Causality Sum Rules in Conventional Scattering Matrices

Summary

This paper, Causality Sum Rules in Conventional Scattering Matrices by Ning Han, Rui Zhao, Shuxing Yang, Mingzhu Li, Hongsheng Chen, and Yihao Yang, presents a theoretical framework that formulates causality sum rules directly in the conventional multichannel scattering matrix S(ω), which is the standard experimental and computational description of photonic and electromagnetic devices. The authors note that while passivity is explicit in the conventional incoming-outgoing matrix (S†(ω)S(ω) ≤ I on the real-frequency axis), causality is not directly visible and is usually formulated only after transforming the response into auxiliary variables such as Green functions, local conservation formulations, or volume T-operator approaches.

The core innovation is the introduction of a domain-delay correction that removes the geometry-induced temporal advance arising from the finite scattering domain and the choice of channel reference surfaces. This reference-dependent contribution appears as a phase factor e(-iωT a), where T a is the earliest-arrival delay operator defined as T a = diag(τ α) α, with τ α ≥ 0 denoting the earliest arrival time of output channel α. The domain-delay operator is D a(ω) = e(iωT a), and the domain-delayed scattering matrix is defined as S̃(ω) = D a(ω)S(ω). Because D a(ω) is unitary on the real-frequency axis, this transformation preserves the real-frequency power balance (S̃†(ω)S̃(ω) = S†(ω)S(ω)), so passivity is unchanged. The transformation only removes the reference-dependent temporal advance and restores the causal analytic structure.

The key mathematical result is that the corrected scattering matrix S̃(ω) belongs to the class of operator Schur functions: it is analytic in the upper half-plane and satisfies S̃(ω) op ≤ 1. This follows from the combination of causality (providing Hardy-class analytic continuation after the time-advance correction) and passivity (preserving the contractive property on the real-frequency axis). Applying the matrix Cayley transform, W(ω) = i[1 + S̃(ω)][1 − S̃(ω)]−1, maps the contractive scattering representation into an operator Herglotz function satisfying Im W(z) ≥ 0. The Herglotz representation then converts the analyticity and passivity constraints into a spectral integral relation whose low-frequency expansion is determined by the geometric delay operator T a.

From this construction, the authors derive two universal causality sum rules. The first is a projected sum rule for coherent channel superpositions within a common-delay subspace. For a unit incident superposition v that lies in an eigenspace of the delay operator (T a v = τ v v), the coherent return amplitude is s v(ω) = ⟨v, S(ω)v⟩. The sum rule states:

∫0∞ [−ln⟨v, S(ω)v⟩ / ω2] dω ≤ (π/2)⟨v, T a v⟩

This limits the bandwidth-integrated logarithmic suppression of return into the selected channel superposition v by its domain-delay time τ v. Deep suppression over a broad band requires a correspondingly large delay budget. For spherical waves, T a = (2a/c)I, so the result holds for every superposition in the truncated channel space.

The second is a determinant sum rule providing a basis-independent constraint on the total multichannel response. For a finite-dimensional N-channel system, the determinant magnitude satisfies det S = ∏σ j(S), where σ j(S) are the singular values. The sum rule states:

∫0∞ [−lndet S(ω) / ω2] dω ≤ (π/2) tr T a

This determinant bound constrains the aggregate logarithmic attenuation of the multichannel scattering operator, capturing the collective contraction of all singular channels including interchannel coupling and mode conversion. It is invariant under unitary transformations of the input and output channel bases.

The framework recovers established scalar limits. For a single-channel passive system with S(ω) = Γ(ω), a planar absorber of thickness d backed by a perfect conductor has earliest round-trip arrival time τ v = 2d/c, giving ∫0∞ [−lnΓ(ω)/ω2] dω ≤ πd/c, which directly recovers the Rozanov absorber bound. In the narrowband approximation, this becomes ln R0 B/ω0 ≤ πd/λ0. For spherical-wave scattering of a passive scatterer enclosed by a sphere of radius a, choosing a single spherical multipole basis vector recovers the Bernland-Gustafsson spherical-multipole scattering bound, but the present result extends this constraint from individual multipole channels to arbitrary coherent superpositions within a common-delay eigenspace.

The paper derives three directly measurable physical consequences. Consequence 1 is coherent-return suppression: for spherical-wave scattering with T a = (2a/c)I, the suppression of any coherent multipole superposition is limited by physical size: ∫0∞ [−ln⟨v, S(ω)v⟩/ω2] dω ≤ πa/c. For a coherent return satisfying s v(ω) ≤ ρ v over a fractional bandwidth β = B/ω0, the suppression depth D v = 20 log10(1/ρ v) satisfies D v ≤ (20π/ln10)k0a/β. For example, a 60-dB return suppression over a 10% fractional bandwidth requires k0a ≈ 0.22.

Consequence 2 is aggregate and geometric-mean attenuation. For a spherical-wave basis truncated at multipole order n = N max, the finite truncation contains N = 2N max(N max + 2) open scattering channels. With T a = (2a/c)I, the determinant sum rule gives ∫0∞ [−lndet S(ω)/ω2] dω ≤ Nπa/c. In finite-band form, the aggregate determinant suppression depth D det = 20 log10(1/Δ0) satisfies D det B/ω0 ≤ 27.3 Nk0a. After normalization by channel number, the geometric-mean singular-value depth D g = −20 log10 σ g satisfies D g B/ω0 ≤ (20π/ln10)k0a, which is independent of N. For N = 70, a determinant depth of 80 dB corresponds to a geometric-mean depth of approximately 1.1 dB.

Consequence 3 is suppressed-channel count. For a suppression threshold 0 < ρ σ < 1, if at least m singular values satisfy σ j(S) ≤ ρ σ throughout an operating band, then m ≤ πNk0a / (ln ρ0 B/ω0), together with m ≤ N. When the scattering matrix contains all power-carrying exterior output channels, singular values relate to absorption eigenchannels through A j = 1 − σ2 j. Requiring at least m eigenchannels to satisfy A j ≥ A0 throughout the operating band gives m ≤ 2πNk0a / (ln(1 − A0) B/ω0). For a channel-complete N = 30 model, maintaining A j ≥ 0.99 in all 30 eigenchannels over a 10% fractional bandwidth requires k0a ≥ 7.3 × 10−2.

The paper also presents a conditional phase-delay extension for lossless multiport systems. For lossless systems, det S = 1, so the logarithmic attenuation bounds become trivial. The relevant causal quantity is the accumulated scattering phase. For a lossless multiport system with det S(ω) = e(iΘ(ω)), the Wigner-Smith time-delay operator is Q(ω) = −iS†∂ ω S, giving the phase-delay identity Θ′(ω) = tr Q(ω). Under a modal-count hypothesis H B (assuming the delay-corrected determinant contains no singular inner factor and its Blaschke phase accumulation satisfies the spectral count), the band-averaged delay trace satisfies tr Q B ≤ 4π2N(l/λ0), where l is the effective propagation length and λ0 = 2πc/ω0. In terms of fractional bandwidth β, tr Qω0 ≤ 4π2N(l/λ0)/β. For a target mean normalized delay of qω0 = 106 over a 1% fractional bandwidth, the required delay-bandwidth product is qB = 104, requiring l/λ0 ≈ 2.5 × 102. The paper emphasizes this is a conditional estimate relying on the modal-count hypothesis, and sharp resonances can introduce additional Blaschke phase accumulation beyond the assumed spectral count, so it should not be applied to arbitrary high-Q resonant or slow-light systems.

The paper also describes a hybrid AI-human discovery workflow. Human researchers defined the open-ended scientific question without prescribing a solution path. Qiushi Engine, an autonomous research system designed for sustained, thousand-step scientific reasoning, conducted the exploration, developed derivations and numerical checks, and generated a traceable record of data, code, figures, research notes, and reports. The system identified the core theoretical route and produced the initial result, which the authors subsequently verified, refined into a rigorous theory, and physically interpreted, retaining full responsibility for scientific validation and oversight.

Based on this paper, I can improve AI systems in the following specific ways:

1. Causal-Aware Physical Modeling

• Improvement: Train AI models to enforce causality sum rules as hard constraints during optimization of scattering/absorption devices. The AI can use the projected sum rule (∫0∞ [−ln⟨v,S(ω)v⟩/ω2]dω ≤ πa/c) to reject designs that violate physical limits before simulation.

• Capability: AI can now design broadband absorbers, filters, or metasurfaces with guaranteed physical feasibility, avoiding trial-and-error iterations that produce non-causal or unphysical solutions.

2. Bandwidth-Delay Budget Optimization

• Improvement: Incorporate the delay operator T a and the determinant sum rule (∫0∞ [−lndet S/ω2]dω ≤ π tr T a) into reinforcement learning reward functions. The AI can explicitly trade off suppression depth, bandwidth, and physical size (k0a) as a multi-objective optimization.

• Capability: AI can automatically generate device geometries that achieve a specified dB suppression over a given fractional bandwidth with minimal size, using the sum rule as an analytical upper bound to prune the search space.

3. Channel-Aware Inverse Design

• Improvement: Use the suppressed-channel count bound (m ≤ πNk0a / (ln ρ0 B/ω0)) to guide AI in selecting the minimum number of scattering channels needed for a target absorption eigenchannel performance. The AI can dynamically truncate multipole orders (N max) based on this constraint.

• Capability: AI can design channel-complete systems (e.g., N=30) that maintain high absorption (A j ≥ 0.99) across a specified band, knowing exactly the required electrical size (k0a) a priori, reducing simulation cost.

4. Phase-Delay-Aware Resonator Design

• Improvement: For lossless multiport systems, the AI can use the conditional phase-delay bound (tr Qω0 ≤ 4π2N(l/λ0)/β) to avoid proposing ultra-high-Q resonators that violate the modal-count hypothesis. The AI can flag designs where Blaschke phase accumulation would exceed the spectral count.

• Capability: AI can design delay lines, slow-light structures, or phase shifters with realistic Q-factor limits, preventing over-optimistic predictions that fail in practice.

5. Hybrid AI-Human Verification Loop

• Improvement: Implement a two-stage pipeline: (a) AI generates candidate physical laws or sum rules from data (as done here), then (b) a human-verification module checks analyticity, passivity, and causality constraints using the domain-delay correction framework. The AI can automatically test if a proposed bound holds under unitary transformations.

• Capability: AI can discover new universal bounds in electromagnetics or photonics, with built-in mathematical rigor checks (e.g., verifying Schur function properties, Herglotz representation validity), reducing false positives in automated theory discovery.

6. Numerical Validation with Sum-Rule Consistency

• Improvement: Add a post-processing layer to AI simulation outputs that computes the left-hand side of the sum rules (e.g., ∫0∞ [−lndet S/ω2]dω) and compares to the right-hand side. If violated, the AI flags the simulation as non-causal or numerically inaccurate.

• Capability: AI can self-audit its own electromagnetic simulations, detecting mesh errors, truncation artifacts, or non-physical boundary conditions that would otherwise corrupt design optimization.

7. Basis-Invariant Feature Extraction

• Improvement: Use the determinant sum rule (basis-independent) as a loss function term for autoencoders or generative models that compress scattering data. The AI can learn latent representations that preserve the causal constraint across arbitrary channel bases.

• Capability: AI can generate compact, physically consistent models of complex multiport devices, useful for surrogate modeling in larger system-level optimizations.

8. Early-Arrival Delay Estimation

• Improvement: Train a neural network to predict the earliest-arrival delay operator T a from geometry (e.g., bounding sphere radius, channel reference surfaces) using the paper's definition τ α ≥ 0. This can be a pretrained module for any AI design tool.

• Capability: AI can instantly compute the delay budget for arbitrary 3D geometries, enabling real-time feasibility checks during interactive design.

9. Narrowband-to-Broadband Extrapolation

• Improvement: Use the sum rules to train AI to extrapolate narrowband scattering measurements to broadband behavior. The AI can enforce that the extrapolated S(ω) satisfies the integral bounds, preventing overfitting to limited data.

• Capability: AI can predict full-spectrum performance from sparse frequency samples, with rigorous uncertainty bounds derived from the sum-rule inequalities.

10. Multi-Channel Coherent Control

• Improvement: For AI-driven wavefront shaping or coherent perfect absorption, use the projected sum rule to constrain optimization of the input superposition v. The AI can maximize suppression of a chosen channel while respecting the delay limit τ v.

• Capability: AI can design coherent control protocols that achieve deep suppression (e.g., 60 dB) over a specified bandwidth, with the required electrical size (k0a ≈ 0.22 for 10% bandwidth) known analytically, enabling rapid system-level design.

Scattering matrices are the standard experimental and computational description of photonic and electromagnetic devices. Passivity is explicit in the conventional incoming-outgoing matrix, whereas causality sum rules are usually formulated only after transforming the response into auxiliary variables. Here we show that these rules can be written directly in the conventional scattering matrix by removing the time advance introduced by the reference domain. Using the earliest-arrival delay of each channel, we define a domain-delayed matrix that preserves real-frequency passivity while restoring the causal time origin. Under explicit analyticity, transparency, and regularity assumptions, this matrix becomes a Schur function, enabling a Cayley-Herglotz construction. The resulting projected and determinant bounds constrain coherent channel superpositions and aggregate multichannel loss. The framework recovers Rozanov's absorber limit and spherical-multipole sum rules, while extending causality bounds to measurable quantities including insertion loss, suppressed singular-value channels, and conditional lossless delay-bandwidth trade-offs. Our work directly connects fundamental causality theory with experimentally accessible scattering data. The initial theoretical route is autonomously explored by Qiushi Engine, an AI research system for open-ended scientific discovery, and subsequently verified, refined, and developed by the authors, demonstrating a hybrid AI-human discovery workflow.