betaFeedback

WorksPhysics and Astronomy

An LLM-assisted theoretical model for aeroengine installation noise scattering

Abstract

A theoretical model is developed for the canonical problem of aeroengine fan noise impinging on, and scattering from, a nearby semi-infinite rigid plate. The model chains two Wiener–Hopf problems in different coordinate systems. The first stage solves the duct-radiation problem analytically in axisymmetric cylindrical coordinates. The second stage expands the field incident on the plate into a double angular spectrum of plane waves: the axial spectrum is inherited from the first stage, while the spanwise spectrum is generated by a Weyl–Sommerfeld expansion of the cylindrical harmonics. Each spectral component is then scattered by the plate. The mathematical crux is the coordinate transformation between the axisymmetric duct frame and the Cartesian plate frame. The core physical insight–that the Weyl–Sommerfeld identity provides an exact spectral bridge between these systems–originates entirely with the authors. The heavy algebra of translating between the two spectral representations, together with the associated numerical verification, was carried out with the assistance of a large language model (LLM), under strict human oversight and independent numerical cross-checking. This division of labour enables the LLM to handle tedious symbolic manipulation without compromising rigour. Results are presented for spinning modes radiating from a cylindrical duct with a subsonic jet into a quiescent ambient. The case study demonstrates that, under systematic verification, LLMs can accelerate tedious algebra in theoretical fluid mechanics, contributing a concrete empirical example to the broader discussion on AI's role in theoretical research.

Introduction

The Leiden Declaration [13] cautions against unreliable AI-generated proofs, particularly in safety-critical domains. Here we present a counterbalancing case study from classical fluid mechanics using Wiener–Hopf techniques [22, 1]. The essential physical insight originates entirely with the authors. We employ the large language model (LLM) solely for tedious symbolic manipulation and numerical verification (mainly in § 2.3), under our strict oversight. This transparent collaboration – where every derivation is independently validated – demonstrates how and when LLMs can accelerate theoretical studies without compromising rigour, thereby offering a concrete empirical example from our work to the broader discussion on AI’s role in applied mathematics and fluid dynamics.

In practice, the fan noise radiated from the open nozzle of an aeroengine bypass duct interacts with the wing, the pylon, or the fuselage, and the scattered field can differ markedly from the free radiation. A canonical building block of this installation effect is the scattering of duct radiation by a nearby rigid surface. The two separate modelling ingredients of the problem are classical in fluid mechanics and aeroacoustics. The radiation of a duct mode from the open end of a semi-infinite cylindrical pipe was solved by the Wiener–Hopf technique by Munt [20, 21] for the hollow duct discharging a subsonic jet, extending the still-medium analysis of Levine & Schwinger [14]; Cargill [4] examined the low-frequency interplay between the unsteady jet flow and the radiated sound; Rienstra [25] extended the solution to the annular bypass duct in a uniform subsonic flow; and a systematic numerical implementation, including efficient Cauchy-integral factorisations and contour deformations, was given by Gabard & Astley [7]. Peake’s group further carried the technique to coaxial multi-stream exhausts (Veitch & Peake [30]).

Next, the diffraction of a plane wave – or a more general incident gust – by the edge of a semi-infinite rigid plate is the canonical Wiener–Hopf problem ([22, 5]). Its aeroacoustic relevance was first established by the trailing-edge noise analysis of Ffowcs Williams & Hall [6]; Rienstra [24] subsequently incorporated mean-flow effects into the diffraction solution. Howe [9] introduced the first analytical model for serrated-edge noise, and Ayton & Peake [2] developed high-frequency aerofoil-scattering theory. More recently, the matrix Wiener–Hopf formulation for spanwise-periodic serrated edges by Huang [10] provides the direct theoretical foundation for the second stage of the present model.

Both lines have remained active. On the duct side, recent work includes thermoacoustic effects on spinning modes (Sugimoto [28]), slowly varying modes in sheared lined ducts (Rienstra [26]), a geometrical-theory-of-diffraction treatment of the duct termination benchmarked against the Wiener–Hopf solution (Jing & Du [12]), and the extension to elliptic cross-sections (Wang & Huang [31]). On the plate side, Baddoo & Ayton [3] solved cascade scattering with complex boundary conditions, and Hales et al. [8] modelled the interaction of anisotropic turbulence with a rigid leading edge.

On the applied side, installation noise – the scattering of engine noise by the airframe – has been computed numerically at ONERA with high-order computational aeroacoustics, with experimental validation for an over-wing nacelle configuration (Lorteau et al. [18]) and for a boundary-layer-ingesting rear-fuselage concept (Lorteau et al. [17]), following the earlier study of Redonnet et al. [23]; model-scale installed jet–pylon–wing experiments are reported by Meloni et al. [19]. Such full-geometry computations are expensive, and a theoretical model of the simplified duct–plate combination – which captures the salient physical features – remains a powerful tool for industrial applications while simultaneously posing a challenge to academia, thereby motivating this work.

Here the incident field on the plate is not a plane wave but the full, radially and azimuthally structured, partly evanescent near field of the duct exit. The purpose of the present work is to construct such a coupled model. The mathematical difficulty of the coupling is, at its core, a change of coordinate system: the duct’s Wiener–Hopf solution was established in axisymmetric cylindrical coordinates and is naturally a one-dimensional axial spectrum of cylindrical harmonics, while the plate problem was developed in Cartesian coordinates and is naturally a two-dimensional spectrum of plane-wave gusts; a coupled model requires an exact translator between these two spectral and the associated coordinates. We achieve this by chaining the two Wiener–Hopf problems through an exact spectral bridge.

The first stage is left entirely untouched: the duct problem is solved once, in axisymmetric cylindrical coordinates, and is exploited only through its output, the half-range axial spectrum F+(u)F_{+}(u) of the vortex-sheet displacement. The bridge then re-expands the incident field on the plate plane into a double angular spectrum of plane waves: the axial dependence of each duct spectral component is already that of a convected gust, so the axial spectrum is inherited analytically, while the spanwise (cylindrical-cross-section) dependence is expanded by a Weyl–Sommerfeld identity for the Hankel harmonics Hm(1)(kr)eimθH_{m}^{(1)}(kr)e^{im\theta}, valid in the half-space containing the plate. second stage is therefore exact. Superposition over the spectrum yields the far-field directivity of the scattered field. We note that this is a one-way coupling: we assume the scattered field is not re-scattered by the duct, an approximation consistent with the plate being relatively far from the duct exit. Should this one-way approximation prove insufficient in other configurations, a straightforward iterative scheme between the two stages can be employed to recover the full coupling.

On the other hand, the LLM-assisted theoretical research has produced visible successes in the last two years, from mathematical discovery by program search (Romera-Paredes et al. 2024) to olympiad-level geometry [29] and interactive derivation assistance in applied mathematics [15]. We state plainly, however, that the core contribution of the paper is not the use of an LLM but our recognition that a Weyl–Sommerfeld identity translates exactly between the cylindrical modal waves of the duct and the Cartesian gust incident on the plate, so that two classical Wiener–Hopf solutions can be chained without approximation. That idea, and the chaining strategy, are from human. Because both modelling stages were already available as validated implementations, the most important (and theoretically difficult) kernel factorisations, the pole tracking and the causality arguments are taken from our previous works [16, 10] – which sidesteps the present immaturity of LLMs in rigorous complex-variable reasoning. The LLM proved particularly helpful for the tedious algebraic manipulations required to bridge the two coordinate systems and for the subsequent verifications detailed in § 3. We expect this division of labour – human supplies the key mathematical idea, the LLM the mechanical execution under systematic verification – to be a productive pattern for similar multi-model coupling problems.

The remaining part of this paper is organised as follows. § 2 states the geometry and conventions and develops the model: the stage-one duct solution is summarised in § 2.2, the spectral bridge by LLM in § 2.5, the stage-two plate Wiener–Hopf problem in § 2.9, and the far-field asymptotics, including the image-source identity, in § 2.12. Readers who are interest mainly in the underlying wave mechanics may directly skip this section and proceed to the results and discussions in the following sections. § 3 describes the numerical implementation and the validation details. § 4 presents near- and far-field results for a spinning mode (m,n)=(5,1)(m,n)=(5,1) at normalised angular frequency ω=20\omega=20, in both a fully quiescent and a with-jet configuration, and the full-circle directivity of a higher-order mode (13,1)(13,1) at ω=28.3\omega=28.3. § 5 concludes the whole work.

Theoretical model

Statement of the problem

In this problem set-up, all lengths are non-dimensionalised by the duct radius DD, so that the duct radius is unity and the dimensionless frequency is ω=2πfD/c0\omega=2\pi fD/c_0 with ff the physical frequency and c0c_0 the speed of sound. The time dependence exp⁡(−iωt)\exp(-i\omega t) is suppressed throughout for brevity. A global Cartesian frame (x,y,z)(x,y,z) is used, with xx along the duct axis and the plate (positive downstream), yy normal to the plate (positive towards the duct side) and zz spanwise. The duct is semi-infinite, with wall r=1r=1 and occupying x≤−2x\leq-2; the exit plane is x=−2x=-2, with exit centre (x,r)=(−2,0)(x,r)=(-2,0). The duct may be annular with a centrebody of radius r=hr=h, the hollow duct corresponding to h=0h=0 (which is assumed in the rest of this paper). Cylindrical coordinates (r,θ)(r,\theta) are taken about the duct axis, r=y2+z2r=\sqrt{y^2+z^2}, θ=atan2⁡(z,y)\theta=\operatorname{atan2}(z,y) measured from the +y+y axis. The duct-local axial coordinate is xd=x+2x_d=x+2 (the subscript dd stands for ‘duct’ and marks duct-frame quantities throughout), so that the exit is xd=0x_d=0.

The plate is rigid, of vanishing thickness and semi-infinite: it lies in the plane y=−dy=-d with d=2d=2 for the rest studies, occupies x≤0x\leq0, and its straight leading edge is the line x=0x=0, y=−dy=-d (the zz axis shifted by −d-d). We write y~=y+d\tilde{y} = y + d, so that the plate surface is y~=0\tilde{y} = 0; the distance between the exit plane and the edge is L=2L = 2. Consequently, the plate lies outside the duct and outside the shed vortex sheet, in the uniform ambient medium. The geometry, the two coordinate systems and the whole set-up are summarised in Figure 1.

Geometry of the coupled problem

Figure 1. Geometry of the coupled problem. The semi-infinite circular duct of radius R=1R = 1 occupies x≤−2x \le-2; the duct-local cylindrical coordinates (xd,r,θ)(x_d,r,\theta) have their origin at the exit centre O1=(−2,0,0)O_1 = (-2,0,0), with the xdx_d axis along the duct axis and the r,θr,\theta directions as drawn on the exit face. The semi-infinite rigid plate of vanishing thickness lies in the plane y=−2y = -2 and occupies x≤0x \le0; its straight leading edge x=0x = 0 is marked by the thick line. The plate-local Cartesian frame is attached at O2=(0,−2,0)O_2 = (0,-2,0). The streamwise gap between the exit plane and the edge is L=2L = 2 and the vertical offset is d=2d = 2.

The configuration consists of a single cylindrical duct (no centrebody, h=0h = 0) discharging a subsonic jet of Mach number M1M_1 into a quiescent ambient (M0=0M_0 = 0), for simplicity the jet-to-ambient density ratio being unity; the fully quiescent limit M1=0M_1 = 0 is contained as a special case. The stage-one solution includes the jet and the vortex sheet shed from the duct lip, with a Kutta condition applied at the lip; the jet enters the coupling only through the output spectrum F+(u)F^{+}(u). The stage-two plate problem, by contrast, treats the medium surrounding the plate as quiescent and uniform: the downstream vortex sheet is presumably not advected over the plate. This is a deliberate one-way-coupling simplification, discussed further in § 5. A uniform external flow M0≠0M_0 \ne0 is accommodated stage by stage by the Lorentz transformation of Gabard & Astley [7] and the Prandtl–Glauert transformation of Huang [10], and is not pursued here. Three configurations are computed in § 4: a fully quiescent baseline (M0=M1=0M_0 = M_1 = 0), in which no vortex sheet is shed and no shear-layer dynamics is present – chosen deliberately to remove the shear layer from the coupled problem, so that the model can later be validated directly against numerical simulation of the pure acoustic radiation–scattering process – and two with-jet cases (M1=0.353M_1 = 0.353 and 0.3380.338). A plug-flow velocity profile is adopted for the jet and the ambient stream, which is a common idealisation in aeroacoustic modelling of aeroengine noise radiation (e.g., [20, 25]). The present configurations are idealised from our prior industrial problems for the Trent-series aeroengine (Huang et al. [11]).

Two different Fourier conventions are used, one per stage, matching the source formulations from their original references:

(a) stage 1 (duct, the convention of Gabard & Astley [7]): forward kernel exp⁡(−iωuxd)\exp(-\mathrm{i}\omega u x_d), inverse transform ψ=(ω/2π)∫Γβexp⁡(+iωuxd) du\psi= (\omega/2\pi)\int_{\Gamma}\beta\exp(+\mathrm{i}\omega u x_d)\,\mathrm{d}u;

(b) stage 2 (plate, the convention of Huang [10]): forward kernel exp⁡(+iαx)\exp(+\mathrm{i}\alpha x), inverse transform ψ=(1/2π)∫Γβexp⁡(−iαx) dα\psi= (1/2\pi)\int_{\Gamma}\beta\exp(-\mathrm{i}\alpha x)\,\mathrm{d}\alpha.

Consequently, an axial dependence exp⁡(+iωux)\exp(+\mathrm{i}\omega ux) produced by the duct solution corresponds, in

SymbolMeaning
(m,μ)(m,\mu)incident duct mode: azimuthal order mm, axial wavenumber ωμ\omega\mu
xdx_dduct-local axial coordinate, the exit is xd=0x_d=0
αmn\alpha_{mn}internal radial wavenumber, root of the eigencondition Λ(α)=0\Lambda(\alpha)=0
F+(u)F_{+}(u)stage-one Wiener–Hopf solution (half-range transform of the vortex-sheet displacement)
K(u)K(u), K~±(u)\widetilde{K}_{\pm}(u)stage-one Wiener–Hopf kernel and its factorisation
λ0(u)\lambda_0(u)external radial wavenumber over ω\omega; λ0=1−u2\lambda_0=\sqrt{1-u^2} when M0=0M_0=0
uu, ν\nu, α\alphaspectral variables: axial (duct convention), spanwise, streamwise (plate convention)
γd(ν)\gamma_d(\nu)(λ0ω)2−ν2\sqrt{(\lambda_0\omega)^2-\nu^2}, wall-normal wavenumber of an incident component
k^\hat{k}, γp(α)\gamma_p(\alpha)k^=ω2−ν2\hat{k}=\sqrt{\omega^2-\nu^2}; γp=α2−k^2\gamma_p=\sqrt{\alpha^2-\hat{k}^2}, stage-two kernel
γ±(α)\gamma_{\pm}(\alpha)α±k^\sqrt{\alpha\pm\hat{k}}, explicit factorisation of the stage-two kernel
M0M_0the Mach number of the ambient environment
M1M_1the Mach number of the jet flow from the duct

caption: tbl:1. Principal symbols.

:::

plate problem, to the incident pole αp=−ωu\alpha_p=-\omega u. This sign bookkeeping is essential and is verified numerically in § 3. The other main symbols are collected in table 1.

Stage 1: radiation from the duct exit

The first stage is the duct-radiation model of Munt [20], Rienstra [25] and Gabard & Astley [7], used here without modification; we nevertheless rederive it in some detail, because the notation and the output quantity F+(u)F_{+}(u) are the interface on which the whole coupled model rests. The derivation follows Gabard & Astley [7], equations of which are cited as (G3.nn).

Governing equations and boundary, edge and radiation conditions

The incident duct mode, normalised as in appendix B of Gabard & Astley [7], is

ϕ0=Ψ(r)eiωμxd+imθ,Ψ(r)=Ym′(αh)Jm(αr)−Jm′(αh)Ym(αr),(1)\phi_0=\Psi(r)e^{i\omega\mu x_d+im\theta},\qquad\Psi(r)=Y_m'(\alpha h)J_m(\alpha r)-J_m'(\alpha h)Y_m(\alpha r), \tag*{(1)}

where the radial wavenumber α=αmn\alpha=\alpha_{mn} satisfies the annular-duct eigencondition

Λ(α)≡Ym′(αh)Jm′(α)−Jm′(αh)Ym′(α)=0,(2)\Lambda(\alpha)\equiv Y_m'(\alpha h)J_m'(\alpha)-J_m'(\alpha h)Y_m'(\alpha)=0, \tag*{(2)}

which reduces to Jm′(α)=0J_m'(\alpha)=0 for the hollow duct (h=0h=0); the radial mode number n=1,2,…n=1,2,\ldots counts the roots in increasing order, n=1n=1 being the lowest radial solution. Since the geometry and the boundary conditions are axisymmetric, the diffracted field retains the azimuthal and time dependence of the incident mode; writing the total field as

References

  1. [1]ABRAHAMS, I. D., HUANG, X., KISIL, A., MISHURIS, G., NIEVES, M., ROGOSIN, S. & SPITKOVSKY, I. 2021 Reinvigorating the Wiener–Hopf technique in the pursuit of understanding processes and materials. Nat. Sci. Rev. 8 (2), nwaa225.
  2. [2]AYTON, L. J. & PEAKE, N. 2013 On high-frequency noise scattering by aerofoils in flow. J. Fluid Mech. 734, 144–182.DOI
  3. [3]BADDOO, P. J. & AYTON, L. J. 2020 Acoustic scattering by cascades with complex boundary conditions: compliance, porosity and impedance. J. Fluid Mech. 898, A16.DOI
  4. [4]CARGILL, A. M. 1982 Low-frequency sound radiation and generation due to the interaction of unsteady flow with a jet pipe. J. Fluid Mech. 121, 59–105.DOI
  5. [5]CRIGHTON, D. G., DOWLING, A. P., FFOwCS WILLIAMS, J. E., HECKL, M. & LEPPINGTON, F. G. 1992 Modern Methods in Analytical Acoustics. London: Springer-Verlag.DOI
  6. [6]FFOWCS WILLIAMS, J. E. & HALL, L. H. 1970 Aerodynamic sound generation by turbulent flow in the vicinity of a scattering half plane. J. Fluid Mech. 40 (4), 657–670.DOI
  7. [7]GABARD, G. & ASTLEY, R. J. 2006 Theoretical model for sound radiation from annular jet pipes: far- and near-field solutions. J. Fluid Mech. 549, 315–341.DOI
  8. [8]HALES, A. D. G., AYTON, L. J., JIANG, C., MAHGOUB, A., KISLER, R., DIXON, R., DE SILVA, C., MOREAU, D. & DOOLAN, C. 2023 A mathematical model for the interaction of anisotropic turbulence with a rigid leading edge. J. Fluid Mech. 970, A29.DOI
  9. [9]HOWE, M. S. 1991 Aerodynamic noise of a serrated trailing edge. J. Fluids Struct. 5 (1), 33–45.DOI
  10. [10]HUANG, X. 2017 Theoretical model of acoustic scattering from a flat plate with serrations. J. Fluid Mech. 819, 228–257.DOI
  11. [11]HUANG, X., CHEN, X. X., K., MA Z. & ZHANG, X. 2008 Efficient computation of spinning modal radiation through an engine bypass duct. AIAA J. 46 (2), 1413–1422.DOI
  12. [12]JING, X. & DU, Z. 2026 Edge diffraction of in-duct modal waves by unflanged or flanged circular duct terminations. J. Fluid Mech. 1034, A33.DOI
  13. [13]LEIDEN DECLARATION 2026 Leiden declaration on artificial intelligence and mathematics. Endorsed by the International Mathematical Union.
  14. [14]LEVINE, H. & SCHWINGER, J. 1948 On the radiation of sound from an unflanged circular pipe. Phys. Rev. 73, 383–406.
  15. [15]LIU, J., SHEN, L. & WEI, G.-W. 2024 ChatGPT for computational topology. Found. Data Sci. 6 (2), 221–250.DOI
  16. [16]LIU, X., JIANG, H. B., HUANG, X. & CHEN, S. Y. 2016 Theoretical model of scattering from flow ducts with semi-infinite axial liner splices. J. Fluid Mech. 786, 62–83.DOI
  17. [17]LORTEAU, M., LE GARREC, T., DAROUKH, M. & POLACSEK, C. 2022 Acoustic assessment of BLI effects on Airbus Nautilus engine integration concept – Part II: noise radiation. In 28th AIAA/CEAS Aeroacoustics Conference, AIAA 2022-2992.DOI
  18. [18]LORTEAU, M., WIART, L., KOPIEV, V. & DENISOV, S. L. 2019 Numerical study, with experimental validation, of fan noise installation effects in over-wing nacelle configuration using the immersed boundary method. In 25th AIAA/CEAS Aeroacoustics Conference.DOI
  19. [19]MELONI, S., PROENÇA, A. R., LAWRENCE, J. L. T. & CAMUSSI, R. 2021 An experimental investigation into model-scale installed jet–pylon–wing noise. J. Fluid Mech. 929, A4.DOI
  20. [20]MUNT, R. M. 1977 The interaction of sound with a subsonic jet issuing from a semi-infinite cylindrical pipe. J. Fluid Mech. 83, 609–640.DOI
  21. [21]MUNT, R. M. 1990 Acoustic transmission properties of a jet pipe with subsonic jet flow: I. the cold jet reflection coefficient. J. Sound Vib. 142 (3), 413–436.DOI
  22. [22]NOBLE, B. 1958 Methods Based on the Wiener–Hopf Technique for the Solution of Partial Differential Equations. London: Pergamon Press.
  23. [23]REDONNET, S., DESQUESNES, G., MANOHA, E. & PARZANI, C. 2010 Numerical study of acoustic installation effects with a computational aeroacoustics method. AIAA J. 48 (5), 929–937.DOI
  24. [24]RIENSTRA, S. W. 1981 Sound diffraction at a trailing edge. J. Fluid Mech. 108, 443–460.DOI
  25. [25]RIENSTRA, S. W. 1984 Acoustic radiation from a semi-infinite annular duct in a uniform subsonic mean flow. J. Sound Vib. 94 (2), 267–288.DOI
  26. [26]RIENSTRA, S. W. 2021 Slowly varying modes in a two-dimensional duct with shear flow and lined walls. J. Fluid Mech. 906, A23.DOI
  27. [28]Sugimoto, N. 2022 Thermoacoustic effects on the propagation of non-planar sound in a circular duct. J. Fluid Mech. 945, A26.DOI
  28. [29]Trinh, T. H., Wu, Y., Le, Q. V., He, H. & Luong, T. 2024 Solving olympiad geometry without human demonstrations. Nature 625, 476–482.DOI
  29. [30]Veitch, B. & Peake, N. 2008 Acoustic propagation and scattering in the exhaust flow from coaxial cylinders. J. Fluid Mech. 613, 275–307.DOI
  30. [31]Wang, R. & Huang, X. 2025 Analytical study of sound radiation from semi-infinite elliptic ducts with uniform subsonic flow. J. Fluid Mech. 1019, A9.DOI
  31. [32]Watson, G. N. 1966 A Treatise on the Theory of Bessel Functions, 2nd edn. Cambridge University Press.

Paper details

vlnc:W_6zt38sch
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Physics and Astronomy

Kimi (multiple versions), a large language model (LLM)

View statement
Looking for related papers…
Contents