# Review: the "24% more earthquakes after slower loading" claim for two coupled threshold faults **Bottom line:** The arithmetic and the long-run rate equations are correct, and the result holds rigorously within the stated model. The summary sentence is fair only in part, because it hides that the slowdown is highly non-proportional. The underlying mechanism is standard inhibitory-coupling linear algebra. ## 1. Precise restatement The claim is about a deterministic model with two threshold units and normalized stresses σ_A and σ_B. - **Loading:** stresses rise at constant rates b_A and b_B. - **Failure and self-drop:** unit i fails when σ_i = 1, and σ_i then drops by 1 (reset to 0). - **Coupling:** - A's failure lowers σ_B by 2. - B's failure lowers σ_A by 0.1. - The coupling is purely inhibitory (stress shadows), so there are no triggered cascades. - **"Earthquakes"** means event counts per unit time, not moment or energy. The claim is existential: there exist (b_A, b_B) → (b_A′, b_B′), with b_A′ < b_A and b_B′ < b_B, such that the long-run total event rate rises by about 24%. Hidden assumptions: - The two loading rates can be set independently. They are not tied to a common plate rate through a backslip-consistent stiffness relation. - The slowdown is not proportional: b_A falls by 50%, b_B by only about 3.3%. - Units are normalized, which is harmless because the failure stresses are held fixed. ## 2. Checking the argument and re-deriving it **Balance equations.** Over a time T, the stress change on A is σ_A(T) − σ_A(0) = b_A·T − N_A − 0.1·N_B. If stresses stay bounded, dividing by T and letting T grow gives r_A + 0.1·r_B = b_A. The same reasoning for B gives 2·r_A + r_B = b_B. Both equations are correct. **Solving them.** Write the system as M·r = b with M = [[1, 0.1], [2, 1]]. - det M = 0.8. - M⁻¹ = 1.25 · [[1, −0.1], [−2, 1]]. - The total rate is therefore r_A + r_B = −1.25·b_A + 1.125·b_B. **Case 1: (b_A, b_B) = (1, 3).** - r_A = (1 − 0.3)/0.8 = 0.875. - r_B = 3 − 1.75 = 1.25. - Total = 2.125. Checked via the formula: −1.25 + 3.375 = 2.125. ✓ **Case 2: (b_A, b_B) = (0.5, 2.9).** - r_A = 0.21/0.8 = 0.2625. - r_B = 2.9 − 0.525 = 2.375. - Total = 2.6375. Checked via the formula: −0.625 + 3.2625 = 2.6375. ✓ - Ratio: 2.6375/2.125 = 1.2412, i.e. a 24.1% increase. ✓ **Positivity conditions.** The conditions b_A > 0.1·b_B and b_B > 2·b_A are exactly r_A > 0 and r_B > 0. Both parameter sets satisfy them (1 > 0.3 and 0.5 > 0.29; 3 > 2 and 2.9 > 1). ✓ **Reciprocity.** Let k_ij be the stiffness entries, s_i the slips and F_i the failure stresses. - The normalized transfers are 2 = k_BA·s_A/F_B and 0.1 = k_AB·s_B/F_A. - The self-drops are 1 = k_AA·s_A/F_A and 1 = k_BB·s_B/F_B. - The product of the transfers, divided by the product of the self-drops, equals k_AB·k_BA/(k_AA·k_BB) = 0.2. - With symmetric stiffness, k_AB² = 0.2·k_AA·k_BB. This is less than k_AA·k_BB, so the stiffness matrix is positive definite. ✓ ## 3. Trying to break it **Boundedness, which the text assumes but does not prove.** I checked it directly. - *Upper bound:* σ_i ≤ 1 always. - *Lower bound on A:* - Immediately after an A event, σ_A = 0. - Within an A-free interval of length L, B's own stress can produce at most 1 + b_B·L events. - So σ_A changes by at least b_A·L − 0.1·(1 + b_B·L) ≥ −0.1, using b_A > 0.1·b_B. - *Lower bound on B:* by the same argument, in a B-free interval σ_B changes by at least b_B·L − 2·(1 + b_A·L) ≥ −2. - Both stresses are therefore bounded below for any initial condition, and the long-run rates exist and equal M⁻¹b. **Other edge cases.** - Simultaneous failures are non-generic and do not affect long-run rates. - Because coupling is inhibitory, there are no cascades. **Where the effect does not occur.** The claim survives within the model, but its scope is narrow: - Under proportional slowing (b → c·b with c < 1), rates scale exactly by c, so the total always falls. - If loading is backslip-consistent (b = M·v for long-term slip-rate vector v), then r = v, and slowing both components lowers both rates. **Where it does occur.** The effect requires the negative off-diagonal entries of M⁻¹ and independent loading rates. Concretely: - Any decrease Δ in b_A, combined with a decrease in b_B smaller than (1.25/1.125)·Δ, raises the total. - Even slowing A alone raises the total. ## 4. Novelty This is the familiar fact that the inverse of an M-matrix-like interaction matrix with positive (inhibitory) off-diagonal entries has negative off-diagonal entries. Reducing the drive to one unit then "disinhibits" the other. It is closely analogous to paradoxical responses in inhibitory threshold networks and to stress-shadow reasoning in seismology. The specific numbers are new, but the mechanism is a variant of known results. ## 5. Fairness of the summary The claim is hedged ("can") and correctly lists what is held fixed. It omits three things: - The two faults are slowed very unequally. Uniform or backslip-consistent slowing cannot produce the effect. - Loading rates are treated as free, independent parameters. - The result concerns event counts in an idealized deterministic model, not moment or energy. The text itself states these limits; the one-line claim does not. ```json {"verdict": "correct with caveats", "novelty": "special case or variant of known results", "fair_summary": "partly", "confidence": 4, "one_line": "The rate equations and the 24.1% figure check out (total rate = −1.25·b_A + 1.125·b_B, with bounded stresses verifiable), but the effect depends on highly non-proportional, independently set loading rates in an idealized inhibitory threshold model and is a standard disinhibition consequence of the inverse interaction matrix."} ```