Boundedness of solutions in feedback systems with antithetic controllers

1. Problem Statement

The paper addresses the fundamental control theory problem of global boundedness for a specific class of nonlinear feedback systems arising in synthetic biology.

• System Model: The study focuses on a four-dimensional dynamical system (Equation 1) representing the closed-loop interconnection of a linear plant ($x_2, x_3$) and an "antithetic controller" ($x_1, x_4$). The system is defined by:
  $$\begin{aligned} \dot{x}_1 &= \alpha_1 - \alpha_2 x_1 x_4 \\ \dot{x}_4 &= \alpha_7 x_3 - \alpha_8 x_1 x_4 \\ \dot{x}_2 &= \alpha_3 x_1 - \alpha_4 x_2 \\ \dot{x}_3 &= \alpha_5 x_2 - \alpha_6 x_3 \end{aligned}$$
  where all parameters $\alpha_k > 0$ and initial conditions are non-negative.
• Context: This architecture implements integral feedback via sequestration (annihilation) reactions between two molecular species ($x_1$ and $x_4$). While it is known that such systems admit a unique equilibrium and exhibit robust perfect adaptation, and that bounded trajectories satisfy Poincaré–Bendixson properties (ruling out chaos), it was previously unknown whether all trajectories remain bounded for any initial condition.
• Challenge: The damping term $\alpha_2 x_1 x_4$ in the $\dot{x}_1$ equation is not a direct feedback signal but is generated through a cascade of stable subsystems ($x_1 \to x_2 \to x_3 \to x_4$). This introduces delays and attenuation, making classical Lyapunov-based stability analysis difficult to apply directly.

2. Methodology

The authors employ a time-domain analysis based on differential inequalities and comparison principles, explicitly avoiding the construction of a Lyapunov function. The core logic follows a "small-gain" intuition in the time domain: if a state grows too large and persists, the delayed feedback will eventually become strong enough to reverse the growth.

The proof strategy is divided into three main logical steps:

A. Forward Completeness and Invariance

• Invariance: The non-negative orthant $\mathbb{R}^4_{\ge 0}$ is shown to be forward invariant using the Bony–Brezis criterion (checking that vector fields point inward at the boundaries).
• Global Existence: By integrating the differential inequalities successively ($\dot{x}_1 \le \alpha_1$, $\dot{x}_2 \le \alpha_3 x_1$, etc.), the authors derive polynomial upper bounds for all states over any finite interval. This proves that solutions cannot escape to infinity in finite time (forward completeness).

B. The "Self-Correcting" Mechanism (Lemma 2)

The central technical contribution is proving that persistent excursions of the regulated state $x_1$ are impossible.

• Hypothesis: Assume $x_1(t) \ge L$ for a duration $T$.
• Cascade Propagation:
  1. Since $x_1$ is large, $x_2$ grows (after a delay $\Delta_2$).
  2. The growth of $x_2$ forces $x_3$ to grow (after a delay $\Delta_3$).
  3. The growth of $x_3$ drives $x_4$ to grow (after a delay dependent on $L$).
• Threshold Crossing: The authors define specific time delays ($\Delta_2, \Delta_3, \Delta_4$) and show that if $x_1$ remains above a sufficiently high threshold $L_0$ for a waiting time $T_0$, the product $x_1(T_0)x_4(T_0)$ will exceed $\alpha_1/\alpha_2$.
• Result: Once $x_1 x_4 > \alpha_1/\alpha_2$, the derivative $\dot{x}_1 = \alpha_1 - \alpha_2 x_1 x_4$ becomes strictly negative.
• Persistence: Using a "first-hitting-time" argument on the product $p(t) = x_1(t)x_4(t)$, they prove that once this product exceeds the threshold, it cannot drop below it while $x_1$ remains large, ensuring $\dot{x}_1$ stays negative.

C. Global Boundedness (Propositions 1 & 2)

• Boundedness of $x_1$: By contradiction, if $x_1$ were to exceed a bound $M_1$, it would have to stay above a high threshold $L$ for time $T_0$. However, Lemma 2 guarantees that in such a scenario, $\dot{x}_1$ becomes negative, preventing the overshoot. Thus, $x_1$ is globally bounded.
• Boundedness of $x_2, x_3, x_4$:
  • $x_2$ and $x_3$ are bounded via standard linear comparison arguments since their inputs ($x_1, x_2$) are bounded.
  • $x_4$ is bounded using an auxiliary function $W(t) = x_4 + c x_2 + d x_3$. The authors construct constants $c$ and $d$ such that the cross-terms cancel, leaving a differential inequality where $W$ is driven by a negative feedback term when $x_4$ becomes too large.

3. Key Contributions

1. Resolution of the Boundedness Conjecture: The paper definitively proves that all solutions of the antithetic feedback system are bounded for any non-negative initial condition, closing a gap left open by previous literature.
2. Lyapunov-Free Proof Technique: The authors demonstrate that boundedness can be established using direct differential inequalities and time-domain delay analysis, offering a more transparent interpretation of the feedback mechanism than complex Lyapunov functions.
3. Mechanism Interpretation: The work provides a clear physical interpretation of the "antithetic" controller: it acts as a time-domain small-gain system. The forward propagation of a large signal through the cascade is eventually counteracted by a delayed negative feedback loop, preventing unbounded growth.
4. Generalizability: The authors note that the technique extends to more general stable positive linear systems, not just the specific second-order plant used in the example.

4. Results

• Theoretical: Theorem 1 establishes that for any initial state in the non-negative orthant, the trajectory $(x_1(t), x_2(t), x_3(t), x_4(t))$ remains within a compact set for all $t \ge 0$.
• Numerical Validation: Section 3 provides a numerical example with specific parameters ($\alpha_2=\alpha_8=30$, others $=1$).
  • The system exhibits sustained oscillations (limit cycles).
  • Theoretical bounds were calculated (e.g., $M_{x1} \approx 3.14$).
  • Simulations confirmed that trajectories remain within these bounds, though the theoretical bounds for downstream states ($x_2, x_3, x_4$) were conservative (significantly larger than observed values) due to the accumulation of worst-case estimates in the cascade.

5. Significance

• Synthetic Biology: This result provides rigorous theoretical assurance for the safety and stability of synthetic biological circuits utilizing antithetic integral feedback. It confirms that these circuits will not exhibit "runaway" behavior (unbounded protein production or degradation) even under extreme initial conditions or disturbances.
• Control Theory: The paper offers a novel analytical framework for nonlinear systems with multiplicative feedback and distributed delays. By bypassing the need for Lyapunov functions, it opens a pathway for analyzing other complex biological and engineering systems where constructing a global Lyapunov function is intractable.
• Robustness: The boundedness result reinforces the robustness of integral feedback in biological contexts, ensuring that the "internal model" principle holds even in the presence of nonlinearities and delays inherent in molecular sequestration.