Ellipsoidal Tubes in Geometry & Control

Updated 28 May 2026

Ellipsoidal tubes are geometric constructs with ellipsoidal cross-sections that appear in both complex geometry—offering canonical complexification of convex domains—and robust control for uncertainty management.
They ensure key properties such as linear convexity, C-convexity, and complete Kobayashi hyperbolicity, enabling rigorous function theory and metric preservation.
In robust MPC, ellipsoidal tubes facilitate constraint tightening and computationally efficient second-order cone programming formulations for real-time adaptive control.

Ellipsoidal tubes are geometric and analytic constructions characterized by an ellipsoidal cross-section, appearing in both complex geometry and robust control theory. In complex analysis, Lempert's "elliptic tubes" provide a canonical complexification of real convex domains and their associated geometric structures, preserving projective invariance and supporting rich function-theoretic properties. In robust control and model predictive control (MPC), ellipsoidal tubes encode the propagation of uncertainty in the state-space and allow for constraint tightening and robust feasibility guarantees. Both contexts leverage the tractability and projective invariance of ellipsoidal sets, but their mathematical formalizations and applications are distinct.

1. Elliptic Tubes in Projective and Complex Geometry

Given a properly convex domain $D \subset \mathbb{R}P^n$ in real projective space, its elliptic tube $D^e \subset \mathbb{C}P^n$ is defined as the union of all complexified real line segments within $D$ . For each projective line $L \subset \mathbb{R}P^n$ and every segment $I \subset D \cap L$ , one forms the closed circle $I^e$ (the complexification) with diameter $I$ inside the complexified line $L^c \cong \mathbb{C}P^1$ , and sets

$D^e = \bigcup\{ I^e \subset \mathbb{C}P^n \mid I \subset D \text{ is a closed segment}\}.$

An equivalent dual cone formulation exists: lifting to cones $\mathfrak{D} \subset \mathbb{R}^{n+1}$ , one obtains

$D^e \subset \mathbb{C}P^n$ 0

where $D^e \subset \mathbb{C}P^n$ 1 is the dual cone of positive functionals. This construction extends projective invariance to the complex setting and naturally produces a "complex thickening" of $D^e \subset \mathbb{C}P^n$ 2 that retains strong convexity properties (Alessandrini et al., 2010).

2. Convexity, C-convexity, and Hyperbolicity Properties

Ellipsoidal tubes $D^e \subset \mathbb{C}P^n$ 3 satisfy several stringent convexity properties:

Linear Convexity: For any open or compact convex $D^e \subset \mathbb{C}P^n$ 4, $D^e \subset \mathbb{C}P^n$ 5 is a dual complement and hence linearly convex in $D^e \subset \mathbb{C}P^n$ 6, i.e., every point outside $D^e \subset \mathbb{C}P^n$ 7 is separated by a complex hyperplane [Corollary 3.3 in (Alessandrini et al., 2010)].
C-convexity: If $D^e \subset \mathbb{C}P^n$ 8 has a $D^e \subset \mathbb{C}P^n$ 9 boundary and is properly convex, then $D$ 0 is C-convex—its intersection with every complex line is connected and simply connected. At every boundary point of $D$ 1, the set of supporting complex hyperplanes is nonempty and connected, which characterizes C-convexity following [Passare–Sigurdsson]. This extends to all properly convex $D$ 2 by exhaustion arguments [Theorem 4.6 and Corollary 4.7, (Alessandrini et al., 2010)].
Complete Kobayashi Hyperbolicity: $D$ 3 is a complete Kobayashi-hyperbolic domain; the Kobayashi distance $D$ 4 coincides with the Hilbert metric $D$ 5 on the real slice $D$ 6. Each point $D$ 7 projects uniquely to $D$ 8 minimizing the Kobayashi distance [Theorem 4.1, Corollary 4.8, (Alessandrini et al., 2010)].

These properties ensure that $D$ 9 supports a rich function theory, with strong separation, tautness, and hyperconvexity.

3. Ellipsoidal Tubes in Robust Model Predictive Control

In robust MPC, an ellipsoidal tube is a sequence of ellipsoidal sets $L \subset \mathbb{R}P^n$ 0 around a nominal trajectory $L \subset \mathbb{R}P^n$ 1 capturing the evolution and propagation of reachable states under uncertainty. The canonical form is:

$L \subset \mathbb{R}P^n$ 2

with dynamics:

$L \subset \mathbb{R}P^n$ 3

where perturbations $L \subset \mathbb{R}P^n$ 4 are modeled by bounding ellipsoids, and controllers propagate the tube by updating $L \subset \mathbb{R}P^n$ 5 using linearizations and ellipsoidal over-approximations of model nonlinearities (Heinlein et al., 16 Sep 2025, Buerger et al., 24 Jan 2025, Buerger et al., 5 Mar 2026).

Ellipsoidal tubes provide an efficient mechanism to guarantee that all state and input constraints will be met for any realization within the propagated uncertainty, using tightened constraints built on Minkowski sums and support function bounds. Their geometry enables scalable second-order cone programming formulations and online feasibility via backtracking line search (Buerger et al., 24 Jan 2025, Buerger et al., 5 Mar 2026).

4. Multi-Stage Partitioning and Scenario Trees with Ellipsoidal Tubes

Traditional ellipsoidal tube MPC employs a single feedback law for all possible uncertainty paths, often resulting in conservatism. Scenario-based methods instead branch on uncertainty realizations but suffer high computational burden and weaker guarantees.

Recent advances merge these approaches by partitioning uncertainty ellipsoids at key tree nodes via hyperplanes:

At split stage $L \subset \mathbb{R}P^n$ 6, the ellipsoid $L \subset \mathbb{R}P^n$ 7 is bisected into two sets by a hyperplane $L \subset \mathbb{R}P^n$ 8. Each halfspace-intersected ellipsoid is over-approximated by a minimal-volume covering ellipsoid using Löwner–John ellipsoid formulas, which provide analytic expressions for the center and shape of the covering (Heinlein et al., 16 Sep 2025).
Each branch then independently propagates its ellipsoidal tube and controller, reducing conservatism.
Complexity scales exponentially with partition depth $L \subset \mathbb{R}P^n$ 9 (i.e., $I \subset D \cap L$ 0 scenarios), motivating small $I \subset D \cap L$ 1 and limiting splitting to early time steps, which still provides significant recourse with tractable computation.

Numerical studies in human–robot systems show that multi-stage ellipsoidal partitioning delivers substantial improvements in closed-loop cost compared to single-tube approaches for similar constraint satisfaction and with manageable computational burden (Heinlein et al., 16 Sep 2025).

5. Algorithmic and Computational Properties

Ellipsoidal tube-based MPC relies on the interplay between ellipsoidal set propagation, constraint tightening, and recursive feasibility recovery. Key computational features include:

Convex SOCP Formulation: All relevant update and constraint tightening steps (tube propagation, Minkowski sums, etc.) admit a second-order cone programming (SOCP) formulation, ensuring polynomial scaling with system size and uncertainty dimension (Buerger et al., 24 Jan 2025, Buerger et al., 5 Mar 2026).
Recursive Feasibility: A backtracking line-search over nominal trajectories guarantees that feasible tubes persist at each time step, ensuring robust operation even in the presence of model mismatch or aggressive linearizations.
Parameter Adaptation: Online set-membership parameter estimation can be incorporated, successively shrinking uncertainty polytopes and associated ellipsoidal tubes as new data accumulates, improving closed-loop performance (Buerger et al., 5 Mar 2026, Buerger et al., 24 Jan 2025).
Computational Scaling: For systems with $I \subset D \cap L$ 2 uncertain parameters, computational demand empirically grows as $I \subset D \cap L$ 3 in test cases, and the number of decision variables in single-tube methods is linear in the system state and input dimension.

6. Connections Between Geometric and Control-Theoretic Tubes

The concept of ellipsoidal tubes provides a projectively invariant, analytic, and computationally tractable framework for "thickening" geometric objects or trajectories. In complex geometry, elliptic tubes serve as a canonical complexification maintaining convexity and hyperbolicity, with direct links to classical tube domains in $I \subset D \cap L$ 4 when $I \subset D \cap L$ 5 is an ellipsoid—the tube $I \subset D \cap L$ 6 is the unit ball in $I \subset D \cap L$ 7 and supports the classical Bergman–Poincaré metric (Alessandrini et al., 2010).

In robust control, ellipsoidal tubes encapsulate all admissible state evolutions subject to uncertainties, supporting robust constraint satisfaction and efficient online computation. The mathematical techniques (support functions, duality, ellipsoidal over-approximation) that underpin both domains highlight a deep structural connection: both utilize the favorable analytic properties of ellipsoidal sets for tractable extensions—into the complex domain or into high-dimensional, uncertain trajectory spaces.

7. Illustrative Examples and Applications

Complexification of Convex Projective Manifolds: For $I \subset D \cap L$ 8, a convex real projective manifold, the complexification $I \subset D \cap L$ 9 is a complete Kobayashi-hyperbolic manifold, topologically homeomorphic to the tangent bundle $I^e$ 0 via a projectively invariant map built from the geometry of elliptic tubes [Theorem 5.3, (Alessandrini et al., 2010)].
Control of Human–Robot Systems: Ellipsoidal tube MPC, especially with scenario partitioning, manages uncertainty propagation and safety constraints in multi-actor systems such as human–robot interaction, offering quantifiable reductions in worst-case cost and tractable real-time implementation (Heinlein et al., 16 Sep 2025).
Adaptive Robust Control: Incorporating parameter estimation and adaptive tube radii, robust NMPC schemes with ellipsoidal tubes realize closed-loop input-to-state practical stability and average cost bounds under model and disturbance uncertainty (Buerger et al., 24 Jan 2025, Buerger et al., 5 Mar 2026).