On the wake and flapping dynamics of different aspect ratio flags Gaétan Raynaud, Karen Mulleners Abstract The flapping of flags is a classical problem involving fast and large amplitude deformations of a thin flexible plate and unsteady flow phenomena. We perform systematic time and space-resolved measurements of the deformation and drag acting on flapping flags for various aspect ratios and mass ratios. Bending waves travel from the root to the tip at a speed close to the incoming flow and the typical wavelength of the waves scales with the length of the flag. With smaller aspect ratio, the local dynamic pressure exerted by the fluid on the flag is reduced, lowering the wave propagation speed, and reducing the tip frequency. The effect of aspect ratio on the vortex formation in the near wake is analysed using flow field measurements. We identified two characteristic length scales, the ratio of the flag area over its perimeter L ∗ L^{*} and the square root of area H ​ L \sqrt{HL} that scale the circulation shed during a cycle. Changes in aspect ratio and mass ratio generate a wide scattering of the mean drag coefficient, ranging from 0 0 to 0.55 0.55 . We discuss a kinematic-based model for the mean drag coefficient. This model uses the mass ratio and the typical tip speed, which depends linearly on the wave speeds, to predict the mean drag coefficient without any fitting parameter. † † affiliation: Unsteady Flow Diagnostics Laboratory, Institute of Mechanical Engineering, École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, 1015, Switzerland 1 Introduction Flags are pieces of soft material used to display symbols, signals, or insignia and their typical waving in the wind makes them recognisable (Encyclopaedia Britannica, Various 1910 ). The flapping flag problem and the various applications of flutter, the dynamic instability causing the motion of flags, have attracted research interest over many years. Comprehensive reviews of the key contributions in this field are provided by Shelley & Zhang 2011 and Yu et al. 2019 . Flutter designates a dynamic fluid-structure instability where small deformations of an object in a fluid flow grow in amplitude to a limit cycle oscillation with constant frequency ( Paidoussis 2014 ) . The onset of flutter occurs when the flow speed exceeds a critical flow velocity such that the restoring effects of bending stiffness and flow-induced tension can no longer balance the inertial response of the flag, and the flag extracts energy from the flow ( Connell & Yue 2007 ; Shelley & Zhang 2011 ) . The stability and the dynamics of flapping flags are governed by the interplay between inertia, bending rigidity, and fluid dynamic forces. The relative strengths of these contributing factors are described by several non-dimensional parameters. The main dimensionless parameters that are relevant here are the reduced velocity, the mass ratio, the aspect ratio, and the Reynolds number. Following the notations of Eloy et al. 2007 , these parameters are defined as: U^{*}={\textrm{U}}_{\scalebox{0.65}{$\infty$}}L\sqrt{\frac{{\rho}_{\scalebox{0.65}{s}}e}{D}},~M^{*}=\frac{{\rho}_{\scalebox{0.65}{f}}L}{{\rho}_{\scalebox{0.65}{s}}e},~\frac{H}{L},\text{ and }Re=\frac{{\textrm{U}}_{\scalebox{0.65}{$\infty$}}L}{\nu}, (1) for a flag of length L , height H , thickness e , density {\rho}_{\scalebox{0.65}{s}} , and bending rigidity D in an incoming fluid flow of velocity {\textrm{U}}_{\scalebox{0.65}{$\infty$}} , fluid density {\rho}_{\scalebox{0.65}{f}} , and kinematic viscosity \nu . The reduced velocity U^{*} compares the characteristic period of free vibrations of a beam in vacuum \sim L^{2}\sqrt{{\rho}_{\scalebox{0.65}{s}}e/D} to the convective time of the surrounding flow L/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} ( Virot et al. 2013 ) . The mass ratio M^{*} compares the magnitude of the fluid mass displaced during a flapping cycle of the flag \sim{\rho}_{\scalebox{0.65}{f}}HL^{2} to the mass of the flag itself {\rho}_{\scalebox{0.65}{s}}eLH . The mass ratio can also be seen as the scaling factor between the normalised pressure forces and the normalised acceleration of the flag in the spanwise-averaged momentum balance ( Eloy et al. 2007 ) . Large mass ratios correspond to flags in a dense fluid, or light flags. The aspect ratio H/L compares the height to the length of a flag. Flags in the limit of low aspect ratio H/L\rightarrow 0 behave like ribbons ( Lemaitre et al. 2005 ) . Two-dimensional conditions are approximated by large aspect ratio flags H/L\gg 1 and have been a classical simplification in the study of flag stability ( Alben 2022 ; Connell & Yue 2007 ; Michelin et al. 2008 ) . The stability boundary of the flag is marked by the critical reduced velocity. Analytical models of flags have focused on predicting the critical reduced velocity as a function of mass ratio and aspect ratio. The critical velocity is most sensitive to the mass ratio and the value of the mass ratio governs which flapping mode becomes unstable at the flapping onset ( Yu et al. 2019 ) . At low mass ratio ( M^{*}<\mathcal{O}(1) ), the critical reduced velocity U^{*} is a decreasing function of M^{*} . For higher mass ratios ( M^{*}>1 ), the critical reduced velocity settles around U^{*}\sim$10$ and different branches described the evolution of U^{*} with M^{*} depending on the unstable flapping mode that is excited ( Michelin et al. 2008 ; Eloy et al. 2008 ) . The minimum values of the critical velocity for different branches are found for infinite aspect ratio flags. Finite aspect ratio flags are subject to three-dimensional flow effects that reduce the pressure forces and delay the onset of flapping to higher reduced velocities. The critical reduced velocity tends to increase with decreasing aspect ratio H/L ( Eloy et al. 2007 ; Kumar et al. 2024 ) . Overall, analytical models tend to underestimate the critical velocity compared to experimental results ( Yu et al. 2019 ) , and critical velocities obtained from experiments show some inherent fluctuations, some of which are due to planeity defects ( Eloy et al. 2012 ) . Flags maintain large amplitude flapping for reduced velocities beyond the onset velocity, which is referred to as the post-critical regime ( Tang & Païdoussis 2007 ) . The spatial characteristics of post-critical flapping are extracted from the flapping amplitude envelope. The first person to experimentally study the amplitude envelope through stroboscopic imaging and to identify a travelling deformation wave propagating towards the tip Taneda 1968 . Depending on mass ratio, different flapping envelopes are observed for different flapping modes ( Taneda 1968 ) . Necks are used to describe the spatial evolution of the flapping envelope. Necks are regions of the flag where the envelope has locally a smaller amplitude. The number of necks along the flag increases with higher order flapping modes and change the bending distribution along the flag. An accurate localisation of the high bending regions along the flag is important to design energy harvesters based on piezzo-electric patches, where the typical milli-Watt output power is relevant to supply wireless sensor networks ( Doaré & Michelin 2011 ; Yu & Liu 2016 ) and for battery-less flow sensors ( Liu et al. 2012 ) . The main temporal characteristic of post-critical flapping is the flapping frequency. The tip frequency f increases linearly with the inverse of the convective time {\textrm{U}}_{\scalebox{0.65}{$\infty$}}/L ( Yu et al. 2019 ) . The ratio fL/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} corresponds to the Strouhal number and typically ranges between 0.11 with several groups of values for each flapping mode ( Taneda 1968 ) . The tip velocity decreases, with lower dimensionless frequency fL/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} , for decreasing aspect ratios at constant mass ratios ( Huang & Sung 2010 ; Kumar et al. 2024 ) . An explanation for the slowdown of the tip velocity is that pressure difference is reduced with lower aspect ratio flags due to flow structures at the edges, reducing the driving force of the flapping ( Huang & Sung 2010 ) . More details on the origin of the flapping time scales is relevant for precise diagnostics of respiratory wheezes ( Gavriely et al. 1989 ) , orinasal snoring and sleep apnoea syndromes ( Huang 1995 ; Aurégan & Depollier 1995 ) where the frequency of pressure waves is perceived as an audible sound. Understanding the effect of aspect ratio and mass ratio on the flapping timescales also matters because human tissues come in a variety of shapes and thicknesses. During the large amplitude flapping, vorticity is accumulated along the surface on both side of the flags and released alternatingly during the upstroke and downstroke motion of the tip ( Kumar et al. 2021 ) . Three-dimensional flow structures are characterised by an alternating release of horse-shoe shaped vortex tubes from the trailing edge, and the wake structures weaken with decreasing aspect ratio ( Huang & Sung 2010 ; Kumar et al. 2024 ) . The time-varying wake produced during flutter serves practical applications such as micro-scale mixers on lab-on-chip devices ( Rips & Mittal 2019 ) , or low-Reynolds number heat dissipation systems ( Shoele & Mittal 2014 ) . The wake past a fluttering leaflet of an aortic valves prosthesis generates higher levels of shear stress compared to a non-fluttering leaflet, contributing to the formation of thrombus ( Bornemann & Obrist 2025 ) . Functional relationships between the shape properties and wake quantities would help design biomedical implants and mixing devices matching the desired flow properties. A sudden increase in drag force is observed as the flag starts flapping ( Taneda 1968 ) . The origin of the drag increase has been associated with the dynamically-induced tension of the oscillating flag ( Moretti 2003 ) , and with flow reorientation in the alternating wake ( Müller 2003 ; Rips & Mittal 2019 ) . Strong variations are observed in the mean and fluctuating drag coefficients and the efforts to model the drag have focused on quantities derived from the tip amplitude and frequency ( Virot et al. 2013 ) . Increased forces compared to a non-flapping flag is responsible for material damage, and can deteriorate the production quality in fast-operating industrial printing press ( Watanabe et al. 2002 ) . Overall, a better understanding of the interaction between flag deformation, wake parameters, and force would be desirable to better characterise the post-critical regime and its dependency with the flag shape. In the present work, we study the effect of aspect ratio on the post-critical flapping of rectangular flags. Variations in flapping timescales, tip speed, and wave propagation speeds are investigated through systematic measurements of deformation at and after the onset of flapping. Flow measurements of the close wake show how these changes in timescales and speeds affect the formation of the vortex wake. Mean drag data is measured and presented for various reduced velocities. We discuss and test a predictive model for the mean drag coefficients based on the tip velocity. 2 Experimental methods 2.1 Rectangular flags Rectangular flags were made out of white paper (PAPETERIA A4 80\text{\,}\mathrm{g}\text{\,}{\mathrm{m}}^{-2} ). Flags were designed with alignment holes to be clamped on a flag pole and held with the long paper side along the stream-wise direction ( fig. 1 a). The rigidity of the A4 paper sheets in the long direction was measured using an elasto-gravity bending procedure, resulting in D=Ee^{3}/12(1-{\nu}_{\scalebox{0.65}{s}}^{2})=$3.4(0.4)e-4\text{\,}\mathrm{N}\text{\,}\mathrm{m}$ where E and {\nu}_{\scalebox{0.65}{s}} are the Young modulus and the Poisson ratio. A total of 48 shapes were cut using a desktop laser cutter with various heights H ranging from 4\text{\,}\mathrm{cm}19.6\text{\,}\mathrm{cm} and lengths L ranging from 10\text{\,}\mathrm{cm}20\text{\,}\mathrm{cm} ( fig. 1 b). In this study, we consider the aspect ratios $0.22$18 . In this post-critical region, the dimensionless tip frequency is also approximately constant fL/{\textrm{U}}_{\scalebox{0.65}{$\infty$}}=$0.53(0.01)$ when varying the reduced velocity U^{*} . When the velocity is reduced starting from the post-critical regime, the flag restabilises. The velocity at which the flag restabilises is called the offset velocity and is typically smaller than the onset velocity. We systematically extracted the critical reduced offset velocity for all tested flags as the lowest flow velocity with a significant flapping amplitude A/L>0.1 . The extraction of the flapping offset is more repeatable and less affected by planeity defects than the flapping onset velocity ( Eloy et al. 2012 ) . The reduced offset velocity {U}_{\scalebox{0.65}{offset}}^{*} lies between 12.78 and 17.17 for different mass ratios $1.4${r}_{\scalebox{0.65}{2}} is extracted to get the corresponding wave speed c and wavelength \lambda=c/f . (d) Wave length \lambda/L as a function of aspect ratio for all tested flags. The dashed horizontal lines indicate the dimensionless wavelengths of the three first cantilevered beam modes. The properties of the travelling wave are extracted based on the full spatiotemporal deformation data of the centreline. We can visually represent the transverse deformation y/L along the centreline as a function of the curvilinear coordinates s/L and normalised convective time t{\textrm{U}}_{\scalebox{0.65}{$\infty$}}/L in the form space-time diagrams. The space-time diagrams for two example flags, H/L=$0.28$ and H/L=$0.91$ , are represented in fig. 4 a and b, respectively. The mass ratio for both example flag is M^{*}=$2.27$ . Both space-time diagrams feature diagonal patterns indicating a growing amplitude in time and space. The vertical spacing between two consecutive y=0 contour lines corresponds to half of the dimensionless flapping period. The horizontal spacing corresponds to half the dimensionless wavelength, i.e. \lambda/2L , and represents the distance along the flag between two opposite signed crests. The slope of the diagonals corresponds to \lambda f/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} , with \lambda f the wave speed which we denote by c . To systematically and reliably extract the wave length and wave speed, we correlate the temporal evolution of the measured flag deflection at different locations along the centreline. We introduce the correlation delay \tau ( fig. 4 c) as a measure of the propagation delay of the wave pattern along the flag. The time delay \tau at a given position s with respect to a reference position {s}_{\scalebox{0.65}{ref}} is obtained using the maximum of the cross correlation of y(s,t) and {y}_{\scalebox{0.65}{ref}}(t)=y({s}_{\scalebox{0.65}{ref}},t) : \tau(s)=\text{argmax}_{\tilde{\tau}}\int y({s}_{\scalebox{0.65}{ref}},t)y(s,t+\tilde{\tau})\differential t\quad. (3) We integrate over the recorded duration covering at least 80 cycles of the slowest flapping flags and up to 450 cycles for the fastest flapping flags. The reference position is chosen for all the flags at {s}_{\scalebox{0.65}{ref}}/L=0.5 . The results in fig. 4 c are shifted such that the first measurable position along s has a zero delay \tau to facilitate the comparison. The correlation delay increases non-linearly in the first half of the centreline and linearly for curvilinear positions beyond the radius of second moment of area s>{r}_{\scalebox{0.65}{2}} . The wave speed is extracted for each of the measurement series by linearly fitting the correlation delay as a function of s/L for s>{r}_{\scalebox{0.65}{2}} (shaded area in fig. 4 c). The wave length is then calculated as \lambda=c/f . The resulting wave length values for all flags are summarised in fig. 4 d. The wavelength is approximately constant for the full dataset when normalised by the length of the flag \lambda/L=$1.87(0.08)$ . The lowest mass ratio series have slightly lower wavelengths. The dimensionless wavelength is larger than the wavelength of the second cantilevered beam mode shape ( m=$2$ in fig. 4 d), but significantly smaller than the first mode ( m=$1$ ). Low mass ratio flags were characterised by a larger contribution of third mode shape ( fig. 3 i). The wavelength of the third mode is smaller than the wavelength of the first and second mode ( fig. 4 d) and its larger contribution in low mass ratio flags explains the decrease of wavelength with decreasing mass ratio. The main spatial characteristics of the post-critical flag flapping are the amplitude envelope, the tip amplitude, and the wavelength of the travelling deformation wave. All flags tested in this study show the same double-neck flapping mode in the post-critical regime. The spatial characteristics non-dimensionalised by the length of the flag are independent of the aspect ratio, and are predominantly governed by the mass ratio. An increase in the mass ratio leads to a shift in the necks toward the tip and a decrease in the nondimensional tip amplitude. The dimensionless wavelength remains approximately constant and only slightly decreases for the lowest mass ratios tested. Next, we will focus on how the aspect ratio and mass ratio affect the temporal characteristics of the deformation speed, including the wave speed, oscillation frequency, and Strouhal numbers. 3.3 Temporal characteristics of the deformation waves The results of the extracted wave speeds from the spatial-correlation delay plots ( fig. 4 c) are presented in fig. 5 a for flags of different aspect ratio, at different mass ratios, and reduced velocities in the post-critical regime ( 181 are observed because the wave speed is measured along the curvilinear abscissa of the flag and the motion travels faster in curvilinear coordinates than in the streamwise direction. Dimensionless wave speed increases both with aspect ratio and mass ratio. Waves travel faster for taller flags when we fix the mass ratio. The dimensionless wave speed c/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} starts from 0.61 at H/L=0.5 and increases to c/{\textrm{U}}_{\scalebox{0.65}{$\infty$}}=$1.04$ when the aspect ratio reaches H/L=1.92 for the example case of M^{*}=$1.4$ . The wave speed also increases with mass ratio when we fix the aspect ratio. Higher mass ratio flapping flags has less inertia, and resist less to the driving force of the flow. Analytical models often assume that deformations and the associated flow structures are convected at the same speed as the external flow ( Argentina & Mahadevan 2005 ) . Shelley et al. 2005 also observed a linear increase of the wave speed with flow velocity. They found c/{\textrm{U}}_{\scalebox{0.65}{$\infty$}}\approx$0.66$ , which is on the lower side of our data range. The value c/{\textrm{U}}_{\scalebox{0.65}{$\infty$}}\approx$0.66$ was obtained for a flag of aspect ratio H/L=$1.46$ and mass ratio M^{*}=$19.8$ , which lies outside our data range ( 1.4}_{\scalebox{0.65}{z}} , according to the model proposed by Eloy et al. 2007 : {<\tilde{p}>}_{\scalebox{0.65}{z}}(\tilde{s},\tilde{t})=\frac{1}{M^{*}}\left[\frac{\partial^{2}\tilde{y}(\tilde{s},\tilde{t})}{\partial\tilde{t}^{2}}+\frac{1}{U^{*2}}\frac{\partial^{4}\tilde{y}(\tilde{s},\tilde{t})}{\partial\tilde{x}^{4}}\right]\quad, (4) where {<.>}_{\scalebox{0.65}{z}} stands for spanwise averaged quantities, and \tilde{.} stands for dimensionless quantities: \tilde{y}=\frac{y}{L},~\tilde{s}=\frac{s}{L},~\tilde{x}=\frac{x}{L},~\tilde{t}=\frac{t{\textrm{U}}_{\scalebox{0.65}{$\infty$}}}{L},~\tilde{p}=\frac{p}{{\rho}_{\scalebox{0.65}{f}}{\textrm{U}}_{\scalebox{0.65}{$\infty$}}^{2}}\quad. (5) Equation 4 is the result of the linearised conservation of transverse momentum equation (more details on the derivation are included in appendix appendix B ). To obtain a single representative value of the pressure difference per flapping flag at a given U^{*} , we compute the overall euclidean norm of the calculated {<\tilde{p}>}_{\scalebox{0.65}{z}}(\tilde{s},\tilde{t}) over both \tilde{s} and \tilde{t} . We denote this overall dynamic pressure difference by \left|{<\tilde{p}>}_{\scalebox{0.65}{z}}\right|_{s,t} and show the results for our entire dataset in fig. 5 b. The estimated overall dynamic pressure difference increases with increasing aspect ratio and decreasing mass ratio. It is more sensitive to variations in aspect ratio than in mass ratio. Our data is compared in fig. 5 b to predictions by Eloy et al. 2007 and Argentina & Mahadevan 2005 . The predictions by Eloy et al. 2007 stem from a linear stability analysis of the three-dimensional potential flow around a flag of finite height in the Fourier space. This approach yields a single spanwise-averaged pressure difference as a function of the normalised streamwise wave number {\bar{k}}_{\scalebox{0.65}{x}}=\pi H/\lambda , which they fitted by {<\tilde{p}>}_{\scalebox{0.65}{z}}\left({\bar{k}}_{\scalebox{0.65}{x}}\right)~\approx~1-\frac{1}{\displaystyle 2\pi\frac{L}{\lambda}\frac{H}{L}+\exp{\left(\left(\pi/4-2 \right)\pi\frac{L}{\lambda}\frac{H}{L} \right)}}\quad. (6) The fit is shown in fig. 5 b by the dashed line. Argentina & Mahadevan 2005 followed a different approach to determine three-dimensional or aspect ratio effect on pressure. They numerically calculate a correction factor for the pressure A , which is the ratio of the spanwise-averaged non-circulatory potential of a plate of finite height and the non-circulatory potential of an infinite-height plate facing a constant flow. We extract the values of A(H/L) from figure 5b in Argentina & Mahadevan 2005 and include it in fig. 5 b for further comparison. Our experimental data is in good agreement with the theoretical prediction by Eloy et al. 2007 and the numerically obtained correction factor by Argentina & Mahadevan 2005 . Our pressure estimates lie between the values predicted by both models for low and high aspect ratios and is smaller in the intermediate region 0.5}_{\scalebox{0.65}{z}}}\quad. (7) This scaling indicates that there is a contribution of the mass ratio and the dynamic pressure difference on the wave speed. This scaling confirms our experimental observation of increasing wave speed with increasing mass ratio ( fig. 5 a) and with increasing dynamic pressure difference, which occurs when the aspect ratio increases ( fig. 5 b). The wave propagation speed is directly linked to the flapping frequency c=\lambda f . The flapping frequency is typically non-dimensionalised using a characteristic length scale and the flow velocity to form a Strouhal number. In literature, both the tip to tip amplitude 2A and the flag length L are used to define a Strouhal number. In fig. 5 c, we show the amplitude-based Strouhal number {St}_{\scalebox{0.65}{A}}=2fA/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} against aspect ratio for all our data. The amplitude-based Strouhal number increases from 0.100.39 with increasing aspect ratio, with most of the data point above 0.20 . We compare our data in fig. 5 c to results from three-dimensional numerical simulations by Huang & Sung 2010 , who also observed an increase of {St}_{\scalebox{0.65}{A}} with aspect ratio. The values of {St}_{\scalebox{0.65}{A}} from the simulations are lower in magnitude due to a different flapping regime ( M^{*}=$1$ and 0.25 corresponding to the single neck flutter). In fig. 5 d, we show the length-based Strouhal number {St}_{\scalebox{0.65}{L}}=fL/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} against aspect ratio for all our data. The dimensionless tip frequency fL/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} increases with aspect ratio, and to a lesser extent with mass ratio for our tested flags, which are all in the double-neck flutter regime. On the lower side, fL/{\textrm{U}}_{\scalebox{0.65}{$\infty$}}=$0.40$ and a half stroke occurs in 1.25 convective times ( L/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} ). On the higher side, fL/{\textrm{U}}_{\scalebox{0.65}{$\infty$}}=$0.55$ such that half a stroke of the tip occurs in only 0.91 convective times. Experimental values of fL/{\textrm{U}}_{\scalebox{0.65}{$\infty$}}\sim 0.4-0.6 are reported in Virot et al. 2013 for a two-neck flutter and most of the references in the literature report fL/{\textrm{U}}_{\scalebox{0.65}{$\infty$}}\approx 0.23 for single neck flutter. Overall, the magnitude of the dimensionless tip-frequency depends on the flutter mode, and increases with aspect ratio within a flutter regime. The relationship between tip frequency and wave speed can be rewritten as \frac{c}{{\textrm{U}}_{\scalebox{0.65}{$\infty$}}}=\frac{\lambda}{L}\times\frac{fL}{{\textrm{U}}_{\scalebox{0.65}{$\infty$}}}. (8) In fig. 5 e, we plot the measured wave speed against the measured tip frequency. For our dataset, c/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} indeed increases approximately linearly with fL/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} , with a slope equal to \lambda/L . The range of length-based Strouhal number within a given flutter regime is explained by variations in the wave propagation speed. Larger variations in the length-based Strouhal number are associated with variations in the flutter regime. For a flutter regime with more necks and a lower wavelength, we expect higher length-based Strouhal numbers. Figure 5: (a) Dimensionless wave speed c/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} obtained from the correlation delay plots against aspect ratio H/L , coloured by mass ratio M^{*} . (b) Estimate of the overall dynamic pressure difference \|<\bar{p}_{n}>_{\bar{z}}\|_{s,t} against aspect ratio compared with the correction factor A(H/L) from Argentina & Mahadevan 2005 and the spanwise averaged pressure correction {<\tilde{p}>}_{\scalebox{0.65}{z}} from Eloy et al. 2007 . (c) Experimental values of the amplitude-based Strouhal number {St}_{\scalebox{0.65}{A}}=2fA/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} (circles) as a function of aspect ratio. For comparison, we included results from three-dimensional numerical simulations at Re=500 (rectangles) and the infinite aspect-ratio limit (dashed line) from Huang & Sung 2010 . (d) Experimental values of the length-based Strouhal number fL/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} (circles) as a function of aspect ratio. For comparison, we included experimental results from Virot et al. 2013 (triangles). (e) Dimensionless wave speed c/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} against flapping frequency fL/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} . The dashed line indicates the linear relationship between c/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} and fL/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} for the mean measured dimensionless wavelength \lambda/L=1.87 . 3.4 Effect of aspect ratio on wake quantities Figure 6: Snapshots of phase-averaged dimensionless vorticity \omega L/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} for three flag with different aspect ratio: (a) H/L=$1.25$ , (b) H/L=$0.875$ , (c) H/L=$0.5$ . The selected snapshot correspond to four instants during the upstroke: (1) t/T=0 when the tip deflection {y}_{\scalebox{0.65}{tip}} reaches its minimum value, (2) t/T=0.16 , (3) t/T=0.32 shortly after the tip passes the midline and (4) t/T=0.48 . Ridges of the negative finite time Lyapunov exponent are overlaid on the vorticity field. To characterise how the vortex formation varies with aspect ratio, we performed flow measurements at mid-height for a series of flags with the same mass ratio M^{*}=$2.27$ . In fig. 6 , we show snapshots of vorticity for three different flags with different aspect ratios ( H/L=$1.25$ , 0.875 and 0.5 for fig. 6 ). For each flag, we selected four snapshots covering half a flapping cycle. We focus on the upward stroke of the flag, where the tip moves from negative y to positive y , which respectively correspond to the bottom and the top in fig. 6 . The upward stroke begins when the tip is at the most extreme deflection (first row of fig. 6 ). Negative vorticity is present along the outside of the curvature of the flag close to the tip (indicated by arrow 1 in fig. 6 a1). When the flag start to move back toward the centerline, the negative vorticity is continuously released into the wake (second and third row of fig. 6 ). When the flag reaches its maximum deflection on the other side, there is no more negative vorticity released at the tip (fourth row of fig. 6 ). Before switching to the downward motion, positive vorticity accumulates near the tip (indicated by arrow 2 fig. 6 a4) and that phase corresponds to the start of the second half cycle. For the three different aspect ratio flags, negative and positive vorticity are alternatingly released in the wake during upstroke and downstroke. For the tallest flag, the vorticity released in the wake rolls up into a coherent concentrated vortex core. With decreasing aspect ratio, the intensity of the vorticity in the wake decreases and the wake vortices are less coherent. We highlight the topology of the vortex street in the wake using the negative finite time Lyapunov exponent (nFTLE) ( Green et al. 2011 ) . The nFTLE ridges are overlaid on the vorticity fields to depict the attractor lines in the flow ( fig. 6 ). The nFTLE ridges coincide with regions of increased shear. As the vortex street develops in the wake, the nFTLE ridges undulate in streamwise direction. The wavelength of the nFTLE ridges in the streamwise direction is decreases with increasing aspect ratio. The wavelength of the undulating nFTLE ridge is approximately twice the distance between two consecutive vortex cores, which is approximately {\textrm{U}}_{\scalebox{0.65}{$\infty$}}/2f if we assume that the vortices are convected downstream with a velocity equal to {\textrm{U}}_{\scalebox{0.65}{$\infty$}} . As taller flags flap faster, the wake wavelength \sim{\textrm{U}}_{\scalebox{0.65}{$\infty$}}/f decreases and consecutive vortices are shed closer to each other. The nFTLE ridge rolls-up in the close wake ( 10.5 in fig. 7 a). At the tip reversal around t/T=0.5 , positive vorticity is released, and we observe the same evolution of the circulation as in the first half cycle, but with opposite sign. For both positive and negative circulation, the maximum value occurs around or slightly after the tip reversal. Maxima of circulation nearly double from {\Gamma}_{\scalebox{0.65}{max}}/{\textrm{U}}_{\scalebox{0.65}{$\infty$}}L=$0.72$ to 1.35 with aspect ratios increasing from H/L=$0.375$ to 1.25 ( fig. 7 ). The variations of the non-dimensional circulation we observe for flags of the same length, reduced velocity, and Reynolds number suggest that L is not the only characteristic dimension to scale the flow quantities. We propose two alternative length scales to account for the variations in circulation: the ratio of the area over the perimeter L^{*}=LH/(2L+2H) and the square root of the area \sqrt{LH} . The area-to-perimeter ratio L^{*} is a fourth of the hydraulic diameter which is used to scale the wake dynamics behind flat plates ( Fernando & Rival 2016 ) . The square root of area \sqrt{LH} performed better at scaling circulation, hydrodynamic impulse, and kinetic energy for accelerated polygonal disks ( Caverly & Nedić 2025 ) . The square root of area is also more robust in the case of fractal objects in steady motions where the perimeter to area ratio can become extremely large without having a measurable influence on the flow properties ( Nedić et al. 2013 ) . We observe a collapse of the temporal evolution of the circulation for the different aspect ratios when replacing L by L^{*}=LH/(2L+2H) ( fig. 7 b). The same holds when we L by \sqrt{LH} . For different aspect ratios, we obtain a similar mean maximum value of {\Gamma^{*}}_{\scalebox{0.65}{max}}={\Gamma}_{\scalebox{0.65}{max}}/{\textrm{U}}_{\scalebox{0.65}{$\infty$}}L^{*}=$4.74(0.30)$ and {\Gamma^{\dagger}}_{\scalebox{0.65}{max}}={\Gamma}_{\scalebox{0.65}{max}}/{\textrm{U}}_{\scalebox{0.65}{$\infty$}}\sqrt{LH}=$1.16(0.06)$ ( fig. 7 c). The standard deviations are 7\text{\,}\mathrm{\%} and 5\text{\,}\mathrm{\%} , for the two scaling approaches respectively. The collapse of the different curves show that the right length scale for the mid-plane circulation should include the height to account for three-dimensional effects. Figure 7: Evolution of the phase-averaged (a) dimensionless circulation \Gamma/{\textrm{U}}_{\scalebox{0.65}{$\infty$}}L and (b) the scaled circulation \Gamma/{\textrm{U}}_{\scalebox{0.65}{$\infty$}}L^{*} as a function of dimensionless time t/T . The instants t/T=$0$ and 0.5 corresponds to the moment of minimum and maximum tip deflection. (c) Maxima of the positive and negative circulation using L^{*} and \sqrt{LH} for the aspect ratio scaling. 3.5 Link between mean drag coefficient and flapping dynamics The mean drag force {\bar{F}}_{\scalebox{0.65}{x}} at the root of the flag is measured systematically for different flow velocities. We take the example of three flags of equal area to analyse how the drag varies with flow velocity ( fig. 8 a). The mean drag is close to zero before the flapping onset as the only contribution to drag comes from the friction drag. A jump to a significant positive mean drag value occurs after the flapping onset when the pressure drag increase due to the flapping. With increasing flow velocity, the mean drag increases following a quadratic increase with flow velocity as we expect for bluff bodies ( fig. 8 a). The mean drag significantly differs between the three flags of same area when compared at the same flow velocity. The three flags have different lengths, so their flapping onset occurs at a different flow velocity. The short flag starts flapping before the tall flag such that the drag of the short flag is higher around {\textrm{U}}_{\scalebox{0.65}{$\infty$}}=$7\text{\,}\mathrm{m}\text{\,}{\mathrm{s}}^{-1}$ . When the tallest flag finally starts flapping, the mean drag increases faster with {\textrm{U}}_{\scalebox{0.65}{$\infty$}} than for the other flags and overtakes the values of the other flags for {\textrm{U}}_{\scalebox{0.65}{$\infty$}}>$8\text{\,}\mathrm{m}\text{\,}{\mathrm{s}}^{-1}$ . The mean drag coefficient {\bar{C}}_{\scalebox{0.65}{x}}=\frac{{\bar{F}}_{\scalebox{0.65}{x}}}{\frac{1}{2}{\rho}_{\scalebox{0.65}{f}}HL{\textrm{U}}_{\scalebox{0.65}{$\infty$}}^{2}} (9) of the three example flags undergoes a transition from 0 to a plateau at post-critical velocities. The plateau values vary between 0.15 for H/L=$0.59$ to 0.35 for H/L=$1.33$ ( fig. 8 b). The individual values of mean drag coefficient for the 48 flags tested display a wide distribution ranging from {\bar{C}}_{\scalebox{0.65}{x}}=$00.55$ with no clear relationship with the reduced velocity U^{*} ( fig. 8 c). The absence of clear trend suggests that the choice of typical area or velocity are not the relevant scales to predict the mean force acting on the flag. The projected area 2AH , which takes into account the difference in wake width, is not a suitable replacement for the flag area HL as shown in the appendix ( appendix C ). The projected area slightly stabilises the mean drag coefficient in the transition region ( fig. 10 a) but does not reduce the overall scattering of the drag coefficient values ( fig. 10 b). At fixed reduced velocity U^{*} , the mean drag coefficient {\bar{C}}_{\scalebox{0.65}{x}} tends to increase with aspect ratio. The scattering of mean drag coefficient suggests that the typical aerodynamic force normalisation with the surface area and the square of the flow velocity fails to scale the mean drag of flapping flags. To explain the distribution of mean drag coefficient, we revisit the work of Thoma 1939 and derive a semi-empirical prediction using the measured flag tip velocity. Thoma 1939 presented a model for a fully flexible rope oscillating under the action of normal forces and derived a two-dimensional estimation of the mean tension at the root of the rope. The mean drag per unit height was derived as {\bar{F}}_{\scalebox{0.65}{x}}=\frac{1}{2}m{\overline{v^{2}}}_{\scalebox{0.65}{tip}}\quad, (10) where \overline{.} stands for a time-averaged quantity, and m={\rho}_{\scalebox{0.65}{s}}eH is the mass per unit length of the rope. For flags, {\rho}_{\scalebox{0.65}{s}}eHL is the total mass. The right-hand side of eq. 10 represents the tension induced by the in-plane kinetic energy of the flag. In dimensionless quantities, eq. 10 can be formulated as {\bar{C}}_{\scalebox{0.65}{x}}=\displaystyle\frac{\frac{1}{2}{\rho}_{\scalebox{0.65}{s}}eH{\overline{v^{2}}}_{\scalebox{0.65}{tip}}}{\frac{1}{2}{\rho}_{\scalebox{0.65}{f}}HL{\textrm{U}}_{\scalebox{0.65}{$\infty$}}^{2}}=\frac{{\rho}_{\scalebox{0.65}{s}}e}{{\rho}_{\scalebox{0.65}{f}}L}\times\frac{{\overline{v^{2}}}_{\scalebox{0.65}{tip}}}{{\textrm{U}}_{\scalebox{0.65}{$\infty$}}^{2}}\quad, (11) which can be rewritten as {\bar{C}}_{\scalebox{0.65}{x}}=\frac{1}{M^{*}}\frac{{\overline{v^{2}}}_{\scalebox{0.65}{tip}}}{{\textrm{U}}_{\scalebox{0.65}{$\infty$}}^{2}}\quad. (12) The mean drag coefficient is proportional to the ratio of the mean square tip velocity and the incoming flow velocity squared with the mass ratio as the proportionality factor. In fig. 8 d, we combine the deformation measurement {v}_{\scalebox{0.65}{tip}} and the load cell data {\bar{C}}_{\scalebox{0.65}{x}} . The two-dimensional kinematic model of eq. 12 describes our data well. The influence of the aspect ratio on the mean drag coefficient is indirectly included in the form of the reduced tip speed. Following Moretti 2003 , we can simplify the tip speed by assuming that the tip displacement is described by {y}_{\scalebox{0.65}{tip}}=-A\cos{2\pi f t} . The simplified tip velocity takes the form \overline{{v}_{\scalebox{0.65}{tip}}^{2}}=2\pi^{2}f^{2}A^{2} . The mean drag coefficient estimate can be rewritten as {\bar{C}}_{\scalebox{0.65}{x}}=\frac{\pi^{2}}{2}\times\frac{{St}_{\scalebox{0.65}{A}}^{2}}{M^{*}}\quad, (13) where the amplitude-based Strouhal number {St}_{\scalebox{0.65}{A}}=2fA/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} , shown in fig. 5 c, accounts for the increasing tip speed with increasing aspect ratio. In fig. 8 e, we show the measured mean drag coefficient against the simplified estimate based on {St}_{\scalebox{0.65}{A}} from eq. 13 . The average discrepancies between the model and the measured data \left|{\bar{C}}_{\scalebox{0.65}{x}}-(\pi^{2}/2)\times{St}_{\scalebox{0.65}{A}}^{2}/M^{*}\right| is 0.023 . The simplified model is more accurate than the model using the direct tip velocity ( eq. 12 ) which has an average discrepancy \left|{\bar{C}}_{\scalebox{0.65}{x}}-{v}_{\scalebox{0.65}{tip}}^{2}/{\textrm{U}}_{\scalebox{0.65}{$\infty$}}^{2}M^{*}\right| of 0.039 . Overall, an increase in aspect ratio induces an increase in tip speed ( fig. 8 d), an increase in amplitude-based Strouhal number ( fig. 8 e), and an increase in mean drag coefficient. Figure 8: (a) Mean drag as function of flow velocity {\textrm{U}}_{\scalebox{0.65}{$\infty$}} for three flags of equal area HL=$192\text{\,}\mathrm{c}\mathrm{m}\mathrm{{}^{2}}$ and aspect ratio H/L=$0.590.981.33$ . (b) Corresponding mean drag coefficient {\bar{C}}_{\scalebox{0.65}{x}} . (c) Mean drag coefficient {\bar{C}}_{\scalebox{0.65}{x}} for flags with 1.4}_{\scalebox{0.65}{z}}, (15) where the \tilde{~} stands for dimensionless quantities normalised by L , {\textrm{U}}_{\scalebox{0.65}{$\infty$}} and {\rho}_{\scalebox{0.65}{f}} : \tilde{y}=\frac{y}{L},~\tilde{x}=\frac{x}{L},~\tilde{t}=\frac{t{\textrm{U}}_{\scalebox{0.65}{$\infty$}}}{L},~\tilde{p}=\frac{p}{{\rho}_{\scalebox{0.65}{f}}{\textrm{U}}_{\scalebox{0.65}{$\infty$}}^{2}}. (16) We numerically differentiate our time and space resolved data \tilde{y}(\tilde{s},\tilde{t}) and use the small deformation approximation \tilde{s}\approx\tilde{x} to replace \tilde{y}(\tilde{x},\tilde{t}) by \tilde{y}(\tilde{s},\tilde{t}) . The second derivative in time and fourth derivative in space provide the left hand side of eq. 15 , which is equal to the spanwise-averaged pressure estimate {<\tilde{p}>}_{\scalebox{0.65}{z}} multiplied by the mass ratio M^{*} on the right hand side. The spanwise-averaged pressure estimate {<\tilde{p}>}_{\scalebox{0.65}{z}} is a space-time field of the \tilde{s} and \tilde{t} variables. To compare different measurement time series, we take the euclidean norm in space and time over the recorded duration {T}_{\scalebox{0.65}{r}} : \|<\tilde{p}>_{z}\|_{s,t}=\left(\frac{L}{{T}_{\scalebox{0.65}{r}}{\textrm{U}}_{\scalebox{0.65}{$\infty$}}}\int_{\tilde{t}=0}^{\tilde{t}={T}_{\scalebox{0.65}{r}}{\textrm{U}}_{\scalebox{0.65}{$\infty$}}/L}\int_{\tilde{s}=0}^{\tilde{s}=1}<\tilde{p}>^{2}_{z}d\tilde{s}d\tilde{t}\right)^{1/2}. (17) Figure 9: Details on the phase-averaged mode coefficients corresponding to the three first cantilevered beam mode shapes. We simplify eq. 15 to highlight the link between dynamic pressure and wave speed. We follow the approach of Moretti 2003 that consists of (1) using an analytical approximation for the deformation and (2) neglecting the bending term of the momentum equation. The analytical approximation for the deformation \tilde{y} is a sinusoidal travelling wave modulated by a space-varying amplitude \tilde{y}(\tilde{s},\tilde{t})=\frac{A\tilde{x}}{L}\sin\left(\frac{2\pi}{\tilde{\lambda}}(\tilde{c}\tilde{t}-\tilde{x})\right), (18) where A is the tip amplitude, \tilde{\lambda}=\lambda/L is the dimensionless wavelength and \tilde{c}=c/{\textrm{U}}_{\scalebox{0.65}{$\infty$}} is the dimensionless wave speed. The simplification of a linear amplitude growth in eq. 18 is justified by Moretti 2003 citing the envelope data from Watanabe et al. 2002 . The linear approximation does not reproduce the two dips around the necks compared to the amplitude growth data presented in fig. 3 c and e, but this simplification captures the overall behavior of the increasing envelope from root to tip. The expression for the deformation of eq. 18 is injected in the left hand side of momentumm balance of eq. 15 . The bending stiffness is neglected, which corresponds to high U^{*} so that the term \frac{1}{U^{*2}}\frac{\partial^{4}\tilde{y}}{\partial\tilde{x}^{4}} is neglected compared to the acceleration \frac{\partial^{2}\tilde{y}}{\partial\tilde{t}^{2}} in eq. 15 . The momentum balance writes as \frac{\partial^{2}\tilde{y}}{\partial\tilde{t}^{2}}=-\frac{A\tilde{x}}{L}\left(\frac{2\pi\tilde{c}}{\tilde{\lambda}}\right)^{2}\sin\left(\frac{2\pi}{\tilde{\lambda}}(\tilde{c}\tilde{t}-\tilde{x})\right)=M^{*}{<\tilde{p}>}_{\scalebox{0.65}{z}} (19) for which we take the euclidean norm over space and the recorded duration \frac{A}{L}\left(\frac{2\pi\tilde{c}}{\tilde{\lambda}}\right)^{2}\norm{ \tilde{x} \sin\left(\frac{2\pi}{\tilde{\lambda}}(\tilde{c}\tilde{t}-\tilde{x})\right) }_{s,t}=M^{*}\norm{ \kindex{<\tilde{p}>}{z} }_{s,t}. (20) The right hand side is our pressure estimate introduced in eq. 17 and the left hand side contains the properties of the wave approximation. The norm of the \norm{ \tilde{x} \sin\left(\frac{2\pi}{\tilde{\lambda}}(\tilde{c}\tilde{t}-\tilde{x})\right) }_{s,t} does depend of \tilde{\lambda} but is overall of the order of unity. We rewrite eq. 20 using the dimensional values and taking the norm of the envelope times the sine as a constant factor: \frac{1}{M^{*}}\left(\frac{c}{{\textrm{U}}_{\scalebox{0.65}{$\infty$}}}\right)^{2}\frac{A}{L}\propto\left(\frac{\lambda}{L}\right)^{2}{<\tilde{p}>}_{\scalebox{0.65}{z}}. (21) The constant dimensionless wavelength \lambda/L=$1.87$ and neglecting the tip amplitude variations (assuming A/L\approx 0.3 ) simplifies this relationship into: c/{\textrm{U}}_{\scalebox{0.65}{$\infty$}}\propto\sqrt{M^{*}{<\tilde{p}>}_{\scalebox{0.65}{z}}}. (22) Overall, the dimensionless wave speed varies as the square root of the mass ratio times the square root of the dynamic pressure and the dynamic pressure is weakened by lower aspect ratio flags as predicted in analytical models. Appendix C Force data Figure 10: Projected area drag coefficient for (a) the three flags of same area in fig. 8 a and (b) for all the different flags and flow velocities. The choice of scaling area for the drag coefficient is traditionally the non-deformed surface HL . The frontal area seen by the flow is estimated using the flapping amplitude and the height 2A\times H . The projected-area mean drag coefficient is defined as \bar{C}_{x,A}=\frac{{\bar{F}}_{\scalebox{0.65}{x}}}{\frac{1}{2}{\rho}_{\scalebox{0.65}{f}}{\textrm{U}}_{\scalebox{0.65}{$\infty$}}^{2}2AH}=\bar{C}_{x}\times\frac{L}{2A}, (23) and the evolution with reduced velocity for the three examples of fig. 8 a and the full dataset are displayed in fig. 10 a and b. Values of \bar{C}_{x,A} before the onset of flapping are heavily scattered due to the small values both for the amplitude and the mean drag. The transition region shortly after the flapping onset is closer to the plateau because the amplitude also varies during this transition region. The values of the projected-area mean drag coefficient at the plateau are still distinct ( \bar{C}_{x,A}=$0.250.430.5$ ). The difference in flapping amplitude or projected area does not explain the scattering of mean drag coefficient.