sections are closed under Lie brackets, and integrable if every point lies in an immersed submanifold LL with TyL=D(y)T_yL=\mathcal{D}(y) for all y∈Ly\in L. By the Frobenius theorem [23], D\mathcal{D} is integrable if and only if it is involutive, and then every point has a neighbourhood WW and a submersion ζ:W→Rn−r\zeta:W\to\mathbb{R}^{n-r} with ker⁡dζ=D∣W\ker\mathrm{d}\zeta=\mathcal{D}|_W. The symbol ⊕\oplus denotes the internal direct sum of subspaces, and co⁡A\operatorname{co}\mathcal{A} the convex hull of a set A\mathcal{A}. For K⊂MK\subset M, x∈Kx\in K and a chart (O,φ)(O,\varphi) at xx, the contingent cone of KK at xx, adapted from [24], is

TK(x):={v∈TxM:∃xk∈K∩O, hk↓0 with hk−1(φ(xk)−φ(x))→dφx(v)}.T_K(x):=\left\{v\in T_xM:\exists x_k\in K\cap O,\ h_k\downarrow0\text{ with }h_k^{-1}\bigl(\varphi(x_k)-\varphi(x)\bigr)\to\mathrm{d}\varphi_x(v)\right\}.

The same argument gives dHx(TK(x))=TH(K)(H(x))\mathrm{d}H_x(T_K(x))=T_{H(K)}(H(x)) for a diffeomorphism HH between open sets containing KK, and TK(x)=TxMT_K(x)=T_xM for xx in the interior of KK. For a control system x˙=F(x,u)\dot{x}=F(x,u) on MM with input set U⊂RmU\subset\mathbb{R}^m, an input is a measurable μ:[0,∞)→U\mu:[0,\infty)\to U, and ψ(⋅,x0,μ)\psi(\cdot,x_0,\mu) denotes the corresponding Carathéodory solution with ψ(0,x0,μ)=x0\psi(0,x_0,\mu)=x_0. A set Ω⊂M\Omega\subset M is control invariant if for every x0∈Ωx_0\in\Omega there is an input μ\mu such that ψ(t,x0,μ)\psi(t,x_0,\mu) is defined and lies in Ω\Omega for all t≥0t\ge0. We write F(x)={F(x,u):u∈U}⊂TxM\mathcal{F}(x)=\{F(x,u):u\in U\}\subset T_xM. The following lemma is the only viability result used in the paper.

Lemma 1 (Viability in a chart): Let F(x,u)=f(x)+g(x)uF(x,u)=f(x)+g(x)u with ff and gg smooth, and let U⊂RmU\subset\mathbb{R}^m be nonempty, convex and compact. Let (O,φ)(O,\varphi) be a chart and K⊂OK\subset O nonempty and compact. If

F(x)∩co⁡TK(x)≠∅for all x∈K,(1)\mathcal{F}(x)\cap\operatorname{co}T_K(x)\ne\varnothing\quad\text{for all }x\in K, \tag*{(1)}

then KK is control invariant.

Problem Formulation

Given the interconnected dynamics on a manifold

x˙=F(x,u)=f(x)+g(x)u,u∈U⊂Rm,(2)\dot{x}=F(x,u)=f(x)+g(x)u,\qquad u\in U\subset\mathbb{R}^m, \tag*{(2)}

where x∈Mx\in M, f∈Γ(TM)f\in\Gamma(TM), g(x)=[g1(x) ⋯ gm(x)]g(x)=[g_1(x)\ \cdots\ g_m(x)] with gj∈Γ(TM)g_j\in\Gamma(TM), and UU is nonempty, convex and compact. Subsystems are not blocks of the state vector since MM may not be a product and coordinates may not respect interconnection. We first formally define our notion of tangent bundle splitting.

Definition 1 (Complementary tangent-bundle splitting): For smooth, constant rank nin_i, distributions Di⊂TM\mathcal{D}_i\subset TM, i∈J:={1,…,N}i\in\mathcal{J}:=\{1,\ldots,N\} define a complementary tangent bundle splitting if TxM=⨁i∈JDi(x)T_xM=\bigoplus_{i\in\mathcal{J}}\mathcal{D}_i(x) for every x∈Mx\in M. Each vx∈TxMv_x\in T_xM has a unique decomposition vx=∑i∈Jviv_x=\sum_{i\in\mathcal{J}}v_i with vi∈Di(x)v_i\in\mathcal{D}_i(x) and ∑i∈Jni=n\sum_{i\in\mathcal{J}}n_i=n. The splitting fixes subsystem directions without reference to coordinates, while the dynamics and the inputs in (2) may still couple the subsystems. On Rn=∏i∈JRni\mathbb{R}^n=\prod_{i\in\mathcal{J}}\mathbb{R}^{n_i}, let Ji:Rni→RnJ_i:\mathbb{R}^{n_i}\to\mathbb{R}^n be the injection onto the ii-th block, Ei:=Ji(Rni)E_i:=J_i(\mathbb{R}^{n_i}) the ii-th coordinate subspace, Pi:=blkdiag⁡(0n1,…,Ini,…,0nN)P_i:=\operatorname{blkdiag}(0_{n_1},\ldots,I_{n_i},\ldots,0_{n_N}) the projection onto EiE_i along the other blocks, and ei,ae_{i,a}, a=1,…,nia=1,\ldots,n_i, the standard basis of EiE_i. To impose constraints subsystem by subsystem, we need coordinates z=(z1,…,zN)z=(z_1,\ldots,z_N), zi∈Rniz_i\in\mathbb{R}^{n_i}, in which motion along Di\mathcal{D}_i changes ziz_i only, that is, a diffeomorphism

Φ:U→V1×⋯×VN,dΦx(Di(x))=Ei,x∈U, i∈J,(3)\Phi:U\to V_1\times\cdots\times V_N,\qquad \mathrm{d}\Phi_x(\mathcal{D}_i(x))=E_i,\quad x\in U,\ i\in\mathcal{J}, \tag*{(3)}

where U⊂MU\subset M and Vi⊆RniV_i\subseteq\mathbb{R}^{n_i} are open and TΦ(x)(V1×⋯×VN)T_{\Phi(x)}(V_1\times\cdots\times V_N) is identified with Rn\mathbb{R}^n. In such a chart, subsystem constraint sets Ωi⊂Vi\Omega_i\subset V_i, each nonempty and compact, combine into the candidate set

Ω:=Φ−1(∏i∈JΩi)⊂U,(4)\Omega:=\Phi^{-1}\left(\prod_{i\in\mathcal{J}}\Omega_i\right)\subset U, \tag*{(4)}

which is compact and contained in the chart domain. The construction is local and does not assume that MM is globally a product. Our objective, shown in Figure 1 is to determine when the given splitting supports a local product representation of the constraints and when the resulting set is control invariant.

Adapted product chart for $N=2$

Figure 1. Adapted product chart for N=2N = 2. Thin curves are integral curves of D1\mathcal{D}_1 (blue) and D2\mathcal{D}_2 (teal); Φ\Phi maps them to coordinate lines, so dΦx(Di(x))=Ei\mathrm{d}\Phi_x(\mathcal{D}_i(x)) = E_i. The product set Ω1×Ω2\Omega_1 \times\Omega_2 pulls back to Ω=Φ−1(Ω1×Ω2)\Omega= \Phi^{-1}(\Omega_1 \times\Omega_2), whose faces {z1∈∂Ω1}\{z_1 \in\partial\Omega_1\} are integral curves of D1c=D2\mathcal{D}_1^c = \mathcal{D}_2 and whose faces {z2∈∂Ω2}\{z_2 \in\partial\Omega_2\} are integral curves of D1\mathcal{D}_1. On the face Σ1\Sigma_1, the component F2F_2 is tangent to the face, so only F1F_1, equivalently F^1\widehat{F}_1, determines whether the trajectory leaves Ω\Omega.

Problem 1: Given the network (2), and the splitting in Definition 1,

  1. determine when charts of the form (3) exist;

  1. give conditions on the subsystem velocities, imposed through the sets Ωi\Omega_i and met by a single admissible input at each x∈Ωx\in\Omega, under which (4) is control invariant;

  1. describe the resulting invariant set and the invariance condition on MM, independently of the chart, so that the safety guarantee does not depend on the coordinates used.

Control Invariance from Tangent-Bundle Splitting

For the network (2) and the complementary splitting introduced in Definition 1, we first describe the associated subsystem projections and conditions under which the distributions admit compatible local product coordinates. These coordinates decompose the state only under an integrability condition characterised in Section III-B.

Compatible subsystem structure

The first step, smoothly assigning each tangent vector its subsystem components, requires no integrability.

Lemma 2 (Smooth subsystem projections). Let TM=⨁i∈JDiTM = \bigoplus_{i\in\mathcal{J}} \mathcal{D}_i be a complementary splitting with inclusions ιi:Di↪TM\iota_i : \mathcal{D}_i \hookrightarrow TM. For i∈Ji \in\mathcal{J}, let Πi:TM→Di\Pi_i : TM \to\mathcal{D}_i restrict on each TxMT_xM to the projection onto Di(x)\mathcal{D}_i(x) along Dic(x)\mathcal{D}_i^c(x). Then each Πi\Pi_i is a smooth bundle homomorphism over id⁡M\operatorname{id}_M, and

Πi∘ιi=id⁡Di,Πi∘ιj=0 (i≠j),∑i∈Jιi∘Πi=id⁡TM.\Pi_i \circ\iota_i = \operatorname{id}_{\mathcal{D}_i},\qquad\Pi_i \circ\iota_j = 0\ (i \ne j),\qquad\sum_{i\in\mathcal{J}}\iota_i \circ\Pi_i = \operatorname{id}_{TM}.

In particular, Π^i:=ιi∘Πi\widehat{\Pi}_i := \iota_i \circ\Pi_i satisfies Π^i∘Π^j=δijΠ^i\widehat{\Pi}_i \circ\widehat{\Pi}_j = \delta_{ij}\widehat{\Pi}_i.

Proof. The identities hold fibrewise by uniqueness of the decomposition v=∑j∈Jvjv = \sum_{j\in\mathcal{J}} v_j, vj∈Dj(x)v_j \in\mathcal{D}_j(x), and Π^i∘Π^j=ιi∘(Πi∘ιj)∘Πj=δijΠ^i\widehat{\Pi}_i \circ\widehat{\Pi}_j = \iota_i \circ(\Pi_i \circ\iota_j) \circ\Pi_j = \delta_{ij}\widehat{\Pi}_i. Smoothness is local. Around any point of MM, fix a chart (W,φ)(W,\varphi) on which each Di\mathcal{D}_i has a smooth local frame (X1i,…,Xnii)(X^i_1,\ldots,X^i_{n_i}), and let A:W→Rn×nA: W \to\mathbb{R}^{n\times n} have as columns the coordinate representations of X11,…,Xn11,…,X1N,…,XnNNX^1_1,\ldots,X^1_{n_1},\ldots,X^N_1,\ldots,X^N_{n_N}. Then AA is smooth, and A(x)A(x) is invertible for every x∈Wx\in W by the direct-sum hypothesis. In the induced chart of TWTW, Π^i\widehat{\Pi}_i reads (x,v)↦(x,A(x)PiA(x)−1v)(x,v) \mapsto(x,A(x)P_iA(x)^{-1}v), which is smooth because A−1=adj⁡(A)/det⁡AA^{-1}=\operatorname{adj}(A)/\det A. Hence Π^i\widehat{\Pi}_i is smooth, and so is Πi\Pi_i, since Di\mathcal{D}_i is an embedded submanifold of TMTM containing ιi(TM)\iota_i(TM).

Applying these projections gives the subsystem velocities

Fi(x,u)=Πi(x)F(x,u)∈Di(x),F_i(x,u)=\Pi_i(x)F(x,u)\in\mathcal{D}_i(x),
F(x,u)=∑i∈JFi(x,u).(5)F(x,u)=\sum_{i\in\mathcal{J}}F_i(x,u). \tag*{(5)}

This decomposition is chart-free and exists for every splitting.

Adapted product charts

Interpreting the distributions as local subsystem state spaces requires additional integrability and compatibility. A state of subsystem ii is a submersion ziz_i from an open subset of MM onto an open subset of Rni\mathbb{R}^{n_i} whose rate depends only on the ii-th component. For every tangent vector vv over its domain,

dzi(v)=dzi(Π^iv).(6)\mathrm{d}z_i(v)=\mathrm{d}z_i(\widehat{\Pi}_i v). \tag*{(6)}

By a rank count, (6) holds iff ker⁡dzi=Dic\ker\mathrm{d}z_i=\mathcal{D}_i^c. Collecting states of all subsystems gives a chart of the form (3).

Definition 2 (Locally product-compatible splitting). A diffeomorphism Φ=(z1,…,zN):U→V1×⋯×VN\Phi=(z_1,\ldots,z_N):U\to V_1\times\cdots\times V_N satisfying (3) is an adapted product chart. A complementary splitting is locally product compatible if every point of MM has an open neighbourhood that is the domain of an adapted product chart.

For a diffeomorphism Φ=(z1,…,zN)\Phi=(z_1,\ldots,z_N), condition (3) is equivalent to ker⁡dzi=Dic\ker\mathrm{d}z_i=\mathcal{D}_i^c for every ii, that is, to each ziz_i being a state of subsystem ii in the sense of (6). The coordinate ziz_i is constant along Dic\mathcal{D}_i^c: motion in the directions of the other subsystems does not change it (see Figure 1). The next result characterises when such charts exist.

Proposition 1 (Product-compatible splittings). Let TM=⨁i∈JDiTM=\bigoplus_{i\in\mathcal{J}}\mathcal{D}_i be a smooth complementary splitting, ni:=rank⁡Din_i:=\operatorname{rank}\mathcal{D}_i, and Dic:=⨁j∈J∖{i}Dj\mathcal{D}_i^c:=\bigoplus_{j\in\mathcal{J}\setminus\{i\}}\mathcal{D}_j. The following are equivalent:

i) the splitting is locally product compatible,

ii) Dic\mathcal{D}_i^c is involutive for every i∈Ji\in\mathcal{J},

iii) around every point of MM there are local frames {Xi,a}a=1ni\{X_{i,a}\}_{a=1}^{n_i} of Di\mathcal{D}_i such that [Xi,a,Xj,b]∈Γ(Di⊕Dj)[X_{i,a},X_{j,b}]\in\Gamma(\mathcal{D}_i\oplus\mathcal{D}_j) for all i,j∈Ji,j\in\mathcal{J} and all a,ba,b,

iv) around every point of MM there are local frames {Xi,a}a=1ni\{X_{i,a}\}_{a=1}^{n_i} of Di\mathcal{D}_i such that [Xi,a,Xj,b]=0[X_{i,a},X_{j,b}]=0 for all i,j∈Ji,j\in\mathcal{J} and all a,ba,b. Moreover, the condition in iii) holds for one choice of local frames iff it holds for every choice.

Proof. i) ⇒\Rightarrow iv): take the coordinate fields Xi,a:=dΦ−1(ei,a)X_{i,a}:=\mathrm{d}\Phi^{-1}(e_{i,a}) of an adapted chart. iv) ⇒\Rightarrow iii): immediate. iii) ⇒\Rightarrow ii): for X=∑afaXi,aX=\sum_a f^aX_{i,a} and Y=∑bgbXj,bY=\sum_b g^bX_{j,b}, their bracket satisfies

[X,Y]=∑a,b(fagb[Xi,a,Xj,b]+faXi,a(gb)Xj,b−gbXj,b(fa)Xi,a)∈Γ(Di⊕Dj),[X,Y]=\sum_{a,b}\left(f^ag^b[X_{i,a},X_{j,b}]+f^aX_{i,a}(g^b)X_{j,b}-g^bX_{j,b}(f^a)X_{i,a}\right)\in\Gamma(\mathcal{D}_i\oplus\mathcal{D}_j),

proving frame independence. By Lemma 2, every local section of Dkc\mathcal{D}_k^c is a sum of local sections of Dj\mathcal{D}_j, j≠kj\ne k, and Di+Dj⊆Dkc\mathcal{D}_i+\mathcal{D}_j\subseteq\mathcal{D}_k^c for i,j≠k.Finally,ii)i,j\ne k. Finally, ii)\Rightarrow$ i): fix x∗∈Mx^*\in M. By the Frobenius theorem, for each ii there is a submersion ziz_i from a neighbourhood of x∗x^* into Rni\mathbb{R}^{n_i} with ker⁡dzi=Dic\ker\mathrm{d}z_i=\mathcal{D}_i^c. Let Φ:=(z1,…,zN)\Phi:=(z_1,\ldots,z_N) on the intersection of these neighbourhoods. For j≠ij\ne i, Di⊆Djc\mathcal{D}_i\subseteq\mathcal{D}_j^c gives dzj(Di)={0}\mathrm{d}z_j(\mathcal{D}_i)=\{0\},

while ker⁡dzi∩Di={0}\ker dz_i \cap\mathcal{D}_i = \{0\} makes dzi∣Didz_i|_{\mathcal{D}_i} injective, hence bijective onto Rni\mathbb{R}^{n_i}. Thus dΦx(Di(x))=Eid\Phi_x(\mathcal{D}_i(x)) = E_i for all ii, and dΦxd\Phi_x is invertible. By the inverse function theorem, Φ\Phi maps an open W∋x∗W \ni x^* diffeomorphically onto an open set; choose open ViV_i with Φ(x∗)∈V1×⋯×VN⊆Φ(W)\Phi(x^*) \in V_1 \times\cdots\times V_N \subseteq\Phi(W) and set U=W∩Φ−1(V1×⋯×VN)U = W \cap\Phi^{-1}(V_1 \times\cdots\times V_N). ▪

Remark 1: For N=2N = 2, condition ii) states that D1\mathcal{D}_1 and D2\mathcal{D}_2 are involutive. For N≥3N \ge3, involutivity of every Di\mathcal{D}_i is necessary but not sufficient. On R3\mathbb{R}^3, the line fields D1=span⁡{∂x}\mathcal{D}_1 = \operatorname{span}\{\partial_x\}, D2=span⁡{∂y+x∂z}\mathcal{D}_2 = \operatorname{span}\{\partial_y + x\partial_z\} and D3=span⁡{∂z}\mathcal{D}_3 = \operatorname{span}\{\partial_z\} are complementary and involutive, but [∂x,∂y+x∂z]=∂z∉D1⊕D2=Dc[\partial_x,\partial_y + x\partial_z] = \partial_z \notin\mathcal{D}_1 \oplus\mathcal{D}_2 = \mathcal{D}^c.

In adapted coordinates, the projections of Lemma 2 become constant coordinate projections, so the velocity decomposition (5) becomes a decomposition into coordinate blocks.

Lemma 3 (Coordinate form of projections): Let Φ=(z1,…,zN):U→V1×⋯×VN\Phi= (z_1,\ldots,z_N): U \to V_1 \times\cdots\times V_N be an adapted product chart, with zi=(zi,1,…,zi,ni)z_i = (z_{i,1},\ldots,z_{i,n_i}). Then, for x∈Ux \in U, dΦx∘Π^i(x)=Pi∘dΦxd\Phi_x \circ\hat{\Pi}_i(x) = P_i \circ d\Phi_x, or equivalently

Π^i=∑a=1ni∂∂zi,a⊗dzi,a.(7)\hat{\Pi}_i = \sum_{a=1}^{n_i} \frac{\partial}{\partial z_{i,a}} \otimes dz_{i,a}. \tag*{(7)}

Conversely, a diffeomorphism Φ:U→V1×⋯×VN\Phi: U \to V_1 \times\cdots\times V_N satisfying the second identity for all i∈Ji \in\mathcal{J} is an adapted product chart.

Proof: Fix x∈Ux \in U and v∈TxMv \in T_xM. Adaptedness gives ∂∂zj,b∣x=(dΦx)−1ej,b∈Dj(x)\left.\frac{\partial}{\partial z_{j,b}}\right|_x = (d\Phi_x)^{-1}e_{j,b} \in\mathcal{D}_j(x), so v=∑j∈J∑b=1njdzj,b(v)∂∂zj,b∣xv = \sum_{j\in\mathcal{J}}\sum_{b=1}^{n_j} dz_{j,b}(v)\left.\frac{\partial}{\partial z_{j,b}}\right|_x is the decomposition of vv along the splitting. By its uniqueness, Π^i(x)v=∑a=1nidzi,a(v)∂∂zi,a∣x\hat{\Pi}_i(x)v = \sum_{a=1}^{n_i} dz_{i,a}(v)\left.\frac{\partial}{\partial z_{i,a}}\right|_x, and applying dΦxd\Phi_x gives the first identity. Conversely, taking images in the second identity gives dΦx(Di(x))=Pi dΦx(TxM)=Eid\Phi_x(\mathcal{D}_i(x)) = P_i\,d\Phi_x(T_xM) = E_i. ▪

Control invariance of product sets

Let Φ:U→∏i∈JVi\Phi: U \to\prod_{i\in\mathcal{J}} V_i be an adapted product chart and consider the candidate set (4). Since Ω⊂U\Omega\subset U, the transformation rule for contingent tangent cones gives

dΦx(TΩ(x))=TΦ(Ω)(Φ(x)),x∈Ω,(8)d\Phi_x(T_\Omega(x)) = T_{\Phi(\Omega)}(\Phi(x)), \qquad x \in\Omega, \tag*{(8)}

with Φ(Ω)=∏i∈JΩi\Phi(\Omega) = \prod_{i\in\mathcal{J}}\Omega_i, so tangency conditions on MM become conditions on a product set in Rn\mathbb{R}^n.

First, we study the properties of contingent cones under the adapted chart. We introduce the notion of geometric derivability for the elements in the cotingent cone.

Definition 3 (Geometric derivability (Definition 6.1 [24])): Let K⊆RnK \subseteq\mathbb{R}^n and zˉ∈K\bar{z} \in K. A vector w∈TK(zˉ)w \in T_K(\bar{z}) is derivable if there exist ε>0\varepsilon> 0 and ξ:[0,ε]→K\xi: [0,\varepsilon] \to K with ξ(0)=zˉ\xi(0) = \bar{z} and lim⁡t↓0(ξ(t)−zˉ)/t=w\lim_{t\downarrow0}(\xi(t)-\bar{z})/t = w. The set KK is geometrically derivable at zˉ\bar{z} if every w∈TK(zˉ)w \in T_K(\bar{z}) is derivable.

In the above definition, no continuity of ξ\xi on (0,ε](0,\varepsilon] is required. Clarke regular sets, in particular convex sets, are geometrically derivable [24 Ch. 6]; so are finite unions of closed convex sets, because the contingent cone of a finite union of closed sets at zˉ\bar{z} is the union of the contingent cones of the members that contain zˉ\bar{z}. The following lemma states the regularity property used to decompose the tangent cone of a product set.

Lemma 4 (Tangent cones of product sets): Let K=K1×⋯×KNK = K_1 \times\cdots\times K_N with Ki⊆RniK_i \subseteq\mathbb{R}^{n_i}, and let z=(z1,…,zN)∈Kz = (z_1,\ldots,z_N) \in K. Then

TK(z)⊆TK1(z1)×⋯×TKN(zN)⊆co⁡TK(z).(9)T_K(z) \subseteq T_{K_1}(z_1) \times\cdots\times T_{K_N}(z_N) \subseteq\operatorname{co} T_K(z). \tag*{(9)}

If, moreover, KiK_i is geometrically derivable at ziz_i for every i∈Ji \in\mathcal{J} except at most one, then

TK(z)=TK1(z1)×⋯×TKN(zN).(10)T_K(z) = T_{K_1}(z_1) \times\cdots\times T_{K_N}(z_N). \tag*{(10)}

Proof: If vi∈TKi(zi)v_i \in T_{K_i}(z_i), there are hk↓0h_k \downarrow0 and zik∈Kiz_i^k \in K_i with (zik−zi)/hk→vi(z_i^k-z_i)/h_k \to v_i; since zj∈Kjz_j \in K_j for j≠ij \ne i, the points z+1i(zik−zi)z + \mathbf{1}_i(z_i^k-z_i) lie in KK, so vi∈TK(z)v_i \in T_K(z). If v∈TK(z)v \in T_K(z), there are hk↓0h_k \downarrow0 and zk∈Kz^k \in K with (zk−z)/hk→v(z^k-z)/h_k \to v, and the ii-th block gives (zik−zi)/hk→vi(z_i^k-z_i)/h_k \to v_i with zik∈Kiz_i^k \in K_i; this is the first inclusion in (9). For the second, v=∑i∈Jλiviv = \sum_{i\in\mathcal{J}}\lambda_i v_i is a finite sum of elements of TK(z)T_K(z), hence lies in co⁡TK(z)\operatorname{co} T_K(z), because the convex hull of a cone is a convex cone.

For (10), we show ⊇\supseteq. Let i0∈Ji_0 \in\mathcal{J} be such that KiK_i is geometrically derivable at ziz_i for all i≠i0i \ne i_0, and let v∈∏i∈JTKi(zi)v \in\prod_{i\in\mathcal{J}}T_{K_i}(z_i). Choose hk↓0h_k \downarrow0 and zi0k∈Ki0z_{i_0}^k \in K_{i_0} with (zi0k−zi0)/hk→vi0(z_{i_0}^k-z_{i_0})/h_k \to v_{i_0}. For i≠i0,i∈Ji \ne i_0, i \in\mathcal{J}, let εi∈(0,1]\varepsilon_i \in(0,1] and zik:=ξi(hk)z_i^k := \xi_i(h_k) for all kk large enough that hk≤εih_k \le\varepsilon_i. Then zk:=(z1k,…,zNk)∈Kz^k := (z_1^k,\ldots,z_N^k) \in K and (zk−z)/hk→v(z^k-z)/h_k \to v, so v∈TK(z)v \in T_K(z). ▪

Main contribution

For x∈Ux \in U, the subsystem velocities in adapted coordinates are

F^i(x,u):=dπi∣Φ(x)(dΦxF(x,u))∈TziVi,(11)\hat{F}_i(x,u) := d\pi_i|_{\Phi(x)}(d\Phi_xF(x,u)) \in T_{z_i}V_i, \tag*{(11)}

so that dΦxF(x,u)=(F^1(x,u),…,F^N(x,u))d\Phi_xF(x,u) = (\hat{F}_1(x,u),\ldots,\hat{F}_N(x,u)) and z˙i=F^i(x,u)\dot{z}_i = \hat{F}_i(x,u) along trajectories in UU. By Lemma 5, F^i(x,u)\hat{F}_i(x,u) is the ii-th coordinate block of the projected velocity dΦxF(x,u)d\Phi_xF(x,u).

Theorem 5 (Control invariance of a local product set): Let Φ=(z1,…,zN):U→V1×⋯×VN\Phi= (z_1,\ldots,z_N): U \to V_1 \times\cdots\times V_N be an adapted product chart and let Ω\Omega be as in (4). Suppose that for every x∈Ωx \in\Omega there exists u∈Uu \in U such that

F^i(x,u)∈TΩi(zi(x))for all i∈J.(12)\hat{F}_i(x,u) \in T_{\Omega_i}(z_i(x)) \quad\text{for all } i \in\mathcal{J}. \tag*{(12)}

Then:

  1. F(x,u)∈co⁡TΩ(x)F(x,u) \in\operatorname{co}T_\Omega(x) for the same uu; if, in addition, every Ωi\Omega_i except at most one is geometrically derivable at zi(x)z_i(x), then F(x,u)∈TΩ(x)F(x,u) \in T_\Omega(x);

  2. Ω\Omega is control invariant.

Proof: i) Fix x∈Ωx \in\Omega and a corresponding uu, and write z=Φ(x)z = \Phi(x). By (11) and (12), dΦxF(x,u)∈∏i∈JTΩi(zi)d\Phi_xF(x,u) \in\prod_{i\in\mathcal{J}}T_{\Omega_i}(z_i), which is contained in co⁡TΦ(Ω)(z)\operatorname{co}T_{\Phi(\Omega)}(z) by Lemma 4 and equals TΦ(Ω)(z)T_{\Phi(\Omega)}(z) under the geometric derivability assumption. Since dΦxd\Phi_x is a linear isomorphism, (8) gives dΦx−1co⁡TΦ(Ω)(z)=co⁡TΩ(x)d\Phi_x^{-1}\operatorname{co}T_{\Phi(\Omega)}(z) = \operatorname{co}T_\Omega(x), and i) follows. ii) By i), Condition (1) holds on Ω\Omega. Since Ω\Omega is compact and contained in the domain of the chart Φ\Phi, Lemma 1 applies with (O,φ)=(U,Φ)(O,\varphi) = (U,\Phi). ▪

The input in (12) is common to all subsystems: feasibility of each condition with a different input does not imply (12). Theorem 5 is stated in the chart. We now describe Ω\Omega and condition (12) on MM. For x∈Ωx \in\Omega, let $A(x) = {i \in \mathcal{J}: z_i(x) \in ∂Ωi\partial\Omega_i} be the set of active subsystems, and let Σi={x∈Ω:zi(x)∈∂Ωi}\Sigma_i=\{x\in\Omega:z_i(x)\in\partial\Omega_i\} be the ii-th face of Ω\Omega.

Proposition 2 (The candidate set on MM): Let Φ=(z1,…,zN)\Phi=(z_1,\ldots,z_N) be an adapted product chart and let Ω\Omega be as in (4). Then:

i) Ω=⋂i∈Jzi−1(Ωi)\Omega=\bigcap_{i\in\mathcal{J}}z_i^{-1}(\Omega_i), and every level set of ziz_i is an integral manifold of Dic\mathcal{D}_i^c. The set ∂Ω\partial\Omega is ⋃i∈JΣi\bigcup_{i\in\mathcal{J}}\Sigma_i, and each face Σi\Sigma_i is a union of pieces of integral manifolds of Dic\mathcal{D}_i^c.

ii) for x∈Ωx\in\Omega and v∈TxMv\in T_xM, Π^i(x)v∈TΩ(x)\widehat{\Pi}_i(x)v\in T_\Omega(x) iff dzi∣xv∈TΩi(zi(x))d z_i|_xv\in T_{\Omega_i}(z_i(x)), and v∈TΩ(x)v\in T_\Omega(x) implies Π^i(x)v∈TΩ(x)\widehat{\Pi}_i(x)v\in T_\Omega(x) for every i∈Ji\in\mathcal{J};

iii) if A(x)={i}A(x)=\{i\}, then TΩ(x)={v∈TxM:Π^i(x)v∈TΩ(x)}⊇Dic(x)T_\Omega(x)=\{v\in T_xM:\widehat{\Pi}_i(x)v\in T_\Omega(x)\}\supseteq\mathcal{D}_i^c(x); if A(x)=∅A(x)=\varnothing, then TΩ(x)=TxMT_\Omega(x)=T_xM.

Proof: i) Since Ω⊂U\Omega\subset U, Φ(Ω)=∏iΩi\Phi(\Omega)=\prod_i\Omega_i is equivalent to Ω=⋂izi−1(Ωi)\Omega=\bigcap_i z_i^{-1}(\Omega_i). Each ziz_i is a submersion with ker⁡dzi=Dic\ker d z_i=\mathcal{D}_i^c, so its level sets are tangent to Dic\mathcal{D}_i^c. Since Φ\Phi is a homeomorphism onto the open set V1×⋯×VNV_1\times\cdots\times V_N and each Ωi\Omega_i is closed, ∂Ω=Φ−1(⋃i∂Ωi×∏j≠iΩj)=⋃iΣi\partial\Omega=\Phi^{-1}\left(\bigcup_i\partial\Omega_i\times\prod_{j\ne i}\Omega_j\right)=\bigcup_i\Sigma_i, and Σi=⋃c∈∂Ωi(Ω∩zi−1(c))\Sigma_i=\bigcup_{c\in\partial\Omega_i}(\Omega\cap z_i^{-1}(c)). For ii), Let z=Φ(x)z=\Phi(x). By Lemma 3, dΦxΠ^i(x)v=ȷidzi∣xvd\Phi_x\widehat{\Pi}_i(x)v=\jmath_i d z_i|_xv. By Lemma 4, ȷiw∈TΦ(Ω)(z)\jmath_iw\in T_{\Phi(\Omega)}(z) iff w∈TΩi(zi)w\in T_{\Omega_i}(z_i): sufficiency is ȷiTΩi(zi)⊆TΦ(Ω)(z)\jmath_iT_{\Omega_i}(z_i)\subseteq T_{\Phi(\Omega)}(z), and necessity is the first inclusion in (9). With (8), this gives the equivalence. If v∈TΩ(x)v\in T_\Omega(x), then dΦxv∈TΦ(Ω)(z)⊆∏jTΩj(zj)d\Phi_xv\in T_{\Phi(\Omega)}(z)\subseteq\prod_jT_{\Omega_j}(z_j), so dzi∣xv∈TΩi(zi)d z_i|_xv\in T_{\Omega_i}(z_i) and Π^i(x)v∈TΩ(x)\widehat{\Pi}_i(x)v\in T_\Omega(x).

Lastly, for iii), j∉A(x)j\notin A(x), zjz_j is an interior point of Ωj\Omega_j, so TΩj(zj)=RnjT_{\Omega_j}(z_j)=\mathbb{R}^{n_j} and Ωj\Omega_j is geometrically derivable at zjz_j. If A(x)={i}A(x)=\{i\}, (10) and (8) give TΩ(x)={v:dzi∣xv∈TΩi(zi)}T_\Omega(x)=\{v:d z_i|_xv\in T_{\Omega_i}(z_i)\}, which equals {v:Π^i(x)v∈TΩ(x)}\{v:\widehat{\Pi}_i(x)v\in T_\Omega(x)\} by (b) and contains ker⁡dzi∣x=Dic(x)\ker d z_i|_x=\mathcal{D}_i^c(x). If A(x)=∅A(x)=\varnothing, then xx is an interior point of Ω\Omega.

Blockwise control feasibility

To obtain a blockwise control structure compatible with the tangent bundle splitting, suppose that

U=U1×⋯×UN,g(x)u=∑i∈JGi(x)ui,(13)U=U_1\times\cdots\times U_N,\qquad g(x)u=\sum_{i\in\mathcal{J}}G_i(x)u_i, \tag*{(13)}

where u=(u1,…,uN)u=(u_1,\ldots,u_N), each Ui⊂RmiU_i\subset\mathbb{R}^{m_i} is nonempty, convex and compact, ∑i∈Jmi=m\sum_{i\in\mathcal{J}}m_i=m, and each Gi(x):Rmi→TxMG_i(x):\mathbb{R}^{m_i}\to T_xM satisfies

Im⁡Gi(x)⊆Di(x),x∈U.(14)\operatorname{Im}G_i(x)\subseteq\mathcal{D}_i(x),\qquad x\in U. \tag*{(14)}

Equivalently, Πj(x)Gi(x)=0\Pi_j(x)G_i(x)=0 for j≠ij\ne i. Then dzi∣xGj(x)=0d z_i|_xG_j(x)=0 for j≠ij\ne i, so F^i\widehat{F}_i depends on the input only through uiu_i, and we write

F^i(x,ui):=dzi∣x(f(x)+Gi(x)ui).(15)\widehat{F}_i(x,u_i):=d z_i|_x\left(f(x)+G_i(x)u_i\right). \tag*{(15)}

Corollary 1 (Blockwise feasibility): Let Φ\Phi and Ω\Omega be as in Theorem 5, and assume (13) and (14). For x∈Ωx\in\Omega, let

Ki(x):={ui∈Ui:F^i(x,ui)∈TΩi(zi(x))}.K_i(x):=\{u_i\in U_i:\widehat{F}_i(x,u_i)\in T_{\Omega_i}(z_i(x))\}.

If Ki(x)≠∅K_i(x)\ne\varnothing for every x∈Ωx\in\Omega and every i∈A(x)i\in A(x), then Ω\Omega is control invariant.

Proof: Fix x∈Ωx\in\Omega. Choose ui∈Ki(x)u_i\in K_i(x) for i∈A(x)i\in A(x), and ui∈Uiu_i\in U_i arbitrary otherwise. Then u:=(u1,…,uN)∈Uu:=(u_1,\ldots,u_N)\in U by (13), and F^i(x,u)=F^i(x,ui)\widehat{F}_i(x,u)=\widehat{F}_i(x,u_i) by (15), so the ii-th condition in (12) holds for i∈A(x)i\in A(x); otherwise, it holds because TΩi(zi(x))=RniT_{\Omega_i}(z_i(x))=\mathbb{R}^{n_i}. Theorem 5 applies. ▪

Remark 2 (Decentralised feedback): Corollary 1 reduces the viability condition (1) to separate input choices when state constraints have an adapted product form, input constraints are independent, and (14) holds. This separates control channels without requiring independent subsystem dynamics or ensuring decentralised feedback, since Ki(x)K_i(x) may depend on the full state. Writing z=(zi,z−i)z=(z_i,z_{-i}), a decentralised selection ui=κi(zi)u_i=\kappa_i(z_i) must satisfy F^i(Φ−1(zi,z−i),κi(zi))∈TΩi(zi)\widehat{F}_i(\Phi^{-1}(z_i,z_{-i}),\kappa_i(z_i))\in T_{\Omega_i}(z_i) for all zi∈∂Ωiz_i\in\partial\Omega_i, z−i∈∏j≠iΩjz_{-i}\in\prod_{j\ne i}\Omega_j. Theorem 5 guarantees admissible input trajectories. Additional regularity conditions are needed to ensure that a state-dependent selection yields well-posed closed-loop dynamics.

Illustrative example

Consider the following Kuramoto oscillator comprising of two interconnected subsystems

θ˙1=a1+ksin⁡(θ2−θ1)+u1,\dot{\theta}_1=a_1+k\sin(\theta_2-\theta_1)+u_1,
θ˙2=a2+ksin⁡(θ1−θ2)+u2,\dot{\theta}_2=a_2+k\sin(\theta_1-\theta_2)+u_2,

defined on T2=S1×S1\mathbb{T}^2=\mathbb{S}^1\times\mathbb{S}^1 subject to input constraints ui∈Ui=[−uˉi,uˉi]u_i\in U_i=[-\bar{u}_i,\bar{u}_i], with uˉi>0\bar{u}_i>0 and i=1,2i=1,2. The parameters of the system are: a1,a2∈Ra_1,a_2\in\mathbb{R} are the drift and k≥0k\geq0 the coupling strength. The distributions Di=span⁡{∂θi}\mathcal{D}_i=\operatorname{span}\{\partial_{\theta_i}\}, i∈{1,2}i\in\{1,2\}, satisfy TT2=D1⊕D2T\mathbb{T}^2=\mathcal{D}_1\oplus\mathcal{D}_2, and [∂θ1,∂θ2]=0[\partial_{\theta_1},\partial_{\theta_2}]=0. The adapted chart Φ:U→(−π,π)2\Phi:U\to(-\pi,\pi)^2, mapping phase arcs to intervals, assigns each phase a unique zi∈(−π,π)z_i\in(-\pi,\pi), see Figure 2 top and middle. We now construct the adapted chart by defining the angular variables z1=θ1z_1=\theta_1, and z2=θ2−βsin⁡(θ1)z_2=\theta_2-\beta\sin(\theta_1) modulo 2π2\pi and β>θ1\beta>\theta_1. This transformation is a smooth diffeomorphism on T2\mathbb{T}^2 with inverse θ1=z1\theta_1=z_1, θ2=z2+βsin⁡(z1)\theta_2=z_2+\beta\sin(z_1) modulo 2π2\pi. The associated vector fields and distributions are: X1=∂θ1+βcos⁡θ1∂θ2X_1=\partial_{\theta_1}+\beta\cos\theta_1\partial_{\theta_2}, X2=∂θ2X_2=\partial_{\theta_2} and let D1=span⁡{X1}\mathcal{D}_1=\operatorname{span}\{X_1\} and D2=span⁡{X2}\mathcal{D}_2=\operatorname{span}\{X_2\}. These satisfy TT2=D1⊕D2T\mathbb{T}^2=\mathcal{D}_1\oplus\mathcal{D}_2 following linear independence of X1X_1 and X2X_2; additionally, we have [X1,X2]=0[X_1,X_2]=0 and dΦx(X1(θ))=(1,0)d\Phi_x(X_1(\theta))=(1,0) and dΦx(X2(θ))=(0,1)d\Phi_x(X_2(\theta))=(0,1). Therefore, Φ\Phi is adapted to the splitting because its differential maps each distribution Di(x)\mathcal{D}_i(x) onto the corresponding coordinate direction in the zz coordinates. The new dynamics are:

Adapted coordinates for the Kuramoto oscillator

Figure 2. Adapted coordinates for the Kuramoto oscillator in (16). (top) Invariant set Ω\Omega on T2\mathbb{T}^{2} with two trajectories (teal) and (purple) starting from the initial conditions z(1)∈Φ(Ω)z^{(1)} \in\Phi(\Omega) (teal) and z(2)∈Φ(Ω)z^{(2)} \in\Phi(\Omega) (purple) respectively towards zeqz_{\mathrm{eq}} (black). (bottom) Adapted chart image Φ(Ω)\Phi(\Omega). For both Ω\Omega and Φ(Ω)\Phi(\Omega), the vector field (red) shows that all flows remain within these sets.

z˙1=b1(z)+u1,\dot{z}_1=b_1(z)+u_1,
z˙2=βcos⁡(z1)(b1(z)+u1)+b2(z)+u2,\dot{z}_2=\beta\cos(z_1)(b_1(z)+u_1)+b_2(z)+u_2,

where b1(z)=a1+ksin⁡(z2−βsin⁡(z1)−z1)b_1(z)=a_1+k\sin(z_2-\beta\sin(z_1)-z_1), b2(z)=a2+ksin⁡(z1−z2+βsin⁡(z1))b_2(z)=a_2+k\sin(z_1-z_2+\beta\sin(z_1)), and ui∈[−uˉi,uˉi]u_i\in[-\bar{u}_i,\bar{u}_i] and k≥0k\geq0. Under this transformation, each input map, however, is aligned with its corresponding distribution, i.e., u1u_1 to D1\mathcal{D}_1 and u2u_2 to D2\mathcal{D}_2 respectively.

The constraints on T2\mathbb{T}^2 for α∈(0,π2)\alpha\in(0,\frac{\pi}{2}) are ∣θ1∣≤α|\theta_1|\leq\alpha and ∣θ2−βsin⁡θ1∣≤α|\theta_2-\beta\sin\theta_1|\leq\alpha. The set Ω\Omega is clearly nonconvex, but the adapted charts transform it into a square. The inputs ui=−aiu_i=-a_i satisfy the tangent conditions whenever ∣ai∣≤uˉi|a_i|\leq\bar{u}_i. Since the input maps are aligned with the splitting, Corollary 1 provides control invariance for the entire set. We note that sets

[−α,α]×[c,d]⊂T2[-\alpha,\alpha] \times[c,d] \subset\mathbb{T}^{2} with c≤dc \le d cannot be made invariant. For invariance to hold, [c,d][c,d] must be contained in every interval [βsin⁡(θ1)−α,βsin⁡(θ1)+α][\beta\sin(\theta_{1})-\alpha,\beta\sin(\theta_{1})+\alpha], θ1∈[−α,α]\theta_{1} \in[-\alpha,\alpha]. Since sin⁡(θ1)\sin(\theta_{1}) is increasing on this interval, a nonempty intersection requires βsin⁡(α)−α≤α−βsin⁡(α)\beta\sin(\alpha)-\alpha\le\alpha-\beta\sin(\alpha). Due to this, if βsin⁡(α)>α\beta\sin(\alpha) > \alpha, then no nonempty rectangle contained in Ω\Omega can keep the full range [−α,α][-\alpha,\alpha] of the first phase.

The example shows how adapted coordinates enable subsystem tangency conditions to certify the entire curved region without restricting it to a rectangle in original coordinates. The synchronisation conditions in [18] do not directly certify this region, while the invariant set construction presented in [12] only applies to coupled linear systems.

Conclusion and Future Work

This paper addressed the issue of design of safety sets from a distributed point of view. We generalised the Euclidean notion of coordinates, i.e., assignment of vector field coordinates to agents, using distributions of the tangent bundle. Using these distributions, we have constructed control invariant sets with a product structure through a tangent bundle splitting and adapted coordinates. The torus example shows how adapted coordinates represent an entire curved constraint region as a product, whereas no contained rectangle in the original coordinates spans the full first phase range under the stated conditions. The construction, therefore, preserves the full constraint region while allowing for subsystem boundary checks. Future work will address the selection of compatible distributions and constraint sets, and extension to larger regions with complex or learned geometry.

References

  1. [1]F. Blanchini, “Set invariance in control,” Automatica, vol. 35, no. 11, pp. 1747–1767, 1999.DOI
  2. [2]J.-P. Aubin, Viability Theory, ser. Modern Birkhäuser Classics. Boston: Birkhäuser, 2009.
  3. [3]B. Hernandez, P. Baldivieso-Monasterios, and P. Trodden, “Distributed MPC: Guaranteeing Global Stability from Locally Designed Tubes,” IFAC-PapersOnLine, vol. 50, no. 1, pp. 11 829–11 834, jul 2017.DOI
  4. [4]P. Trodden, J. M. Maestre, and H. Ishii, “On the limits of robust control under adversarial disturbances,” IEEE Transactions on Automatic Control, pp. 1–16, 2026.arxiv.org/abs/2507.03630
  5. [5]P. R. Baldivieso-Monasterios and P. A. Trodden, “Coalitional predictive control: Consensus-based coalition forming with robust regulation,” Automatica, vol. 125, p. 109380, 2021.arxiv.org/abs/2001.10240
  6. [6]E. Masero, P. R. Baldivieso-Monasterios, J. M. Maestre, and P. A. Trodden, “Robust coalitional model predictive control with plug-and-play capabilities,” Automatica, vol. 153, p. 111053, 2023.DOI
  7. [7]A. D. Ames, X. Xu, J. W. Grizzle, and P. Tabuada, “Control barrier function based quadratic programs for safety critical systems,” IEEE Transactions on Automatic Control, vol. 62, no. 8, pp. 3861–3876, 2017.DOI
  8. [8]P. Rabiee and J. B. Hoagg, “Soft-minimum and soft-maximum barrier functions for safety with actuation constraints,” Automatica, vol. 171, p. 111921, Jan. 2025.DOI
  9. [9]S. Prajna, A. Jadbabaie, and G. J. Pappas, “A framework for worst-case and stochastic safety verification using barrier certificates,” IEEE Transactions on Automatic Control, vol. 52, no. 8, pp. 1415–1428, 2007.DOI
  10. [10]H. Wang, A. Papachristodoulou, and K. Margellos, “Distributed safe control design and probabilistic safety verification for multi-agent systems,” Automatica, vol. 179, p. 112393, 2025.DOI
  11. [11]S. Bansal, M. Chen, S. Herbert, and C. J. Tomlin, “Hamilton-Jacobi reachability: A brief overview and recent advances,” 2017 IEEE 56th Annu. Conf. Decis. Control. CDC 2017, vol. 2018–January, pp. 2242–2253, 2018.arxiv.org/abs/1709.07523
  12. [12]P. Nilsson and N. Ozay, “Synthesis of separable controlled invariant sets for modular local control design,” in 2016 American Control Conference (ACC). IEEE, 2016, pp. 5656–5663.arxiv.org/abs/1510.02990
  13. [13]A. Pirastehzad, A. van der Schaft, and B. Besselink, “Modular verification and design of interconnected linear systems using (γ, δ)-similarity,” Automatica, vol. 190, p. 113046, Aug. 2026.DOI
  14. [14]V. Jurdjevic, Geometric Control Theory. Cambridge University Press, 1996.DOI
  15. [15]F. Bullo and A. D. Lewis, Geometric Control of Mechanical Systems: Modeling, Analysis, and Design for Simple Mechanical Control Systems. Springer, 2005, vol. 49.DOI
  16. [16]M. de Sa, P. Ong, and A. D. Ames, “From bundles to backstepping: Geometric control barrier functions for safety-critical control on manifolds,” arXiv preprint arXiv:2510.20202, 2025.
  17. [17]T. Ibuki, S. Wilson, A. D. Ames, and M. Egerstedt, “Distributed collision-free motion coordination on a sphere: A conic control barrier function approach,” IEEE Control Systems Letters, vol. 4, no. 4, pp. 976–981, 2020.DOI
  18. [18]F. Dörfler and F. Bullo, “Synchronization and transient stability in power networks and nonuniform Kuramoto oscillators,” SIAM Journal on Control and Optimization, vol. 50, no. 3, pp. 1616–1642, 2012.
  19. [19]T. Lee, “Geometric tracking control of the attitude dynamics of a rigid body on SO(3),” in Proceedings of the 2011 American Control Conference. IEEE, 2011, pp. 1200–1205.
  20. [20]M. Abdelgali, M. Krstic, and J. I. Poveda, “Non-holonomic gradient play: Leafwise nash equilibria, stability, and deception,” Sep. 2026.arxiv.org/abs/2609.38051
  21. [21]M. P. do Carmo, Riemannian Geometry, ser. Mathematics: Theory & Applications. Boston: Birkhäuser, 1992.
  22. [22]J. M. Lee, Introduction to Riemannian Manifolds, 2nd ed., ser. Graduate Texts in Mathematics. Cham: Springer, 2018, vol. 176.DOI
  23. [23]J. M. Lee, Introduction to Smooth Manifolds, 2nd ed., ser. Graduate Texts in Mathematics. New York: Springer, 2013, vol. 218.
  24. [24]R. T. Rockafellar and R. J.-B. Wets, Variational Analysis. Springer, 1998.DOI

Paper details

vlnc:W_nd4eyu6j
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Engineering

Not recorded for this paper

View statement
Looking for related papers…
Contents