betaFeedback

WorksPhysics and Astronomy

Compressible unsteady aerodynamics of finite-chord porous aerofoils

Abstract

A unified linear theory is developed for the unsteady loading of finite-chord aerofoils, rigid or porous, in compressible subsonic flow. The formulation combines Possio's integral operator with a convective permeable boundary condition, allowing chordwise-varying admittance while retaining the wake and unsteady Kutta condition. The loading exponents at aerofoil edges and admittance discontinuities keep their incompressible form in terms of the local permeability parameter, so weighted-Jacobi collocation carries over. The solution is verified against published incompressible results and an independent compressible formulation. Gust and heave responses and their indicial counterparts reveal that permeability controls the sensitivity of unsteady loading to compressibility. As permeability increases, the material impedance rather than the surrounding flow sets the pressure jump, but only gradually: at Mach number 0.7 the gust load on a weakly permeable surface changes by a third to a half, and closed-form steady and high-frequency limits for resistive surfaces show that it vanishes only when the permeability parameter greatly exceeds the Mach number. The porous-to-impermeable load ratio, by which a treatment is judged, cannot be obtained by combining incompressible porous and compressible rigid theories, which misjudge it by up to a factor of two. Incompressible theory overestimates this ratio for an acoustically compact chord and underestimates it for a non-compact one, by factors of 1.4 to 4.3 at Mach numbers 0.5 to 0.7 and reduced frequencies 5 to 20, overstating the load reduction. Since compactness depends on Mach number times reduced frequency, the error persists at low speed, exceeding 10 per cent at Mach number 0.05 for reduced frequencies above 10.

Introduction

An aerofoil in a non-uniform or unsteady stream carries an unsteady load, and the classical theory of that load is the foundation of aeroelasticity, of rotor and propeller load prediction, and of every engineering model of turbulence-interaction noise [6, 16, 31]. It comprises Theodorsen [36] for a section oscillating in a uniform stream, Sears [35] and von Kármán and Sears [38] for a section meeting a convected gust, Wagner [39] and Küssner [26] for the corresponding indicial responses, and Possio [34] for the compressible kernel. All of it assumes an impermeable surface.

Permeable surfaces are of interest because the loading they carry at an edge is weaker. Ffowcs Williams and Hall [10] showed that the efficiency with which a half-plane scatters near-field pressure into sound is controlled by the inverse-square-root singularity of the loading at a rigid edge; softening that singularity is the mechanism behind every porous edge treatment, and porous leading and trailing edges have been measured to reduce broadband noise appreciably [13, 9, 7]. Porosity is one of a family of passive edge treatments, alongside serrations, brushes, and compliant and graded surfaces, but the family does not act through a single mechanism. A large part of the benefit of serrations comes from destructive interference between sources distributed along an oblique edge, so it is a property of the radiation and is retained in rotor predictions only when the acoustic calculation keeps the edge geometry and the spanwise phase of the loading [25, 8, 30, 29]. Porous, poroelastic and graded edges act instead on the edge singularity itself [22, 2], so their benefit is carried by the unsteady load, and the reduction predicted in application depends on how the treatment is represented in the loading model rather than in the acoustic analogy that follows it [27]. The same singularity that sets the scattering, moreover, is what generates leading-edge suction, so for this class of treatment the aerodynamic price is fixed by the same quantity as the benefit. Both statements are statements about loading, and both therefore belong to unsteady aerodynamics rather than to acoustics. That is the subject of this paper: what a permeable surface does to the unsteady load on a finite chord in compressible flow, before any acoustic analogy is applied.

Analytical treatments of permeable surfaces divide by geometry. Jaworski and Peake [22] solved the semi-infinite problem by Wiener–Hopf, with the static porosity condition of Howe [20], in which the flow through the surface is proportional to the pressure jump across it. They showed that, in certain parametric limits, the far-field acoustic power of a porous edge scales with the sixth power of the flow speed rather than the fifth power found for a rigid edge, a result that motivated much of what followed. The finite-chord problem is different in kind: the chord introduces a second edge, a shed wake and the Kutta condition, none of which the half-plane carries. Hajian and Jaworski [18] solved the steady finite-chord problem for a permeability that varies along the chord, and Baddoo et al. [4] extended the analysis to unsteady flow. They derived porous counterparts of the four classical unsteady functions: the Theodorsen and Sears functions for harmonic oscillation and sinusoidal gust encounter, and the Wagner and Küssner functions for the corresponding indicial responses. The present analysis borrows its essential device from Baddoo et al. [4]. At a permeable leading edge the load is still singular, but more weakly than the inverse square root of a rigid edge. Its exponent falls below 1/21/2 by an amount set by the local permeability, and returns to 1/21/2 only where the surface is impermeable. A Chebyshev (Glauert) expansion builds in the rigid exponent and cannot represent this behaviour, whereas a family of Jacobi polynomials whose weights carry the correct exponents at the two edges can. Ayton et al. [3] treated chordwise-varying porosity by a Wiener–Hopf-based approximation over the same frequency range.

Every one of those finite-chord treatments is incompressible, and the question is whether that matters. For a rigid aerofoil the answer has long been known. Consider a section of chord 2b2b in a stream of speed UU, with sound speed c0c_0 and Mach number M=U/c0M = U/c_0, responding at angular frequency ω\omega, so that its reduced frequency is σ=ωb/U\sigma= \omega b/U and the acoustic wavenumber is k0=ω/c0k_0 = \omega/c_0. Incompressible gust theory requires the chord to be acoustically compact, k0b=Mσ≪1k_0b = M\sigma\ll1; beyond that the loading is no longer in phase along the chord, and the compressible response departs from Sears’s function by more than a correction [17, 33, 1, 24]. The incompressible theories are the limit M→0M \to0 at fixed σ\sigma and so describe a compact chord at any speed, since compactness is a condition on MσM\sigma, not on MM. The surfaces to which permeable edges are applied lie well outside that range, because they move fast through disturbed flow. A rotor blade meets its own wake, a neighbouring rotor’s wake and the ambient turbulence once per revolution. At M=0.5M = 0.5–0.70.7, ingesting turbulence or passing through a tip vortex, it is loaded at σ=5\sigma= 5–2020, so MσM\sigma lies between 2.52.5 and 1414, and on the advancing side of a high-speed rotor, where the local Mach number reaches 0.60.6–0.90.9, the impulsive part of the loading dominates what is radiated [23]. A wing or fan blade at high subsonic speed that meets atmospheric turbulence or the wakes of surfaces upstream is in the same position: a gust whose wavelength equals the chord has σ=π\sigma= \pi, so at M=0.5M = 0.5–0.70.7 the product MσM\sigma is already 1.6–2.2, and shorter gusts give larger values. Even a low-speed aerofoil is non-compact once σ\sigma approaches 1/M1/M, about 20 at M=0.05M = 0.05, which is within the range that turbulence-interaction and blade–vortex-interaction loading models must cover. In each case the unsteady load is the input to aeroelastic and acoustic predictions. For rotors and propellers it drives both the tonal and the broadband parts of the signature, frequency-domain load models are what make the prediction tractable at design time [15], and the spread in predicted hovering-rotor noise that follows from gust-induced unsteady loading is large enough that the loading model, not the acoustic propagation, sets the uncertainty [14]. A treatment of such a surface therefore has to be assessed in compressible flow, at finite chord, with the wake retained.

For a permeable aerofoil it is not known whether incompressible theory remains adequate outside the compact range, and there is no reason to expect the rigid criterion Mσ≪1M\sigma\ll1 to carry over. The pressure jump across a rigid surface is set entirely by the surrounding flow, which is where compressibility acts. Across a permeable surface it is set partly by the surrounding flow and partly by the material impedance, which fixes the pressure jump needed to drive fluid through the surface, and the balance between the two shifts with permeability. How permeability controls the sensitivity of the unsteady loading to compressibility, and how compressibility changes the load of a porous aerofoil relative to that of an impermeable one, which is how a treatment is judged, are the questions a compressible theory has to answer.

At the compressible end, Hales et al. [19] formulated a semi-infinite permeable edge in compressible flow with the grazing-flow conductivity of Howe et al. [21]; being a half-plane theory it carries neither the finite chord nor the wake. The combination that is missing (Table 1) is the subject of this paper: a finite chord in compressible flow, with an admittance (the reciprocal of the material impedance) that varies arbitrarily along the chord, and with the wake and the Kutta condition.

finite chordcompressiblewake, Kuttaarbitrary admittance
Jaworski and Peake [22]nopartialnono
Hajian and Jaworski [18] (steady)yesnon/ayes
Baddoo et al. [4]yesnoyesyes
Ayton et al. [3]yesnoyesyes
Hales et al. [19]noyesnono
present workyesyesyesyes

Table 1. Table 1: Analytical treatments of a permeable aerofoil surface. “Partial” records a theory formulated for compressible radiation from an incompressible near field.

This paper develops a linear theory for the unsteady loading of finite-chord porous aerofoils in compressible subsonic flow. It combines Possio’s integral operator with a convective permeable boundary condition, allows the admittance to vary along the chord, and retains the shed wake and the unsteady Kutta condition. Three things change in passing from the incompressible formulation to the compressible one, and one does not. The integral operator becomes Possio’s rather than Cauchy’s, and its compressible part acts on the jump in velocity potential itself and not only on the bound vorticity, so the compressible solver must carry both where the incompressible one closes on the vorticity alone. The permeable boundary condition becomes convective, so that the permeability parameter (§ 2.2) carries the Mach number. And the reduced frequencies are stretched by the Prandtl–Glauert transformation, so that the ratio of the chord to the acoustic wavelength enters as well as its ratio to the hydrodynamic one. What does not change is the local behaviour at the aerofoil edges and at admittance discontinuities. The non-Cauchy part of Possio’s operator is bounded, so it cannot alter the exponents that a Riemann–Hilbert analysis assigns there, and the loading exponents retain their incompressible functions tional form when expressed in terms of the local permeability parameter (§ 2.5). This is what allows the weighted-Jacobi collocation method of Baddoo et al. [4], including their treatment of interior admittance discontinuities, to be carried over at the cost of a rigid calculation. The solution is verified against published incompressible results and against an independent compressible formulation in physical variables (§ 4). The formulation contains the classical theories as limits: with λ=0\lambda= 0 it is Possio’s problem, and for ke→0k_e \to0 it reduces to the functions of Theodorsen, Sears, Wagner and Küssner; with λ≠0\lambda\ne0 and ke→0k_e \to0 it is the incompressible porous theory of Baddoo et al. [4]. Each limit is verified in § 4.

The theory is then used to examine how permeability controls the sensitivity of the unsteady loading to compressibility (§ 6). Incompressible and compressible responses are compared for the same material and reduced frequency, both where the chord is acoustically compact and where it is not, and at low as well as high Mach number, since compactness is set by MσM\sigma rather than by MM (§ 6.3). The two routes by which the Mach number reaches a permeable surface, the boundary condition and the integral operator, are separated. The shortcut of combining the existing incompressible porous and compressible rigid theories is tested against the full solution, and the condition under which it is adequate is derived. The harmonic gust and heave responses, the porous counterparts of the Sears and Theodorsen functions, are computed together with their indicial counterparts, the porous Küssner and Wagner functions, and the steady and high-frequency limits are obtained in closed form. The aim is compressible loading predictions for assessing porous aerofoils and for supplying aerodynamic inputs to aeroelastic and aeroacoustic models.

The paper is organised as follows. Section 2 formulates the problem and derives the edge and junction exponents; § 3 describes the solution; § 4 verifies it against classical, closed-form and published results; § 6 compares the compressible and incompressible responses; and § 7 draws conclusions.

Formulation

The problem

Figure 1 shows the configuration and the notation used below. Three features of the configuration drive everything that follows. The permeable region has ends, so the problem carries interior discontinuities in boundary condition as well as the two aerofoil edges, and § 2.5 shows that each carries its own algebraic singularity with exponents that are not independent. The trailing edge carries a shed wake and the unsteady Kutta condition, so the circulation is fixed by the solution. And the admittance is a function of position, not a constant: a manufactured insert has a definite extent, and the grazing flow that sets the admittance of a perforate varies along it.

The configuration and the three exponents

Figure 1. The configuration and the three exponents. A thin section of chord c=2bc = 2b in a stream of Mach number MM carries a permeable region of length ℓ\ell, across which the admittance enters through λ(ξ)\lambda(\xi) alone; here the region begins at the leading edge and ends at a junction sjs_j, but the formulation admits any distribution, including a permeable trailing edge. The loading is singular at the leading edge and at the junction, bounded at the trailing edge, and each exponent is an inverse cotangent of the local λ\lambda (§ 2.5). The wake and the unsteady Kutta condition make the circulation part of the solution rather than a free parameter. A semi-infinite formulation cannot represent this, and it is what makes the chord enter the answer.

A thin aerofoil of chord c=2bc = 2b lies along −b≤x1≤b-b \le x_1 \le b, x2=0x_2 = 0, in a uniform stream UU of an ideal fluid of density ρ0\rho_0 and sound speed c0c_0, at Mach number M=U/c0M = U/c_0 and Prandtl–Glauert factor βM=1−M2\beta_M = \sqrt{1-M^2}. The surface is permeable, with a transfer admittance that may vary along the chord. The disturbance is harmonic with the convention e−iωte^{-i\omega t}; we write k0=ω/c0k_0 = \omega/c_0 for the acoustic wavenumber and k1=ω/Uk_1 = \omega/U for the hydrodynamic one. The canonical forcing is the frozen transverse gust w=w0exp⁡[i(k1x1−ωt)]w = w_0 \exp[i(k_1 x_1 - \omega t)], but the formulation admits any imposed upwash, and § 2.4 lists the three that are used here.

The Prandtl–Glauert transformation (appendix § 14) maps the problem to a Helmholtz problem of wavenumber k0/βMk_0/\beta_M on a plate of half-length b/βMb/\beta_M. Normalising the chordwise coordinate on that stretched half-length, ξ∈[−1,1]\xi\in[-1,1], the two reduced wavenumbers each pick up βM\beta_M twice,

ke=k0bβM2,kˉ=k1bβM2,ke=Mkˉ.(1)k_e = \frac{k_0 b}{\beta_M^2}, \qquad\bar{k} = \frac{k_1 b}{\beta_M^2}, \qquad k_e = M\bar{k}. \tag*{(1)}

Both are dimensionless. kek_e is the phase, in radians, that a sound wave of the transformed problem accumulates across its half-chord b/βMb/\beta_M, and so measures acoustic compactness; kˉ\bar{k} is the corresponding phase of the convected gust. Equivalently, the chord spans keβM2/πk_e\beta_M^2/\pi acoustic wavelengths. The classical reduced frequency is σ=ωb/U=k1b=βM2kˉ\sigma= \omega b/U = k_1b = \beta_M^2\bar{k}, and results are reported in σ\sigma so that they may be read against the classical functions directly.

The permeable boundary condition

The surface is modelled as a homogenised sheet of vanishing thickness across which the pressure jump Δp\Delta p, upper surface minus lower, drives a seepage velocity vsv_s, positive in the x2x_2 direction. Writing the normalised transfer admittance A=1/Zt\mathcal{A} = 1/Z_t through

vs=−A Δpρ0c0,(2)v_s = -\frac{\mathcal{A}\,\Delta p}{\rho_0 c_0}, \tag*{(2)}

so that fluid passes from the side at higher pressure to the side at lower pressure. Mass conservation across the sheet requires the normal velocity of the fluid relative to the surface, the imposed upwash ww plus the disturbance velocity ∂ϕ/∂x2\partial\phi/\partial x_2, to equal the seepage velocity on both faces, which gives

∂ϕ∂x2∣x2=0=−w+vs=−w+Ac0(∂∂t+U∂∂x1)[φ],(3)\left.\frac{\partial\phi}{\partial x_2}\right|_{x_2=0} = -w + v_s = -w + \frac{\mathcal{A}}{c_0}\left(\frac{\partial}{\partial t} + U\frac{\partial}{\partial x_1}\right)[\varphi], \tag*{(3)}

the second form following from the linearised momentum relation Δp=−ρ0(∂t+U∂x1)[φ]\Delta p = -\rho_0(\partial_t + U\partial_{x_1})[\varphi], with [φ][\varphi] the jump in velocity potential. Written this way the condition is convective: the seepage is proportional to the material rate of change of the potential jump seen by the mean flow, and not to the potential jump itself. The convective operator sits in the momentum relation rather than in the material law (2.2), which is why much of the porous-media literature, working at U=0U = 0, sees only ∂x2φ∝[φ]\partial_{x_2}\varphi\propto[\varphi]. The two conditions coincide only when U=0U = 0. The convective derivative comes from the linearised Bernoulli relation, which carries it in incompressible flow as well, so it is not removed by letting M→0M \to0 at fixed UU.

With harmonic time dependence and lengths scaled on the stretched half-chord, the material enters the integral equation of § 2.3 through a single dimensionless group, the permeability parameter

λ=2AMβM,(4)\lambda= \frac{2\mathcal{A}M}{\beta_M}, \tag*{(4)}

which multiplies the pressure jump there. A rigid surface has λ=0\lambda= 0, a fully transparent one λ→∞\lambda\to\infty, and a chordwise distribution of material is a function λ(ξ)\lambda(\xi). Appendix § 13 gives the derivation and shows that λ\lambda, being a ratio of the seepage coefficient to the Cauchy coefficient, does not depend on how the unknown is normalised. It also shows how the condition relates to the static condition of Jaworski and Peake [22] and Ayton et al. [3]: holding the porosity parameter μ=−ik0A\mu= -ik_0\mathcal{A} fixed as M→0M \to0 keeps λ=2iμ/(βMk1)\lambda= 2i\mu/(\beta_M k_1) finite, but the static condition is recovered only when, in addition, the convective derivative is dropped.

Relation to the porosity parameter of Baddoo et al. [4]. Their ψ=4/(Φ+2ikρe)\psi= 4/(\Phi+ 2ik\rho_e), with Φ\Phi the dimensionless resistance, ρe\rho_e the effective density of the pore fluid and k=ωb/Uk = \omega b/U their reduced frequency, equal to σ\sigma, is the same quantity as λ\lambda, written in the time convention e+ikte^{+ikt} and for a specific material law. The identification is not an analogy: it is an equality of the coefficient multiplying the loading in the two governing equations, it is confirmed numerically in § 4 across the full unsteady range, and it is why their exponent formula and ours are the same function. In the present convention the same material is the conjugate, ψ↦4/(Φ−2ikρe)\psi\mapsto4/(\Phi- 2ik\rho_e).

Reduction to a singular integral equation

Representing the aerofoil and its wake by a dipole sheet whose strength is the jump in potential, taking the pressure jump Δp(ξ)\Delta p(\xi) as the unknown, and enforcing (3) on the mean surface yields, after the Prandtl–Glauert transformation,

λ(ξ)Π(ξ)+1π∫−11Π(s)ξ−s ds+∫−11K(ξ−s)Π(s) ds=W(ξ),(5)\lambda(\xi)\Pi(\xi) + \frac{1}{\pi}\int_{-1}^{1}\frac{\Pi(s)}{\xi-s}\,\mathrm{d}s + \int_{-1}^{1}\mathcal{K}(\xi-s)\Pi(s)\,\mathrm{d}s = \mathcal{W}(\xi), \tag*{(5)}

where Π\Pi is the pressure jump, non-dimensionalised so that the Cauchy coefficient is 1/π1/\pi, and K\mathcal{K} is a bounded remainder. The bar on the integral sign denotes the Cauchy principal value,

p.v.⁡∫−11Π(s)ξ−s ds≡lim⁡ϵ→0+(∫−1ξ−ϵ+∫ξ+ϵ1)Π(s)ξ−s ds,\operatorname{p.v.}\int_{-1}^{1}\frac{\Pi(s)}{\xi-s}\,\mathrm{d}s \equiv\lim_{\epsilon\to0^+}\left(\int_{-1}^{\xi-\epsilon}+\int_{\xi+\epsilon}^{1}\right)\frac{\Pi(s)}{\xi-s}\,\mathrm{d}s,

which exists because Π\Pi is Hölder-continuous on any closed subinterval of (−1,1)(-1,1) that contains no admittance junction, and is integrable across the junctions (§ 2.5). The operator is the Possio kernel (Possio [34]): appendix § 14 derives (5) from the convected wave equation and shows that its singular part is exactly the Glauert Cauchy kernel, while the remainder consists of the regularised Hankel function H1(1)(z)+2i/(πz)H_1^{(1)}(z)+2i/(\pi z), which is bounded, together with terms whose kernels are at worst logarithmically singular. Each therefore maps a Hölder-continuous Π\Pi to a bounded function, which is all the analysis below requires; K\mathcal{K} is a bounded operator rather than a bounded convolution kernel. It includes the wake, which enters through the circulation; at the trailing edge the chord and wake contributions each carry a logarithm, and appendix § 14 shows that the two cancel, so that K\mathcal{K} stays bounded there too.

Two auxiliary conditions close the problem: Kelvin’s theorem, relating the bound circulation to the shed vorticity, and the unsteady Kutta condition, which requires the loading to be bounded at the trailing edge. On a rigid aft section the latter is Π∼(1−ξ)1/2\Pi\sim(1-\xi)^{1/2}, or a permeable one it is Π∼(1−ξ)+α\Pi\sim(1-\xi)^{+\alpha} with α\alpha given by (10) below, which is weaker but still bounded.

Provenance. The equation is assembled from two independent ingredients and is not inherited whole from any previous paper. The integral operator is the compressible thin-aerofoil operator of Possio [34], unchanged. The term λ(ξ)Π(ξ)\lambda(\xi)\Pi(\xi) is what the permeable boundary condition (3) contributes, and is derived in appendix § 13. The relation to the incompressible equation of Baddoo et al. [4] is a change of dependent variable together with the loss of the Hankel remainder: they carry the bound vorticity γ\gamma, so their algebraic term appears as γ+ik∫−1ξγ\gamma+ ik\int_{-1}^{\xi}\gamma, a Volterra contribution which is exactly their pressure jump, and writing the equation in Π\Pi from the outset absorbs it. Setting ke=0k_e=0 removes both Hankel terms, and what remains of K\mathcal{K} is precisely the bounded term generated by the change of dependent variable; the present equation and theirs are therefore the same equation written in different variables. Both are dominant-singular equations with a variable coefficient, which is why they share edge exponents.

The forcing

W\mathcal{W} is the forcing of (5) and the only place the incident flow enters. It is the physical upwash w(ξ)w(\xi), normalised on its amplitude, with the sign it carries in (3). Equation (5) is written for the transformed potential of appendix § 14, φ=Ψe−iMkeξ\varphi=\Psi e^{-iMk_e\xi}, so ww enters it multiplied by eiMkeξe^{iMk_e\xi}:

W(ξ)=−w(ξ)eiMkeξ,w(ξ)={eiσ(ξ+1)convected gust (Sears),1heave (Theodorsen),1−iσ(ξ−ξp)pitch about ξ=ξp.(6)\mathcal{W}(\xi)=-w(\xi)e^{iMk_e\xi},\qquad w(\xi)= \begin{cases} e^{i\sigma(\xi+1)} & \text{convected gust (Sears)},\\ 1 & \text{heave (Theodorsen)},\\ 1-i\sigma(\xi-\xi_p) & \text{pitch about }\xi=\xi_p. \end{cases} \tag*{(6)}

The heave upwash is normalised on the plunge velocity and the pitch upwash on UU times the pitch amplitude, whose two terms are the incidence and the pitch rate. The sign is the statement that the surface must cancel the imposed upwash, up to whatever seepage λ\lambda permits. For the gust the two exponentials combine, σ+Mke=kˉ\sigma+Mk_e=\bar{k}, into −ei(kˉξ+σ)-e^{i(\bar{k}\xi+\sigma)}: in the transformed variables the gust convects at kˉ\bar{k}, the wavenumber at which the wake convects in every case. In the heave and pitch problems too, kˉ\bar{k} enters through the wake and through the loading Π=−∂m/∂ξ+ikˉm\Pi=-\partial m/\partial\xi+i\bar{k}m of (19), not through the forcing. The gust phase is referred to the leading edge, ξ=−1\xi=-1, and not to the midchord: this costs nothing in the frequency domain but is essential when the transfer function is inverted to the time domain, since a midchord reference makes the response begin before the gust arrives. The indicial responses are not a fourth forcing: they are the heave and gust responses carried into the time domain (§ 3.4). In every case the Kutta and Kelvin closures of § 3.2 are unchanged.

The same factor appears, inverted, in the output. Π\Pi is the transformed loading; the physical pressure jump, upper surface minus lower, is Π(ξ)e−iMkeξ\Pi(\xi)e^{-iMk_e\xi}, and the lift is

L=−∫−11Π(ξ)e−iMkeξ dξ(7)L=-\int_{-1}^{1}\Pi(\xi)e^{-iMk_e\xi}\,\mathrm{d}\xi \tag*{(7)}

the sign because a lifting section carries the lower pressure on its upper surface. The steady lift of appendix § 17, CL=−2∫Π dξC_L=-2\int\Pi\,\mathrm{d}\xi, is the same statement. With the normalisation of appendix § 13, Δp=(2/βM)ρ0Uw0Πe−iMkeξ\Delta p=(2/\beta_M)\rho_0Uw_0\Pi e^{-iMk_e\xi}, the lift per unit span is (2/βM)ρ0Uw0bL(2/\beta_M)\rho_0Uw_0bL, so that on the basis used in § 6, the incompressible quasi-steady lift 2πρ0Uw0b2\pi\rho_0Uw_0b of the rigid plate, the load is L/(πβM)L/(\pi\beta_M). The factor has unit modulus, so it leaves the magnitude of the loading, and with it every exponent of § 2.5, unchanged; it does not enter the permeable term either, since (2) relates the seepage to the pressure jump at the same point and the factor multiplies both. It does change the lift once MkeMk_e is of order one, and a check that it has been carried consistently is the high-frequency limit of the heave response, which must approach the acoustic (piston) value, 2βM/(πM)2\beta_M/(\pi M) times the compressible quasi-steady lift (Table 2).

testreferenceerror
Basis at α=β=1/2\alpha=\beta=1/2cot⁡(ϑ/2)\cot(\vartheta/2), sin⁡(ϑ/2)\sin(\vartheta/2)5.8×10−135.8\times10^{-13}
Possio symbol, ∣ke∣=0.5|k_e|=0.5–101012ike2−ν2\frac{1}{2}i\sqrt{k_e^2-\nu^2}, closed form7.3×10−167.3\times10^{-16}
the same, as assembled in § 3the same, by quadrature1.7×10−51.7\times10^{-5}
Sears function, σ=0.1\sigma=0.1–88[J0+iJ1]C(σ)−iJ1[J_0+iJ_1]C(\sigma)-iJ_14.3×10−54.3\times10^{-5}
Heave response, σ=0.02\sigma=0.02–88C(σ)−iσ/2C(\sigma)-i\sigma/2 (incl. apparent mass)1.4×10−41.4\times10^{-4}
Heave, σ=40\sigma=40–5050, M=0.3M=0.3–0.70.7piston limit, and (26) for λ0=1\lambda_0=1–30302×10−32\times10^{-3} rel.
Steady lift, λ=0.02\lambda=0.02–200200CL=4παiβC_L=4\pi\alpha_i\beta, closed form (27)2.7×10−42.7\times10^{-4}
Seepage dragCd,seep=αiCLC_{d,\mathrm{seep}}=\alpha_i C_L, closed form (28)4.0×10−44.0\times10^{-4}
Porous quasi-steady, λ=0\lambda=0–882β2\beta, from the unsteady solver8.9×10−48.9\times10^{-4}
Junction invariance on a rigid platemust be exactly unity3.5×10−103.5\times10^{-10}
Edge exponent read off the solutionβ\beta (9), λ=0\lambda=0–88, ke≤6.7k_e\le6.73×10−33\times10^{-3} rel.
Junction exponent read off the solutionδ\delta (11), M=0M=0 and 0.50.58×10−38\times10^{-3} rel.
Generalised basis at α=1/2\alpha=1/2the rigid-trailing-edge basisbit-identical
Mode count, σ=10\sigma=10, M=0.7M=0.7NN default against N=160N=160<10−5<10^{-5}
Steady lift, λ=0\lambda=0–1010Baddoo et al. [4] eq. (3.1), solved independently6.5×10−56.5\times10^{-5}
Unsteady, ke→0k_e\to0, k=0.01k=0.01–1010Baddoo et al. [4], their published code0.09%0.09\%
Rigid, M=0.3M=0.3–0.70.7, σ=0.2\sigma=0.2–1010physical-variable solution, appendix G8×10−48\times10^{-4}
Permeable, λ0=1\lambda_0=1, 33, σ=0.5\sigma=0.5–1010the same8×10−68\times10^{-6}
Independent reimplementationMATLAB vs PYTHON3.1×10−103.1\times10^{-10}
Loading/potential identity(19), evaluated on every solve5.0×10−45.0\times10^{-4}

Table 2. Table 2: Verification tests. αi\alpha_i is the incidence, not the trailing-edge exponent. C(σ)=H1(1)(σ)/[H1(1)(σ)−iH0(1)(σ)]C(\sigma)=H_1^{(1)}(\sigma)/[H_1^{(1)}(\sigma)-iH_0^{(1)}(\sigma)] is Theodorsen’s function in the e−iωte^{-i\omega t} convention, Jn=Jn(σ)J_n=J_n(\sigma), and the Sears phase is referred to midchord. The second row is the most sensitive test of the quadrature; the last is the only one available at no cost on every solve.

An oblique gust, with spanwise wavenumber k3k_3, does not require a new formulation either: the factor eik3x3e^{\mathrm{i}k_3x_3} is a parameter of the two-dimensional problem and enters only by replacing the acoustic wavenumber with k~e=(b/βM)k02/βM2−k32\tilde{k}_e = (b/\beta_M)\sqrt{k_0^2/\beta_M^2-k_3^2}, the branch chosen so that Im⁡k~e≥0\operatorname{Im}\tilde{k}_e \ge0. This matters for the acoustic use of the theory and is recorded here for completeness; all results below are for k3=0k_3 = 0.

Edge conditions

Equation (5) is a dominant-singular equation with a variable coefficient. Its solution is not smooth, and the local exponents are set by the coefficient.

Near an edge the bounded part of the kernel is subdominant and (5) reduces to the dominant equation on a half-line with constant coefficient. This is a Riemann–Hilbert problem [32, 12]. Writing the sectionally analytic function Ω(z)=(2πi)−1∫Π(s)/(s−z) ds\Omega(z) = (2\pi\mathrm{i})^{-1}\int\Pi(s)/(s-z)\,\mathrm{d}s and applying the Plemelj formulae gives Ω+=GΩ−+\Omega^+ = G\Omega^- + (forcing) with

G=λ+iλ−i.(8)G = \frac{\lambda+\mathrm{i}}{\lambda-\mathrm{i}}. \tag*{(8)}

Setting λ=cot⁡(πβ)\lambda= \cot(\pi\beta) makes G=exp⁡(2πiβ)G = \exp(2\pi\mathrm{i}\beta), and the canonical solution then behaves as (1+ξ)−β(1+\xi)^{-\beta} at the leading edge. Hence

β=1πcot⁡−1λ(−1)=12−arctan⁡λ(−1)π.(9)\beta= \frac{1}{\pi}\cot^{-1}\lambda(-1) = \frac{1}{2} - \frac{\arctan\lambda(-1)}{\pi}. \tag*{(9)}

At the trailing edge the same construction with the admissible (bounded) branch gives the exponent of the vanishing rather than of the singular behaviour,

Π∼(1−ξ)+α,α=1πcot⁡−1λ(1),(10)\Pi\sim(1-\xi)^{+\alpha}, \qquad\alpha= \frac{1}{\pi}\cot^{-1}\lambda(1), \tag*{(10)}

so that a rigid aft section, λ(1)=0\lambda(1) = 0, recovers the classical (1−ξ)1/2(1-\xi)^{1/2}, and a permeable one vanishes more slowly. Both exponents are local: each is evaluated with the value of λ\lambda at its own edge, and neither depends on the rest of the distribution.

Two limits check (9). As λ→0\lambda\to0, β→1/2\beta\to1/2 and the classical inverse-square-root singularity is recovered. As λ→∞\lambda\to\infty, β→0\beta\to0 and the loading is bounded: a fully transparent edge scatters nothing and carries no suction. Between them β\beta falls monotonically (Figure 2). For complex λ\lambda, as for a material with inertia, (9) is analytic except at the branch points λ=±i\lambda= \pm\mathrm{i}. These lie on the imaginary axis and are reached only by a lossless inertive layer, at the single frequency where ∣λ∣=1|\lambda| = 1; any resistance, Re⁡λ>0\operatorname{Re}\lambda> 0, keeps λ\lambda clear of them.

Figure 2. (a) The leading-edge exponent β=π−1cot⁡−1λ\beta= \pi^{-1}\cot^{-1}\lambda of (9). (b) The loading singularity (1+ξ)−β(1+\xi)^{-\beta} it produces. A Glauert basis imposes β=1/2\beta= 1/2, valid only for λ≪1\lambda\ll1.

Compressibility does not change the exponents, and this is not automatic. The compressible kernel of (5) is not a Cauchy kernel. What the argument above uses is that its non-Cauchy part K\mathcal{K} is bounded (appendix § 14), so it is subdominant in the local analysis and cannot enter GG. The exponents are therefore fixed by the dominant singular part alone, are independent of kek_e, and coincide in form with the incompressible exponents of Baddoo et al. [4]; Hajian and Jaworski [18] reached the corresponding steady result by the same route. The Mach number reaches the exponents only through λ\lambda, and more weakly than λ=2AM/βM\lambda= 2\mathcal{A}M/\beta_M suggests. Since A=ρ0c0/Zd\mathcal{A} = \rho_0c_0/Z_d the explicit MM cancels, and for the same UU, ω\omega and material what remains is the Prandtl–Glauert factor, λ=λ0/βM\lambda= \lambda_0/\beta_M, whatever the material (§ 2.6).

The junction exponent. At an interior point sjs_j where λ\lambda jumps, such as the junction between a porous insert and the rigid remainder or between two materials, the same argument applies with a piecewise-constant coefficient. The Riemann–Hilbert coefficient jumps from Gleft=exp⁡(2πiβleft)G_{\mathrm{left}} = \exp(2\pi\mathrm{i}\beta_{\mathrm{left}}) to Gright=G_{\mathrm{right}} = exp⁡(2πiβright)\exp(2\pi i\beta_{\mathrm{right}}), and the canonical function acquires a local exponent

Π∼∣ξ−sj∣−δ,δ=12πilog⁡GrightGleft=βright−βleft,(11)\Pi\sim\lvert\xi-s_j\rvert^{-\delta},\qquad\delta=\frac{1}{2\pi i}\log\frac{G_{\mathrm{right}}}{G_{\mathrm{left}}}=\beta_{\mathrm{right}}-\beta_{\mathrm{left}}, \tag*{(11)}

with βleft,right=π−1cot⁡−1λ(sj∓)\beta_{\mathrm{left,right}}=\pi^{-1}\cot^{-1}\lambda(s_j^{\mp}). For the common case of a porous insert ending on a rigid aft section this is δ=1/2−β>0\delta=1/2-\beta>0: moving downstream into a less permeable section makes the loading singular, and the junction acts as the leading edge of the aft section. Moving downstream into a more permeable section, as from a rigid forward section onto a porous aft one, gives δ=β−1/2<0\delta=\beta-1/2<0, and the loading vanishes there as ∣ξ−sj∣1/2−β\lvert\xi-s_j\rvert^{1/2-\beta}, as at a trailing edge, as found by Baddoo et al. [4]. For λ→0\lambda\to0 on both sides, δ→0\delta\to0 and the junction disappears, which is why inserting a junction into a rigid plate must change nothing; this property is used as a numerical test in § 4. For λ→∞\lambda\to\infty on the forward side, δ→1/2\delta\to1/2, the inverse-square-root singularity of a rigid leading edge. Since 2δ<12\delta<1 the squared loading remains integrable and the energy is finite.

The permeability parameter

Because λ\lambda is the only route by which the material enters, its parametric dependence can be read off before (5) is solved. Two statements follow, and both are testable.

The semi-infinite version of (5) admits the Wiener–Hopf kernel

K~(χ)=γ(χ)−iλ(χ+kˉ),γ(χ)=χ2−ke2,(12)\widetilde{K}(\chi)=\gamma(\chi)-i\lambda(\chi+\bar{k}),\qquad\gamma(\chi)=\sqrt{\chi^2-k_e^2}, \tag*{(12)}

in which χ\chi is the chordwise wavenumber scaled on bb, for the transform ∫f(ξ)eiχξ dξ\int f(\xi)e^{i\chi\xi}\,\mathrm{d}\xi, and the material enters only through iλi\lambda. The permeable term vanishes at χ=−kˉ\chi=-\bar{k}, the wavenumber of the frozen gust in the transformed variables, because a disturbance convected with the stream carries no pressure jump and so drives no seepage. This is a property of the kernel, not of the finite-chord response, which depends on the whole kernel and on the edges. Using k0=Mk1k_0=Mk_1,

λ=2iμβMk1=iμUπβMf,μ=−ik0A,(13)\lambda=\frac{2i\mu}{\beta_M k_1}=\frac{i\mu U}{\pi\beta_M f},\qquad\mu=-ik_0\mathcal{A}, \tag*{(13)}

in which the Mach number appears only through βM\beta_M. Substituting (13) into (12) gives K~=γ(χ)+(2μ/βMk1)(χ+kˉ)\widetilde{K}=\gamma(\chi)+(2\mu/\beta_Mk_1)(\chi+\bar{k}), which tends to ∣χ∣+2μb+(2μ/k1)χ\lvert\chi\rvert+2\mu b+(2\mu/k_1)\chi as M→0M\to0 at fixed μ\mu and k1k_1. The last term is the convective part of the boundary condition; the static porous half-plane kernel of Jaworski and Peake [22], ∣χ∣+2μb|\chi|+2\mu b, is recovered only when it is dropped, as at U=0U=0. The group controlling how porous a surface appears to a gust is the ratio of the porosity parameter μ\mu to the hydrodynamic wavenumber k1=ω/Uk_1=\omega/U, not to the acoustic one. For a material whose μ\mu does not depend on frequency, such as an inertive perforate, ∣λ∣|\lambda| falls as 1/f1/f, and ∣λ∣=1|\lambda|=1, where the seepage and flow terms of (5) carry equal weight and β=1/4\beta=1/4, defines a porous cut-off frequency fc=∣μ∣U/(πβM)f_c=|\mu|U/(\pi\beta_M), linear in speed, above which the surface appears increasingly rigid. For a lossless perforate this is the branch point λ=±i\lambda=\pm i of §2.5; a resistive layer, whose μ\mu is proportional to frequency, has no cut-off.

What compressibility does and does not do to λ\lambda. Write the material law through the dimensional transfer impedance Zd=−Δp/vsZ_d=-\Delta p/v_s, so that A=ρ0c0/Zd\mathcal{A}=\rho_0c_0/Z_d. Then (4) becomes

λ=2ρ0UβMZd=λ0βM,λ0=2ρ0UZd,(14)\lambda=\frac{2\rho_0U}{\beta_M Z_d}=\frac{\lambda_0}{\beta_M},\qquad\lambda_0=\frac{2\rho_0U}{Z_d}, \tag*{(14)}

and the explicit factor MM in (4) is cancelled by A∝c0\mathcal{A}\propto c_0. When the compressible and incompressible theories are compared for the same flow, at the same UU, ω\omega and material with the sound speed as the only difference, λ0\lambda_0 is therefore fixed and the whole Mach dependence of the boundary condition is the Prandtl–Glauert factor, for any material: λ\lambda rises by 15%15\% at M=0.5M=0.5 and by 40%40\% at M=0.7M=0.7. That is the comparison made in §5. Everything else compressibility does acts through the kernel and the Prandtl–Glauert transformation, that is through ke=Mσ/βM2k_e=M\sigma/\beta_M^2, kˉ=σ/βM2\bar{k}=\sigma/\beta_M^2 and the 1/βM1/\beta_M of the quasi-steady lift. The factor 1/βM1/\beta_M in λ\lambda has the same size as the Prandtl–Glauert rise of the rigid lift and the opposite effect, since a larger λ\lambda means a weaker load, and §5 shows that the two cancel in the steady, strongly permeable limit. The frequency dependence of λ0\lambda_0 is set by the material. A resistive layer, Zd=RZ_d=R, has a real λ0\lambda_0 independent of frequency. An inertive perforate, Zd=−iωmaZ_d=-i\omega m_a with mam_a an effective mass per unit area, has λ0=2iρ0b/(σma)\lambda_0=2i\rho_0b/(\sigma m_a), imaginary and varying as 1/σ1/\sigma. A layer with both, the form of the material law of Baddoo et al. [4], passes from the first behaviour to the second as the frequency rises. If the Mach number is instead raised by raising UU in the same fluid, λ0∝U\lambda_0\propto U for a resistive layer, and λ0\lambda_0 is fixed at fixed σ\sigma for an inertive one.

Acoustic non-compactness. The kernel carries the second compressible parameter, ke=Mσ/βM2k_e=M\sigma/\beta_M^2 of (1), the acoustic phase across the half-chord. It enters through the Hankel functions of (B.2): the upwash induced at ξ\xi by the loading at ss arrives with the phase of a sound wave that has crossed the distance between them, so for ke∣ξ−s∣≳1k_e|\xi-s|\gtrsim1 the kernels oscillate like eike∣ξ−s∣e^{ik_e|\xi-s|} and different parts of the chord act on one another with phase lags of order a radian or more. The term in ke2H0(1)k_e^2H_0^{(1)}, which acts on the potential jump itself, has no incompressible counterpart. Incompressible flow is the limit c0→∞c_0\to\infty, in which k0=0k_0=0, the Hankel kernels reduce to the Cauchy kernel and the retardation disappears; what remains is the hydrodynamic scale kˉ\bar{k} of the convected gust and the wake. The two compressible parameters are independent. The Mach number enters the boundary condition through λ0/βM\lambda_0/\beta_M, and the forcing through the phase eiMkeξe^{iM k_e\xi} of (6); kek_e enters the kernel and does not depend on the material at all. At low Mach number the first is negligible and the second is not, since ke≈Mσk_e\approx M\sigma is of order one whenever σ\sigma is of order 1/M1/M (§5.3).

Material-class scaling. Equation (13) makes ∣λ∣|\lambda| depend on the material only through μ\mu, and the two classes scale oppositely. For an inertive perforate μ\mu is a purely geometric inverse length, λ\lambda is imaginary, and

∣λ∣=∣μ∣cπβMfcU,(15)|\lambda|=\frac{|\mu|c}{\pi\beta_M}\sqrt{\frac{fc}{U}}, \tag*{(15)}

a function of the Strouhal number fc/Ufc/U and the perforate geometry alone. For a resistive bulk layer Zt≃rh/(ρ0c0)Z_t \simeq rh/(\rho_0 c_0) carries no factor of k0k_0, so μ∝k0\mu\propto k_0 and

∣λ∣=2ρ0UβMrh,(16)|\lambda| = \frac{2\rho_0 U}{\beta_M r h}, \tag*{(16)}

frequency-independent and linear in UU; this is the parameter by which Geyer et al. [13] organised sixteen porous aerofoils. A perforate therefore collapses on fc/Ufc/U and a bulk medium on ρ0U/(rh)\rho_0U/(rh); the two predictions are of opposite sign and follow from the same boundary condition.

Solution

The weighted-Jacobi basis

Equation (5) is solved by collocation in a weighted-Jacobi basis built on the exponents of § 2.5. The idea of replacing the Chebyshev (Glauert) family by Jacobi polynomials whose weights carry the porosity-dependent exponents is due to Baddoo et al. [4], who introduced it for the incompressible problem and treated an interior discontinuity by splitting the chord there into two sections, each with its own weighted-Jacobi expansion carrying the junction exponent (their § 3.3). What is new here is that the method transfers unchanged to the compressible kernel, for the reason given in § 2.5. At a junction we keep a single basis on the whole chord and add two junction modes (18) that carry the same local exponent. The basis is

g0=(1−ξ)α(1+ξ)−β,gn=(1−ξ)α(1+ξ)1−βPn−1(α,1−β)(ξ),gK=(1+ξ2)1−β,(17)g_0 = (1-\xi)^\alpha(1+\xi)^{-\beta}, \qquad g_n = (1-\xi)^\alpha(1+\xi)^{1-\beta}\mathcal{P}_{n-1}^{(\alpha,1-\beta)}(\xi), \qquad g_K = \left(\frac{1+\xi}{2}\right)^{1-\beta}, \tag*{(17)}

with β\beta from (9) and α\alpha from (10), and the last member representing the wake. The aerofoil modes g0g_0 and gng_n all carry (1−ξ)α(1-\xi)^\alpha and so vanish at the trailing edge; the wake mode does not (gK(1)=1g_K(1)=1), because it carries the vorticity leaving the edge. The Kutta condition is therefore a property of the combination the solve selects rather than of each member separately, and it is satisfied without an extra equation (§ 3.2). At α=β=1/2\alpha=\beta=1/2 the family collapses exactly onto Glauert’s, g0=cot⁡(ϑ/2)g_0=\cot(\vartheta/2) and gK=sin⁡(ϑ/2)g_K=\sin(\vartheta/2) with ξ=−cos⁡ϑ\xi=-\cos\vartheta, which is the first of the verification tests of § 4. At each junction two further modes are added,

q0=(1−ξ)α(1+ξ)1−β∣ξ−sj∣−δ,q1=q0sgn⁡(ξ−sj),(18)q_0=(1-\xi)^\alpha(1+\xi)^{1-\beta}|\xi-s_j|^{-\delta}, \qquad q_1=q_0\operatorname{sgn}(\xi-s_j), \tag*{(18)}

spanning the two-sided local solution of (11). Their envelope vanishes at both plate edges, so the leading-edge exponent and the Kutta condition are untouched.

Figure 3 shows what the exponents do to the basis. At β=1/2\beta=1/2 the members are the classical Glauert modes, reproduced here to 6×10−136\times10^{-13} (Table 2); at β=1/4\beta=1/4 every member is visibly weaker at the nose while remaining identical at the trailing edge. The difference cannot be produced by adding Glauert modes: no finite combination of functions singular as (1+ξ)−1/2(1+\xi)^{-1/2} reproduces (1+ξ)−1/4(1+\xi)^{-1/4}, which is why the exponent must be built in rather than resolved. The same argument applies at a porous trailing edge, where α<1/2\alpha<1/2, and at a junction, where the singularity is interior.

Figure 3. The basis, each mode normalised by its largest value away from the nose. (a) At β=1/2\beta= 1/2 the weighted-Jacobi family reduces exactly to Glauert’s. (b) At β=1/4\beta= 1/4 the nose behaviour is weaker, while the trailing edge is unchanged. (c) The two junction modes {(xref)}(3.2) carrying δ=1/2−β\delta= 1/2 - \beta. The trailing-edge exponent α\alpha enters the same way, through the weight (1−ξ)α(1-\xi)^\alpha common to every aerofoil mode.

The linear system

The working unknown. Equation (5) is written in Π\Pi, but Π\Pi is a derivative combination of the jump in velocity potential across the sheet. Let m(ξ)m(\xi) be that jump, scaled by the same constant as Π\Pi so that Γ=m(1)\Gamma=m(1) is the circulation. Substituting into Δp=−ρ0(∂t+U∂x1)[φ]\Delta p=-\rho_0(\partial_t+U\partial_{x_1})[\varphi] with the e−iωt\mathrm{e}^{-\mathrm{i}\omega t} convention gives

Π(ξ)=−∂m∂ξ+ikˉ m(ξ),(19)\Pi(\xi)=-\frac{\partial m}{\partial\xi}+\mathrm{i}\bar{k}\,m(\xi), \tag*{(19)}

in which γ=−∂m/∂ξ\gamma= -\partial m/\partial\xi is the bound vortex-sheet strength. The two terms carry opposite signs: the first is the quasi-steady loading a stationary vortex sheet would carry, the second the contribution of the unsteady rate of change of the potential jump. Their partial cancellation is why the unsteady lift falls below its quasi-steady value as kˉ\bar{k} grows, and the Sears and Theodorsen checks of {(xref)}§4 test exactly this combination. The potential jump is expanded in preference to Π\Pi for three reasons.

  1. Both mm and ∂m/∂ξ\partial m/\partial\xi appear. In compressible flow the normal velocity the sheet induces at ξ\xi has two parts: a vortex-sheet part driven by ∂m/∂ξ\partial m/\partial\xi, which supplies the Cauchy kernel together with an H1(1)H_1^{(1)} correction, and a compressibility part driven by mm itself, of order ke2H0(1)k_e^2 H_0^{(1)}. The second vanishes as ke→0k_e \to0, which is why the incompressible problem closes on the vorticity alone and the compressible one does not.

  1. The wake and Kelvin’s theorem are statements about Γ\Gamma, which is a functional of mm and is carried here as an explicit unknown.

  1. The time-domain inversion of {(xref)}§3.4 and the radiation integral both act on mm, so solving in mm needs no further differentiation of the answer.

Expansion. The basis {(xref)}(3.1) and the junction modes {(xref)}(3.2) span ∂m/∂ξ\partial m/\partial\xi, which carries the singular part of Π\Pi and therefore the exponents of {(xref)}§2.5; the expansion of mm itself uses their antiderivatives, taken from the trailing edge, Gn(ξ)=−∫ξ1gnG_n(\xi) = -\int_\xi^1 g_n, GK(ξ)=−∫ξ1gKG_K(\xi) = -\int_\xi^1 g_K and Qjk(ξ)=−∫ξ1qkQ_{jk}(\xi) = -\int_\xi^1 q_k, all of which vanish at ξ=1\xi= 1 so that m(1)=Γm(1) = \Gamma. Then

\begin{aligned} m(\xi) &= \Gamma+ \sum_{n=0}^{N} c_n G_n(\xi) + c_K G_K(\xi) + \sum_j \left(c_{j0}Q_{j0}(\xi) + c_{j1}Q_{j1}(\xi)\right),\\ \frac{\partial m}{\partial\xi} &= \sum_{n=0}^{N} c_n g_n(\xi) + c_K g_K(\xi) + \sum_j \left(c_{j0}q_0(\xi;s_j) + c_{j1}q_1(\xi;s_j)\right), \end{aligned} \qquad c_K = i\bar{k}\Gamma, \tag*{(20)}\end{aligned}

and Π\Pi follows pointwise from {(xref)}(3.3). The wake coefficient is not independent: the wake carries mw(ξ)=Γeikˉ(ξ−1)m_w(\xi) = \Gamma e^{i\bar{k}(\xi-1)} for ξ>1\xi> 1, so continuity of ∂m/∂ξ\partial m/\partial\xi across ξ=1\xi= 1, which means no concentrated vortex at the edge, fixes cKgK(1)=cK=ikˉΓc_K g_K(1)=c_K=i\bar{k}\Gamma. That single tie makes the Kutta condition automatic rather than imposed, because (19) then gives

Π(1)=−cK+ikˉΓ=0(21)\Pi(1)=-c_K+i\bar{k}\Gamma=0 \tag*{(21)}

identically, for every Γ\Gamma, every porosity and every trailing-edge exponent α>0\alpha>0. The GnG_n are available in closed form through a Jacobi identity; the QjkQ_{jk} are not, because qkq_k carries the interior algebraic singularity, and they are evaluated once per junction by cumulative quadrature on a graded grid. The unknown vector is {c0,…,cN,Γ,cj0,cj1}\{c_0,\ldots,c_N,\Gamma,c_{j0},c_{j1}\}: N+2+2njN+2+2n_j numbers for njn_j junctions, with N≃1.6kˉ+2ke+12N\simeq1.6\bar{k}+2k_e+12 set automatically and raised where the admittance distribution has interior kinks (§ 4).

Collocation and closure. Substituting (20) into (5) and enforcing it at nc=N+1+2njn_c=N+1+2n_j points ξi=−cos⁡ϑi\xi_i=-\cos\vartheta_i, ϑi=(i−12)π/nc\vartheta_i=(i-\frac{1}{2})\pi/n_c, gives ncn_c rows; the variable coefficient enters simply as λ(ξi)\lambda(\xi_i) on the diagonal, which is why an arbitrary λ(ξ)\lambda(\xi) costs nothing. Kelvin’s circulation theorem supplies the one remaining row. For a closed material contour that at t→−∞t\to-\infty lay in undisturbed fluid and encloses the aerofoil together with all the vorticity it has shed, the circulation cannot change ([5]), so

∫−11∂m∂ξ dξ⏟bound+(−Γ)⏟shed=0⟺m(−1)=0.(22)\underbrace{\int_{-1}^{1}\frac{\partial m}{\partial\xi}\,\mathrm{d}\xi}_{\text{bound}}+\underbrace{(-\Gamma)}_{\text{shed}}=0\quad\Longleftrightarrow\quad m(-1)=0. \tag*{(22)}

The two auxiliary conditions act at opposite ends of the plate and play distinct roles: Kutta fixes the behaviour of the solution at ξ=+1\xi=+1, through the basis; Kelvin fixes the amount of circulation at ξ=−1\xi=-1, through the one extra row. The result is a square (N+2+2nj)(N+2+2n_j) system, solved directly.

Two consequences follow. First, the cost is that of the classical rigid problem: the same number of unknowns, the same quadrature, one dense solve. Porosity changes the exponents in the basis functions and the diagonal, not the size or the structure of the matrix. Second, a basis without the right exponents fails in a specific way: omitting (18) removes the only columns able to represent ∣ξ−sj∣−δ|\xi-s_j|^{-\delta}, and the remaining columns then fit the collocation data as well as they can, so the error stops falling well short of zero, at a level set by the missing function. That is the 1–5 % stall found in § 4, and it is a statement about the span of the basis rather than about the number of modes.

Quadrature

No closed-form Hilbert transform of the weighted-Jacobi family is needed. The Cauchy singularity is removed by subtraction,

∫−11g(ξ)s−ξ dξ=∫−11g(ξ)−g(s)s−ξ dξ+g(s)log⁡1+s1−s,(23)\int_{-1}^{1}\frac{g(\xi)}{s-\xi}\,\mathrm{d}\xi = \int_{-1}^{1}\frac{g(\xi)-g(s)}{s-\xi}\,\mathrm{d}\xi +g(s)\log\frac{1+s}{1-s}, \tag*{(23)}

after which Gauss–Legendre panels graded towards the singular points suffice. The panels must additionally be subdivided so that no panel spans more than Δϑ=πnp/(4N)\Delta\vartheta=\pi n_p/(4N), with npn_p Gauss points per panel, otherwise the cos⁡(Nϑ)\cos(N\vartheta) oscillation of the high-order modes is aliased; without this refinement the junction-invariance test of § 4 returns a 7 % error. Details are in appendix D.

Indicial responses

The Wagner and Küssner functions are the time-domain counterparts of the Theodorsen and Sears functions, and they are what an aeroelastic, gust-load or blade–vortex-interaction calculation actually requires. They are obtained here by inverting the computed transfer function rather than by solving a separate problem. Let L(σ)L(\sigma) be the lift response to the harmonic forcing of (6) at reduced frequency σ\sigma, normalised on its quasi-steady value. The indicial response to a step of the same kind is

ϕ(τ)=2π∫0∞Re⁡L(σ)σsin⁡(στ) dσ,(24)\phi(\tau) = \frac{2}{\pi}\int_{0}^{\infty}\frac{\operatorname{Re} L(\sigma)}{\sigma}\sin(\sigma\tau)\,\mathrm{d}\sigma, \tag*{(24)}

τ=Ut/b\tau= Ut/b being semichords travelled, the standard inversion [6] valid because LL is the transfer function of a causal, stable, real system. Two properties of the present formulation matter for it. The gust phase is referred to the leading edge (6), so the Küssner response starts at τ=0\tau= 0 rather than at τ=−1\tau= -1; and the apparent-mass part of the heave response, which is an impulse at τ=0\tau= 0 and −iσ/2-i\sigma/2 in the rigid limit, is removed before inverting, since Wagner’s function describes what is left. Both the subtraction and the tail of (24) beyond the computed range are given in appendix § 18; the end-to-end accuracy of the scheme is the 0.007 with which it reproduces the classical Wagner function (§ 6.6).

Verification

Table 2 lists the verification tests. They fall into three groups: tests of the machinery that the rigid theory shares (the first six rows), tests of the porous terms against closed forms and against the exponents they are built on (the next eight), and tests of the assembled solver against independent implementations (the next five); the last row is a consistency check made on every solve. The sixth row is the only classical check made at finite Mach number: at high reduced frequency the loading of a heaving plate approaches the acoustic value Δp=2ρ0c0w\Delta p = 2\rho_0 c_0 w, which is 2βM/(πM)2\beta M/(\pi M) times the compressible quasi-steady lift. The test is sensitive to the Prandtl–Glauert factor of (6) and (7): omitted from both, it returns 1.28 at M=0.5M = 0.5 in place of 1.10. For a permeable plate the same argument, with flow through the material as a second path in parallel with the piston motion, gives the limit (26) of § 6.2, which the solver reproduces to 10−410^{-4} at σ=50\sigma= 50.

The compressible operator. All the classical checks except the piston limit are made at ke→0k_e \to0, where the ke2H0(1)k_e^2 H_0^{(1)} term and the regularised H1(1)H_1^{(1)} remainder vanish, so they do not test the compressible part of the operator. That part is verified directly, through its Fourier symbol: on an unbounded sheet, a potential jump m=eiνsm = e^{i\nu s} must induce the normal velocity w=12irmw = \frac{1}{2}i r m, with r=(ke2−ν2)1/2r = (k_e^2 - \nu^2)^{1/2} and Im⁡r≥0\operatorname{Im} r \ge0, which is the exact response of the Helmholtz problem (B.1). Applying (B.2) to m=eiνsm = e^{i\nu s}, with γ=−iνm\gamma= -i\nu m and the transforms

∫−∞∞H0(1)(ke∣u∣)e−iνu du=2r,∫−∞∞H1(1)(ke∣u∣)sgn⁡(u)e−iνu du=−2iνker,\int_{-\infty}^{\infty} H_0^{(1)}(k_e|u|)e^{-i\nu u}\,\mathrm{d}u = \frac{2}{r}, \qquad\int_{-\infty}^{\infty} H_1^{(1)}(k_e|u|)\operatorname{sgn}(u)e^{-i\nu u}\,\mathrm{d}u = \frac{-2i\nu}{k_e r},

gives (−iν2+ike2)/(2r)=12ir(-i\nu^2 + ik_e^2)/(2r) = \frac{1}{2}ir, as required. Because the identity must hold for every ν\nu, it tests each Hankel term and their relative coefficient separately, which a comparison with a tabulated lift coefficient would do only indirectly. It is satisfied to 7×10−167 \times10^{-16} in closed form, and to 1.7×10−51.7 \times10^{-5} by quadrature over ∣ke∣=0.5|k_e| = 0.5–10 and ν=0.1\nu= 0.1–8, both directly and through the split into a Cauchy principal value, the regularised H1(1)H_1^{(1)} and ke2H0(1)k_e^2 H_0^{(1)} that § 3 assembles. End to end, the solver approaches the classical Sears function continuously as ke→0k_e \to0 at fixed kˉ\bar{k}: the difference falls from 5.5×10−25.5 \times10^{-2} at ke=0.1k_e = 0.1 to 1.3×10−31.3 \times10^{-3} at ke=10−2k_e = 10^{-2}, at the ke2log⁡kek_e^2 \log k_e rate of the small-argument expansion of the kernel, and reaches the solver’s own floor of 4×10−54 \times10^{-5} by ke=10−3k_e = 10^{-3}. Figure 4 shows the resulting agreement with the Sears function over σ=0.1\sigma= 0.1–8.

Figure 4. The rigid, incompressible limit against the Sears function. Lines, the classical function [35]; symbols, the present solver at ke→0k_e \to0, with the phase referred to midchord. The difference is below 5×10−55 \times10^{-5} over σ=0.1\sigma=0.1–88.

An independent solution at finite Mach number. The Fourier-symbol test verifies the operator and the ke→0k_e\to0 tests verify its limit, but neither tests the assembled compressible solution, and the Prandtl–Glauert factor of (6) and (7), which is invisible at M=0M=0, has so far been tested only in the high-frequency heave limit. The assembled solution is therefore compared with a second solution of the same physical problem (appendix G). Its unknown is the physical pressure jump rather than the transformed potential, it has no wake and no Prandtl–Glauert transformation, and the kernel is applied through its exact Fourier symbol in physical variables; only the edge exponents of § 2.5 are common to the two methods. The absolute difference in the lift is at most 8×10−48\times10^{-4} for a rigid plate over M=0.3M=0.3–0.70.7 and σ=0.2\sigma=0.2–1010, and at most 8×10−68\times10^{-6} for permeable plates with λ0=1\lambda_0=1 and 33 over σ=0.5\sigma=0.5–1010, in magnitude and phase and for both the gust and the heave forcing. The larger rigid-plate difference comes from truncating the wavenumber integral of the second method (appendix G), not from the first.

The incompressible limit: comparison with Baddoo et al. [4]. Setting ke→0k_e\to0 in (5) removes the ke2H0(1)k_e^2H_0^{(1)} terms and the Possio kernel collapses to the Cauchy kernel alone, so the equation becomes the incompressible singular integral equation of Baddoo et al. [4] with λ\lambda in the role of their porosity parameter; the exponents (9), (10) and (11) are independent of kek_e and so are already common to both. Their results are therefore the ke→0k_e\to0 limit of the present solver, and the results of § 6 are its continuation in kek_e. That claim is checked in three increasingly demanding ways.

First, structurally, in the steady limit: their equation (3.1) was discretised from scratch in their variable, the bound vorticity, and in their basis, sharing no code with § 3, and its circulation reproduces Γ(ψ)/Γ(0)=2β\Gamma(\psi)/\Gamma(0)=2\beta to 6.5×10−56.5\times10^{-5}.

Second, in the unsteady quasi-steady limit, where the solver is asked for limits it was not constructed to satisfy: with λ=0\lambda=0 a heave forcing returns C(σ)−iσ/2C(\sigma)-\mathrm{i}\sigma/2 to 3.2×10−43.2 \times10^{-4} over σ≤5\sigma\le5, so the circulatory and non-circulatory parts of the classical response are both recovered; with λ>0\lambda>0 the quasi-steady limit of the unsteady solution returns 2β2\beta to 8.9×10−48.9 \times10^{-4} over λ=0\lambda=0–88, for heave and gust forcing alike.

Third, and most stringently, against their published code itself (https://github.com/baddoo/unsteady-porous-aerofoils), run on the porosity profiles exactly as their figure 10 is generated: an impermeable leading edge, Ψ=1/Φ\Psi=1/\Phi ramping linearly to 0.50.5 over 0.80.8 of the chord from a∈{0.25,0,−0.25,−0.5}a \in\{0.25, 0, -0.25, -0.5\} and constant thereafter, so that the trailing edge is porous, and ρe=1.5\rho_e=1.5. Their ψ\psi is written for the time convention e+ikte^{+\mathrm{i}kt}, so in the present convention the same material is its complex conjugate; the porosity supplied to the present solver and the lift it returns were both converted accordingly before the two codes were compared. Figure 5 shows the result. The ratio ∣Lporous/Limpermeable∣\lvert L_{\mathrm{porous}}/L_{\mathrm{impermeable}}\rvert, which is free of any normalisation or phase convention, agrees with their code to within 0.09%0.09\% on all four profiles over k=0.01k=0.01–1010, with no growth in kk (0.050.05, 0.070.07, 0.070.07 and 0.09%0.09\% for the four ramps); the impermeable control agrees to 2.4×10−52.4 \times10^{-5}. This is point-by-point agreement with an independent implementation of the limiting case, on configurations chosen by its authors rather than by us, and it includes a porous trailing edge, α≈0.15\alpha\approx0.15, which exercises (10). The features Baddoo et al. [4] report are reproduced with it: ∣S∣\lvert S\rvert grows with permeable length, and the porous phase curves cross the impermeable one, here at k≈1.4k \approx1.4–2.22.2, the crossover moving to lower frequency as the permeable length grows (Figure 5c). On the impermeable curve the present solver reproduces their closed form (6.3) to 4.6×10−64.6 \times10^{-6}.

Figure 5. The present solver at ke→0k_e \to0 (lines) against the published code of Baddoo et al. [4] (open symbols), on the porosity profiles of their figure 10: an impermeable leading edge, Ψ=1/Φ\Psi= 1/\Phi rising linearly from zero at x=ax = a to 0.5 at x=a+0.8x = a + 0.8, ρe=1.5\rho_e = 1.5; black, impermeable. Each curve is normalised at k=10−2k = 10^{-2}, as their figure is. (a) Argand diagram, (b) magnitude and (c) phase of the porous Sears function, laid out as their figure 10. (d) The relative difference between the two codes in the ratio ∣Lporous/Limpermeable∣|L_{\mathrm{porous}}/L_{\mathrm{impermeable}}|, which is independent of normalisation and time convention; the dotted line marks 0.1 %. Both codes at their converged basis size: N=60N = 60 here, N=⌊25+20k⌋N = \lfloor25 + 20\sqrt{k}\rfloor in theirs.

Convergence and the junction modes. The decisive convergence result concerns the junction modes (Figure 6). For a permeable insert with λ=3\lambda=3 over the forward 30%30\% of the chord at σ=2\sigma=2, the lift computed without the junction modes wanders at the 11–5×10−25 \times10^{-2} level and does not settle as NN is raised from 2020 to 200200: the expansion converges, but not to the right answer, because a global basis cannot represent an algebraic singularity at an interior point. With them the error settles at a few times 10−310^{-3}, an order of magnitude better and still falling, and the remaining level is the price of representing ∣ξ−sj∣−δ\lvert\xi-s_j\rvert^{-\delta} with two modes rather than a local expansion. Raising the quadrature resolution of the junction modes by a factor of six changes nothing, so this is a basis limit and not a quadrature one. The

Figure 6. (a) Convergence of the lift with and without the junction modes (18), for a permeable insert over the forward 30 % of the chord, λ=3\lambda= 3, σ=2\sigma= 2; the reference is the same solver at N=280N = 280. (b) The complementary test: a junction placed on a rigid plate, where δ=0\delta= 0, must change nothing.

Results

asymmetry between the two curves is the numerical statement of (11): a basis that does not carry the junction exponent cannot converge to the solution, which is why Baddoo et al. [4] split the chord at a discontinuity and why the junction modes are added here. The complementary test is exact rather than approximate, and it is the sharpest test of the quadrature: a junction placed on a rigid plate, where δ=0\delta= 0 and the modes must vanish, must change nothing, so any residual is discretisation error. It changes the lift by 3.5×10−103.5 \times10^{-10} over kˉ=0.5\bar{k} = 0.5–4040. A smoothly varying λ(ξ)\lambda(\xi) with interior kinks, such as the ramps of Figure 5, converges more slowly than a piecewise-constant insert with explicit junction modes, so the default mode count must be raised there. N=60N = 60 suffices over k≤10k \le10, and raising the count from the default to 60 takes the comparison of Figure 5 from 0.63 % to 0.09 %. The remaining difference is at the level expected of two independent spectral discretisations rather than a property of either.

Results

All results below are for a uniformly permeable plate or for a permeable insert of length ℓ\ell at the leading edge, with λ\lambda quoted directly rather than through a particular material, so that they apply to any surface with the same λ\lambda. Unless stated otherwise the surface is a resistive layer: λ0=2ρ0U/R\lambda_0 = 2\rho_0 U/R is real and independent of frequency, and the compressible and incompressible responses are compared at the same UU, ω\omega and RR, so that λ=λ0/βM\lambda= \lambda_0/\beta_M by (14) and the sound speed is the only difference. A layer with inertia as well as resistance is considered in §5.7. Reduced frequency is the classical σ=ωb/U\sigma= \omega b/U throughout. Loads at finite Mach number are reported on a single basis, the incompressible quasi-steady lift of the rigid plate, 2πρ0Uw0b2\pi\rho_0Uw_0b per unit span, rather than on the compressible one, 2πρ0Uw0b/βM2\pi\rho_0Uw_0b/\beta_M. The Prandtl–Glauert rise is therefore kept, so that a rigid plate at low frequency reads 1/βM1/\beta_M rather than unity, and a change with Mach number is a change in the load itself.

The loading at a permeable edge

Figure 7 shows the chordwise loading. Three things in it are worth separating, because they are usually run together.

Figure 7. Chordwise loading, uniformly permeable plate, at (a) σ=1\sigma= 1 and (b) σ=10\sigma= 10; solid M=0M = 0, dashed M=0.5M = 0.5 with the same material, so that λ=λ0/βM\lambda= \lambda_0/\beta_M; colour and the legend give λ0\lambda_0. (c) A permeable insert over the forward 30 % of the chord, λ=3\lambda= 3, σ=2\sigma= 2: the nose singularity is weakened and a junction singularity ∣ξ−sj∣−δ\lvert\xi- s_j \rvert^{-\delta} appears at the dotted line.

The exponents are in the solution, not only in the basis. Fitting ∣Π∣∼(1+ξ)−β|\Pi| \sim(1+\xi)^{-\beta} over the three decades 10−5≤1+ξ≤10−210^{-5} \le1+\xi\le10^{-2} returns β=0.5002\beta= 0.5002, 0.3527, 0.1478 and 0.0396 for λ=0\lambda= 0, 0.5, 2 and 8, against 0.5000, 0.3524, 0.1476 and 0.0396 from (9), which is agreement to three figures; at σ=10\sigma= 10 and M=0.5M = 0.5, where ke=6.7k_e = 6.7, the agreement is still within 2 %. That the fit reproduces (9) at finite kek_e is the numerical counterpart of the analytical statement of §2.5: the compressible part of the kernel is subdominant at the edge and does not touch the exponent. At the junction of a finite insert the same fit returns δ=0.3943\delta= 0.3943 against 0.3976 from (11) at M=0M = 0, and 0.4096 against 0.4105 at M=0.5M = 0.5.

The same material gives a weaker singularity at higher Mach number. Because λ(M)=λ0/βM\lambda(M) = \lambda_0/\beta_M, a surface with λ0=2\lambda_0 = 2 presents λ=2.31\lambda= 2.31 at M=0.5M = 0.5, and the fitted exponent falls from 0.148 to 0.130. Nothing about the material has changed; the exponent has, by 12 %. A rigid edge has a Mach-independent singularity and a permeable one does not, which is a statement no static permeability condition can make.

The loading redistributes as much as it falls. Panels (a) and (b) are plotted logarithmically because the interesting comparison is not the peak but the shape: as λ\lambda rises, the nose loading collapses while the aft loading is left almost untouched, so the centre of pressure moves aft and the lift falls by less than the edge loading does. The finite insert in panel (c) makes the mechanism explicit. The insert weakens the nose, but the junction then carries ∣ξ−sj∣−δ\lvert\xi- s_j \rvert^{-\delta} and behaves as a second, weaker leading edge. That is why a treated nose does not simply remove the edge but moves a reduced version of it downstream. At ℓ/c=0.3\ell/c = 0.3, λ=3\lambda= 3 and σ=2\sigma= 2 the lift is 0.208 of the rigid quasi-steady value against 0.279 for a rigid plate at the same σ\sigma, a 25 % reduction for a treatment covering 30 % of the chord.

The porous Sears function at finite Mach number

What is held fixed as the Mach number varies is the flow speed, the frequency and the material, so that λ=λ0/βM\lambda= \lambda_0/\beta_M by (14) while ke=Mσ/βM2k_e = M\sigma/\beta_M^2 grows from zero. Figure 8 is the result, and it separates cleanly into what is classical and what is not.

Figure 8. The porous Sears function at finite Mach number for a resistive layer, compared at the same flow speed, frequency and material so that λ=λ0/βM\lambda= \lambda_0/\beta M. Every load is referred to the incompressible quasi-steady lift of the rigid plate. (a) ∣L∣|L| for λ0=3\lambda_0 = 3 (solid) and for a rigid plate (dashed), coloured by Mach number. (b) The fraction of the rigid load that survives the treatment; the M=0M = 0 curve is what an incompressible theory would predict. (c) The Mach factor ∣L(M)∣/∣L(0)∣|L(M)|/|L(0)| for λ0=0\lambda_0 = 0, 1 and 3, at M=0.7M = 0.7 (solid) and 0.5 (dashed). (d) The two channels at λ0=3\lambda_0 = 3 against Mach number: the full response (filled symbols), the response with only λ(M)\lambda(M) varying, the response with only the kernel varying, and the rigid plate, for which only the kernel acts; circles σ=0.5\sigma= 0.5, triangles σ=5\sigma= 5.

The rigid response is the classical one. Compressibility raises the rigid load at low frequency by the Prandtl–Glauert factor, 1/βM=1.151/\beta_M = 1.15 at M=0.5M = 0.5 and 1.40 at M=0.7M = 0.7, and collapses it once the chord is no longer acoustically compact: at σ=10\sigma= 10 the load at M=0.5M = 0.5 is 0.32 of its incompressible value, and at σ=20\sigma= 20 it is 0.25 (panel a, dashed). Above σ≈5\sigma\approx5 the collapse carries shallow interference minima, at σ≈12.4\sigma\approx12.4 for M=0.3M = 0.3, 8.8 and 18 for M=0.5M = 0.5, and 7.1 and 14.8 for M=0.7M = 0.7, spaced so that 2k0b/(1+M)2k_0b/(1+M), the phase a sound wave accumulates in crossing the chord with the stream, increases by very nearly 2π2\pi from one to the next. None of this is new. It is the behaviour of Possio’s problem that the similarity rules of Graham [17], the approximate theories of Osborne [33], Amiet [1] and Kemp and Homicz [24], and the compressible indicial functions used in aeroelastic and rotor analysis (Leishman [28]) all describe; the present solver reproduces it, and the independent solution of appendix G reproduces the minima to 2×10−32 \times10^{-3}. It is the reference against which the permeable results are read.

The permeable response depends on Mach number through its permeability. Panel (c) shows the Mach factor ∣L(M)∣/∣L(0)∣\lvert L(M)\rvert/\lvert L(0)\rvert for λ0=0\lambda_0 = 0, 1 and 3, and further cases at λ0=0.5\lambda_0 = 0.5 and 8 complete the picture. At M=0.7M = 0.7 and σ=5\sigma= 5 the load is 0.43 of its incompressible value for a rigid plate, 0.53 at λ0=0.5\lambda_0 = 0.5, 0.66 at λ0=1\lambda_0 = 1, 0.86 at λ0=3\lambda_0 = 3 and 0.95 at λ0=8\lambda_0 = 8: compressibility changes the load of a weakly to moderately permeable surface by between a tenth and a half, and that of a very permeable one hardly at all. At λ0=3\lambda_0 = 3 the three compressible curves of panel (a) lie close to the incompressible one; over M≤0.7M \le0.7 and σ≤20\sigma\le20 the load changes by at most 16 % (near σ=3.8\sigma= 3.8 at M=0.7M = 0.7), and by 1–2 % in the quasi-steady limit, where the rigid load changes by 40 %. The sensitivity falls steadily as the surface becomes more permeable at high frequency and in the quasi-steady limit, where the same ratio is available in closed form (§ 6.7) and interpolates between the Prandtl–Glauert factor of the rigid plate and unity. Compressibility therefore matters for a permeable surface, but by an amount and with a frequency dependence that neither the rigid compressible theory nor the incompressible porous theory supplies.

Why: the load of a permeable surface is set by the material. In the strongly permeable limit the surface simply passes the imposed upwash: the disturbance potential vanishes, the seepage velocity in (2.3) balances the upwash, ∣vs∣=∣w∣|v_s| = |w|, and the pressure jump is whatever the material needs to drive it through. By (2.2),

∣Δp∣→ρ0c0∣A∣∣w∣=2ρ0U∣λ0∣∣w∣,(25)|\Delta p| \to\frac{\rho_0 c_0}{|A|}|w| = \frac{2\rho_0 U}{|\lambda_0|}|w|, \tag*{(25)}

the second form following from (2.4) with λ=λ0/βM\lambda= \lambda_0/\beta_M. The flow around the surface does not enter, and with it neither does compressibility: (5.1) contains no MM. On the common basis of § 6, the incompressible quasi-steady lift of the rigid plate, it gives ∣L∣→(2/πλ0)∣sin⁡σ/σ∣|L| \to(2/\pi\lambda_0)|\sin\sigma/\sigma| for the gust (Figure 8) and 2/(πλ0)2/(\pi\lambda_0) for heave (§ 6.4), at every Mach number. How the limit is approached is where compressibility enters, and for heave at high frequency it can be written down. Each face of the plate then radiates as a piston, with pressure ρ0c0\rho_0c_0 times its normal velocity. The imposed upwash is taken up partly by this motion and partly by flow through the material, both driven by the same pressure jump, so the acoustic and material admittances add: eliminating the face velocity between the piston relation and (2.3) gives

∣Δp∣→2ρ0Uλ0+M∣w∣,∣L∣2πρ0Uw0b→2π(λ0+M),(26)|\Delta p| \to\frac{2\rho_0 U}{\lambda_0 + M}|w|,\qquad\frac{|L|}{2\pi\rho_0 U w_0 b} \to\frac{2}{\pi(\lambda_0 + M)}, \tag*{(26)}

which is the piston value 2/(πM)2/(\pi M) for a rigid plate and the seepage limit as M/λ0→0M/\lambda_0 \to0. The argument is local and holds for any material law, with λ0\lambda_0 evaluated at the frequency in question: for a resistive layer λ0\lambda_0 is constant and the load tends to a plateau, whereas for an inertive one λ0→0\lambda_0 \to0 and the load tends to the piston value of a rigid plate (§ 6.8). The solver reproduces (5.2) to four figures at M=0.5M = 0.5 and σ=50\sigma= 50, and also at M=0.7M = 0.7 and σ=20\sigma= 20 for λ0=3\lambda_0 = 3, 8 and 30, where the load lies M/(λ0+M)=19M/(\lambda_0 + M) = 19, 8 and 2 % below the seepage limit. The high-frequency Mach sensitivity is therefore of first order in M/λ0M/\lambda_0, whereas the steady one (§ 6.7) is of second order. For the gust the leading term is approached more slowly as σ\sigma grows, since the edges then carry more of the load, but its Mach independence is inherited: at λ0=30\lambda_0 = 30 the gust load changes by at most 1.4 % between M=0M = 0 and 0.7. The same limit explains the shoulder in every permeable curve near σ=π\sigma= \pi, where the chord average of the gust, and with it the leading term, vanishes. A rigid plate is the opposite extreme: its load is set entirely by the flow around it, which is where compressibility acts.

The two channels act in opposite directions. Panel (d) separates them at fixed σ\sigma, on the same basis. The boundary-condition channel, in which λ\lambda rises as 1/βM1/\beta_M while the kernel is held incompressible, lowers the permeable load by 22–24 % at M=0.7M = 0.7, almost independently of σ\sigma. The kernel channel, which carries the Prandtl–Glauert rise, raises it by 23 % at σ=0.5\sigma= 0.5 and by 5 % at σ=5\sigma= 5; at σ=5\sigma= 5 the same kernel, acting on a rigid plate, lowers the load by 57 %. At λ0=3\lambda_0 = 3 the two cancel in part. The full change is 4 % at σ=0.5\sigma= 0.5, 7 % at σ=2\sigma= 2 and 14 % at σ=5\sigma= 5, where the kernel’s rise has faded, and the product of the two channels reproduces the full response to within 4 % for σ≤2\sigma\le2 and 6 % at σ=5\sigma= 5. The cancellation improves as the surface becomes more permeable: the full change is 2–5 % at λ0=8\lambda_0 = 8 over σ≤20\sigma\le20, and it is exact in the steady, strongly permeable limit (§ 6.7).

What the treatment is worth. The surviving fraction ∣Lporous/Lrigid∣\lvert L_{\mathrm{porous}}/L_{\mathrm{rigid}}\rvert, the load of the treated plate divided by that of the rigid plate at the same σ\sigma and MM, is shown in panel (b). It is a ratio whose numerator hardly depends on Mach number and whose denominator carries the whole classical collapse. It therefore inherits that collapse. At low reduced frequency the incompressible theory is pessimistic: it gives 0.236 at σ=0.1\sigma= 0.1 where M=0.5M = 0.5 gives 0.217, and 0.332 against 0.301 at σ=1\sigma= 1. The largest difference is the quasi-steady one: the compressible fraction is then the incompressible one times 2β(λ0/βM)/2β(λ0)2\beta(\lambda_0/\beta_M)/2\beta(\lambda_0), which is 0.87 at M=0.5M = 0.5 and 0.73 at M=0.7M = 0.7. The curves cross near σ=2.6\sigma= 2.6 at M=0.5M = 0.5 (1.5 at M=0.7M = 0.7, 4.9 at M=0.3M = 0.3), and above that the incompressible theory is increasingly optimistic: at σ=10\sigma= 10 it predicts that 7.6 % of the rigid load survives where M=0.5M = 0.5 gives 21.6 %. Over the non-compact range σ=5\sigma= 5–20, it underestimates the surviving fraction by a factor of between 1.4 and 4.2 at M=0.5M = 0.5 and between 2.0 and 4.3 at M=0.7M = 0.7. The peaks of the compressible curves above σ=5\sigma= 5 sit just below the interference minima of the rigid response.

Can existing theories be combined? Because the permeable load depends on Mach number less than the rigid one, it is tempting to form the surviving fraction from two theories that already exist: the treated load from an incompressible porous theory, which is the ke→0k_e \to0 limit of (2.5) solved by Baddoo et al. [4], and the reference load from a compressible rigid one. Figure 9 tests that shortcut against the full solution. The rigid load is then exact, so what is plotted is the error of the porous numerator, in two versions: the incompressible theory used as it stands, with the same λ0\lambda_0 (solid), and with the permeability parameter corrected by hand to λ0/βM\lambda_0/\beta_M (dashed). At low frequency, where the chord is compact, the first version is within about 20 %. Over the non-compact range σ=5\sigma= 5–20 it overestimates the surviving fraction, and so understates the benefit, by factors of 1.45–1.96 at λ0=0.5\lambda_0 = 0.5 and 1.26–1.52 at λ0=1\lambda_0 = 1 when M=0.7M = 0.7 (1.34–1.68 and 1.20–1.35 at M=0.5M = 0.5), by 12–16 % at λ0=3\lambda_0 = 3 and by at most 6 % at λ0=8\lambda_0 = 8. Correcting λ\lambda reduces the error for a weakly permeable surface but overcorrects a strongly permeable one, whose surviving fraction it underestimates by up to 35 % at λ0=8\lambda_0 = 8 and M=0.7M = 0.7. It applies the boundary-condition channel of Figure 8(d) without the kernel channel that offsets it, and in the seepage limit it returns βM\beta_M times the true load. The two versions err in opposite directions, and which of them is closer to the full solution depends on λ0\lambda_0, MM and σ\sigma, so the better one cannot be identified without computing the compressible solution itself.

Image not extracted

Figure 9. Can existing theories be combined? The surviving fraction ∣Lporous/Lrigid∣|L_{\mathrm{porous}}/L_{\mathrm{rigid}}| estimated from an incompressible porous theory and a compressible rigid one, divided by the full compressible result, for resistive layers at (a) M=0.5M=0.5 and (b) M=0.7M=0.7. Solid, the incompressible porous load with the same λ0\lambda_0; dashed, with λ0/βM\lambda_0/\beta_M. Values above unity understate the load reduction, values below overstate it; the shaded band is ±10%\pm10\%.

When the shortcut is adequate follows from the high-frequency limit (5.2). For heave at high frequency the incompressible porous load exceeds the compressible one by exactly (λ0+M)/λ0(\lambda_0 + M)/\lambda_0, so the shortcut is within 10 % only if λ0≳10M\lambda_0 \gtrsim10M, that is λ0≳5\lambda_0 \gtrsim5–7 at M=0.5M = 0.5–0.7; in steady flow the error is of second order in M/λ0M/\lambda_0, by (5.6). Only for a strongly permeable surface, therefore, can compressibility be left to the rigid reference. For weakly and moderately permeable surfaces the compressible porous theory is needed for the load itself, and for every permeability it is needed for what a ratio of lift magnitudes does not contain: the phase, the chordwise loading and the Mach-dependent edge exponent of § 6.1, and the indicial responses of § 6.6.

Low Mach number, high frequency

The comparisons so far vary the Mach number at reduced frequencies of up to 20. The compressible parameter of the kernel is ke=Mσ/βM2k_e = M\sigma/\beta_M^2 (§ 2.6), so a chord is acoustically non-compact at low Mach number too once σ\sigma approaches 1/M1/M. Figure 10 takes M=0.05M = 0.05, at which λ=λ0/βM\lambda= \lambda_0/\beta_M exceeds λ0\lambda_0 by 0.1 % and the phase MkeM k_e of (6) is at most 0.3, so that the boundary condition and the forcing are effectively incompressible, and extends the gust response to σ=120\sigma= 120, where ke=6k_e = 6 and the chord spans about two acoustic wavelengths. What separates each pair of curves is then the acoustic non-compactness of the chord alone.

Image not extracted

Figure 10. Acoustic non-compactness at low Mach number: resistive layers at M=0.05M = 0.05, compared with the incompressible solution at the same UU, ω\omega and material (the same solver at ke=10−3k_e = 10^{-3}). (a) The surviving fraction ∣Lporous/Lrigid∣|L_{\mathrm{porous}}/L_{\mathrm{rigid}}| for a uniformly permeable plate and a leading-edge insert over 30 % of the chord, λ0=3\lambda_0 = 3; solid M=0.05M = 0.05, dashed incompressible. (b) The ratio of the compressible to the incompressible surviving fraction for uniform plates and for inserts over 15 % and 30 % of the chord, and the ratio of the rigid loads (dash-dot); the upper axis gives kek_e.

The rigid load. Compressibility raises the rigid load by 12 % at σ=10\sigma= 10 and by 25 % at σ=20\sigma= 20, and then collapses it, to 0.61 of its incompressible value at σ=40\sigma= 40 and 0.26 at σ=120\sigma= 120 (panel b, dash-dot): the low-Mach counterpart of the behaviour of § 6.2, and the regime of the high-frequency theory of Amiet [1].

A uniformly permeable plate. Its load is set by the material, as in § 6.2, and the seepage limit (25) contains no kek_e: the compressible and incompressible loads agree to 3 % over the whole range at λ0=1\lambda_0 = 1 and to 1 % at λ0=3\lambda_0 = 3. The surviving fraction therefore inherits the non-compactness of the rigid reference entirely. The compressible fraction is 0.88–0.89 times the incompressible one at σ=10\sigma= 10 and 0.78 times it at σ=20\sigma= 20, and then 1.6, 3.0 and 3.7 times it at σ=40\sigma= 40, 60 and 120, for both permeabilities.

A leading-edge insert. Here the porous load is itself non-compact. The leading edge, the junction and the rigid aft section carry loads that act on one another with the retarded phase ke∣ξ−s∣k_e|\xi-s|, and over σ=20\sigma=20–120120 the lift of the insert departs from its incompressible value by factors between 0.550.55 and 1.021.02 for ℓ/c=0.3\ell/c=0.3, and between 0.180.18 and 1.331.33 for ℓ/c=0.15\ell/c=0.15, whose load nearly cancels near σ=38\sigma=38. The surviving fraction follows neither the incompressible curve nor the rigid reference. For the 30% insert it is 0.71 of the incompressible value at σ=20\sigma=20 and 1.8–3.0 times it over σ=60\sigma=60–120120; for the 15% insert it is 1.3–2.1 times it over σ=60\sigma=60–120120, and 0.28 of it at the near-cancellation.

Non-compactness therefore makes compressibility a low-speed effect as well. At M=0.05M=0.05 the surviving fraction departs from the incompressible prediction by more than 10% from σ≈10\sigma\approx10–1616, that is ke≈0.5k_e\approx0.5–0.80.8, and by factors of 1.3 to 3.8 over σ=60\sigma=60–120120 in every configuration computed. An incompressible porous theory describes the loading only where Mσ≪1M\sigma\ll1, whatever the speed.

The porous Theodorsen function

For heave the physical upwash is the constant w=1w=1 of (6), and the rigid incompressible response is Theodorsen’s C(σ)−iσ/2C(\sigma)-\mathrm{i}\sigma/2, which the solver returns to 1.4×10−41.4\times10^{-4} (Table 2). The porous responses are shown in Figure 11. As in § 6.2, the surface is a resistive layer, with λ0\lambda_0 real and independent of frequency. For an inertive perforate λ0∝i/σ\lambda_0\propto\mathrm{i}/\sigma vanishes at high frequency, the surface then behaves as a rigid one, and the high-frequency end of these curves is not reached (§ 6.8).

Image not extracted

Figure 11. Heave response of a permeable plate, λ0\lambda_0 independent of frequency; solid M=0M = 0, dashed M=0.5M = 0.5. (a) ∣L∣|L| referred to the incompressible quasi-steady lift of the rigid plate; the rigid incompressible curve follows σ/2\sigma/2 at high frequency, the rigid compressible one saturates at the piston value, and the permeable ones tend to the high-frequency limit 2/(π(λ0+M))2/\left(\pi(\lambda_0+M)\right) of (26), which at M=0M = 0 is the seepage limit 2/(πλ0)2/(\pi\lambda_0), marked by the short bars (solid M=0M = 0, broken M=0.5M = 0.5). (b) The surviving fraction ∣Lporous/Lrigid∣|L_{\mathrm{porous}}/L_{\mathrm{rigid}}|: porosity acts as a low-pass filter on the load. (c) The two ends of the frequency range against λ0\lambda_0: the quasi-steady ratio is the steady closed form 2β2\beta, while at σ=50\sigma= 50 the surviving fraction is one to two orders of magnitude smaller at M=0M = 0, and only 14–27 % smaller at M=0.5M = 0.5, where compressibility has already capped the rigid response.

Porosity removes the apparent mass almost entirely. The rigid response grows without bound, ∣L∣→σ/2|L|\to\sigma/2, because accelerating a rigid plate accelerates the fluid attached to it. A permeable plate cannot sustain the pressure jump that would require: the seepage term λΠ\lambda\Pi grows with the loading, and the response tends to a constant instead. At M=0M=0 and σ=50\sigma=50 the rigid value is 25.0 and the porous ones are 0.630, 0.212 and 0.0795 for λ0=1\lambda_0=1, 3 and 8: fractions of 2.5×10−22.5\times10^{-2}, 8.5×10−38.5\times10^{-3} and 3.2×10−33.2\times10^{-3}. The constants are the seepage limit ((25)), 2/(πλ0)=0.6372/(\pi\lambda_0)=0.637, 0.212 and 0.0796, to within 1%; at high frequency the load of a resistive plate is set by the material, which has no inertia to add, and not by the fluid that a rigid plate would have to accelerate. A layer with inertia keeps part of its

Indicial responses

apparent mass (§ 6.8). At the other end of the frequency range the quasi-steady limit is the steady closed form ((27)), and the solver returns it to four figures: at σ=10−4\sigma= 10^{-4}, 0.5000, 0.2048 and 0.0792 against 2β=0.50002\beta= 0.5000, 0.2048 and 0.0792; at M=0.5M = 0.5 the ratio to the rigid plate at the same Mach number is the same closed form evaluated with λ0/βM\lambda_0/\beta_M, to the same accuracy. Between those two ends lies the whole content of panel (c): for a resistive layer in incompressible flow porosity is far more effective against the non-circulatory load than against the circulatory one: at λ0=3\lambda_0 = 3 the apparent mass falls to 0.3 % of its rigid value while the circulation keeps 20 %. For flutter this is the useful direction, since apparent mass is what couples plunge to pitch inertia, and it is the opposite of what the steady lift ratio alone would suggest.

Compressibility caps the rigid response and moves the permeable one much less. At M=0.5M = 0.5 the rigid heave response no longer grows like σ/2\sigma/2: it saturates at the acoustic (piston) value 2/(πM)=1.272/(\pi M) = 1.27 of § 2.4, the classical result that at high reduced frequency each element of the surface radiates as a piston. The permeable responses tend to the high-frequency limit ((26)) instead (panel a): at σ=50\sigma= 50 they are 0.42, 0.18 and 0.075 for λ0=1\lambda_0 = 1, 3 and 8, equal to 2/(π(λ0+M))2/(\pi(\lambda_0 + M)) to four figures, against 0.630, 0.212 and 0.0795 at M=0M = 0. The change, M/(λ0+M)M/(\lambda_0 + M), is 33, 14 and 6 %, and it is smaller the more permeable the surface, as in § 6.2. The apparent mass that porosity would have removed has therefore already been removed by compressibility, and the surviving fractions at σ=50\sigma= 50 rise to 0.33, 0.14 and 0.059, thirteen to nineteen times the incompressible fractions, so that the benefit is an order of magnitude smaller than the incompressible calculation reports. This is the Sears result of § 6.2 again: a nearly Mach-independent permeable load measured against a rigid load that carries the whole of the classical compressibility effect.

Indicial responses

Figure 12 gives the porous analogues of the Wagner and Küssner functions, obtained from (24) as appendix F describes. Each is normalised by its own quasi-steady value, so the curves show how fast the load settles, not how large it is; the size is the steady factor 2β2\beta of (27), and the two statements should be read together.

Image not extracted

Figure 12. Indicial responses, each normalised by its own quasi-steady value. (a) The Wagner analogue: the response to a step in incidence, with the apparent-mass impulse removed as appendix § 18 describes. (b) The Küssner analogue: the response to a sharp-edged gust, with τ\tau measured from the instant the gust front reaches the leading edge. Solid M=0M=0, dashed M=0.5M=0.5; symbols, Jones’s approximations to the classical rigid functions at M=0M=0.

The rigid limit. At M=0M = 0 and λ=0\lambda= 0 the computed Wagner function agrees with Jones’s classical approximation to 0.007 over τ≤40\tau\le40, including the correct initial value ϕ(0+)=1/2\phi(0^+) = 1/2, which is not imposed anywhere and emerges from the high-frequency limit of the computed transfer function. The Küssner analogue agrees to 0.035 for τ≥2\tau\ge2 and departs by up to 0.095 below τ=1\tau= 1, where Jones’s two-exponential fit is itself least reliable.

Porosity shortens the transient as well as reducing the load. A permeable plate reaches 90 % of its steady gust response in τ=7.4\tau= 7.4 semichords at λ0=1\lambda_0 = 1 and 3.5 at λ0=3\lambda_0 = 3, against 14.3 for a rigid plate: the treatment removes most of the wake memory along with most of the load. The mechanism is the same one that sets 2β2\beta: less circulation is shed, so the wake exerts less influence on what follows. The consequence is that a permeable section responds to a passing vortex with a load that is both smaller and shorter, which is not what a quasi-steady estimate scaled by 2β2\beta would give.

Compressibility slows the response and changes how it starts. At M=0.5M = 0.5 the rigid Wagner analogue begins near the piston value, since the first instant after a step in incidence is the high-frequency limit of § 6.4; the computation gives 1.08 of the steady lift, against 1.10 in the limit. It then falls within two semichords to 0.58 and recovers more slowly than the incompressible function, reaching 90 % of its steady value at τ=18\tau= 18 rather than 13. The Küssner analogue is delayed: it has reached 0.10 of its steady value at τ=0.25\tau=0.25, against 0.22 incompressibly, and 90 % at τ=19.4\tau=19.4 against 14.3, because the gust front takes a finite time to make itself felt through a non-compact chord. The permeable cases inherit the delay in weaker form: at λ0=3\lambda_0=3 the Küssner rise time lengthens from 3.5 to 3.9 semichords. They do not inherit the raised start. A permeable plate has almost no apparent mass even at M=0M=0, so its Wagner analogue already begins near its steady value, and compressibility lowers that start, to 0.88 at λ0=3\lambda_0=3, because the piston-limited high-frequency response is then below the permeable steady lift. For a gust-encounter or blade–vortex interaction calculation the practical consequence is that the transient a permeable section produces at flight Mach number is neither the incompressible porous one nor the compressible rigid one.

The steady limit: lift and seepage drag

At ω=0\omega=0 the same boundary condition is a complete steady thin-aerofoil theory, and it closes in terms of the single exponent (9). For a uniformly porous plate at incidence αi\alpha_i the Riemann–Hilbert solution is a single power law and (appendix § 17)

CLCL,rigid=2β=1−2πarctan⁡λ,(27)\frac{C_L}{C_{L,\mathrm{rigid}}}=2\beta=1-\frac{2}{\pi}\arctan\lambda, \tag*{(27)}
Cd,seep=αiCL.(28)C_{d,\mathrm{seep}}=\alpha_i C_L. \tag*{(28)}

Equation (28) is written for a flat plate at incidence, where the physical upwash is the constant w=αiw=\alpha_i and W=−αi\mathcal{W}=-\alpha_i. For a general mean line the streamwise force is the loading worked against that upwash, which in the normalisation of appendix § 17, where CL=−2∫Π dξC_L=-2\int\Pi\,\mathrm{d}\xi, is

Cd,seep=2∫−11Π(ξ)W(ξ) dξ,(29)C_{d,\mathrm{seep}}=2\int_{-1}^{1}\Pi(\xi)\mathcal{W}(\xi)\,\mathrm{d}\xi, \tag*{(29)}

of which (5.4) is the constant-W\mathcal{W} case. The distinction matters wherever camber is used to buy the penalty back: a cambered section at its zero-lift incidence carries loading that integrates to zero but is not zero pointwise, and since the dissipation is quadratic in the loading, the cancellation that removes the lift leaves the drag untouched. Seepage drag vanishes with zero loading, not with zero lift.

Three statements follow, and all three are consequences of the edge exponent rather than of the material as such. First, the lift ratio is exactly twice the edge exponent, so one number sets the loss of lift and the weakening of the singularity together (Figure 13): half the lift is gone at λ=1\lambda= 1, where β=1/4\beta= 1/4. Second, a rigid plate at incidence carries a streamwise pressure-force component αiCL,rigid\alpha_i C_{\mathrm{L,rigid}} that is cancelled exactly by leading-edge suction, and that suction exists only because the loading is singular as (1+ξ)−1/2(1+\xi)^{-1/2}; a permeable edge carries (1+ξ)−β(1+\xi)^{-\beta} with β<1/2\beta< 1/2, so (1+ξ)Π2→0(1+\xi)\Pi^2 \to0 at the nose, there is no suction, and (28) says the streamwise component is dissipated entirely by seepage, at 2β2\beta times the suction 2παi22\pi\alpha_i^2 that the corresponding rigid edge would have developed. Third, the drag integrand behaves as (1+ξ)−2β(1+\xi)^{-2\beta}, integrable if and only if β<1/2\beta< 1/2, so the porosity-dependent exponent is what keeps the energy budget finite. The drag therefore falls as λ\lambda increases (Figure 13b): the seepage velocity grows, but the pressure jump that drives it falls faster, and the largest penalty, close to the full suction 2παi22\pi\alpha_i^2, is paid by a slightly permeable edge.

Image not extracted

Figure 13. The steady limit, uniformly permeable plate. (a) The lift ratio: the closed form 2β2\beta of (27) (line) against the collocation solver (symbols), over four decades of λ\lambda, from 0.02 to 200. (b) The seepage drag per unit incidence squared, 4πβ4\pi\beta from (28); the dotted line is the leading-edge suction 2π2\pi that a rigid edge would have recovered. Solver and closed forms agree to 4×10−44 \times10^{-4}.

Mach number. In compressible flow CL,rigid=2παi/βMC_{\mathrm{L,rigid}} = 2\pi\alpha_i/\beta_M, and (27) holds with the local value of λ\lambda. When the compressible and incompressible theories are compared at the same flow speed and material, λ=λ0/βM\lambda= \lambda_0/\beta_M by (14) for any material, and the lift therefore varies with Mach number as

CL(M)CL(0)=1−(2/π)arctan⁡(λ0/βM)1−(2/π)arctan⁡λ0(30)\frac{C_{\mathrm{L}}(M)}{C_{\mathrm{L}}(0)} = \frac{1-(2/\pi)\arctan(\lambda_0/\beta_M)} {1-(2/\pi)\arctan\lambda_0} \tag*{(30)}

which is the Prandtl–Glauert factor 1/βM1/\beta_M for a rigid plate and tends to unity as λ0→∞\lambda_0 \to\infty, where CL→4αi/λ0C_{\mathrm{L}} \to4\alpha_i/\lambda_0. In that limit the Prandtl–Glauert rise of the circulation is cancelled exactly by the rise of λ\lambda, and the lift is the seepage load (25) of a surface that passes the incidence upwash. At M=0.7M = 0.7 the steady lift rises by 10.6, 1.7 and 0.25 % for λ0=1\lambda_0 = 1, 3 and 8, against 40 % for a rigid plate, and the unsteady solver reproduces (30) to three figures in its quasi-steady limit. For a resistive layer the limit is the Darcy drop Δp=Rw\Delta p = Rw, which contains no Mach number.

For a permeable insert of length ℓ\ell at the leading edge the solver returns something simpler than the local analysis would suggest (Figure 14). Over ℓ/c=0.05\ell/c = 0.05–1 and λ=0.5\lambda= 0.5–10 the computed lift satisfies

Image not extracted

Figure 14. Steady penalty of a leading-edge insert. (a) Lift against treated length: symbols, the collocation solver with junction modes; lines, the linear law (31). (b) Seepage drag against lift, both per unit incidence; the identity Cd,seep=αiCLC_{d,\mathrm{seep}} = \alpha_i C_L is the dotted diagonal.

CLCL,rigid=1−ℓc(1−2β)(31)\frac{C_{\mathrm{L}}}{C_{\mathrm{L,rigid}}} = 1 - \frac{\ell}{c}(1-2\beta) \tag*{(31)}

to 10−510^{-5}. It is a linear interpolation between the rigid plate and the uniformly permeable closed form (27), with no dependence on where between the two the junction sits beyond its position. The lift penalty is therefore proportional to the treated chord fraction, which is the design statement one would want and is not obvious from a loading distribution that carries two algebraic singularities. The law is exact: for a constant upwash the steady Riemann–Hilbert problem with a piecewise-constant λ\lambda has a closed-form solution whose lift is linear in the junction position (appendix E), and the solver reproduces it to 10−510^{-5}.

The seepage-drag identity survives the junction. With the forcing normalised to unit incidence, Cd,seep=αiCLC_{d,\mathrm{seep}} = \alpha_i C_L becomes the statement that the two computed coefficients are equal, and they are: to four figures for a uniform plate, and to within 7 % for a partial insert, the largest departures being for short inserts at λ=3\lambda= 3, where the drag integrand is sampled right up to the junction singularity and converges most slowly.

A layer with inertia

A real perforate or porous layer has inertia as well as resistance, since the fluid in the pores must be accelerated, and the material law of Baddoo et al. [4] carries an effective density for that reason. Writing Zd=R−iωmaZ_d = R - \mathrm{i}\omega m_a in (14) gives

λ0(σ)=λR1−iσ/σm,λR=2ρ0UR,σm=RbUma,(32)\lambda_0(\sigma) = \frac{\lambda_R}{1-\mathrm{i}\sigma/\sigma_m}, \qquad\lambda_R = \frac{2\rho_0 U}{R}, \qquad\sigma_m = \frac{Rb}{U m_a}, \tag*{(32)}

resistive below σm\sigma_m and inertive above it, where ∣λ0∣|\lambda_0| falls as 1/σ1/\sigma. Figure 15 shows the gust response for λR=3\lambda_R = 3 and σm=2\sigma_m = 2: the resistive layer of §5.2 below σ≈2\sigma\approx2, and above it a surface whose effective permeability falls to ∣λ0∣=1.1|\lambda_0| = 1.1 at σ=5\sigma= 5 and 0.420.42 at σ=14\sigma= 14.

Image not extracted

Figure 15. Gust response of a resistive–inertive layer, λ0=λR/(1−iσ/σm)\lambda_0 = \lambda_R/(1 - i\sigma/\sigma_m) with λR=3\lambda_R = 3 and σm=2\sigma_m = 2, against the resistive layer with λ0=3\lambda_0 = 3 and the rigid plate. (a) The Mach factor ∣L(M)∣/∣L(0)∣|L(M)|/|L(0)| of each surface at M=0.7M = 0.7 (solid) and 0.50.5 (dashed). (b) The fraction of the rigid load at the same Mach number that survives, for the resistive–inertive layer at M=0M = 0, 0.50.5 and 0.70.7; dotted, the resistive layer at M=0M = 0.

Permeability still controls the sensitivity to compressibility. The Mach factor of panel (a) follows the resistive layer where the two have the same λ0\lambda_0, lying between 0.96 and 1.01 at M=0.7M = 0.7 for σ≤2\sigma\le2, and moves towards the rigid plate as ∣λ0∣|\lambda_0| falls: at M=0.7M = 0.7 it is 0.83 at σ=5\sigma= 5 and 0.65 at the top of the range, against 0.86–0.90 for the resistive layer and 0.2–0.4 for a rigid plate. The same statement covers both materials: the sensitivity of the load to compressibility is set by how permeable the surface is at the frequency in question. It also defeats the shortcut of § 6.2: at σ=20\sigma= 20 and M=0.7M = 0.7 the incompressible load of this layer is 1.5 times the compressible one, because the surface has become weakly permeable where the chord is non-compact.

Inertia removes most of the high-frequency benefit. Incompressibly the fraction of the rigid gust load that survives levels off at 0.27–0.31 for σ≥5\sigma\ge5, where the resistive layer falls to 0.05 (panel b). At M=0.5M = 0.5 and 0.7 it rises to between 0.4 and 1.0, 1.5 to 3.3 times the incompressible value, and near the interference minima of the rigid response the treatment achieves nothing. The heave response makes the same point: at M=0M = 0 this layer keeps 17 % of the rigid plate’s apparent mass at σ=20\sigma= 20–50, against 0.85 % at σ=50\sigma= 50 for the resistive layer with λ0=3\lambda_0 = 3, because its admittance falls with frequency and the high-frequency load is again supported by the flow. At M=0.5M = 0.5 it keeps 82–96 % of the rigid piston load over the same range, against 14 % for the resistive layer. The removal of apparent mass found in § 6.4 is a property of resistive layers.

Limits of validity

The results of §§§ 6.2–§ 6.7 are for resistive layers, and § 6.8 shows how inertia changes them; the formulation itself admits any admittance, complex and varying along the chord. The formulation is linear, subsonic and thin-aerofoil, so it inherits the limits of the Prandtl–Glauert transformation and the Possio kernel: a thin section, attached flow, small incidence, and no transonic region, which in practice means M≲0.7M \lesssim0.7. Nothing in the edge analysis degrades as MM rises, since the exponents are independent of kek_e, and the solver remains cheap there: at M=0.6M = 0.6 and kˉ=20\bar{k} = 20, so ke=12k_e = 12, the lift is converged to six figures at N=140N = 140. Two further conditions belong to the material rather than to the aerodynamics. The homogenised admittance requires many apertures per hydrodynamic wavelength, which is a statement about the perforate pitch and the frequency, and becomes demanding at the reduced frequencies of § 6.3, where σ=120\sigma= 120 puts the hydrodynamic wavelength at 2πb/σ≈0.05b2\pi b/\sigma\approx0.05b; and for a bulk medium the pore Reynolds number must be of order unity or below, since the leading correction to Darcy’s law is Forchheimer’s [11], in which the resistance grows with the seepage velocity and hence with the gust amplitude, an amplitude dependence that a linear frequency-domain theory cannot carry. The seepage limit (25) makes that condition more pressing rather than less: a surface that passes the imposed upwash carries seepage velocities of the order of the gust velocity itself.

Conclusions

A linear theory has been developed for the unsteady loading of finite-chord porous aerofoils in compressible subsonic flow. It combines Possio’s integral operator, written with its Cauchy singularity separated from a bounded remainder, with a convective permeable boundary condition, allows chordwise-varying admittance, and retains the shed wake and the unsteady Kutta condition; the material enters through the permeability parameter λ=2AM/βM\lambda= 2AM/\beta_M alone. Because the non-Cauchy part of the operator is bounded, the loading exponents at the aerofoil edges and at admittance discontinuities retain their incompressible functional form when expressed in terms of the local permeability parameter, and are independent of the reduced frequency. The weighted-Jacobi collocation method therefore carries over, with the trailing-edge exponent carried as a parameter and junction modes added where the admittance jumps. The solution reproduces the published incompressible code of Baddoo et al. [4] to 0.09 % and agrees with an independent compressible formulation in physical variables to 10−310^{-3} or better.

The harmonic gust and heave responses, together with their indicial counterparts, show that permeability controls the sensitivity of the unsteady loading to compressibility. The results are for resistive layers, compared at the same flow speed, frequency and material, so that λ=λ0/βM\lambda= \lambda_0/\beta_M for any material. The pressure jump across a permeable surface is set partly by the surrounding flow and partly by the material impedance, and compressibility acts on the first. As permeability increases the material governs an increasing share of the load, and the Mach-number dependence associated with the surrounding flow weakens: at M=0.7M = 0.7 and σ=5\sigma= 5 the gust load is 0.43 of its incompressible value for a rigid plate, 0.53 at λ0=0.5\lambda_0 = 0.5, 0.66 at λ0=1\lambda_0 = 1, 0.86 at λ0=3\lambda_0 = 3 and 0.95 at λ0=8\lambda_0 = 8. The weakening is gradual rather than complete: a weakly permeable surface keeps much of the Mach dependence of a rigid plate, and only a strongly permeable one loses it. The Mach number reaches the load through two channels that act in opposite directions. The boundary condition, through which a given material appears more permeable by 1/βM1/\beta_M, lowers the load by a fifth to a quarter at M=0.7M = 0.7, nearly independently of frequency; the integral operator, which carries the Prandtl–Glauert rise, raises it at low frequency and, unlike its effect on a rigid plate, does not collapse it at high frequency. In the time domain a permeable section responds faster as well as more weakly, reaching 90 % of its steady gust response in 3.5 semichords at λ0=3\lambda_0 = 3 against 14.3 for a rigid one, and compressibility lengthens these rise times to 3.9 and 19.4 semichords at M=0.5M = 0.5.

Closed-form steady and high-frequency limits describe this transition for resistive surfaces. In steady flow the lift varies with Mach number as (30), between the Prandtl–Glauert factor of a rigid plate and unity, and tends to CL=4αi/λ0C_L = 4\alpha_i/\lambda_0 as permeability increases, with a residual Mach dependence of second order in M/λ0M/\lambda_0. At high frequency the heave load tends to 2/(π(λ0+M))2/(\pi(\lambda_0 + M)) of the incompressible rigid quasi-steady lift: the material and acoustic (piston) resistances of the two faces act in parallel, since both are driven by the same pressure jump, giving the piston value for a rigid plate and the pressure drop set by the material impedance, ρ0c0w/A\rho_0 c_0 w/\mathcal{A}, for a strongly permeable one, with a residual Mach dependence of first order in M/λ0M/\lambda_0. Together the two limits state when compressibility can be ignored for the porous load: only when λ0≫M\lambda_0 \gg M. The same limit explains why, for a resistive layer in incompressible flow, porosity removes the apparent mass far more effectively than the circulation: the high-frequency load is set by a material that has no inertia to add. A layer with inertia as well as resistance becomes less permeable as the frequency rises; its Mach sensitivity then moves towards that of a rigid plate, it keeps part of its apparent mass, 17% in the case computed, and it loses most of its high-frequency benefit.

Although the porous load becomes less sensitive to Mach number, its ratio to the impermeable load remains strongly affected by compressibility, because the impermeable load follows the classical behaviour: a Prandtl–Glauert rise at low frequency and a collapse once the chord is acoustically non-compact, to a quarter of its incompressible value at σ=20\sigma= 20 and M=0.5M = 0.5. Incompressible theory overestimates the ratio at low frequency, most of all in steady flow, where the compressible ratio is the incompressible one times 2β(λ0/βM)/2β(λ0)2\beta(\lambda_0/\beta_M)/2\beta(\lambda_0), 0.87 at M=0.5M = 0.5 and 0.73 at M=0.7M = 0.7 for λ0=3\lambda_0 = 3; it underestimates the ratio once the chord is non-compact. For the gust cases examined at M=0.5M = 0.5–0.70.7 and σ=5\sigma= 5–2020 it underestimates the ratio by factors of 1.4–4.3 (7.6 % against 21.6 % at σ=10\sigma= 10 and M=0.5M = 0.5), thereby overstating the load reduction achieved by the treatment; for the layer with inertia the factor reaches 3.3, and near the interference minima of the rigid response the treatment achieves nothing. The heave response shows the same structure: the unbounded apparent mass of a rigid plate is capped by compressibility at its piston value, and the fraction of the rigid high-frequency load that a permeable plate retains at λ0=3\lambda_0 = 3 rises from under 1 % incompressibly to 14 % at M=0.5M = 0.5.

It follows that the existing theories cannot simply be combined. Taking the porous load from incompressible theory and the rigid load from compressible theory overestimates the surviving fraction, over σ=5\sigma= 5–2020 at M=0.7M = 0.7, by factors of up to 1.96 at λ0=0.5\lambda_0 = 0.5 and 1.52 at λ0=1\lambda_0 = 1; correcting the permeability parameter by 1/βM1/\beta_M overcorrects a strongly permeable surface, underestimating its surviving fraction by up to 35 % at λ0=8\lambda_0 = 8; and for the layer with inertia, whose permeability falls with frequency, the shortcut is 1.5 times too high at σ=20\sigma= 20. The high-frequency limit sets the condition for the shortcut, λ0≳10M\lambda_0 \gtrsim10M for 10 % accuracy, and even where that holds a ratio of lift magnitudes supplies neither the phase, nor the chordwise loading and the Mach-dependent edge exponent on which scattering depends, nor the indicial responses. A theory that carries permeability and compressibility together is therefore required for the load of a weakly or moderately permeable aerofoil, and for everything beyond the lift magnitude at any permeability.

Nor is compressibility only a high-speed effect. The kernel carries the acoustic compactness ke=Mσ/βM2k_e = M\sigma/\beta_M^2 separately from the Mach number of the boundary condition, and at M=0.05M = 0.05 the surviving fraction departs from its incompressible value by more than 10 % from σ≈10\sigma\approx10, and by factors of up to 3.8 at σ=120\sigma= 120, where the chord spans about two acoustic wavelengths. For a uniformly permeable plate this comes from the rigid reference alone; for a leading-edge insert the porous load is itself non-compact, the leading edge, the junction and the aft section acting on one another with a retarded phase. An incompressible porous theory therefore describes the loading only where Mσ≪1M\sigma\ll1, at any speed.

In the steady limit the lift ratio is CL/CL,rigid=2βC_L/C_{L,\mathrm{rigid}} = 2\beta and the seepage drag is Cd,seep=αiCLC_{d,\mathrm{seep}} = \alpha_i C_L, so the loss of lift and the weakening of the edge singularity are one number, and a permeable edge develops no leading-edge suction. For a leading-edge insert the lift penalty is, to 10−510^{-5}, linear in the treated fraction of the chord, CL/CL,rigid=1−(ℓ/c)(1−2β)C_L/C_{L,\mathrm{rigid}} = 1 - (\ell/c)(1 - 2\beta).

The formulation is thus a unified linear theory of the unsteady loading of a thin aerofoil, rigid or permeable, in incompressible or subsonic compressible flow, with the classical and incompressible porous results as its limits. It provides these loads at the cost of a rigid thin-aerofoil calculation, for assessing porous aerofoils and for supplying aerodynamic inputs to aeroelastic and aeroacoustic models.

Acknowledgements

The author used Claude (Anthropic), a large language model, in the versions Claude Opus 5, Claude Opus 5.5 and Claude Fable 5.1, accessed through the Claude desktop application, in the course of this work in September 2026. It was used to check the analytical derivations, including the signs and conventions of the governing equations, to write and run, under the author’s direction, the numerical scripts from which the results and figures were produced, to port and run the published code of Baddoo et al. [4] for the comparison of § 4, to check the bibliography against publisher records, and to assist in drafting and editing the text. Every derivation, numerical result and statement in the paper was examined and verified by the author, who takes full responsibility for its content. The model is not an author of this work.

Funding

This research received no specific grant from any funding agency, commercial or not-for-profit sectors.

Declaration of interests

The author reports no conflict of interest.

Data availability statement

The data and scripts that support the findings of this study are available from the corresponding author upon reasonable request.

Author ORCID

S. Lee, https://orcid.org/0000-0001-8081-2268.

The convective permeable boundary condition

Let φ\varphi be the disturbance velocity potential and [⋅][\cdot] the jump across the sheet. Three ingredients are combined. (i) Linearised momentum across the sheet gives Δp=−ρ0(∂t+U∂x1)[φ]\Delta p=-\rho_0(\partial_t+U\partial_{x_1})[\varphi]. (ii) Mass conservation requires the total normal velocity to be continuous and equal to the seepage, w+∂x2φ=vsw+\partial_{x_2}\varphi=v_s on both faces. (iii) The material law (2), vs=−AΔp/(ρ0c0)v_s=-\mathcal{A}\Delta p/(\rho_0c_0). Eliminating Δp\Delta p between (i) and (iii) and substituting into (ii) gives (3). The order matters: (ii) is the only ingredient carrying boundary information, and the seepage enters as the normal velocity, not as a second statement about Δp\Delta p.

To obtain the coefficient of Π\Pi in (5), write Δp=Cρ0Uw0Πe−iMkεξ\Delta p=C\rho_0Uw_0\Pi e^{-iMk_\varepsilon\xi} for whatever constant CC the normalisation of Π\Pi implies, the exponential being the Prandtl–Glauert factor of appendix § 14. Then (iii) contributes −ACMw0Πe−iMkεξ-\mathcal{A}CMw_0\Pi e^{-iMk_\varepsilon\xi} to the physical normal velocity, and so −ACMw0Π-\mathcal{A}CMw_0\Pi to the transformed one, which appears with a positive sign on the left of (5), while the flat-plate inversion of (ii) contributes (βMC/2π)∫Π(s)/(ξ−s) ds(\beta_M C/2\pi)\int\Pi(s)/(\xi-s)\,\mathrm{d}s after the Prandtl–Glauert transformation. Fixing the Cauchy coefficient at 1/π1/\pi requires C=2/βMC=2/\beta_M, whence λ=ACM=2AM/βM\lambda=\mathcal{A}CM=2\mathcal{A}M/\beta_M as claimed. Because λ\lambda is the ratio of the seepage coefficient to the Cauchy coefficient it is independent of how Π\Pi is normalised. Holding μ=−ik0A\mu=-ik_0\mathcal{A} fixed as M→0M\to0, in which case A=iμ/(Mk1)\mathcal{A}=i\mu/(Mk_1) diverges, keeps λ=2iμ/(βMk1)\lambda=2i\mu/(\beta_M k_1) finite. The resulting condition still contains the convective derivative, which the linearised Bernoulli relation carries in incompressible flow as well; the static porous half-plane condition of Jaworski and Peake [22] and Ayton et al. [3], ∂x2φ∝[φ]\partial_{x_2}\varphi\propto[\varphi], is recovered only when that derivative is dropped, as at U=0U=0.

From the convected wave equation to (2.5)

The operator that follows is classical: it is the subsonic unsteady kernel of Possio [34], and no novelty is claimed for it. It is set down here in a particular form, with the Cauchy singularity separated from a remainder that is shown to be bounded, because that separation is what the edge analysis of § 2.5 and the quadrature of § 3.3 both act on, and the classical representations, which carry the singularity inside the special functions, do not supply it.

Prandtl–Glauert. The disturbance potential satisfies c0−2(∂t+U∂x1)2φ=∇2φc_0^{-2}(\partial_t+U\partial_{x_1})^2\varphi=\nabla^2\varphi, which for e−iωt\mathrm{e}^{-\mathrm{i}\omega t} is βM2φx1x1+φx2x2+2iMk0φx1+k02φ=0\beta_M^2\varphi_{x_1x_1}+\varphi_{x_2x_2}+2\mathrm{i}Mk_0\varphi_{x_1}+k_0^2\varphi=0. Writing φ=Ψeiμx1\varphi=\Psi\mathrm{e}^{\mathrm{i}\mu x_1}, the coefficient of Ψx1\Psi_{x_1} vanishes for μ=−Mk0/βM2\mu=-Mk_0/\beta_M^2, leaving βM2Ψx1x1+Ψx2x2+(k0/βM)2Ψ=0\beta_M^2\Psi_{x_1x_1}+\Psi_{x_2x_2}+(k_0/\beta_M)^2\Psi=0. The stretch X=x1/βMX=x_1/\beta_M then gives the Helmholtz equation

ΨXX+Ψx2x2+κ2Ψ=0,κ=k0/βM,(33)\Psi_{XX}+\Psi_{x_2x_2}+\kappa^2\Psi=0,\qquad\kappa=k_0/\beta_M, \tag*{(33)}

on a plate of half-length b~=b/βM\tilde b=b/\beta_M. Normalising ξ=X/b~\xi=X/\tilde b gives ke=κb~=k0b/βM2k_e=\kappa\tilde b=k_0b/\beta_M^2 as in (1); the same bookkeeping on the gust, whose chordwise wavenumber becomes k1−μ=k1/βM2k_1-\mu=k_1/\beta_M^2, gives kˉ=k1b/βM2\bar k=k_1b/\beta_M^2 and hence ke=Mkˉk_e=M\bar k. Since X/b~=x1/bX/\tilde b=x_1/b, ξ\xi is also the physical chordwise coordinate, and μx1=−Mkeξ\mu x_1=-Mk_e\xi: the two potentials are related by φ=Ψe−iMkeξ\varphi=\Psi\mathrm{e}^{-\mathrm{i}Mk_e\xi}. The material derivative transforms as

(∂∂t+U∂∂x1)φ=Ube−iMkeξ(∂∂ξ−ikˉ)Ψ.\left(\frac{\partial}{\partial t}+U\frac{\partial}{\partial x_1}\right)\varphi=\frac{U}{b}\mathrm{e}^{-\mathrm{i}Mk_e\xi}\left(\frac{\partial}{\partial\xi}-\mathrm{i}\bar k\right)\Psi.

because ω−Uμ=ω/βM2\omega-U\mu=\omega/\beta_M^2. This is why the transformed loading (19) carries kˉ\bar k rather than σ\sigma, and why the physical pressure jump is Πe−iMkeξ\Pi\mathrm{e}^{-\mathrm{i}Mk_e\xi} and a physical upwash enters (5) multiplied by eiMkeξ\mathrm{e}^{\mathrm{i}Mk_e\xi} (§ 2.4).

The dipole sheet. With mm the potential jump and γ=−∂m/∂ξ\gamma=-\partial m/\partial\xi the bound vorticity, the loaded plate is a sheet of normal doublets with potential φ=−∂x2∫mG ds\varphi=-\partial_{x_2}\int mG\,\mathrm{d}s, which jumps by [φ]=φ(0+)−φ(0−)=m[\varphi]=\varphi(0^+)-\varphi(0^-)=m across the sheet, and the normal velocity it induces on itself is w=−∂x22∫mG dsw=-\partial_{x_2}^2\int mG\,\mathrm{d}s, with G=(i/4)H0(1)(κR)G=(\mathrm{i}/4)H_0^{(1)}(\kappa R) the outgoing Green’s function of (B.1). That integral is hypersingular as written; (B.1) itself removes the difficulty, since Gx2x2=−GXX−κ2GG_{x_2x_2}=-G_{XX}-\kappa^2G away from the source, so that w=∫m(GXX+κ2G) dsw=\int m(G_{XX}+\kappa^2G)\,\mathrm{d}s. Integrating the first part by parts once, with the boundary terms vanishing because m(−1)=0m(-1)=0 by Kelvin’s theorem (22) and the sheet continues into the wake at ξ=1\xi=1, and using dH0(1)/dz=−H1(1)\mathrm{d}H_0^{(1)}/\mathrm{d}z=-H_1^{(1)},

w(ξ)=ike4∫−11H1(1)(z)sgn⁡(ξ−s)γ(s) ds+ike24∫−11H0(1)(z)m(s) ds,z=ke∣ξ−s∣,(34)w(\xi)=\frac{\mathrm{i}k_e}{4}\int_{-1}^{1}H_1^{(1)}(z)\operatorname{sgn}(\xi-s)\gamma(s)\,\mathrm{d}s+\frac{\mathrm{i}k_e^2}{4}\int_{-1}^{1}H_0^{(1)}(z)m(s)\,\mathrm{d}s,\qquad z=k_e|\xi-s|, \tag*{(34)}

plus the wake integral. The second term carries ke2k_e^2 and has no incompressible counterpart.

The singular part. Since H1(1)(z)∼−2i/(πz)H_1^{(1)}(z)\sim-2\mathrm{i}/(\pi z), the first kernel tends to 1/[2π(ξ−s)]1/[2\pi(\xi-s)] as z→0z\to0: the classical Glauert kernel, coefficient and all. Splitting H1(1)(z)=−2i/(πz)+[H1(1)(z)+2i/(πz)]H_1^{(1)}(z)=-2\mathrm{i}/(\pi z)+[H_1^{(1)}(z)+2\mathrm{i}/(\pi z)], in which the bracket is bounded, turns (B.2) into a Cauchy principal value plus bounded terms.

Assembly. Adding the seepage contribution (appendix A) and changing the unknown to Π\Pi through Π=γ+ikˉm\Pi=\gamma+\mathrm{i}\bar k m gives (5). The change of dependent variable contributes one further term. On the wake Π=0\Pi=0, so that γ=−ikˉm\gamma=-\mathrm{i}\bar k m there, and the term is −(ikˉ/π)∫−1∞m(s)/(ξ−s) ds-(\mathrm{i}\bar k/\pi)\int_{-1}^{\infty}m(s)/(\xi-s)\,\mathrm{d}s, taken over the chord and the wake together. Over the chord alone it would not be bounded: m(1)=Γm(1)=\Gamma is not zero, and the end point s=1s=1 contributes a logarithm, log⁡∣1−ξ∣\log|1-\xi|, as ξ→1\xi\to1. The wake integral contributes the same logarithm with the opposite sign, because mm is continuous across the trailing edge, m(1−)=m(1+)=Γm(1^-)=m(1^+)=\Gamma, and the two cancel. Over [−1,∞)[-1,\infty) the function mm is Hölder-continuous, vanishes at the leading edge and is continuous at ξ=1\xi=1, so the term is bounded on the whole chord, trailing edge included. The Hankel terms acting on the wake are bounded for the same reason. Hence K\mathcal{K}, which includes the wake through the circulation, is a bounded operator, made up of the regularised H1(1)H_1^{(1)} acting on γ\gamma, the ke2H0(1)k_e^2H_0^{(1)} acting on mm and the Hilbert transform of mm over the chord and the wake, and the dominant singular part of (5) is kek_e-independent, which is why the exponents of § 2.5 are the same at every Mach number. Setting ke=0k_e=0 removes both Hankel terms and leaves the incompressible equation of Baddoo et al. [4].

The trailing-edge exponent in the basis

The basis (17) carries the trailing-edge exponent α\alpha of (10) as a parameter, so that a porous trailing edge is represented exactly rather than through a rigid-edge weight. Three points are worth recording.

First, the wake mode gK=((1+ξ)/2)1−βg_K=((1+\xi)/2)^{1-\beta} carries no (1−ξ)(1-\xi) factor and is therefore unchanged by α\alpha; the tie cK=ikˉΓc_K=i\bar{k}\Gamma follows from continuity of the wake and not from the edge exponent, so (21) holds for every α>0\alpha>0.

Second, the closed-form antiderivatives GnG_n generalise with the Jacobi parameter: the whole-chord integrals are 2α−β+1B(1−β,1+α)2^{\alpha-\beta+1}B(1-\beta,1+\alpha) with BB the beta function, and the orthogonality ∫gn=0\int g_n=0 for n≥2n\ge2 that the solver relies on is preserved. These have been checked against direct quadrature to 10−710^{-7}–10−910^{-9} for real and complex α\alpha.

Third, the generalisation is exact at α=1/2\alpha=1/2: the implementation carrying general α\alpha reproduces the rigid-trailing-edge implementation bit for bit, with zero difference in lift, circulation, loading distribution and every expansion coefficient, over eleven cases spanning rigid plates, porous nose inserts with junctions, complex λ\lambda, kek_e up to 4.4, kˉ\bar{k} up to 101, and heave forcing.

How much does α\alpha matter? Less than one would expect for integrated quantities, and a great deal for local ones. On the profiles of Figure 5, where α≈0.15\alpha\approx0.15, carrying the correct exponent changes the lift by at most 2×10−42\times10^{-4}; on a uniformly permeable plate at λ=3\lambda=3, where α=0.102\alpha=0.102 against the rigid 1/21/2, the lift agrees to five figures. The loading itself does not: within 10−310^{-3} of the trailing edge the two bases differ by a factor of order five, in the direction (10) requires. The reason integrated quantities are insensitive is that the region in which the two exponents differ materially is that small. Local quantities are a different matter, and the loading near a permeable trailing edge is exactly what a trailing-edge scattering calculation needs; that is why the exponent is carried as a parameter rather than fixed.

Quadrature and the aliasing condition

Each influence integral is evaluated on Gauss–Legendre panels graded geometrically towards the singular points ξ=±1\xi=\pm1 and ξ=sj\xi=s_j, with the Cauchy singularity removed by the subtraction (23). Grading alone is insufficient: the basis members oscillate as cos⁡(Nϑ)\cos(N\vartheta), so a panel wider than one half-period aliases, and the resulting error is systematic rather than random. Requiring every panel to satisfy Δϑ≤πnp/(4N)\Delta\vartheta\le\pi n_p/(4N) removes it, and the junction-invariance test of Table 2 falls from 7%7\% to 3.5×10−103.5\times10^{-10} when the condition is imposed. The grading depth is capped so that panel edges never fall below floating-point resolution.

Steady closed forms

At ω=0\omega=0 the dominant equation is λΠ0(ξ)+π−1p.v.⁡∫Π0(τ)/(ξ−τ) dτ=−αi\lambda\Pi_0(\xi)+\pi^{-1}\operatorname{p.v.}\int\Pi_0(\tau)/(\xi-\tau)\,\mathrm{d}\tau=-\alpha_i on [−1,1][-1,1], with bounded loading at the trailing edge. Seeking Π0=A0[(1−ξ)/(1+ξ)]β\Pi_0=A_0[(1-\xi)/(1+\xi)]^\beta, which for a uniformly porous plate carries (1−ξ)+β(1-\xi)^{+\beta} at the trailing edge (the bounded-loading condition in that case, consistent with (10) at λ(1)=λ\lambda(1)=\lambda), and using the Tricomi identity Tricomi [37]

1π∫−11(1−τ1+τ)βdττ−ξ=cot⁡(πβ)(1−ξ1+ξ)β−csc⁡(πβ)(35)\frac{1}{\pi}\int_{-1}^{1}\left(\frac{1-\tau}{1+\tau}\right)^\beta\frac{\mathrm{d}\tau}{\tau-\xi}=\cot(\pi\beta)\left(\frac{1-\xi}{1+\xi}\right)^\beta-\csc(\pi\beta) \tag*{(35)}

reduces the equation to the algebraic pair λ=cot⁡(πβ)\lambda=\cot(\pi\beta) and A0=−αisin⁡(πβ)A_0=-\alpha_i\sin(\pi\beta): the ξ\xi-dependent terms cancel identically only at that β\beta, which is (9), and the constant terms then fix A0A_0. Integrating with ∫−11[(1−ξ)/(1+ξ)]βdξ=2πβ/sin⁡(πβ)\int_{-1}^{1}[(1-\xi)/(1+\xi)]^\beta\mathrm{d}\xi=2\pi\beta/\sin(\pi\beta) gives ∫Π0dξ=−2πβαi\int\Pi_0\mathrm{d}\xi=-2\pi\beta\alpha_i, so that CL=−2∫Π0dξ=4πβαiC_L=-2\int\Pi_0\mathrm{d}\xi=4\pi\beta\alpha_i and hence (27). The seepage dissipation is −∫vsΔp dx1-\int v_s\Delta p\,\mathrm{d}x_1, which is positive and by the same identity equals αiCL\alpha_iC_L, that is, 2β2\beta times the leading-edge suction 2παi22\pi\alpha_i^2 of the corresponding rigid plate, not equal to it, because a permeable edge with β<1/2\beta<1/2 develops no suction.

The insert law (31). Let the leading-edge insert occupy −1<ξ<sj-1<\xi<s_j with a constant λ\lambda and exponent β\beta, the rest of the chord being rigid. Writing Ω(z)=(2πi)−1∫Π0(τ)/(τ−z) dτ\Omega(z)=(2\pi i)^{-1}\int\Pi_0(\tau)/(\tau-z)\,\mathrm{d}\tau, the Plemelj formulae turn the steady equation into (λ−i)Ω+−(λ+i)Ω−=−αi(\lambda-i)\Omega^+-(\lambda+i)\Omega^-=-\alpha_i on the chord, that is Ω+=GΩ−−αi/(λ−i)\Omega^+=G\Omega^--\alpha_i/(\lambda-i) with GG of (8). For a constant upwash this has the constant particular solution Ω=−iαi/2\Omega=-i\alpha_i/2, whatever λ\lambda is, because (λ−i)−(λ+i)=−2i(\lambda-i)-(\lambda+i)=-2i. The canonical function

X(z)=(z+1)−β(z−sj)β−1/2(z−1)1/2X(z)=(z+1)^{-\beta}(z-s_j)^{\beta-1/2}(z-1)^{1/2}

has X+/X−=e2πiβ=GX^+/X^-=e^{2\pi i\beta}=G on the insert, X+/X−=−1=GX^+/X^-=-1=G on the rigid part, no jump off the chord, the edge behaviour (1+ξ)−β(1+\xi)^{-\beta}, ∣ξ−sj∣β−1/2|\xi-s_j|^{\beta-1/2} and (1−ξ)1/2(1-\xi)^{1/2} of (9), (11) and the Kutta condition, and X→1X\to1 at infinity since its exponents sum to zero. Hence Ω(z)=−(iαi/2)[1−X(z)]\Omega(z)=-(i\alpha_i/2)[1-X(z)] solves the problem and vanishes at infinity; it is the only such solution, since the difference of two solutions divided by XX is entire and vanishes at infinity. As z→∞z\to\infty, Ω∼−(2πiz)−1∫Π0 dξ\Omega\sim-(2\pi iz)^{-1}\int\Pi_0\,\mathrm{d}\xi and X=1+κ1/z+O(z−2)X=1+\kappa_1/z+O(z^{-2}) with κ1=−β−(β−12)sj−12=−1+(12−β)(1+sj)\kappa_1=-\beta-(\beta-\frac{1}{2})s_j-\frac{1}{2}=-1+(\frac{1}{2}-\beta)(1+s_j), which is linear in sjs_j. Therefore

∫−11Π0 dξ=παiκ1=−παi[1−ℓc(1−2β)],ℓc=1+sj2,\int_{-1}^{1}\Pi_0\,\mathrm{d}\xi=\pi\alpha_i\kappa_1=-\pi\alpha_i\left[1-\frac{\ell}{c}(1-2\beta)\right],\qquad\frac{\ell}{c}=\frac{1+s_j}{2},

which is (31): the rigid plate at sj=−1s_j=-1, (27) at sj=1s_j=1, and exactly linear between. The argument extends to any number of piecewise-constant segments, the lift remaining linear in each junction position; it relies on the upwash being constant and does not carry over to a cambered section.

Inversion to the time domain

What is inverted is the lift transfer function with its apparent-mass part removed. For heave that part is an impulse at τ=0\tau=0 and is not what Wagner’s function describes: writing the computed response as L(σ)∼a−ibσL(\sigma)\sim a-i b\sigma at the top of the sweep, the fitted bb is subtracted, H(σ)=[L(σ)+ibσ]/L(0)H(\sigma)=[L(\sigma)+ib\sigma]/L(0). The fit returns b=0.5006b=0.5006 for a rigid plate, against the exact 1/21/2, and ∣b∣≤0.003|b|\le0.003 for every permeable case computed. A permeable plate has almost no apparent mass (§ 6.4), so for those cases the subtraction is immaterial. At M=0.5M=0.5 it returns ∣b∣<2×10−3|b|<2\times10^{-3} even for a rigid plate, whose apparent mass compressibility has already removed. For the gust no subtraction is needed, and the leading-edge phase reference of (6) is the one Küssner’s function requires. At finite Mach number both transfer functions are formed from the physical lift (7); the transformed integral of Π\Pi would shift the gust reference by the constant phase MkeMk_e and misstate the heave response.

The inversion (24) is then split at the top of the computed range, Σ=25\Sigma=25:

ϕ(τ)=2π[∫σmin⁡ΣRe⁡H(σ)σsin⁡(στ) dσ+c∞(π2−Si⁡(Στ))],c∞=lim⁡σ→ΣRe⁡H,\phi(\tau)=\frac{2}{\pi}\left[\int_{\sigma_{\min}}^{\Sigma}\frac{\operatorname{Re}H(\sigma)}{\sigma}\sin(\sigma\tau)\,\mathrm{d}\sigma+c_\infty\left(\frac{\pi}{2}-\operatorname{Si}(\Sigma\tau)\right)\right],\qquad c_\infty=\lim_{\sigma\to\Sigma}\operatorname{Re}H,

the tail being exact for a transfer function that has settled to a constant. HH is computed on 102 points, of which 48 are geometrically spaced from σmin⁡=10−4\sigma_{\min}=10^{-4} to 0.5, 40 are uniformly spaced up to 8 and 14 are geometrically spaced up to 25. It is splined onto a grid fine enough to carry at least twenty points per half-period of sin⁡(στ)\sin(\sigma\tau) before integration; without that refinement the long-time values are aliased. Three convergence checks were made: lowering σmin⁡\sigma_{\min} from 10−310^{-3} to 10−410^{-4} changes ϕ(40)\phi(40) by 0.02 and ϕ(τ≤10)\phi(\tau\le10) by less than 0.006; raising Σ\Sigma from 25 to 50, for the rigid and λ0=3\lambda_0 = 3 Wagner and Küssner analogues at M=0M = 0 and 0.5, changes ϕ\phi by less than 10−310^{-3} for τ≥1\tau\ge1 in every case and by at most 0.014 below τ=1\tau= 1, the largest change being in the rigid Wagner analogue at M=0.5M = 0.5, whose initial piston value is approached only as Σ\Sigma grows; and the rigid Wagner function is reproduced to 0.007 over τ≤40\tau\le40, which is the end-to-end test of the scheme.

An independent solution in physical variables

The check of § 4 solves the same physical problem by a route that shares no variable, basis or closure with § 3. With b=U=ρ0=1b = U = \rho_0 = 1 and e−iωte^{-i\omega t}, a pressure jump Δp(x1)=p(x1,0−)−p(x1,0+)\Delta p(x_1) = p(x_1,0^-) - p(x_1,0^+) carried on the chord alone induces the normal velocity

v(x1)=12π∫−∞∞K^(ζ)Δp^(ζ)eiζx1 dζ,K^(ζ)=−iν(ζ)2(σ+i0−ζ),ν=[ζ2−(k0−Mζ)2]1/2v(x_1) = \frac{1}{2\pi}\int_{-\infty}^{\infty}\widehat{K}(\zeta)\widehat{\Delta p}(\zeta)e^{i\zeta x_1}\,\mathrm{d}\zeta,\qquad\widehat{K}(\zeta) = \frac{-i\nu(\zeta)}{2(\sigma+i0-\zeta)},\qquad\nu= \left[\zeta^2-(k_0-M\zeta)^2\right]^{1/2}

which follows from the convected wave equation for a potential odd in x2x_2. The branch of ν\nu has Re⁡ν≥0\operatorname{Re}\nu\ge0 and is −i∣ζ2−(k0−Mζ)2∣1/2-i\lvert\zeta^2-(k_0-M\zeta)^2\rvert^{1/2} on the radiating interval, and the +i0+i0, which is causality, makes the pole at ζ=σ\zeta= \sigma contribute −iπ-i\pi times its residue; on the real axis the rest is a principal value, evaluated with a Gauss rule symmetric about the pole. No wake appears, because the pressure is continuous across it, and the Kutta condition is imposed by the basis. The trial functions are (1+x1)−β(1−x1)αPn(α,−β)(x1)(1+x_1)^{-\beta}(1-x_1)^\alpha P_n^{(\alpha,-\beta)}(x_1) with the exponents of § 2.5, their Fourier transforms computed by Gauss–Jacobi quadrature (for a rigid plate, α=β=1/2\alpha= \beta= 1/2, they are Bessel functions in closed form, and both versions were run); the test functions are sin⁡mϑ\sin m\vartheta, x1=−cos⁡ϑx_1=-\cos\vartheta. The permeable condition adds (λβM/2)Δp(\lambda\beta_M/2)\Delta p to the normal velocity, which is the seepage of (2) rewritten with (4); the sign differs from (2) because Δp\Delta p is here lower minus upper. The imposed velocity is the physical upwash, the lift is ∫Δp dx1\int\Delta p\,\mathrm{d}x_1, and nothing is transformed at either end. The ζ\zeta-integral is truncated at ∣ζ∣=2500\lvert\zeta\rvert= 2500–6000; for a rigid plate the integrand decays only as ζ−2\zeta^{-2} and the truncation limits the accuracy to about 10−310^{-3} (the method reproduces the Sears and Theodorsen functions at M=0M = 0 to that level), while for a permeable plate the weaker edge singularity makes it decay faster and the two solutions agree to 10−510^{-5}.

References

  1. [1]R. K. Amiet. High frequency thin-airfoil theory for subsonic flow. AIAA J., 14(8):1076–1082, 1976. doi: 10.2514/3.7187.DOI
  2. [2]L. J. Ayton. Acoustic scattering by a finite rigid plate with a poroelastic extension. J. Fluid Mech., 791:414–438, 2016. doi: 10.1017/jfm.2016.59.DOI
  3. [3]L. J. Ayton, M. J. Colbrook, T. F. Geyer, P. Chaitanya, and E. Sarradj. Reducing aerofoil–turbulence interaction noise through chordwise-varying porosity. J. Fluid Mech., 906:A1, 2021. doi: 10.1017/jfm.2020.746.DOI
  4. [4]P. J. Baddoo, R. Hajian, and J. W. Jaworski. Unsteady aerodynamics of porous aerofoils. J. Fluid Mech., 913:A16, 2021. doi: 10.1017/jfm.2020.1031.DOI
  5. [5]G. K. Batchelor. The Theory of Homogeneous Turbulence. Cambridge University Press, 1953.
  6. [6]R. L. Bisplinghoff, H. Ashley, and R. L. Halfman. Aeroelasticity. Addison-Wesley, 1955.
  7. [7]L. Bowen, A. Celik, B. Zhou, M. F. Westin, and M. Azarpeyvand. The effect of leading edge porosity on airfoil turbulence interaction noise. J. Acoust. Soc. Am., 152(3):1437–1448, 2022. doi: 10.1121/10.0013703.DOI
  8. [8]P. Chaitanya, P. Joseph, S. Narayanan, C. Vanderwel, J. Turner, J. W. Kim, and B. Ganapathisubramani. Performance and mechanism of sinusoidal leading edge serrations for the reduction of turbulence–aerofoil interaction noise. J. Fluid Mech., 818:435–464, 2017. doi: 10.1017/jfm.2017.141.DOI
  9. [9]P. Chaitanya, P. Joseph, and L. J. Ayton. Leading-edge profiles for the reduction of airfoil interaction noise. AIAA J., 58(3):1118–1129, 2020. doi: 10.2514/1.J058456.DOI
  10. [10]J. E. Efowcs Williams and L. H. Hall. Aerodynamic sound generation by turbulent flow in the vicinity of a scattering half plane. J. Fluid Mech., 40(4):657–670, 1970. doi: 10.1017/S0022112070000368.DOI
  11. [11]P. Forchheimer. Wasserbewegung durch Boden. Z. Ver. Dtsch. Ing., 45:1782–1788, 1901.
  12. [12]F. D. Gakhov. Boundary Value Problems. Pergamon, 1966.
  13. [13]T. Geyer, E. Sarradj, and C. Fritzsche. Porous airfoils: noise reduction and boundary layer effects. Intl J. Aeroacoust., 9(6):787–820, 2010. doi: 10.1260/1475-472X.9.6.787.DOI
  14. [14]H. Gill and S. Lee. Uncertainty quantification of hovering rotor noise due to gust-induced unsteady loading using polynomial chaos expansion. J. Am. Helicopter Soc., 71(3):032001, 2026a. doi: 10.4050/JAHS.71.032001.DOI
  15. [15]H. Gill and S. Lee. Frequency-domain rotor and propeller tonal noise models: analytical framework and prediction accuracy. AIAA J., 64(7):3820–3839, 2026b. doi: 10.2514/1.J066263.DOI
  16. [16]S. Glegg and W. Devenport. Aeroacoustics of Low Mach Number Flows. Academic Press, 2017.
  17. [17]J. M. R. Graham. Similarity rules for thin aerofoils in non-stationary subsonic flows. J. Fluid Mech., 43(4):753–766, 1970. doi: 10.1017/S0022112070002719.DOI
  18. [18]R. Hajian and J. W. Jaworski. The steady aerodynamics of aerofoils with porosity gradients. Proc. R. Soc. A, 473(2205):20170266, 2017. doi: 10.1098/rspa.2017.0266.DOI
  19. [19]A. D. G. Hales, L. J. Ayton, A. O. Wills, C. Jiang, C. de Silva, D. Moreau, and C. Doolan. A mathematical model for the interaction of anisotropic turbulence with porous surfaces. J. Fluid Mech., 1001:A16, 2024. doi: 10.1017/jfm.2024.1067.DOI
  20. [20]M. S. Howe. Acoustics of Fluid–Structure Interactions. Cambridge University Press, 1998. doi: 10.1017/CBO9780511662898.DOI
  21. [21]M. S. Howe, M. I. Scott, and S. R. Sipcic. The influence of tangential mean flow on the Rayleigh conductivity of an aperture. Proc. R. Soc. Lond. A, 452:2303–2317, 1996. doi: 10.1098/rspa.1996.0123.DOI
  22. [22]J. W. Jaworski and N. Peake. Aerodynamic noise from a poroelastic edge with implications for the silent flight of owls. J. Fluid Mech., 723:456–479, 2013. doi: 10.1017/jfm.2013.139.DOI
  23. [23]Z. Jia and S. Lee. Impulsive loading noise of a lift-offset coaxial rotor in high-speed forward flight. AIAA J., 58(2):687–701, 2020. doi: 10.2514/1.J058295.DOI
  24. [24]N. H. Kemp and G. Homicz. Approximate unsteady thin-airfoil theory for subsonic flow. AIAA J., 14(8):1083–1089, 1976. doi: 10.2514/3.7188.DOI
  25. [25]J. W. Kim, S. Haeri, and P. F. Joseph. On the reduction of aerofoil–turbulence interaction noise associated with wavy leading edges. J. Fluid Mech., 792:526–552, 2016. doi: 10.1017/jfm.2016.95.DOI
  26. [26]H. G. Küssner. Zusammenfassender Bericht über den instationären Auftrieb von Flügeln. Luftfahrtforschung, 13:410–424, 1936.
  27. [27]S. Lee, L. Ayton, F. Bertagnolio, S. Moreau, T. P. Chong, and P. Joseph. Turbulent boundary layer trailing-edge noise: theory, computation, experiment, and application. Prog. Aerosp. Sci., 126:100737, 2021. doi: 10.1016/j.paerosci.2021.100737.DOI
  28. [28]J. G. Leishman. Validation of approximate indicial aerodynamic functions for two-dimensional subsonic flow. J. Aircraft, 25(10):914–922, 1988. doi: 10.2514/3.45680.DOI
  29. [29]S. Li and S. Lee. Analytic prediction of rotor broadband noise with serrated trailing edges with applications to urban air mobility aircraft. J. Am. Helicopter Soc., 69(3):032005, 2024. doi: 10.4050/JAHS.69.032005.DOI
  30. [30]B. Lyu, M. Azarpeyvand, and S. Sinayoko. Prediction of noise from serrated trailing edges. J. Fluid Mech., 793:556–588, 2016. doi: 10.1017/jfm.2016.132.DOI
  31. [31]S. Moreau. Turbomachinery noise predictions: present and future. Acoustics, 1(1):92–116, 2019. doi: 10.3390/acoustics1010008.DOI
  32. [32]N. I. Muskhelishvili. Singular Integral Equations. Noordhoff, 1953.
  33. [33]C. Osborne. Unsteady thin-airfoil theory for subsonic flow. AIAA J., 11(2):205–209, 1973. doi: 10.2514/3.6730.DOI
  34. [34]C. Possio. L’azione aerodinamica sul profilo oscillante in un fluido comprimibile a velocità iposonora. L’Aerotecnica, 18:441–458, 1938.
  35. [35]W. R. Sears. Some aspects of non-stationary airfoil theory and its practical application. J. Aeronaut. Sci., 8(3):104–108, 1941. doi: 10.2514/8.10655.DOI
  36. [36]T. Theodorsen. General theory of aerodynamic instability and the mechanism of flutter. Technical Report 496, NACA, 1935.
  37. [37]F. G. Tricomi. Integral Equations. Interscience, 1957.
  38. [38]T. von Kármán and W. R. Sears. Airfoil theory for non-uniform motion. J. Aeronaut. Sci., 5(10):379–390, 1938. doi: 10.2514/8.674.DOI
  39. [39]H. Wagner. Über die Entstehung des dynamischen Auftriebes von Tragflügeln. Z. Angew. Math. Mech., 5(1):17–35, 1925. doi: 10.1002/zamm.19250050103.DOI

Paper details

vlnc:W_czc4fhh4
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 6

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents