On the Nyquist Criterion and the Scaled Relative Graph for SISO systems: A Generalized Circle Criterion
summary
The gist
Scaled Relative Graphs (SRGs) provide a novel graphical frequency-domain method for the analysis of nonlinear systems that suffers from a pitfall limiting its applicability in analyzing practical…
In short
The paper addresses a limitation in Scaled Relative Graph (SRG) analysis for nonlinear systems, particularly concerning unstable linear components in feedback loops. The authors introduce an extended SRG by incorporating Nyquist criterion information to resolve this pitfall. This extension allows for a consistent stability and L2-gain analysis across general interconnections of SISO LTI and nonlinear operators, generalizing the circle criterion.
Key concepts
- Extended SRG
- This is a modified version of the SRG that includes both the h-convex hull of Nyquist values (GR) and information from the Nyquist criterion (NR). It serves as an effective tool to compute stability conditions consistently, even when standard SRG calculus fails for unstable linear systems.
- Operator Context Free Grammar (OCFG)
- This is a formal language approach used to represent complex system interconnections. Words in this grammar define a system structure, allowing the authors to derive unique SRG bounds by systematically replacing individual operator components with their corresponding SRG representations.
- Generalized Circle Criterion
- This criterion provides an L2-gain bound for Lur'e systems that is broader than the traditional circle criterion. It applies to arbitrary operators, not just those within a sector, offering a more general method for analyzing stability and control performance.
- Incremental vs. Non-Incremental Bounds
- The paper derives two types of L2-gain bounds: incremental bounds, which are useful for stability analysis using the homotopy theorem, and non-incremental bounds. The latter is derived using a different theorem and is particularly useful for finding Lyapunov functions.
Terminology used across episodes
This episode discusses
- On the Nyquist Criterion and the Scaled Relative Graph for SISO systems: A Generalized Circle Criterion · Paper Radio
- Nonlinear Bandwidth and Bode Diagrams based on Scaled Relative Graphs
- Scaled Relative Graph Analysis of Lur'e Systems and the Generalized Circle Criterion
- A homotopy theorem for incremental stability
- Equilibrium-Independent Control of Continuous-Time Nonlinear Systems via the LPV Framework -- Extended Version
- The Scaled Relative Graph of a Linear Operator
The paper
On the Nyquist Criterion and the Scaled Relative Graph for SISO systems: A Generalized Circle Criterion · Read on arXiv
University of Technology Eindhoven · HUN-REN Institute for Computer Science and Control, Budapest
Scaled Relative Graphs (SRGs) provide a novel graphical frequency-domain method for the analysis of nonlinear systems. However, existing SRG-based methods are limited to stable open-loop dynamics and can lead to false conclusions when applied to unstable Linear and Time-Invariant (LTI) components. In this paper, we identify the source of this pitfall and resolve it by introducing the extended SRG, which augments the SRG of an LTI system with the encirclement information of the Nyquist criterion. We show how the extended SRG can be used for stability and H infinity-performance analysis of LTI systems and the computation of SRGs for unstable LTI systems. Building on these results, we derive graphical stability and incremental L 2-gain conditions for three canonical nonlinear feedback interconnections: the Lur'e system, the controlled Lur'e system, and an LTI plant with Lur'e controller. As a corollary, we derive a user-friendly generalized circle criterion that relates the L 2-gain directly to separation between the Nyquist curve and the critical disk from the classical circle criterion, requiring SRGs only for the proof, not its statement. The proposed approach is validated on several examples involving unstable LTI dynamics and nonlinearities, demonstrating applicability beyond the capabilities of existing SRG methods.
Transcript
Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.
Rosa: We're looking at the paper: "On the Nyquist Criterion and the Scaled Relative Graph for SISO systems".
Dev: Here's the gist of the paper. Scaled Relative Graphs (SRGs) provide a novel graphical frequency-domain method for the analysis of nonlinear systems that suffers from a pitfall limiting its applicability in analyzing practical nonlinear systems.
Rosa: The authors overcome this pitfall by introducing a novel reformulation of the SRG of a linear time-invariant operator and combining the SRG with the Nyquist criterion to obtain a theorem assessing stability and L2-gain performance for general interconnections of nonlinear dynamic systems.
Dev: The Pitfall in Current Analysis The current SRG analysis suffers from a pitfall when applied to unstable LTI systems in a feedback interconnection.
Rosa: This contradiction arises because while the Nyquist criterion correctly predicts instability for an unstable plant, applying the rules of SRG calculus can lead to incorrect results due to limitations regarding the domain of the operator for which the SRG is computed.
Title and authors: Rosa: So we're looking at this paper titled "On the Nyquist Criterion and the Scaled Relative Graph for SISO systems: A Generalized Circle Criterion". It’s about taking something that has been around, like the Scaled Relative Graph Analysis, and fixing a problem where it doesn't work well for practical nonlinear systems.
Dev: Right. The authors are tackling this pitfall by doing a reformulation of the Scaled Relative Graph for linear time-invariant operators and then sticking it together with the Nyquist criterion to get a theorem that assesses stability and L2-gain performance for general interconnections of nonlinear dynamic systems.
Taro: What I'm hearing is they're dealing with unstable linear time-invariant systems in a feedback loop, which seems like where the standard Scaled Relative Graph analysis hits a wall because of some domain issues with the operator.
Rosa: Exactly. The pitfall they found is that even when the Nyquist criterion correctly tells you that an LTI system is unstable, applying the rules of SRG calculus can lead to wrong results because of what inputs are allowed for the operator in that analysis.
Dev: It seems like they resolve this by introducing something called the extended SRG, which they define as combining two things: the h-convex hull of Nyquist information and the encirclement information from the Nyquist criterion.
Taro: So it's an extension where you’re not just looking at the graph itself but also factoring in whether something is outside a certain region based on that circle test.
Rosa: That extension allows them to show that applying SRG calculus directly leads to consistent results when you use this extended version instead. It makes the analysis more robust for these types of systems.
Dev: They then develop a formal language approach, an Operator Context Free Grammar, to represent these general interconnections as words in a language, and each word gets an SRG bound computed by replacing the basic operator with this new extended version.
Taro: I'm curious how that language helps when you have multiple different types of operators—linear ones and nonlinear ones—all connected together in a network.
Rosa: Well, the language lets them define these systems as words, and then they use a rule to substitute each part with its appropriate extended SRG version based on Proposition one. It turns the system definition into something manageable for calculation.
Title and authors: Dev: The main result is stated in Theorem six of this paper: for an operator defined by a word in that language, the gain bound satisfies that the gain is less than or equal to the minimum value of a certain set called C R.
Taro: So, what does this mean for someone looking at stability? Does it give them a direct way to check if the system will blow up, or is it more about how much energy flows through it?
Rosa: It allows them to assess both stability and incremental L2-gain performance using Proposition one with the extended SRG for the linear parts and the standard SRG for the nonlinear parts.
Dev: They also show this result yields a generalization of the celebrated circle criterion that we see in Lur'e systems, provided at least one of those specific SRG components obeys a chord property.
Taro: So this moves beyond just sector-bounded nonlinearities and lets you look at a broader class of nonlinearities for L2-gain bounds. That’s pretty broad for practical use.
Rosa: And they apply this to several examples, including the Lur'e system, the controlled Lur'e plant, and even an LTI plant with a Lur'e controller to get some concrete results.
Dev: For instance, for the Duffing oscillator stabilized by a PD controller using Theorem four for the Lur’e system they get an incremental L2-gain bound of one over r min.
Taro: That sounds like it puts a specific number on how much gain you can expect in that scenario when you're doing control. But what about the non-incremental bounds?
Rosa: They derive those too by swapping out the SRG for the nonlinear part with something called SG0, which gives them an L2-gain bound instead of an incremental one. This is useful if you are trying to find a Lyapunov function.
Dev: And they connect this to a homotopy theorem that gives a well-posedness result with gain bound gamma of T less than or equal to one over r min.
Taro: It sounds like they’re giving us tools for analysis even when stability isn't guaranteed by the classical criteria, which is important when you can't prove it easily.
Rosa: The big picture here is that the extended SRG resolves a major pitfall in state-of-the-art SRG methods by incorporating Nyquist criterion information, enabling treatment of general interconnections including unstable elements.
Dev: It provides a modular analysis framework for arbitrary networks of SISO LTI and nonlinear operators, which is really useful for handling many nonlinearities in large systems.
Title and authors: Taro: So it’s scalable enough to handle complex setups without getting bogged down in the usual pitfalls of these methods. What about the future work they mention?
Rosa: They note that for factors within the word that only contain LTI operators and terminals, you can tighten the bound by taking SRG prime instead of computing it separately for each LTI operator.
Dev: They introduce simplification rules like combining terms or multiplying terms using these extended SRG versions to make the calculations cleaner in practice. It's about efficiency when dealing with large networks.
Taro: So they are optimizing the calculation process itself, which is a good engineering step for making this usable outside of just theoretical proofs.
Rosa: They also show that the SRG analysis of the Lur’e system never ends up being more conservative than the circle criterion, which they prove by showing a contradiction when certain conditions hold.
Dev: That means you can use their method to get a bound that is at least as tight as what you could get from the simpler circle criterion for those systems.
Taro: That’s reassuring if it means we don't end up with unnecessarily conservative bounds in real system design scenarios. So where does this leave things for control theory?
Rosa: It gives a new way to analyze general interconnections, including unstable elements, and it's modular enough for large-scale systems with many nonlinearities. This framework lets you build complex analyses systematically.
Dev: It’s a solid tool for analyzing arbitrary networks of SISO LTI and nonlinear operators where the standard methods fail due to those domain issues we talked about earlier.
Taro: I just hope that as they apply this to more real-world scenarios, they can show how gracefully the system recovers when things misbehave in ways that aren't perfectly modeled by these graphs.
Rosa: That’s what we'll be watching for. It shows how this mathematical framework translates into actual system behavior when you take it outside the controlled environment of a pure simulation or lab test.
Dev: So, to wrap up on "On the Nyquist Criterion and the Scaled Relative Graph for SISO systems: A Generalized Circle Criterion," they've given us an extended SRG that handles unstable systems by adding Nyquist information.
Taro: It seems like a solid step toward making these advanced analysis tools more reliable for real-world, complex robotic or control challenges.
The paper's summary: Rosa: So, this paper takes the Scaled Relative Graph Analysis and fixes a big problem where it breaks down when you're dealing with unstable linear systems in a feedback loop.
Dev: It introduces something called the extended Scaled Relative Graph, which essentially blends the standard graph analysis with information from the Nyquist criterion to get a more consistent picture of stability.
Rosa: The authors show that this extension lets them correctly assess stability and L2-gain performance even when you’re looking at systems that are unstable in certain ways. It resolves that pitfall where the math gets inconsistent because the graph definition doesn't perfectly match what the Nyquist criterion predicts for unstable systems.
Dev: They build a formal way to represent any complicated network of linear and nonlinear operators using something like an Operator Context Free Grammar, which lets them calculate these bounds systematically.
Rosa: The main result is that you get a gain bound that relates directly to the minimum value of this calculated set C R, which covers both stability and how much energy flows through the system.
Dev: And they’ve generalized the circle criterion for Lur'e systems, meaning it works better for nonlinearities than before, even when those nonlinearities aren't perfectly bounded by a sector.
Rosa: They test this on real examples like the Duffing oscillator with a PD controller and show that their analysis is never more conservative than just using the old circle criterion.
Dev: They also give you non-incremental bounds, which are useful if you want to find a Lyapunov function, and they connect those to a well-posedness result for the entire system.
Rosa: So what this means for us in the field is that we now have a framework that can handle general interconnections, including unstable parts, without running into those domain contradictions we used to see with standard SRG analysis.
Dev: It’s modular enough to analyze arbitrary networks of linear and nonlinear operators, which opens up possibilities for analyzing really complex systems with many different kinds of nonlinearities.
Rosa: This isn't just theory; they showed how you can use these tools to get concrete L2-gain bounds for things like the controlled Lur'e plant.
Dev: And that leads us to thinking about how we can make this framework even faster for real-time control loops, which is where my job comes in next.
The paper's improvements: Taro: So, we've heard about how they fixed the fundamental problem with unstable linear systems in feedback loops by adding that extended SRG concept. Now, what are they suggesting to make this even better?
Rosa: They’re actually tightening up the math for efficiency when you’re dealing with networks of linear operators. For any part of the word that only has linear stuff and terminals, you don't have to run a separate calculation for each one anymore.
Dev: They introduce some simplification rules, like if you have two parts connected linearly, they can combine them into one extended SRG version instead of doing two separate computations. It’s about making the math cleaner when you're dealing with big systems.
Taro: That sounds like a practical thing for large networks. Does this mean we can actually calculate these bounds faster than just plugging everything in individually?
Rosa: Yes, it does. It lets you simplify those long operator words down into one more manageable calculation, which is crucial if you're dealing with hundreds of connections.
Dev: And they also showed how to get non-incremental bounds using this extended method instead of the standard SRG for the nonlinear parts. This gives you a different kind of safety check, one that’s useful for finding a Lyapunov function.
Taro: So, we can use one method to find an incremental gain bound and another version of it to find something related to stability proofs? How do those two relate in practice?
Rosa: They connect the non-incremental result back into a homotopy theorem, which gives you a well-posedness result with a specific gain bound. It links the two analysis types together in a way that should give you more confidence in the overall safety case.
Dev: The authors also point out that their analysis of Lur'e systems is never more conservative than just using the circle criterion, provided certain conditions are met regarding those SRG components. That’s a nice piece of information for system designers.
Taro: So it means we can use this SRG method to get bounds that are at least as tight as what we get from the simpler circle criterion for those systems where things behave nicely? That's a good thing for real-world design.
Rosa: Exactly. It’s about ensuring that when you use this more complex tool, you aren't just getting a looser estimate than what the simpler, well-known method provides.
Dev: So we have the core stability fix and now we have efficiency rules and different types of bounds for non-incremental analysis. The next thing to look at is how this framework handles things that are really messy, like those unstable elements we talked about before.
Conclusion: Rosa: So, to wrap up, this paper on "On the Nyquist Criterion and the Scaled Relative Graph for SISO systems: A Generalized Circle Criterion" basically shows how to fix a major issue with standard SRG analysis by adding Nyquist information in an extended version.
Dev: It gives us a solid tool for analyzing unstable linear systems and general interconnections, which is really important because it opens up the door to using these methods where they used to fail.
Taro: I think the real impact is that we can now build larger, more complex robotic systems with many nonlinearities and still get meaningful stability results instead of just hitting a wall.
Rosa: Right, because it’s modular, you can handle huge networks of different operators without having to do separate calculations for every single component.
Dev: We saw how this method provides both incremental gain bounds for control design and non-incremental bounds that help us find Lyapunov functions, which is a double win for safety cases.
Taro: It’s interesting that they proved it’s never more conservative than the circle criterion for Lur'e systems under certain conditions, which gives us confidence in using their method over older tools.
Rosa: Exactly, so we get to use this more powerful graph-based approach knowing it’s at least as reliable as the simpler circle test when things line up correctly.
Dev: It’s a solid foundation for designing controllers where we need to manage those non-linear dynamics in real time with low latency and a clear safety case.
Taro: I just think the next big thing will be seeing how this scales up when you add more complex, dynamic elements, maybe even incorporating those real-world failure modes we’re testing on other papers.
Rosa: Yeah, that’s where we need to watch it from the field side. It shows a way to systematically analyze these complicated networks, which is a big step for autonomous systems operating in the messy world.
Dev: It really proves that even with unstable parts or complex feedback loops, there’s a rigorous mathematical path forward for checking performance and stability.
Taro: So this paper sets up a new language for analyzing these kinds of systems, which should help us tackle those hard autonomy challenges coming down the road.
More episodes
- 2608.13438-ContactGuard: Pre-Contact Execution Monitoring with Action-Conditioned Latent World Models
- 2610.34593-Future Prediction as Training-Only Video Supervision for Pretrained VLA Policies
- 2610.35385-EmbodiedLab: Beyond Manipulation, Benchmarking Embodied Agents That Act to Know
- 2610.36378-Data-Driven Hamiltonian for Neural Optimal Control
- 2610.37275-A Heuristic for Scheduling Autonomous Vehicles at Intersections
- 2610.37288-Dynamic Coordination of Autonomous Vehicles at Intersections
- 2610.37632-Real-Time Coordination of Autonomous Vehicles at Intersections
- 2611.13779-Softness is something you also need! Why compliance helps control: information metrics of a passive walker with compliant ankle
- 2611.13411-Residual-Corrected Physics-Informed Neural Network KKL Observers with Input-to-State Stability Guarantees
- 2611.14488-Is Your Robot's Memory Just a Clock? Temporal Shortcuts in Memory-Augmented VLA Policies