2025-11-24T10:10:18.436123

Computing the Hard Scaled Relative Graph of LTI Systems

Krebbekx, Baron-Prada, Tóth et al.
Scaled Relative Graphs (SRGs) provide a novel graphical frequency-domain method for the analysis of nonlinear systems, where Linear Time-Invariant (LTI) systems are the fundamental building block. To analyze feedback loops with unstable LTI components, the hard SRG is required, since it aptly captures the input/output behavior on the extended $L_2$ space. In this paper, we develop a systematic computational method to exactly compute the hard SRG of LTI systems, which may be unstable and contain integrators. We also study its connection to the Nyquist criterion, including the multivariable case, and demonstrate our method on several examples.
academic

Computing the Hard Scaled Relative Graph of LTI Systems

Basic Information

  • Paper ID: 2511.17297
  • Title: Computing the Hard Scaled Relative Graph of LTI Systems
  • Authors: Julius P. J. Krebbekx, Eder Baron-Prada, Roland Tóth, Amritam Das
  • Institutions: Eindhoven University of Technology, Austrian Institute of Technology, ETH Zurich, HUN-REN Institute for Computer Science and Control
  • Classification: eess.SY (Systems and Control), cs.SY, math.OC
  • Submission Date: November 21, 2025
  • Paper Link: https://arxiv.org/abs/2511.17297

Abstract

Scaled Relative Graphs (SRGs) provide a novel graphical frequency-domain method for nonlinear system analysis, where linear time-invariant (LTI) systems serve as fundamental building blocks. To analyze feedback loops containing unstable LTI components, the hard SRG is required, as it properly captures input/output behavior on the extended L₂ space. This paper develops a systematic computational method to precisely compute the hard SRG of LTI systems, which may be unstable and contain integrators. The article also investigates its connection to the Nyquist criterion, including the multivariable case, and demonstrates the method through multiple examples.

Research Background and Motivation

Research Problem

This paper addresses a core problem in nonlinear system analysis: how to systematically compute the Scaled Relative Graph (SRG) of feedback systems containing unstable LTI components.

Problem Significance

  1. Limitations of Classical Methods: The Nyquist plot is a cornerstone of LTI system analysis but is difficult to systematically generalize to nonlinear system analysis
  2. Practical Needs: Industrial applications frequently require analyzing feedback systems with unstable components
  3. Theoretical Completeness: The existing SRG framework lacks a systematic method for handling unstable systems

Limitations of Existing Methods

  1. Soft SRG: Originally proposed in 3, applicable only to stable SISO systems, problematic for unstable systems
  2. Extended SRG: Addressed SISO unstable systems in 11 by adding Nyquist criterion information, but limited to single-variable cases
  3. Hard SRG: Defined in 12, theoretically capable of handling unstable systems, but lacking systematic computational methods

Research Motivation

Since the primary advantage of hard SRG lies in handling unstable systems in feedback loops, developing methods to compute the hard SRG of unstable LTI systems will significantly advance SRG analysis of nonlinear systems.

Core Contributions

  1. Systematic Computational Method: Proposes a non-approximate algorithm (Theorem 2) that precisely computes the hard SRG of LTI systems, including unstable systems and systems with integrators
  2. Theoretical Equivalence Proof: Proves that for SISO systems, hard SRG is completely equivalent to extended SRG (Theorem 3)
  3. MIMO Extension: Extends the method to multivariable (MIMO) cases, providing a complete computational framework for square systems
  4. Connection to Nyquist Criterion: Demonstrates how hard SRG serves as an alternative to MIMO Nyquist criterion and Generalized Nyquist Criterion (GNC)
  5. Practical Algorithm: Based on transfer function representation, constructs boundaries through intersection and removal of disks centered on the real axis in the complex plane

Method Details

Problem Definition

Given an LTI system G: L²ₑᵖ → L²ₑᵍ with transfer function G(s) ∈ Rᵍˣᵖ(s), compute its hard SRG:

SRGe(G)=⋃T>0SRG(G∣T)\text{SRG}_e(G) = \bigcup_{T>0} \text{SRG}(G|_T)

where G|_T is the truncated operator of G on the time interval 0,T.

Core Theoretical Framework

1. Hyperbolic Convexity Representation

Any h-convex set C has a closure that can be represented as:

cl C=⋂α∈R(DRα(α)∖Drα(α))=:G({rα},{Rα})\text{cl}\,C = \bigcap_{\alpha \in \mathbb{R}} \left(D_{R_\alpha}(\alpha) \setminus D_{r_\alpha}(\alpha)\right) =: \mathcal{G}(\{r_\alpha\}, \{R_\alpha\})

where Dr(x)D_r(x) denotes a closed disk centered at x with radius r.

2. Shifted System Definition

For α ∈ ℝ, define the shifted system:

Gα=(G0(n−q)×p)−(αI0(n−p)×p)G_\alpha = \begin{pmatrix} G \\ 0_{(n-q)\times p} \end{pmatrix} - \begin{pmatrix} \alpha I \\ 0_{(n-p)\times p} \end{pmatrix}

where n = max{p,q}.

Algorithm Core Components

Proposition 2: Computation of Maximum Gain Radius Rα

\bar{\sigma}(G_\alpha) & \text{if } G \in RH_\infty^{q\times p} \\ \infty & \text{otherwise} \end{cases}$$ **Key Insight**: - For stable systems, use the H∞ norm (maximum singular value) - For unstable or non-proper systems, Rα = ∞ #### Proposition 3: Computation of Minimum Gain Radius rα $$r_\alpha = \begin{cases} \underline{\sigma}(G_\alpha) & \text{if } G_\alpha \text{ is minimum-phase} \\ 0 & \text{otherwise} \end{cases}$$ **Key Insight**: - Compute through the inverse system $G_\alpha^{-1}$ and apply Proposition 2 - rα is nonzero only when Gα is minimum-phase (all transmission zeros in the open left half-plane) - Utilize Smith-McMillan form: poles of Gα are zeros of $G_\alpha^{-1}$, and vice versa ### Main Algorithm (Theorem 2) **For square LTI systems** G: L²ₑᵖ → L²ₑᵖ with transfer function G(s) ∈ Rᵖˣᵖ(s): $$\text{SRG}_e(G) = \mathcal{G}(\{r_\alpha\}, \{R_\alpha\})$$ **Algorithm Steps**: 1. For each α ∈ ℝ, construct Gα = G - αI 2. Use Proposition 2 to compute Rα 3. Use Proposition 3 to compute rα 4. Compute the intersection of all annular regions ### Technical Innovations #### 1. Extended L₂ Space Treatment - Traditional methods work on L₂ space, unable to handle divergent signals - Hard SRG is defined on extended space L²ₑ, which includes periodic and divergent signals - Handles instability by considering truncated signals for all T > 0 #### 2. Minimum-Phase Detection - Key innovation: Reduces computation of rα to detecting whether Gα is minimum-phase - Minimum-phase ⟺ all transmission zeros in the open left half-plane - Non-minimum-phase systems have rα = 0, reflecting potential issues in feedback #### 3. Unification with Nyquist Criterion For SISO systems, proves: $$\text{SRG}_e(G) = \text{SRG}'(G)$$ where SRG'(G) is the extended SRG, incorporating soft SRG and Nyquist criterion information. #### 4. Handling Special Cases - **Integrators**: Automatically result in Rα = ∞ - **Non-proper systems**: Rank deficiency detected via Smith-McMillan form - **Time delays**: Result in rα = 0, Rα unaffected ## Experimental Setup ### Example Systems #### SISO Systems 1. **G₁(s) = 1/(s² + s + 1)**: Stable second-order system 2. **G₂(s) = 1/(s(s+1))**: System with integrator 3. **G₃(s) = 3/((s-2)(s/10+1))**: Unstable system (right half-plane pole) #### MIMO Systems 1. **G₄(s)**: 2×2 stable system $$G_4(s) = \begin{pmatrix} \frac{1}{s+1} & \frac{1}{s+2} \\ \frac{1}{s+4} & \frac{1}{s+3} \end{pmatrix}$$ 2. **G₅(s)**: 2×2 unstable system $$G_5(s) = \begin{pmatrix} \frac{s+7}{s-1} & \frac{s-5}{(s+2)^2} \\ \frac{1}{(s+4)^3} & \frac{s}{(s+3)^2} \end{pmatrix}$$ 3. **G₆(s)**: 3×3 system with integrator ### Visualization Method - Plot hard SRG boundaries on the complex plane (gray region) - Overlay soft SRG boundaries (dark gray region) - Mark critical points such as -1 point (for stability analysis) ## Experimental Results ### Main Results #### SISO System Results 1. **G₁**: Hard SRG coincides with soft SRG (system is stable), forming a bounded region 2. **G₂**: Due to integrator, hard SRG extends to infinity, including semi-real axis [0,∞) 3. **G₃**: Hard SRG of unstable system includes additional regions, reflecting Nyquist encirclement information **Verification of Extended SRG Equivalence**: - For G₂, $\mathcal{N}_{G_2} \cap \mathbb{R} = [0,\infty)$ - This perfectly matches the hard SRG in Figure 2b - Verifies correctness of Theorem 3 #### MIMO System Results **G₄ (Stable 2×2 System)**: - Hard SRG boundaries closely match soft SRG boundaries - Both are separated from the -1 point, indicating feedback stability **G₅ (Unstable 2×2 System)**: - Hard SRG displays **two holes** (Figure 2e) - These holes correspond to regions of stable closed-loop configurations - The smaller hole can be used for gain design **G₆ (3×3 System with Integrator)**: - Hard SRG is partially unbounded (due to integrator) - Soft SRG boundaries tend toward ±j∞ in the imaginary direction ### Stability Analysis Application #### Gain Design Example (Based on G₅) The paper demonstrates three gain selection strategies: 1. **Pure Scaling**: k₁ = 0, k₂ = 1/5 - Interval [-6, 0) ⊈ SRGₑ(G₅) - After scaling, -1 point separates from SRG 2. **Flipping and Scaling**: k₁ = 0, k₂ = -2 - Flips the right small hole onto the -1 point - Exploits geometric properties of hard SRG 3. **Translation**: k₁ = -1.5, k₂ = 1 - Shifts the small hole leftward to contain the -1 point - Ensures closed-loop stability Each case satisfies the stability condition: $$\text{dist}(-1, \text{SRG}_e(k_1I + k_2G_5)) \geq r > 0$$ ### Key Findings 1. **Exactness**: For SISO and normal matrix MIMO systems, boundaries are exact (non-approximate) 2. **Numerical Stability**: Finite α sets result in numerical boundaries that always outer-approximate, never underestimate the SRG 3. **Geometric Intuitiveness**: Hard SRG provides intuitive geometric interpretation, visualizing stability margins 4. **Modularity**: Once computed for G, the hard SRG can be reused for different gain designs without recomputation ## Related Work ### SRG Framework Development 1. **Original SRG** [2,3]: Proposed by Ryu et al. and Chaffey et al., based on soft SRG on L₂ space 2. **Extended SRG** [4,11]: Extended to unstable SISO systems by adding Nyquist criterion information 3. **Hard SRG** [12]: Proposed by Chen, defined on L₂[0,T], theoretically capable of handling unstable systems ### SRG Application Domains 1. **Nonlinear Analysis**: Generalized circle criterion [4], small-gain theorem [3] 2. **Reset Control**: Analysis [8] and design [9] 3. **Decentralized Systems**: Stability conditions [10] 4. **Frequency Domain Properties**: Nonlinear Bode plots and bandwidth definitions [5] 5. **Phase Analysis**: Phase lead/lag concepts [6] ### Relationship to Classical Methods 1. **Nyquist Criterion**: Hard SRG is equivalent to extended SRG for SISO cases, which incorporates Nyquist information 2. **MIMO Nyquist Criterion** [13]: Hard SRG provides an alternative method without computing winding numbers 3. **Generalized Nyquist Criterion (GNC)** [14]: Hard SRG avoids manual splicing of characteristic loci ### Advantages of This Work 1. **Systematization**: Provides explicit algorithms rather than heuristic methods 2. **Generality**: Handles stable, unstable, and integrator-containing systems 3. **Exactness**: Non-approximate method providing exact boundaries 4. **Scalability**: Naturally extends to MIMO cases ## Conclusions and Discussion ### Main Conclusions 1. **Algorithm Completeness**: Develops the first systematic method to precisely compute hard SRG of LTI systems, including unstable systems 2. **Theoretical Unification**: Proves equivalence between hard SRG and extended SRG for SISO cases, unifying the two frameworks 3. **Practical Value**: Hard SRG provides a practical alternative to MIMO Nyquist criterion and GNC with the following advantages: - No need to compute winding numbers - SRG can be reused when gains change - Directly provides L₂ gain bounds and stability margins 4. **Geometric Intuitiveness**: Geometric construction based on disk intersection provides clear visualization and intuitive understanding ### Limitations #### 1. Model Dependency - Current method based on transfer function representation, requires parametric system model - Nyquist criterion's advantage is direct use of frequency response measurement data - **Future Need**: Compute rα and Rα directly from input/output data #### 2. Non-Minimum-Phase Detection Challenge - Proposition 2 only requires checking for unstable poles (no counting needed) - Proposition 3 requires detecting non-minimum-phase zeros - **Known Difficulty**: Detecting non-minimum-phase zeros from data is challenging [22] #### 3. Non-Square Systems - Complete theory currently applies only to square systems (p = q) - Wide case (p > q): Gα always has nontrivial algebraic kernel, resulting in rα = 0 - Tall case (p < q): Can use left pseudoinverse, but requires substantial additional technical details #### 4. Well-Posedness Assumption - Hard SRG stability analysis requires assuming well-posedness of feedback system - Soft and extended SRG guarantee well-posedness in addition to gain bounds - This is a disadvantage of hard SRG relative to extended SRG ### Future Directions #### 1. Data-Driven Methods Compute hard SRG directly from frequency response or input/output data, particularly: - Detect unstable poles from data - Identify non-minimum-phase zeros from data #### 2. Non-Square System Extension - Complete theory for tall systems (p < q) - Develop alternative methods for wide systems #### 3. Nonlinear System Applications - Apply results to nonlinear MIMO system analysis [7] - Combine with hard SRG nonlinear feedback analysis framework in [12] #### 4. Computational Optimization - Develop efficient numerical algorithms - Address computational complexity for high-dimensional MIMO systems - Adaptively select α set to balance accuracy and computational cost #### 5. Robustness Analysis - Extend to uncertain systems - Incorporate robust control theory ## In-Depth Evaluation ### Strengths #### 1. Theoretical Rigor - **Complete Mathematical Proofs**: All main results (Propositions 1-3, Theorems 2-3) have rigorous proofs - **Geometric Foundation**: Solid theoretical foundation based on hyperbolic convexity - **Equivalence Proof**: Theorem 3 establishes clear connection to classical methods #### 2. Method Innovation - **Unified Framework**: First to provide unified computational method for stable and unstable LTI systems - **Clever Inverse System Technique**: Computing rα through inverse Gα is an elegant solution - **Geometric Construction**: Disk intersection method is both intuitive and easy to implement #### 3. Practical Value - **Algorithm Implementability**: Method based on standard linear systems tools (singular values, minimum-phase detection) - **Modular Design**: Computed SRG reusable for different gain designs - **Intuitive Visualization**: Geometric representation in complex plane facilitates understanding and application #### 4. Sufficient Examples - Covers SISO and MIMO, stable and unstable, integrator-containing cases - Demonstrates practical gain design application - Clear and effective visualization ### Weaknesses #### 1. Theoretical Limitations - **Square Restriction**: Main results apply only to square systems, limiting application scope - **Well-Posedness**: Cannot guarantee well-posedness like extended SRG - **Normal Matrix Assumption**: For general MIMO systems, boundaries may not be exact #### 2. Practical Challenges - **Model Dependency**: Requires precise transfer function model - **Non-Minimum-Phase Detection**: Challenging from data - **Computational Complexity**: For high-dimensional MIMO systems, requires computation over continuous α #### 3. Experimental Insufficiency - **Missing Computational Time Analysis**: Algorithm efficiency not reported - **Numerical Precision Discussion**: Numerical error from finite α set not quantified - **Lack of Practical GNC Comparison**: Only theoretical comparison, missing practical case studies #### 4. Presentation Issues - **Symbol-Dense**: Extensive mathematical notation may impact readability - **Missing Algorithm Pseudocode**: While method is clear, lacks directly implementable algorithm description - **Figure Explanation**: Some figures (e.g., numerical artifacts in Figure 2b) need more explanation ### Impact Assessment #### Contribution to the Field 1. **Theoretical Completeness**: Fills gap in SRG framework for handling unstable systems 2. **Methodological Progress**: Provides new tools for nonlinear system analysis 3. **Classical Theory Modernization**: Provides modern geometric interpretation of Nyquist criterion #### Practical Value 1. **Control Design**: Directly applicable to feedback system design with unstable components 2. **Stability Analysis**: Provides more intuitive stability margin visualization than traditional methods 3. **Educational Value**: Geometric method aids understanding of feedback system behavior #### Reproducibility - **Theoretical Reproducibility**: All proofs completely given in appendix - **Implementation Feasibility**: Based on standard tools, easy to implement - **Example Clarity**: Provides specific transfer functions for verification - **Missing Code**: No open-source implementation provided (a drawback) ### Applicable Scenarios #### Most Suitable Applications 1. **Complex Feedback System Design**: Particularly systems with unstable components 2. **Robustness Analysis**: When visualizing stability margins is important 3. **Multivariable Systems**: MIMO system gain adjustment and stability analysis 4. **Nonlinear Systems**: As foundation for nonlinear SRG analysis #### Less Suitable Scenarios 1. **Pure Data-Driven**: When precise models unavailable 2. **Non-Square Systems**: When p ≠ q, method incomplete 3. **Real-Time Applications**: When computational complexity is critical constraint 4. **Very High-Dimensional Systems**: Computation may become difficult ### Overall Assessment This is a **high-quality theoretical control paper** making important contributions to the Scaled Relative Graph framework. Main strengths: - Provides first systematic method for handling unstable LTI systems' hard SRG - Rigorous theory with complete proofs - Establishes clear connection to classical Nyquist criterion - Lays foundation for nonlinear system analysis Main improvement areas: - Extend to non-square systems - Develop data-driven versions - Provide open-source implementation and more practical case studies **Recommendation Score**: ⭐⭐⭐⭐☆ (4/5) This work is of significant reference value for researchers in nonlinear control, robust control, and system analysis, particularly those interested in SRG methods or requiring unstable system handling. ## Selected References 1. **[2] Ryu et al., 2022**: "Scaled relative graphs: Nonexpansive operators via 2D Euclidean geometry" - Original mathematical definition of SRG 2. **[3] Chaffey et al., 2023**: "Graphical Nonlinear System Analysis" - Introduction of SRG to control theory 3. **[11] Krebbekx et al., 2025**: "Scaled Relative Graph Analysis of General Interconnections" - Definition of extended SRG 4. **[12] Chen, 2025**: "Soft and Hard Scaled Relative Graphs" - First definition of hard SRG 5. **[13] MacFarlane & Postlethwaite, 1977**: Classical literature on MIMO Nyquist criterion 6. **[14] Desoer & Wang, 1980**: Generalized Nyquist Criterion (GNC)