Lloyd N. Trefethen, Anne E. Trefethen, Satish C. Reddy and Tobin A. Driscoll (1993), Science 261(5121), 578–584 — DOI-10.1126/science.261.5121.578
Key takeaways
From the reading, NOTE-839.
- The anomaly. Eigenvalues predict instability above R = 5772 for Poiseuille flow and never for Couette flow. Transition is seen near R = 1000 and R = 350.
- Three linear quantities, all large. Transient growth sup_t ‖exp(−itL)‖ is R/29.1 (Couette) and R/71.5 (Poiseuille). The maximum resonance is (R/8.12)² and (R/17.4)², where a normal operator would give O(R). The smallest destabilizing operator perturbation is its inverse, O(R⁻²). The abstract’s “factors on the order of 10⁵” is the resonance, not the transient growth.
- The amplified structures are 3D. Streamwise vortices that become streaks, at streamwise wavenumber zero. Two-dimensional perturbations, the traditional focus, amplify far less.
- Pseudospectra unify them. The ε-pseudospectrum is where ‖(ωI − L)⁻¹‖ ≥ 1/ε, equivalently the spectra of all perturbations of L of norm ≤ ε. Its reach into the unstable half-plane bounds transient growth from below.
- Transition is conjectured, not shown. In a 2 × 2 model, linear growth with an energy-neutral quadratic term gives a threshold amplitude O(R⁻³). The authors conjecture that shear-flow transition works the same way, with threshold O(R^γ), γ < −1.
Standing in the record
Read on 2026-10-14 in NOTE-839. It is the short, widely read statement of the non-modal view beside Farrell & Ioannou (LIT-931, LIT-929). Two accounts are filed from it. THEORY-304 holds the shear-flow scalings, with the transition conjecture set outside the claim. THEORY-323, with LIT-931, holds the general statement that the eigenvalues of a non-normal operator govern only the long-run limit.
CLAIM-186 cites it for: “Shear flows judged stable by their eigenvalues turn turbulent because a transient amplification hands the disturbance to nonlinear dynamics.” The paper shows the amplification. The “because” is its conjecture, introduced with “we speculate” and supported by a 2 × 2 model, and CLAIM-186 should cite it as such. The model itself suggests a narrower reading across: a burst becomes lasting only if something feeds the grown disturbance back to where it started, and the disturbance needed for that falls steeply as the amplification grows.
Not held in the Anthology of the SOTA; it carries no instruction for machine-learning practice.