Similarity Transformations: The Stretching Sheet

1. The problem

A sheet at y = 0 stretches with u_w = ax under a transverse field B₀. The boundary-layer equations (see the MHD notes for the last term):

with u = ax, v = 0 at y = 0 and u → 0 as y → ∞.

2. Choosing the similarity variables

The stretching rate a (units 1/s) and viscosity ν (m²/s) combine into exactly one length, √(ν/a), and one velocity scale at station x, namely ax. So scale y by the length and u by ax:

The form of v is not guessed — it is forced by continuity, which this ansatz satisfies identically (check: ∂u/∂x = af′, ∂v/∂y = −af′).

3. The collapse

Compute each momentum term: u∂u/∂x = a²xf′², v∂u/∂y = −a²xff″, ν∂²u/∂y² = a²xf‴, and the Lorentz term −(σB₀²/ρ)axf′. Every term carries the factor a²x — divide it out and x disappears entirely:

with M = σB₀²/(ρa). The two-variable PDE system has become one ODE — exactly, not approximately. That cancellation is the definition of a similarity solution; had any stray x survived, the ansatz would have failed.

4. The exact solution and what research adds

Try f = (1 − e^(−βη))/β: substitution satisfies the equation iff β² = 1 + M, giving wall shear f″(0) = −√(1+M) (full details in WLM-2026-0001). Modern papers perturb this base — nonlinear stretching u_w = ax^n, suction, radiation, nanofluid property models, Casson rheology. The similarity recipe survives; the closed form usually does not, and the ODE goes to the numerical methods.

5. Extended worked examples

Example 1 — verify the exact solution by direct substitution. Claim: f = (1 − e^{−βη})/β solves f‴ + ff″ − f′² − Mf′ = 0 iff β² = 1 + M. Compute (write E ≡ e^{−βη}): f′ = E, f″ = −βE, f‴ = β²E. Substitute: β²E + [(1−E)/β](−βE) − E² − ME = β²E − E + E² − E² − ME = E(β² − 1 − M). Zero for all η exactly when β² = 1 + M ∎. Boundary conditions: f(0) = 0 ✓, f′(0) = 1 ✓, f′(∞) = 0 ✓. Three lines of algebra, and a full nonlinear boundary-layer problem is solved — the reward the similarity machinery was built to earn.

Example 2 — numbers from the solution. At M = 3: β = 2, wall shear f″(0) = −2. Dimensionally, τ_w = μ(∂u/∂y)|₀ = μ a x √(a/ν) f″(0) — the sheet at station x = 0.1 m, stretched at a = 2 s⁻¹ in water, feels τ_w = 10⁻³×2×0.1×√(2/10⁻⁶)×(−2) = −0.566 Pa (resisting the stretch). The boundary-layer thickness scales as √(ν/a)/β: the M = 3 field halves the layer relative to M = 0 — magnetic compression made visible.

Example 3 — the cancellation test failing (instructive!). Try the same ansatz on a sheet stretching nonlinearly, u_w = ax². Then u = ax²F′(η) forces u∂u/∂x = a²x³(2F′² + …) while ν∂²u/∂y² produces a²x²·(…): the powers of x no longer match, the ansatz dies, and one must rescale η itself with x (η ∝ y·x^{(n−1)/2} for u_w = axⁿ) to restore similarity. The lesson: the transformation is not a ritual — each new wall law demands its own scaling, found by forcing the cancellation.

Example 4 — what v is doing. From v = −√(aν)f(η): far from the wall f(∞) = 1/β, so v_∞ = −√(aν)/β — a uniform inflow toward the sheet. The stretching surface acts as a planar sink, entraining fluid at a rate the magnetic field reduces (larger β). This entrainment is the physical reason polymer-sheet cooling baths circulate even when nothing stirs them.

6. Common misconceptions

"Similarity solutions are approximations." The reduction is exact — every discarded x cancelled identically. Approximation enters only where it always did: in the boundary-layer equations themselves and in any numerical solution of the final ODE.

"η is a universal variable you memorise." η is derived, per problem, from the available dimensional ingredients (here a and ν; in Blasius, U, ν and x). Transplanting Blasius's η into the sheet problem produces nonsense; deriving η is the skill being taught.

"The exact solution makes numerics unnecessary." Add one ingredient — nonlinear stretching, suction v_w ≠ 0, a nanofluid coefficient, thermal coupling with dissipation — and closure is gone. The exact case's enduring role is as the benchmark: every solver in this literature must reproduce f″(0) = −√(1+M) before its new results are believed.

"f, η have physical units." Both are dimensionless by construction; all dimensions were quarantined into the scales √(ν/a) and ax. Reporting "f″(0) in Pa" (it happens) signals the scaling was never understood.

7. Where this shows up

Crane's 1970 closed form (the M = 0 case) launched an entire research industry because the configuration is industrially real: polymer films and fibres are drawn from dies at controlled rates, glass sheets are stretched, metals are extruded — and the cooling rate at the surface, governed by exactly this boundary layer, sets product quality. Every added effect in the literature is a manufacturing variable in disguise: suction = porous mould walls; magnetic field = electromagnetic damping of a conducting melt; nanofluid properties = engineered coolants; radiation = high-temperature furnaces. Your own MPhil problem class — MHD hybrid nanofluid over a stretching sheet via bvp4c — sits at the current end of precisely this lineage.

8. Practice problems

P1. Derive v from ψ = √(aν)x f(η) and confirm continuity is identically satisfied. Solution: u = ∂ψ/∂y = axf′; v = −∂ψ/∂x = −√(aν)f; ∂u/∂x + ∂v/∂y = af′ − af′ = 0 ✓.

P2. For M = 8, give β, f″(0), and the entrainment velocity ratio v_∞(M=8)/v_∞(0). Solution: β = 3; f″(0) = −3; ratio = β₀/β₈ = 1/3.

P3. Show the momentum boundary-layer "thickness" where f′ = 0.01 is η = ln(100)/β and evaluate for M = 0 and 3. Solution: E = 0.01 → η = 4.605/β: η = 4.61 (M=0), 2.30 (M=3).

P4. Add uniform suction f(0) = S > 0 to the M-problem and try f = S + (1 − e^{−βη})/β. Find the condition on β. Solution: substitution now gives β² − Sβ − 1 − M = 0 → β = [S + √(S² + 4(1+M))]/2 — the exact solution survives suction, a small research-grade result you just derived.

9. Going deeper

Why do similarity variables exist at all? Because the governing equations and boundary conditions admit a scaling symmetry: the transformation x → λx, y → λ^{1/2}y, ψ → λ^{1/2}ψ maps solutions to solutions. Lie group analysis systematises this — compute the symmetry generators of the PDE, and their invariants are the similarity variables, no inspired guessing required. Bluman & Kumei's framework turns the art of this chapter into an algorithm, and modern computer algebra performs it mechanically; the stretching-sheet reduction is the two-line special case of a theory that classifies every reduction the boundary-layer equations will ever admit.

10. Historical context

The moving-surface boundary layer is younger than the classical one by half a century. Sakiadis (1961) analysed the layer on a continuously issuing surface — inspired by polymer filaments and metal strips emerging from dies — and found it differs genuinely from Blasius (the entrained fluid moves with the wall, not past it). Crane's 1970 paper, barely two pages in ZAMP, treated the linearly stretching sheet and noticed the miracle: the profile e^{−η} solves the nonlinear problem exactly. That closed form — rare as rain in boundary-layer theory — made the configuration irresistible as a host for added physics, and the citations compounded: suction (Gupta & Gupta 1977), magnetic fields (Pavlov 1974; Andersson 1992's exact MHD form is your β = √(1+M)), viscoelasticity, porous media, nanofluids (post-2006), and hybrids thereof. Reviews now count thousands of descendant papers. Your MPhil problem stands in a genealogy traceable, citation by citation, to those two pages.

11. Another way to see it: dimensional analysis first

Before any clever substitution, ask what the answer could depend on. Near a linearly stretching sheet the physical inputs are only a (s⁻¹) and ν (m²/s) — no external length or velocity exists. The unique length is √(ν/a); the unique velocity at station x is ax. Buckingham's theorem then forces the solution shape: u/(ax) must be a function of the one dimensionless coordinate y/√(ν/a) — which is to say u = ax f′(η), η = y√(a/ν), before a single derivative is taken. The similarity ansatz is not guessed; it is the only form dimensional analysis permits. Blasius differs instructively: there an external U and the coordinate x supply the scale √(νx/U), so η carries x inside it and the layer grows downstream, while the stretching sheet's layer has constant thickness — visible immediately from the fact that x could not enter η. Two problems, one method, and the contrast teaches exactly what similarity is: the statement that a problem owns fewer scales than variables.

12. Frequently asked questions

Why does η contain √ν? Because ν (m²/s) needs a square root to become a length when combined with a rate a (1/s): √(ν/a) is the only length constructible — the same √ν that thickens every diffusive layer in this course.

When do dual solutions appear, and are both real? For stretching with assisting physics, the solution is unique. Shrinking sheets (u_w = −ax, with suction) and opposing buoyancy bend the solution curve back on itself: two mathematical solutions at one parameter value. Stability analysis (small-perturbation eigenvalues) labels one stable — the physically observable branch — and one not; papers report both because the fold's location is itself the physical prediction (where steady flow ceases to exist).

Is the constant-thickness layer a paradox — shouldn't layers grow? Growth needs a downstream direction with memory. Here every station is a scaled copy of every other (u_w ∝ x), so the layer settles to the thickness where stretching-induced thinning balances viscous diffusion: √(ν/a)/β, x-independent. The magnetic field shifts the balance, not the logic.

Do experiments confirm any of this? The base flow, yes — laser measurements over stretching films reproduce the exponential profile. The exotic superstructure (hybrid nanofluid + radiation + slip + …) is largely theory racing ahead of measurement; the honest reading of that literature is "systematic exploration of a model family," with the exact solutions of this chapter as its fixed calibration points.

13. Further practice

P5. Purely from dimensional analysis, what must the wall shear stress look like? Solution: τ_w = μ ∂u/∂y|₀ must scale as μ·(ax)/√(ν/a) = ρx√(νa³)·(dimensionless): τ_w ∝ ρ√(ν a³) x — the linear-in-x growth of Example 2, obtained without solving anything.

P6. For the unsteady stretching sheet u_w = ax/(1−λt), what similarity variable keeps the reduction alive? Solution: the only consistent scale is η = y√(a/(ν(1−λt))): time enters through the same square-root, adding an unsteadiness parameter A = λ/a to the ODE — the standard "unsteady stretching" family of the literature, derived by scale-bookkeeping alone.

P7. Verify the suction result of P4 (§8) reduces correctly: S = 0 and M = 0. Solution: β = [0 + √(0+4)]/2 = 1 → f″(0) = −1, Crane recovered ✓ — the limit-check habit, practised until reflexive.

14. Worked exam problem

Problem. Water flows over a sheet stretching at a = 4 s⁻¹ under a field giving M = 1.25. At the station x = 5 cm find (a) β and f″(0), (b) the dimensional wall shear stress, (c) the layer thickness where f′ = 0.01.

Solution. (a) β = √(1+1.25) = 1.5; f″(0) = −1.5 — exactly, from the closed form. (b) τ_w = μ a x √(a/ν)·f″(0) = 10⁻³×4×0.05×√(4/1.004×10⁻⁶)×1.5 = 2×10⁻⁴×1 996×1.5 = 0.60 Pa opposing the stretch. (c) η₀.₀₁ = ln(100)/β = 4.605/1.5 = 3.07; physical thickness = η₀.₀₁√(ν/a) = 3.07×√(2.51×10⁻⁷) = 3.07×5.01×10⁻⁴ = 1.54 mm — constant along the sheet, as the dimensional analysis of §11 said it must be. An exam problem in this field is exactly this: exact solution, one conversion layer of scales, engineering numbers out.

15. Key takeaways

The similarity ansatz u = axf′(η), η = y√(a/ν) is forced by dimensional analysis, satisfies continuity via the stream function, and collapses the boundary-layer PDEs exactly to f‴ + ff″ − f′² − Mf′ = 0. The linear-stretching case solves in closed form, f = (1 − e^{−βη})/β with β = √(1+M) — Crane's M = 0 result generalised — giving f″(0) = −β and a constant-thickness layer. This exact solution is the permanent benchmark of a thousand-paper literature.

16. Where to go next

The −Mf′ term's physics is the Lorentz force chapter; the temperature field rides this flow in heat transfer; the numerics for every non-exact extension are in shooting and bvp4c; nanofluid coefficients come from nanofluid models. The Reynolds number calculator evaluates β, f″(0) and τ_w with visible steps.