**1. Precise restatement** Two coordinates x, y with unit masses (M = I), potential U₀ = ½(x² + a²y²), 0 < a < 1, in thermal equilibrium at temperature T. Hidden assumptions: (i) "normal-mode frequencies" are √eigenvalues of K since M = I; (ii) "equilibrium displacement noise" means the total (frequency-integrated) equilibrium variance ⟨x²⟩ = k_BT(K⁻¹)_xx, not a spectral density at some band; (iii) "spatially uniform static force" means the same generalized force f applied to both coordinates, DC response; (iv) the added element is an ideal, massless, lossless lever coupling realizing the potential ½(x − ay)²; (v) "without bound" means as a → 0 within the family, not for a fixed system. Claim: after adding this spring, both eigenvalues of K strictly increase, ⟨x²⟩ drops from k_BT to exactly (2/3)k_BT, and the static x-displacement under uniform f grows from f to f(2+1/a)/3, which → ∞ as a → 0. **2. Step check and re-derivation** - Original: K₀ = diag(1, a²), K₀⁻¹ = diag(1, 1/a²). Uniform force (f,f): x̄ = f, ȳ = f/a². ⟨x²⟩ = k_BT·1. ✓ - Added stiffness: ½(x−ay)² ⇒ ΔK = [[1,−a],[−a,a²]] = vvᵀ with v = (1,−a), rank-1 PSD. ✓ - K = [[2,−a],[−a,2a²]], det = 4a² − a² = 3a². K⁻¹ = (1/3a²)[[2a², a],[a, 2]] = [[2/3, 1/(3a)],[1/(3a), 2/(3a²)]]. ✓ - ⟨x²⟩ = k_BT·(2/3). ✓ (exactly one third reduction). - x̄ = f(2/3 + 1/(3a)) = f(2+1/a)/3 → ∞ as a→0. ✓ - Eigenvalues: tr = 2+2a², det = 3a², λ± = (1+a²) ± √(1−a²+a⁴). λ₊ > 1 trivially. λ₋ > a² ⇔ 1 > √(1−a²+a⁴) ⇔ a² > a⁴ ⇔ a < 1. ✓ Also follows from interlacing for a rank-1 PSD update, with strictness because v is not orthogonal to either original eigenvector (1,0), (0,1). ✓ - FDT remark: (K⁻¹)_xx is the response to a force on x alone, and that is what sets ⟨x²⟩; the uniform-force response involves the off-diagonal (K⁻¹)_xy = 1/(3a). Consistent. ✓ Numerical check, a = 0.1: K = [[2,−0.1],[−0.1,0.02]]; K⁻¹ = [[0.667, 3.33],[3.33, 66.7]]; x̄ = 4f (vs f), ⟨x²⟩ = 0.667 k_BT; eigenvalues ≈ 0.01497 (>0.01) and 2.005 (>1). All consistent. **3. Attempts to break it** - The math is airtight for the stated model. The only places to push are the hidden assumptions. - "Without bound": for every fixed a the amplification is finite; divergence requires a → 0, i.e. an original system with an arbitrarily soft y mode (compliance 1/a²). Moreover the new soft eigenvalue is λ₋ ≈ 3a²/2 — raised by 50%, but still arbitrarily soft. So while no mode is *softened*, the amplification is entirely due to a mode that *remains* soft and is now coupled into x by the lever. The text's "neither a softened mode ... explains the amplification" is literally true but somewhat misleading. - Signal-referred noise: originally reading y directly gives x̄/σ = (f/a²)/(√k_BT/a) = f/(a√k_BT); the new x gives ≈ (f/3a)/√(2k_BT/3) = f/(a√(6k_BT)), i.e. worse. So the "free lunch" flavor evaporates once one compares to the best original readout — not claimed, but relevant to the sensor framing. - Spectral noise within a bandwidth, or an unequal-mass/lossy lever, would alter the "exactly one third" and could alter the frequency conclusions; these restrictions are implicit. - No counterexample to the stated proposition exists; it is a direct linear-algebra fact. **4. Novelty** This is a special case of standard results: Rayleigh's theorem / Cauchy interlacing (adding a PSD constraint or spring never lowers any natural frequency), the equipartition result ⟨x²⟩ = k_BT(K⁻¹)_xx, and ordinary lever amplification of a compliant degree of freedom (as in mechanical lever anti-springs or lever-coupled suspensions). The specific numbers are a neat pedagogical construction, not a new physical effect. **5. Fairness** The one-sentence claim is correct for the construction but omits that: "exactly one third" is specific to this choice of spring and masses; "without bound" holds only in the limit of an arbitrarily soft pre-existing mode, which persists (merely stiffened by 3/2); "sensor noise" means integrated equilibrium variance of one coordinate under force on that coordinate alone. Read as a general statement about adding springs, it overstates; read as describing this example, it is accurate. {"verdict": "correct with caveats", "novelty": "special case or variant of known results", "fair_summary": "partly", "confidence": 5, "one_line": "All computations (K⁻¹, 2/3 variance, f(2+1/a)/3 response, both eigenvalues strictly rising) check out, but the unbounded amplification relies on a soft original mode that stays soft and is merely lever-coupled into x, so the effect is a pedagogical instance of interlacing plus lever amplification rather than a new or general sensor principle."}