⚡ electrical engineering

Minimum settling-time PI control of pure delay processes under a hard non-overshoot constraint: exact boundary-contact characterization and the role of the MID point

This paper establishes that while the generic multiplicity-induced-dominancy (GMID) point provides the theoretical limit for minimum settling-time PI control of pure delay processes under a strict non-overshoot constraint, the actual optimal gains for any finite tolerance strictly outperform this fixed tuning by approximately 40%.

Original authors: Senol Gulgonul

Published 2026-06-16

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Senol Gulgonul

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to park a very long, heavy truck (the "process") that has a significant delay between when you turn the steering wheel and when the truck actually moves (the "pure delay"). Your goal is to get the truck to stop exactly at a specific spot (the "setpoint") as quickly as possible.

However, you have a strict rule: The truck must never overshoot the line. It cannot swing past the spot and then back up. It has to approach the line and settle within a tiny, acceptable margin of error (like a few inches) without ever crossing it.

This paper solves the mathematical puzzle of how to steer this delayed truck to stop in the absolute minimum time while obeying that "no overshoot" rule.

Here is the breakdown of their discovery using simple analogies:

1. The "Echo" Problem

In a normal system, if you make a mistake, it smooths out. But in a system with pure delay, the reaction is jerky. The authors describe the truck's movement as a series of "echoes."

The Jumps: Every time the truck hits a "delay boundary" (like hitting a second, then a third second), the position jumps up or down suddenly. It's like a drumbeat that gets quieter with every hit.
The Shape: Between these jumps, the truck moves in smooth, curved lines (polynomials).
The Strategy: To stop the truck as fast as possible without crossing the line, you have to time your steering so that these "jumps" land exactly on the edge of your allowed error zone, and the "curves" between them just barely touch the target line without going over.

2. The "Pinned" Solution

The authors found that the perfect steering setting isn't a smooth, continuous curve that changes as you get closer to the goal. Instead, the optimal solution is "pinned" to specific moments in time.

Think of it like a tent being held up by stakes. The "stakes" are the specific moments where the truck's position hits the limit (either the top limit or the bottom limit of the error zone).

The math shows that the best steering settings are determined entirely by how many of these "stakes" you need to hold the tent up.
For some specific error margins (between roughly 4% and 8%), there is a magic, fixed steering setting that works perfectly for all of them. It's like finding a "one-size-fits-all" gear that is perfect for a specific range of speeds.

3. The "Staircase" of Time

If you try to make the error margin smaller and smaller (demanding the truck stop more precisely), the time it takes to stop doesn't shrink smoothly. Instead, it looks like a staircase.

The Flats: Sometimes, even if you demand a tighter precision, the truck can't stop any faster because it's "stuck" waiting for the next delay jump to land exactly on the line. The time stays the same for a range of precision levels.
The Steps: Once you tighten the precision enough, the truck can finally land on a new, earlier "jump," and the time drops suddenly.

4. The "Perfect" Theoretical Point (GMID) vs. Reality

There is a famous, theoretical "perfect" steering setting in control theory called the GMID point (Multiplicity-Induced-Dominancy).

The Theory: This point is mathematically special because it makes the system decay as fast as possible in theory. It's like a theoretical "perfect engine."
The Reality Check: The authors discovered something surprising. While this "perfect engine" is the limit of what you can achieve as you demand infinite precision, it is never actually the best choice for any real-world task.
The Analogy: Imagine a runner who is the fastest in the world in a vacuum. But in a real race with wind and obstacles, a slightly different runner who adjusts their stride to the wind wins every time. The "perfect" theoretical runner is the limit, but the "adapted" runner (the one the authors found) is always faster in practice. The adapted runner beats the theoretical one by about 40% in typical scenarios.

5. The Trade-off: Speed vs. Stability

The paper also compares their "adapted" solution to other common methods (like simple cancellation or integral control).

The Ladder: They built a "ladder" of solutions.
- At the bottom: Slow but very stable (like a cautious driver).
- At the top: The authors' "contact optimum" (the fastest possible stop without overshooting).
The Cost: The fastest method is about 2.5 times faster than the slow, cautious methods. However, it comes with a price: the system becomes slightly more sensitive to noise (like driving a very fast car that is harder to control on a bumpy road).

Summary

The paper proves that for delayed systems, the fastest way to stop without overshooting is to treat the problem like a game of "pinning" the system's position to specific time boundaries. They found a "sweet spot" where the steering settings are constant, and they showed that the famous theoretical "perfect" setting is actually a trap—it's the limit you approach but never reach, because a slightly different, custom-tuned setting is always faster in the real world.

Technical Summary: Minimum Settling-Time PI Control of Pure Delay Processes under Hard Non-Overshoot Constraints

Problem Formulation
This paper addresses the exact problem of minimizing the settling time ( $T_s$ ) for a Proportional-Integral (PI) controller applied to a pure delay process $P(s) = Ke^{-Ls}$ , subject to a hard time-domain constraint of zero overshoot ( $y(t) \le 1$ for all $t$ ) and a settling band tolerance $\delta$ (i.e., $|y(t)-1| \le \delta$ ). Unlike classical approaches that rely on spectral criteria (e.g., Integral of Squared Error or spectral abscissa minimization) or assume retarded systems, this work focuses on the specific challenges of neutral delay systems generated by PI control of relative-degree-zero plants. The authors aim to resolve two open questions: (1) the exact characterization of the constrained optimum in the time domain, and (2) the relationship between this constrained optimum and the spectral optimum defined by the Multiplicity-Induced-Dominancy (MID) principle.

Methodology and Theoretical Framework
The authors employ an "echo" analysis of the closed-loop step response. By normalizing time by $L$ and defining gains $a = KK_p$ and $b = KK_iL$ , the system is characterized by the neutral quasipolynomial $\Delta(s) = s + (as+b)e^{-s}$ .

Echo Structure: The step response is shown to be piecewise polynomial on intervals $[k, k+1)$ , with geometrically decaying jump discontinuities at integer boundaries $t=k$ . The magnitude of these jumps is $(-1)^{k+1}a^k$ .
Boundary-Contact Characterization: The optimization problem is reformulated as a finite-dimensional algebraic problem. The authors prove that the optimal response is "pinned" to the echo grid. The active constraints (contacts) that define the optimum occur exclusively at segment boundaries (kinks or jump landings) or as interior tangencies.
Regime Analysis: The solution space is divided into regimes based on the tolerance $\delta$ . The authors derive explicit polynomial systems for the controller gains $(a, b)$ by equating the number of active contact equations to the number of controller gains.
Spectral Comparison: The paper rigorously analyzes the GMID (Generic Multiplicity-Induced-Dominancy) point, where the characteristic root has maximal multiplicity (triple root at $s=-2$ ), and compares its time-domain performance against the derived contact optima.

Key Contributions and Results

Exact Boundary-Contact Characterization (Theorem 1):
The constrained optimum is characterized by an equioscillation-type structure where active contacts sit at echo boundaries.
- Types of Contacts: The optimal response features specific contact types: boundary grazes at the setpoint ( $y=1$ ), boundary trough anchors at the lower band edge ( $y=1-\delta$ ), and jump landings where an upward jump lands exactly on the band edge.
- Computability: For any $\delta$ , the optimal gains are the roots of an explicit small polynomial system. The optimal settling time $T_s^*(\delta)$ forms a "staircase" function: smooth arcs connected by exact integer flats (e.g., $3L, 5L$ ) caused by jump-landing pinning.
Closed-Form Optimal Regime (Theorem 2):
For a specific range of tolerances $\delta \in [(3-2\sqrt{2})/4, (3-2\sqrt{2})/2] \approx [4.29\%, 8.58\%]$ , the optimal gains are independent of $\delta$ :
- $K_p = \frac{1}{K}(1 - \frac{\sqrt{2}}{2})$
- $K_i = \frac{1}{KL}\frac{\sqrt{2}}{2}$
- $T_i = (\sqrt{2}-1)L$
- The settling time is given by $T_s^*(\delta) = (4 - \sqrt{2} - 2\sqrt{\delta})L$ .
  Outside this window, the gains vary monotonically with $\delta$ .
The Role of the GMID Point:
- Feasibility and Dominance: The GMID point ( $a=e^{-2}, b=4e^{-2}$ ) is proven to satisfy the hard non-overshoot constraint and uniquely maximizes the decay rate (minimizes the spectral abscissa $\alpha$ ) among all non-overshooting gains.
- Limit vs. Optimum: As $\delta \to 0$ , the optimal gains of the constrained problem converge to the GMID point. However, for any finite $\delta$ (down to $10^{-6}$ ), the GMID tuning is strictly suboptimal compared to the $\delta$ -adapted contact optimum.
- Performance Gap: The adapted contact optimum achieves settling times approximately 40% faster than the fixed GMID tuning at $\delta=2\%$ . The GMID point acts as the accumulation point of the minimizers but is never the minimizer itself at finite tolerance. The performance gap is attributed to the secular penalty ( $\Theta(\ln \ln(1/\delta))$ ) introduced by the triple root's $t^2$ factor in the GMID response, which the contact optimum avoids through transient shaping.
Design Ladder and Trade-offs:
The paper quantifies the speed/robustness trade-off ( $M_s$ ) across different tuning strategies:
- Cancellation/I-control: Slowest, lowest $M_s$ (~1.39).
- GMID: Intermediate speed, $M_s \approx 1.50$ .
- Contact Optima: Fastest (approx. 2.5x faster than cancellation methods), but with higher $M_s$ (up to ~1.76).
- FOTD Extension: Numerical results for First-Order-Plus-Time-Delay plants show that the non-monotone settling time behavior persists, where a small lag can actually improve performance over the pure delay limit by smoothing echo extrema.

Significance and Claims
The paper claims to provide the first exact treatment of minimum settling-time PI design for delay plants under hard non-overshoot constraints. It resolves the open question regarding the relationship between spectral optimality (MID) and time-domain optimality for neutral systems, demonstrating that while the MID point is the spectral limit, it is not the time-domain optimum for any finite specification. The work establishes that the optimal control strategy is a "contact configuration" pinned to the echo grid, rendering the problem finitely algebraic and exactly solvable, rather than requiring numerical optimization or relying on spectral abscissa approximations. The findings challenge the assumption that maximizing decay rate (via MID) automatically yields the fastest settling time under hard constraints.