Contribution
For the shear flows where eigenvalue analysis fails worst, the paper puts three linear accounts of the failure on one footing, the pseudospectra of the linearized Navier–Stokes operator. It gives their leading-order scaling with Reynolds number R (Table 1). It then offers, as speculation, a 2 × 2 nonlinear model of how linear transient growth and nonlinear mixing could combine to give a transition threshold that falls faster than 1/R. The transient growth results themselves are credited to earlier work (Gustavsson; Henningson et al.; Butler & Farrell; Reddy & Henningson). The pseudospectral framing and the synthesis are this paper’s.
Key insight
Eigenvalues answer “what happens as t → ∞” and “at what frequencies is the response infinite”. A non-normal operator can have a large response far from its eigenvalues. Its pseudospectra, the regions where ‖(ωI − L)⁻¹‖ exceeds 1/ε, show where. When they reach far into the unstable half-plane the flow, though eigenvalue-stable, amplifies strongly, and a small change to the operator can make it unstable.
Assumptions
- Linearization about laminar flow. dU/dt = −iLU (eq. 1), with U an infinitesimal perturbation of plane Couette or plane Poiseuille flow in an infinite slab.
- The energy norm ‖U‖ = (∫|U|² dx dy dz)^{1/2} (eq. 2). Every amplification figure is in that norm.
- Fourier decomposition in x and z, so each calculation is one-dimensional in y with wavenumbers α and β; the reported quantities are optimized over α and β.
- Discretization. Chebyshev spectral, projected onto dominant eigenmodes, matrices of about 80 × 80 (notes 24–25).
- For the nonlinear section: a 2 × 2 model with a quadratic, energy-conserving nonlinearity standing in for the Navier–Stokes advection term.
Key results
-
The anomaly (pp. 578–579). Eigenvalue analysis gives a critical R = 5772 for Poiseuille flow, where transition is seen at R ≈ 1000, and stability for all R for Couette flow, where transition is seen at R ≈ 350.
-
Three linear quantities, all non-modal and three-dimensional.
- Pseudoresonance: sup over real ω of ‖(ωI − L)⁻¹‖ (eqs. 5–6). For a normal operator with spectrum at distance O(R⁻¹) below the axis it would be O(R); here it is O(R²).
- Transient growth: sup over t ≥ 0 of ‖exp(−itL)‖ (eq. 9), O(R), reached at times O(R). For a normal operator with stable spectrum it would be at most 1.
- Destabilizing perturbations: the ε-pseudospectrum is the union of the spectra of L + E over ‖E‖ ≤ ε (eq. 13). The smallest destabilizing E has norm ε_min = (sup over real ω of ‖(ωI − L)⁻¹‖)⁻¹, which is O(R⁻²) (Fig. 8). It is rank one, ε_min vu*, turning streaks back into vortices.
-
Table 1, leading order as R → ∞ (accurate to 1% for R > 100).
quantity Couette Poiseuille distance of spectrum from real axis (R/2.47)⁻¹ (R/2.47)⁻¹ maximum resonance (R/8.12)² (R/17.4)² transient growth sup ‖exp(−itL)‖ R/29.1 R/71.5 optimal time t_opt R/8.52 R/13.2 lower bound from pseudospectra R/42.6 R/103 The Poiseuille figures “pertain to the highly nonnormal part of the problem, ignoring the R > 5772 mode (TS wave)”. The maxima are at α = 0 or nearly so, with β between 1.18 and 2.04: streamwise vortices that become streaks.
-
Bounds linking the three (eqs. 10–12). ‖L⁻¹‖ ≤ ∫₀^∞‖exp(−itL)‖dt (eq. 11), “typically within a factor of 2 of equality”, so the O(R²) resonance is an O(R) accumulation time times an O(R) transient growth. sup_t ‖exp(−itL)‖ ≥ sup_ε ε⁻¹σ_ε(L), with σ_ε the furthest reach of the ε-pseudospectrum into the upper half-plane (eq. 12); for these flows the bound falls short by a factor of about 1.4.
-
Specific figures. At R = 5000 a streamwise streak is an ε-pseudomode of Poiseuille flow for ε ≈ 1.2 × 10⁻⁵, so it “can be excited by a streamwise vortex five orders of magnitude weaker in amplitude”; and a perturbation of L of norm 1.2 × 10⁻⁵ makes the flow unstable. Amplification by 1000 is possible for all R ≥ 549 (Fig. 2). About 90% of random perturbations of norm 10⁻³ to a 20 × 20 projection of L at R = 5000 render it unstable (note 30). The abstract’s “factors on the order of 10⁵” refers to these resonance and pseudomode figures, not to transient growth, which at R = 5000 is about 70 in norm by Table 1.
-
The 2 × 2 nonlinear model (eq. 14, Table 2, Fig. 10). du/dt = Au + ‖u‖Bu with A = [[−R⁻¹, 1], [0, −2R⁻¹]] and B = [[0, −1], [1, 0]]. B is skew-symmetric, so the nonlinear term moves energy and does not create it. At R = 25 the threshold amplitude is 4.22 × 10⁻⁴. Across R = 12.5 to 100 the threshold ε falls by a factor of about 8 per doubling of R, so ε = O(R⁻³), while the linear growth factor M is O(R). With forcing in place of an initial condition the threshold is O(R⁻⁴). The mechanism (“bootstrapping”): linear growth to Rε moves the state into a non-growing direction, and the quadratic term returns R(Rε)² = R³ε² to the growing one, which sustains growth if it exceeds ε.
-
The conjecture (p. 582). “We conjecture that transition to turbulence of eigenvalue-stable shear flows proceeds analogously to our model in that the destabilizing mechanism is essentially linear in the senses described above and the amplitude threshold for transition is O(R^γ) for some γ < −1.” Note 39 cites bounds −21/4 ≤ γ ≤ −1 from Kreiss, Lundbladh and Henningson.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | For non-normal operators, eigenvalue stability is compatible with transient growth, resonance far from the spectrum and sensitivity of the spectrum to small perturbations | strong | linear algebra, eqs. 5–13 |
| C2 | Linearized Couette and Poiseuille flow show transient growth O(R), resonance O(R²) and destabilizing perturbations O(R⁻²), with streamwise vortex–streak structures | strong | computation, Table 1 and Figs. 2, 4–9 |
| C3 | The three linear accounts are facets of one pseudospectral picture and predict similar flow structures | moderate | eqs. 11–13 and Fig. 9 |
| C4 | In a 2 × 2 model, linear transient growth with energy-neutral quadratic mixing gives a threshold amplitude O(R⁻³) | strong for the model | computation, Table 2, and a scaling argument |
| C5 | Transition in eigenvalue-stable shear flows proceeds by this linear-growth-plus-mixing route with threshold O(R^γ), γ < −1 | weak | conjecture by analogy with the model, labelled as such |
| C6 | Distinct triggers (vibration, inlet disturbance, geometric imperfection) may correspond to the three linear mechanisms | weak | speculation in the Conclusion |
Concepts
- Normal operator. One with orthogonal eigenfunctions (in the energy norm here).
- ε-pseudospectrum Λ_ε(L). The set of ω with ‖(ωI − L)⁻¹‖ ≥ ε⁻¹; equivalently the spectra of all L + E with ‖E‖ ≤ ε (eqs. 7, 13).
- ε-pseudomode. A function u with ‖(L − ω)u‖ ≤ ε‖u‖: nearly an eigenfunction, and excitable to large amplitude by a small input.
- Pseudoresonance. Large response to forcing at frequencies far from any eigenvalue.
- Subcritical transition. Transition to turbulence at R below the eigenvalue critical value.
- 3D perturbation. Here, one depending on z, whether or not it depends on x (note 19).
Connections
It builds on Butler & Farrell (1992) and Reddy & Henningson (1993), named as its foundation, and on Reddy, Schmid & Henningson (1993) for the pseudospectra of the Orr–Sommerfeld operator. It sets aside Squire’s theorem, which justified studying 2D perturbations: “this emphasis on 2D perturbations has been misplaced”. It questions secondary-instability theory as an account of natural transition (note 31). Farrell & Ioannou (LIT-931, LIT-929) cite it as the parallel development in laboratory shear flow.
Bearing on the record
- CLAIM-186 states the paper’s conjecture as its finding. The note of 2026-10-11 says: “Shear flows judged stable by their eigenvalues turn turbulent because a transient amplification hands the disturbance to nonlinear dynamics.” The paper shows the transient amplification (C2). That turbulence follows “because” of it, through nonlinear bootstrapping, it offers as a conjecture from a 2 × 2 model, in a section that opens “we speculate”. The sentence should say that Trefethen et al. conjecture this, and on what model.
- The model supports a narrower form of CLAIM-186’s reading. In the 2 × 2 model the nonlinearity creates no energy; it only rotates the grown state back into the growing direction. The threshold falls as R⁻³ against a linear growth of R. The reading across would be that a burst becomes a lasting change only if something feeds the grown disturbance back to where it started, and that the size of disturbance needed falls steeply with the amplification. That is an analogy the model invites, not a result about arguments.
- LIT-932’s own summary says nonlinearity “can carry the amplified disturbance into turbulence”. That is fair as “can”, in the model, and is corrected on the LIT to mark it as conjecture for the flows.
- Two new THEORY documents. THEORY-323 (with Farrell & Ioannou Part I, LIT-931) holds the general linear-algebra claim. THEORY-304 holds the shear-flow scalings, with the transition conjecture placed under “What this does not say”. No existing THEORY held either.
- No instruction for machine-learning practice; nothing for the anthology.
Limitations
- Linear analysis about laminar flow in infinite plane geometry. Pipe flow and boundary layers are mentioned, not computed.
- The scalings are asymptotic in R, with constants from discretized computations.
- The transition mechanism is conjecture. The authors list ways the Navier–Stokes equations differ from the model: infinitely many degrees of freedom, most without non-normal growth, and nonlinear interactions across wavenumbers, which make the model’s nonlinearity “perhaps unrealistically strong”.
- The physical meaning of operator perturbations is left open: they may be “better understood as a symptom of the nonnormality of the unperturbed problem than as a description of how transition to turbulence actually takes place”.
Open questions
- The exponent γ in the transition threshold O(R^γ) for real flows, which the authors expected simulation to settle “in the coming decade”.
- Which trigger (vibration, inlet disturbance, imperfect geometry) matches which of the three linear mechanisms in experiment.