1. Introduction
Graceful and agile swimming of aquatic animals has driven extensive research in biofluid dynamics and inspired numerous designs for underwater propulsion systems. Although aquatic locomotion encompasses a wide variety of swimming modes, propulsion by body undulation in the form of a travelling wave appears to be predominant (Sfakiotakis et al. Reference Sfakiotakis, Lane and Davies1999; Lauder & Tytell Reference Lauder and Tytell2005). In particular, locomotion displaying an undulation wavelength longer than body length is commonly referred to as oscillatory swimming, as exemplified by tunas and dolphins. Since only the posterior part of the body performs reciprocal oscillations with the anterior portion acting as a stiff base, this type of propulsive appendage can be effectively modelled by a rigid foil undergoing a heave and/or pitch motion (Anderson et al. Reference Anderson, Streitlien, Barrett and Triantafyllou1998; Godoy-Diana et al. Reference Godoy-Diana, Marais, Aider and Wesfreid2009; Floryan et al. Reference Floryan, Van Buren, Rowley and Smits2017). In contrast, locomotion passing the travelling wave along the whole body is generally referred to as undulatory swimming. Examples such as eels and lampreys generate propulsion by undulating large portions of their bodies, typically producing waveforms that span at least one full wavelength. Modelling this type of motion involves generating a backward-propagating wave along the propulsive structure (Smits Reference Smits2019). In this paper, we focus on undulatory swimming.
The pioneering study on modelling undulatory swimming was conducted by Taylor (Reference Taylor1952), who idealised the long animals as a flexible cylinder and estimated the thrust production by the sum of the drag forces acting on the body with a wave travelling backwards. This theory is known as a resistive model and is largely restricted to low Reynolds numbers. Later, Lighthill (Reference Lighthill1960, Reference Lighthill1971) developed the elongated body theory to model the swimming of slender fish. At the same time, Wu (Reference Wu1961, Reference Wu1971) proposed a theory in which the swimming fish is treated as a waving plate. By assuming an inviscid fluid, these two theories and their extensions estimate forces based on the reaction force arising from the acceleration of the surrounding fluids, and thus classified into a reactive model. Subsequently, numerical simulation based on computational fluid dynamics emerges as a powerful tool for studying undulatory locomotion (Carling et al. Reference Carling, Williams and Bowtell1998). Such approaches are capable of handling complex fish geometries and, more importantly, capturing all relevant effects, such as viscous forces. With the advancement of measurement techniques, experimental methods have also become an important means of investigating aquatic locomotion, particularly playing a key role in visualising flow details and analysing fish behaviours under varying environmental conditions (Liao et al. Reference Liao, Beal, Lauder and Triantafyllou2003).
Over the past few decades, significant progress has been made in our understanding of undulatory swimming, driven by extensive research spanning a wide range of aspects. Among these, swimming performance characterised by speed and efficiency remains one of the main concerns. Undulatory swimmers have been shown to switch swimming patterns from efficient mode to fast mode by adjusting the amplitude of the tail-beat and the proportion of tail length participating in the undulation (Kern & Koumoutsakos Reference Kern and Koumoutsakos2006). Studies on the optimisation of undulatory propulsion demonstrate that swimming speed and energetic efficiency are also strongly influenced by body shape (Tokić & Yue Reference Tokić and Yue2012; Van Rees et al. Reference Van Rees, Gazzola and Koumoutsakos2015). Further research suggests that natural undulatory swimmers employ swimming strategies, which are far more sophisticated than previously recognised, to improve locomotor capacity and reduce energy expenditure (Gazzola et al. Reference Gazzola, Argentina and Mahadevan2015; Li et al. Reference Li, Liu, Müller, Voesenek, Van and Johan2021). Quantifying the relationship of swimming speed to undulatory kinematics and fluid properties is another research area that is of considerable interest. Such studies were initially motivated by the experimental observation that the forward speed of a flapping wing increases linearly with the flapping frequency (Vandenberghe et al. Reference Vandenberghe, Zhang and Childress2004). Subsequently, Alben & Shelley (Reference Alben and Shelley2005) confirmed that unidirectional locomotion arises as the dynamically preferred state for a free flapping body when the flapping frequency is sufficiently large. Deng & Caulfield (Reference Deng and Caulfield2016, Reference Deng and Caulfield2018) also conducted systematic studies on the free motion of a two-dimensional elliptical foil and a three-dimensional oblate spheroid undergoing vertical oscillations. Their findings showed that the body initially displays irregular locomotion depending on the aspect ratio, which eventually transitions to unidirectional motion once the oscillation frequency exceeds a critical threshold. Gazzola et al. (Reference Gazzola, Argentina and Mahadevan2014) conducted a comprehensive investigation of undulatory swimmers and derived a fundamental scaling relation that characterises their locomotion. Specifically, they found that the cruising Reynolds number based on swimming speed can be correlated to the swimming number defined by tail-beat amplitude and frequency via a simple power law. This scaling law is consistent with a substantial amount of biological data. Similar power laws are also identified in oscillatory swimming, with different exponents depending on body shape and rigidity (Lin et al. Reference Lin, Wu and Zhang2021; Liu et al. Reference Liu, Liu and Huang2022). In addition, much effort has been devoted to investigating wake formation and vortex manipulation arising from the complex interplay between undulatory swimmers and the fluid environment. A variety of wake patterns induced by bodies with travelling-wave undulation are observed in both numerical simulations and experimental measurements (Kern & Koumoutsakos Reference Kern and Koumoutsakos2006; Moored et al. Reference Moored, Dewey, Smits and Haj-Hariri2012). Interestingly, undulatory swimmers have been shown to enhance their locomotion performance by modulating interactions with ambient flows (Liao et al. Reference Liao, Beal, Lauder and Triantafyllou2003; Alben Reference Alben2009; Zhu et al. Reference Zhu, He and Zhang2014).
The studies mentioned above focus exclusively on homogeneous fluid environments, in which the physical properties of fluid, such as viscosity and density, are assumed to be uniform in space and constant in time. However, fluids in the real world are generally more heterogeneous than those we modelled, which is presumably motivating recent extensive studies on swimming in complex fluids (Spagnolie & Underhill Reference Spagnolie and Underhill2023). The existing literature on this topic is largely limited to the swimming of microorganisms at low Reynolds numbers, and the literature frequently reports inconsistent observations. For example, when swimming in viscoelastic fluids, fluid elasticity has been shown to hinder (Shen & Arratia Reference Shen and Arratia2011) or enhance (Espinosa-Garcia et al. Reference Espinosa-Garcia, Lauga and Zenit2013) the self-propulsion of undulatory swimmers, a seeming inconsistency that was later clarified by Thomases & Guy (Reference Thomases and Guy2014). In the case of swimming in shear-thinning fluids, Gagnon et al. (Reference Gagnon, Keim and Arratia2014) experimentally observed that shear-thinning viscosity does not influence the locomotion of undulatory swimmers, while numerical simulations by Li & Ardekani (Reference Li and Ardekani2015) showed that it can improve swimming speed. These inconsistencies highlight the inherent complexity of swimming in inhomogeneous fluid environments. Fluids stratified in certain properties represent another complex fluid environment that has attracted significant attention. Dandekar et al. (Reference Dandekar, Shaik and Ardekani2019) theoretically studied the undulatory locomotion of marine microorganisms in a density-stratified fluid. They considered an infinitely long Taylor swimming sheet that propels in the direction opposite to gravity and examined flow regimes at both low and finite Reynolds numbers. Their results showed that the effects of density stratification on swimming speed depend sensitively on the stratification strength, whereas the presence of density stratification consistently increases power expenditure. This theoretical method was also applied to undulatory swimming in a viscosity-stratified fluid (Dandekar & Ardekani Reference Dandekar and Ardekani2020). Using the same swimming sheet, they found that the tendency to migrate towards regions of favourable viscosity is governed by relative importance of convection and diffusion of the stratifying agent.
Among various complex fluids, density-stratified fluid is the environment most commonly encountered for relatively large swimmers, such as fish and cetaceans, as oceans are typically stratified in density due to vertical gradients of temperature and salinity. Although the ocean density gradient generally spans a large spatial scale, the fundamental length scale governing the influence of density stratification on the locomotion of marine swimmers is found to be much smaller and comparable to the scale of biological activity (Ardekani & Stocker Reference Ardekani and Stocker2010). In particular, the critical effects of density stratification have been observed in numerous physical processes, ranging from the settling dynamics of centimetre-scale zooplankton (Magnaudet & Mercier Reference Magnaudet and Mercier2020) to the wake evolution of metre-scale bodies (Spedding Reference Spedding2014). It is thus anticipated that the locomotion of both natural undulatory swimmers and bioinspired artificial systems will experience the significant effects of density stratification, especially within the oceanic regions characterised by steep density gradients, known as pycnoclines. Although the movement of small organisms in density-stratified fluids has been well documented in the literature (More & Ardekani Reference More and Ardekani2023), the influence of density stratification on undulatory swimming at relatively high Reynolds numbers, to our knowledge, remains rarely explored. Therefore, addressing the extent to which and the manner in which density stratification affects the fish-like locomotion serves as the primary motivation of this study.
The interplay between mobile structures and density stratification has received sustained attention for decades. In a density-stratified environment, vertically displaced fluid parcels experience a buoyancy-driven restoring force, often resulting in dynamical behaviours in marked contrast to those observed in homogeneous fluids. An astonishing example is that a wedge-shaped object floating in a density-stratified fluid can produce sideways thrust via diffusion-driven flow along its sloping boundaries, consequently giving rise to the motion at constant speed in the direction of the wedge tip (Allshouse et al. Reference Allshouse, Barad and Peacock2010). Due to the growing interest in oceanic transport and mixing processes, extensive work has been conducted on the settling dynamics of objects of various shapes in density-stratified fluids. The effects of density stratification have been indicated to induce oscillation in the settling velocity and even the levitation behaviour of a spherical particle (Doostmohammadi et al. Reference Doostmohammadi, Dabiri and Ardekani2014), enhance radial dispersion while retarding the vertical descent of discs (Lam et al. Reference Lam, Vincent and Kanso2019), increase the drag force acting on the bodies (Zhang et al. Reference Zhang, Mercier and Magnaudet2019) and modify the settling orientation of spheroids (More et al. Reference More, Ardekani, Brandt and Ardekani2021). Similar studies were carried out on the vertical motion of self-propelled swimmers on both individual and collective scales (Doostmohammadi et al. Reference Doostmohammadi, Stocker and Ardekani2012; More & Ardekani Reference More and Ardekani2020; Ouillon et al. Reference Ouillon, Houghton, Dabiri and Meiburg2020), showing that density stratification tends to reduce swimming speed, increase energy expenditure and result in distinctive flow patterns. Moreover, Christin et al. (Reference Christin, Meunier and Le Dizès2021) experimentally examined the influence of density stratification on wake-induced vibrations of a circular cylinder. In a density-stratified fluid, the horizontally towed cylinder with freedom to move in the vertical direction is found to exhibit different occurrence scenarios of oscillating modes compared with homogeneous observations. Le Gal et al. (Reference Le Gal, Castillo Morales, Hernandez-Zapata and Ruiz Chavarria2022) conducted a series of experiments on the dynamics of a neutrally buoyant body immersed in a density-stratified fluid. By imposing pressure oscillations on the system, they observed resonance phenomena and bifurcation towards free horizontal motion for the body, governed by the forcing frequency and amplitude. The locomotion of buoyant bodies in stratified fluids was further theoretically investigated by Voisin (Reference Voisin2024b), who considered three configurations under different forcing conditions and calculated their motion back to equilibrium. It has been demonstrated that the added mass of an oscillating body in stratified fluids is complex, with its real part characterising inertia and its imaginary part accounting for wave damping (Voisin Reference Voisin2024a ). These works highlight the critical role of stratification in governing the dynamics of freely moving bodies.
In this study, we numerically investigate the locomotion of an undulatory swimmer with Reynolds numbers of
$O({10^3})$
in a density-stratified fluid. The swimmer actively generates a travelling wave along its body, resulting in free swimming in the direction perpendicular to gravity. Our previous studies (Wang et al. Reference Wang, Kandel and Deng2024a
, Reference Wang, Kandel and Deng2025) have shown that density stratification can enhance both thrust production and propulsive efficiency for a tethered oscillating foil, driven by the resonance between foil kinematics and stratification effects. However, the tethered foil in an incoming flow cannot fully characterise the propulsion of real swimmers, mainly because it fails to capture the thrust-drag balance in free-swimming condition. Furthermore, in comparison with the combined heaving–pitching motion of the foil, undulating the body into a backward-travelling wave remains more relevant to biological propulsion in nature. By addressing these limitations, we aim to provide a more realistic evaluation of the influence of density stratification on biological propulsion and offer new insights into the physical mechanisms governing undulatory swimming in complex aquatic environments.
2. Methodology
2.1. Problem formulation
We consider an undulatory swimmer in a density-stratified fluid, as illustrated in figure 1(
$a$
). The undisturbed background density is set to vary linearly along the direction of gravitational acceleration
$\boldsymbol{g}$
with a constant gradient
$\gamma$
(
$ {\gt } 0$
),
where
$\rho _0$
is the reference fluid density and
$y$
is the vertical coordinate. To focus on the stratification effects, we model the swimmer as a two-dimensional filament that performs prescribed undulation along its body, as shown in figure 1(
$b$
). Following the undulatory movement introduced by Carling et al. (Reference Carling, Williams and Bowtell1998), the lateral displacement of the filament in the local coordinate system is defined as
Here,
$s$
is the distance along the body measured from the head,
$t$
denotes the time and
$L$
represents the body length. Here
$A$
and
$T$
are the amplitude and cycle of undulation, respectively. Here
$\sigma (s) = s$
is a control function ensuring that the undulation amplitude increases linearly from head to tail. Based on the undulation profile (2.2), the swimmer passes a travelling wave of increasing amplitude along its entire body with a wavelength of one body length. In addition, the two-dimensional swimmer propels itself by undulating the body vertically. These features suggest that the undulatory swimmer considered here is biologically relevant to anguilliform and rajiform species in nature.
(a) Schematic of the present numerical set-up for an undulatory swimmer in a density-stratified fluid. Dimensions in this figure are exaggerated for clarity. (b) Temporal evolution of the swimmer’s shape over one undulation cycle
$T$
with a time interval of
$0.1T$
, determined by (2.2). The solid line denotes the body profile at
$t=T$
, and the dashed lines in lighter colours correspond to later instants.

Figure 1. Long description
Panel A: A schematic diagram of the numerical setup for an undulatory swimmer in a density-stratified fluid. The diagram shows a rectangular domain with a length of 70L and a height of 50L. A swimmer of length L is positioned within this domain, moving to the left with a velocity U. The fluid density varies linearly with depth, denoted as ρ_b(y) = ρ_0 - γy. Gravity acts downward, indicated by the vector g. Panel B: A diagram depicting the temporal evolution of the swimmer’s shape over one undulation cycle. The solid line represents the body profile at a specific time, while the dashed lines in lighter colors correspond to later instants, showing the progression of the swimmer’s shape over time.
The motion of the self-propelled swimmer is governed by Newton’s equation of motion in the global coordinate system
where
$m$
is the total mass of the swimmer,
$I_z$
is the moment of inertia of the swimmer about the
$z$
-axis,
$\boldsymbol{x}_c$
is the position for the centre of mass of the swimmer and
$\theta$
is the angular position of the swimmer. Here
$\boldsymbol{F}_{\!c}$
and
$M_z$
represent the fluid force and torque acting on the swimmer body, respectively.
The undulatory swimmer moves in a fluid that is initially quiescent. To facilitate the analysis of stratification effects, the swimmer in the present simulations is restricted to rectilinear swimming. Specifically, the swimmer is allowed to freely move along the horizontal direction, whereas its translational motion along the
$y$
-axis and rotation about the
$z$
-axis are constrained. Examination of the net lateral force and the net moment on the swimmer shows negligibly small time-averaged values, demonstrating that the assumption of rectilinear cruising is maintained in the present model. Unless otherwise specified, the swimming velocity of the swimmer in this paper refers to that along the horizontal direction. We set the ratio of swimmer density to reference fluid density as
${\rho _s}/{\rho _0} = 1$
. Since the swimmer cruises at a constant depth, we assume it to be neutrally buoyant relative to the surrounding fluid. The undulation of the swimmer starts from rest, achieved by applying an initiation function to (2.2) during the first undulation cycle. In all cases, we allow the swimmer to swim freely from right to left over a distance of 12 body lengths within the central region of the computational domain.
2.2. Numerical methods
The motion of a linearly stratified, Newtonian, incompressible fluid is governed by the conservation equations of mass and momentum, in conjunction with the advection–diffusion equation of the density. These equations are solved under the Boussinesq approximation, which means that density variations are ignored everywhere except in the buoyancy force term. The total density and pressure are decomposed into background and fluctuating components, namely
$\rho = {\rho _b} + \rho '$
and
$p = {p_b} + p'$
. Here, the background density
$\rho _b$
has a linear distribution as previously specified, and the background pressure
$p_b$
satisfies the hydrostatic balance
$\boldsymbol{\nabla } {p_b} = {\rho _b}\boldsymbol{g}$
. The perturbations of density and pressure,
$\rho '$
and
$p'$
, arise from the fluid motion. Therefore, the governing equations for the fluid motion are given by
where
$\boldsymbol{u}$
is the fluid velocity,
$\nu$
is the kinematic viscosity and
$\kappa$
is the density diffusivity. We employ an immersed boundary method to solve the interaction between swimmer and fluid. This approach accounts for the presence of the swimmer by adding an additional forcing term
$\boldsymbol{f}$
in the momentum equation (2.6), which is given by
where
$\boldsymbol{x}$
is the Cartesian coordinate,
$\boldsymbol{X}$
is the physical position of Lagrangian points and
$\delta$
is the Dirac delta function. Here
$\boldsymbol{F}$
represents the Lagrangian forcing exerted on the swimmer by the surrounding fluid, solved using a feedback law (Goldstein et al. Reference Goldstein, Handler and Sirovich1993),
Here,
$a$
and
$b$
are large negative free constants. Here
$\boldsymbol{U}$
denotes the velocity of the swimmer, calculated by
$\boldsymbol{U} = \mathrm{d}\boldsymbol{X}/\mathrm{d}t$
. Here
$\boldsymbol{U}_{\!ib}$
indicates the fluid velocity, obtained by interpolation of the velocity field at the immersed boundary,
As a result, the applied forcing term
$\boldsymbol{f}$
enforces a no-slip boundary condition on the body surface. To ensure that the stratifying agent cannot cross into the body regime, an insulating/impermeable boundary condition must also be imposed on the swimmer surface. Here, we set
$\kappa = 0$
in the body phase and
$\kappa = \nu /Pr$
in the fluid phase, which is equivalent to the no-flux boundary condition
$\boldsymbol{\nabla }\!\rho \boldsymbol{\cdot } \boldsymbol{n} = 0$
on the swimmer surface. This approach has been widely used in the literature on the rigid body immersed in a stratified fluid (Doostmohammadi et al. Reference Doostmohammadi, Dabiri and Ardekani2014; Ardekani et al. Reference Ardekani, Abouali, Picano and Brandt2018; Ouillon et al. Reference Ouillon, Houghton, Dabiri and Meiburg2020; More et al. Reference More, Ardekani, Brandt and Ardekani2021). Since temperature variation is the main cause of density differences in aquatic environments (Sutherland Reference Sutherland2010), the Prandtl number in this study is set to
$Pr = 7$
, which is a typical value for thermally stratified fluids.
We perform all simulations using a finite-volume solver. The first-order implicit Euler scheme is employed to advance the equations in time. The spatial discretisation of the convection and diffusion terms is carried out using second-order upwind and central-difference schemes, respectively. The pressure implicit with splitting of operators algorithm is used to resolve the pressure–velocity coupling. The governing equations are solved on a non-uniform Cartesian grid. The boundary condition on all sides of the computational domain is set to be zero normal gradient for both the velocity field and the density perturbation, i.e.
$\partial \boldsymbol{u}/\partial n = \boldsymbol{0}$
and
$\partial \rho '/\partial n = 0$
. We use a sufficiently large domain size (
$70L \times 50L$
) to prevent spurious reflection of internal waves. The starting position of the swimmer’s head is set at a location
$41L$
from the left-hand boundary and
$25L$
from the bottom boundary of the computational domain.
2.3. Performance metrics and governing parameters
For a self-propelled swimmer, the performance metrics of interest include cruising speed, input power and swimming efficiency. We define the steady-state cruising speed as the average velocity of the swimmer’s head,
where
$U$
represents the instantaneous swimming velocity measured at the head of the swimmer. Once the Lagrangian forcing
$\boldsymbol{F}$
is obtained from (2.9), the total fluid force and torque acting on the swimmer can be calculated by
where
$\boldsymbol{r}_s$
is the position vector from the centre of mass of the swimmer to a point on its surface, and
$\boldsymbol{e}_z$
is the unit vector along the
$z$
-axis. After that,
$U$
can be obtained through the solution of (2.3). The steady-state input power required to maintain body undulation of the swimmer can be defined as
The time-averaged
$\bar U$
and
$\bar P$
over one undulation cycle are calculated after the swimmer reaches a periodically steady state. In addition, we use the cost of transport (
$\textit{COT}$
) to evaluate swimming efficiency, which is given by the equation
The metric
$\textit{COT}$
measures the energy spent per unit distance travelled, which means that a lower
$\textit{COT}$
indicates higher swimming efficiency. In the literature on swimming fish, velocity is commonly expressed in body lengths per second (
$L\,{\mathrm{s}^{ - 1}}$
), whereas power and
$\textit{COT}$
are usually reported in their actual dimensional units (Kendall et al. Reference Kendall, Lucey, Jones, Wang and Ellerby2007; Bale et al. Reference Bale, Hao, Bhalla and Patankar2014; Li et al. Reference Li, Liu, Müller, Voesenek, Van and Johan2021). For comparison with corresponding homogeneous observations, we retain these conventions in this study.
The key dimensionless parameters that govern the problem considered here include the length-based Reynolds number, the amplitude-based Strouhal number and the internal Froude number,
where
$f$
is the undulation frequency (
$f = 1/T$
),
$U_{\textit{ref}}$
represents the reference swimming speed and
$D$
denotes the characteristic length along the gravitational direction. The Brunt–Väisälä frequency
$N$
, which signifies the natural oscillation frequency of fluid parcels in a stratified fluid, is defined as
$N = {(\gamma g/{\rho _0})^{1/2}}$
.
In this study, we vary
$A$
,
$f$
and
$\textit{Fr}$
to evaluate the swimming performance of the self-propelled swimmer under stratification effects. To non-dimensionalise the undulatory parameters, we consider a reference case in which the swimmer cruises at
${A_{ref}} = 0.1\,L$
and
${f_{ref}} = 1\,\mathrm{Hz}$
in a homogeneous fluid, giving rise to a reference swimming speed
$U_{\textit{ref}}$
around
$0.57\,L\,{\mathrm{s}^{ - 1}}$
. This reference case corresponds to a typical anguilliform swimmer cruising in efficient swimming mode (D’aout & Aerts Reference D’aout and Aerts1999; Tytell Reference Tytell2004). The dimensionless amplitude and frequency of the undulation are therefore defined as
${A^*} = A/L$
and
${f^*} = f/{f_{ref}}$
. Taking the constant
$U_{\textit{ref}}$
as the characteristic velocity, we can directly use
$\textit{Fr}$
to measure the stratification strength. The length scale
$D$
in the definition of the internal Froude number is selected as the vertical body height of the swimmer. Choosing a characteristic length along the direction of the density gradient for buoyancy-related parameters is a common method in the literature (Ortiz-Tarin et al. Reference Ortiz-Tarin, Chongsiripinyo and Sarkar2019; More et al. Reference More, Ardekani, Brandt and Ardekani2021), presumably due to the fact that the gravity-aligned length scale can better characterise the dynamical behaviours in density-stratified fluids. Based on biological data from anguilliform swimmers, we set
$D$
to 10 % of body length. Hence, the desired
$\textit{Fr}$
can be easily achieved by adjusting
$g$
in our simulations.
The amplitude and frequency of the swimmer undulation in the present study are confined to the ranges of
$0.1 \le {A^*} \le 0.2$
and
$0.5 \le {f^*} \le 2.0$
, in accordance with the biological limits for anguilliform species (Di Santo et al. Reference Di Santo, Goerig, Wainwright, Akanyeti, Liao, Castro-Santos and Lauder2021; Li et al. Reference Li, Liu, Müller, Voesenek, Van and Johan2021). The typical values of the Brunt–Väisälä frequency
$N$
range from
$10^{ - 4}$
to
$0.3\,{\mathrm{s}^{ - 1}}$
in natural aquatic environments (Thorpe Reference Thorpe2007; Geyer et al. Reference Geyer, Scully and Ralston2008). If we consider a density-stratified fluid characterised by a Brunt–Väisälä frequency of
$N = O(0.1\,{\mathrm{s}^{ - 1}})$
, a swimmer on a length scale
$O(1\,\mathrm{m})$
cruising at velocities ranging from 0.1 to
$1.0\,\mathrm{m}\,{\mathrm{s}^{ - 1}}$
can produce a
$\textit{Fr}$
interval of
$1 - 10$
. Therefore, in this study, the range of the internal Froude number is chosen as
$0.5 \le \textit{Fr} \le 20.0$
, which encompasses a wide spectrum of stratification strengths from strong to weak. The Reynolds number defined with respect to
$U_{\textit{ref}}$
is fixed at 2000, whereas it varies from 800 to 5000 based on
$\bar U$
. It has been indicated that the fundamental mechanisms (Kern & Koumoutsakos Reference Kern and Koumoutsakos2006; Li et al. Reference Li, Liu, Müller, Voesenek, Van and Johan2021) and scaling laws (Lin et al. Reference Lin, Wu and Zhang2021; Liu et al. Reference Liu, Liu and Huang2022) established in such low-Reynolds-number regimes remain largely applicable to high-Reynolds-number turbulent swimming conditions. This suggests that Reynolds numbers of
$O({10^3})$
are sufficient for identifying the underlying physics governing fish swimming, although they may not fully capture the quantitative characteristics associated with turbulence and related effects. It is worth noting that the Reynolds number range considered in this study is selected to characterise a general inertial swimmer in natural environments, rather than to represent any particular fish species. We aim to extend current understanding of swimming in density-stratified fluids to the inertia-dominated regime, complementing the extensive studies on the viscous-dominated locomotion of small organisms (More & Ardekani Reference More and Ardekani2023).
To maintain biological relevance, we provide a summary of the typical Reynolds numbers and tail-beat frequencies of natural swimmers. The data are obtained from the supplementary information of Gazzola et al. (Reference Gazzola, Argentina and Mahadevan2014). Inertial aquatic swimmers exhibit the Reynolds numbers spanning several orders of magnitude, from
$10-10^3$
in larval zebrafish to
$10^7-10^8$
in whales. Medium-sized swimmers, such as dace, trout and mackerel, typically fall within the range of
$10^3-10^6$
, depending on swimming mode. Stingray and sturgeon are mainly observed at the Reynolds numbers of
$10^4-10^5$
, whereas saithe generally operate within
$10^5-10^6$
. Larger marine animals, such as bluefin tuna, sharks and dolphins, can reach Reynolds numbers of the order of
$10^6-10^7$
. In general, tail-beat frequency decreases with increasing body size in aquatic swimmers, ranging from
$80\,\mathrm{Hz}$
in larval zebrafish to
$0.25\,\mathrm{Hz}$
in whales. Intermediate values include more than
$10\,\mathrm{Hz}$
for mackerel, approximately
$5\,\mathrm{Hz}$
for dace and trout, and around
$1\,\mathrm{Hz}$
for bluefin tuna, sharks and dolphins.
The present solver of fluid motion has been successfully applied to a range of density-stratified flow problems and has been extensively validated in our previous studies (Deng & Kandel Reference Deng and Kandel2022; Wang et al. Reference Wang, Wang and Deng2023, Reference Wang, Kandel and Deng2024a , Reference Wang, Kandel, Deng, Caulfield and Dalzielb , Reference Wang, Kandel and Deng2025; Kandel et al. Reference Kandel, Wang and Deng2026). Moreover, our earlier studies (Liu et al. Reference Liu, Wang, Mao and Deng2023; Mao et al. Reference Mao, Wang, Mao and Deng2025) have demonstrated the applicability of the immersed boundary method employed here for modelling fish self-propulsion in homogeneous fluids. Appendix A further presents a series of time step, grid and domain-size independence tests to verify the convergence of the numerical results.
3. Results and discussion
To highlight the influence of density stratification, we focus on several aspects commonly examined in homogeneous swimming, ranging from wake dynamics to swimming performance. Existing studies of biological swimming in stratified fluids have largely been confined to microorganisms that undergo vertical motion at low Reynolds numbers. By contrast, the effects of stratification on moderate-Reynolds-number swimmers moving perpendicular to gravity remain poorly understood. Our results are therefore expected to provide new insights into this unexplored regime.
Instantaneous vorticity contours for a self-propelled undulatory swimmer at varying stratification strengths. The swimmer kinematics are
${A^*} = 0.2$
and
${f^*} = 1.0$
. The vorticity
$\omega _z$
is normalised by
${U_{\textit{ref}}}/L$
, and the case
$\textit{Fr} = \infty$
corresponds to a homogeneous fluid.

3.1. Stratification-driven wake dynamics
The wake dynamics of an undulatory swimmer in stratified fluids are first examined to build physical intuition for the problem addressed in this study. Figure 2 shows the vortical patterns with varying
$\textit{Fr}$
for an undulatory swimmer with fixed kinematics. The homogeneous case (
$\textit{Fr} = \infty$
) is provided as a reference. As shown in figure 2(
$a$
), the wake induced by a self-propelled undulatory swimmer in a homogeneous environment manifests as a typical vortex street, in which the arrangement and coherence of the vortex structures are primarily governed by the swimmer kinematics. The homogeneous simulation presented here is qualitatively consistent with previous observations of two-dimensional swimming wake patterns (Kern & Koumoutsakos Reference Kern and Koumoutsakos2006; Gazzola et al. Reference Gazzola, Argentina and Mahadevan2014).
In the presence of stratification, the wake dynamics of undulatory swimming differ significantly from its homogeneous counterpart. We note that stratification influences the wake in several distinct ways. First, stratification tends to inhibit the vertical motion of flow structures. This can be illustrated through the evolution of the start-up vortices induced by the swimmer body during the initial stage. As the swimmer moves leftward, the start-up vortices in a homogeneous fluid are able to drift freely in the direction of gravity, as indicated by the vortex pairs at the right-hand end of figure 2(
$a$
). In contrast, the motion of the start-up vortices in stratified regimes is confined to the midline region, as shown in figures 2(
$b$
) and 2(
$c$
). Second, the motion of the flow structures in the horizontal direction is enhanced by stratification. From figure 2(
$d$
–
$f$
), we see that the shedding vortices are stretched into the elongated structures arranged horizontally, consequently giving rise to an increasingly chaotic near wake with the increase in stratification strength. This tendency of stratification to suppress vertical motion while promoting horizontal motion has been observed in a wide range of flow problems (Riley & Lelong Reference Riley and Lelong2000). The fundamental mechanism responsible for this dynamical behaviour lies in the buoyancy effects, which act as a restoring force that raises the energetic penalty of vertical displacements and directs the momentum preferentially towards the horizontal direction. Moreover, the flow patterns in stratified fluids are strongly affected by internal waves arising from the interplay between inertia and buoyancy restoring force on perturbed fluid parcels. We note that internal waves become prominent only when the Froude number is sufficiently small. Especially, the wave patterns are mainly concentrated in the downstream wake in weakly stratified fluids (see figure 2
$e$
,
$f$
), while under strongly stratified conditions the wave-like vorticity structures begin to appear in both the upstream and downstream regions of the swimmer (see figure 2
$g$
–
$i$
). Strong stratification further leads to a significant bifurcation in the far wake, as shown in figure 2(
$g$
–
$i$
), due to the dominant buoyancy effects that reorganise the flow disturbances along the internal-wave propagation paths.
Instantaneous vorticity contours for a self-propelled undulatory swimmer at varying undulation frequencies and stratification strengths. The undulation amplitude of the swimmer is fixed at
${A^*} = 0.1$
. Contour colours are the same as in figure 2.

Another notable observation in figure 2 is that the flow patterns remain symmetric with respect to the midline, despite the presence of density gradient. Recalling the transport equation for density (2.7), we see that density perturbations in a linearly stratified fluid under the Boussinesq approximation depend only on the background density gradient
$\gamma$
. Consequently, there is no dynamical distinction between upward and downward density perturbations in a fluid domain with constant
$\gamma$
(Sutherland Reference Sutherland2010). This wake symmetry about the horizontal axis in stably stratified fluids has been observed in both numerical simulations and experimental measurements on the flow configurations of bluff bodies (Madison et al. Reference Madison, Xiang and Spedding2022; Ortiz-Tarin et al. Reference Ortiz-Tarin, Nidhan and Sarkar2023).
Comparison of internal wave patterns between theoretical predictions and numerical simulations for (
$a$
,
$b$
)
$\varOmega = 0.55$
and (
$c$
,
$d$
)
$\varOmega = 1.75$
, where
$\varOmega = {\omega _f} /N$
. Panels (
$a$
) and (
$c$
), predicted by the theory of Lighthill (Reference Lighthill1967), show curves of constant phase of the internal waves for a travelling disturbance. Here, the dashed lines represent the wave system associated with steady motion, while the solid lines indicate waves induced by oscillation. Panels (
$b$
) and (
$d$
), obtained from the present simulations of a self-propelled undulatory swimmer with
${A^*} = 0.1$
and
${f^*} = 1.0$
, show schlieren contours that characterise the magnitude of density gradient. The numerical results are normalised by
${\rho _0}/L$
and displayed using a linear colour scale ranging from 0 (dark) to 0.0006 (light).

Figure 4. Long description
Panel A: A theoretical diagram showing curves of constant phase of internal waves for a traveling disturbance. Dashed lines represent the wave system associated with steady motion, while solid lines indicate waves induced by oscillation. The diagram includes a scale bar labeled 2L. Panel B: A numerical simulation showing schlieren contours that characterize the magnitude of the density gradient for a self-propelled undulatory swimmer with Q equals 0.55. The contours are displayed using a linear color scale ranging from 0 (dark) to 0.0006 (light). Panel C: Another theoretical diagram similar to Panel A but with different wave patterns. Panel D: A numerical simulation similar to Panel B but for a swimmer with Q equals 1.75.
Figure 3 shows the vorticity structures around a self-propelled undulatory swimmer at different
$f^*$
and
$\textit{Fr}$
. The wake pattern in a homogeneous fluid undergoes a clear transition from a single-row vortex street to a double-row configuration as the undulation frequency increases, resulting from the change in the number of vortices shed per undulation cycle. Upon entering a stratified regime, this wake transition with increasing
$f^*$
becomes indistinguishable owing to the effects of stratification on fluid motion, as indicated earlier. In contrast, the flow features in stratified fluids are characterised by variations in the spatial extent of internal-wave coverage, depending on the relative magnitudes of the forcing frequency of the swimmer and the buoyancy frequency of fluid parcels. This behaviour can facilitate our understanding of the interplay between inertia and buoyancy. If the forcing frequency is smaller than the buoyancy frequency, the buoyancy restoring force is able to keep pace with the perturbations of the fluid parcels, inducing periodic oscillations around the swimmer. As a result, we can observe the internal waves near the swimmer body at
${f^*} = 0.5$
for all considered
$\textit{Fr}$
(see figure 3
$a$
). However, when the forcing frequency exceeds the buoyancy frequency, the flow disturbances become evanescent, presenting as an exponential attenuation in wave amplitude with distance from the source (Sutherland Reference Sutherland2010). This accounts for the decrease in critical
$\textit{Fr}$
that induces internal waves around the swimmer body with increasing
$f^*$
, as shown in figure 3(
$b$
–
$d$
).
For a self-propelled swimmer in stratified fluids, the resulting flow pattern is clearly separated by two distinct wave systems, one centred on the swimmer and the other spreading downstream (see figure 3). This feature that becomes increasingly evident with stratification strength is in sharp contrast to the flow patterns induced by a tethered oscillating foil (Wang et al. Reference Wang, Kandel and Deng2024a , Reference Wang, Kandel and Deng2025), in which different wave systems are strongly intermingled and hard to distinguish. In particular, compared with the rapidly dissipating wake in tethered cases (Wang et al. Reference Wang, Kandel and Deng2025), the wake structure associated with free swimming spans a much longer downstream distance. These marked differences between the two configurations presumably result from the fact that the background flow facilitates interactions between flow structures across varying scales and enhance energy dissipation, while disturbances in quiescent fluids remain more coherent and persistent.
To quantify the internal waves generated by a self-propelled undulatory swimmer, we make a comparison of wave patterns between the present numerical simulations and the predictions of classical theory. Lighthill (Reference Lighthill1967) developed a general theory that describes the dynamics of anisotropic dispersive waves induced by travel disturbances. Thereafter, this theory was widely used to validate experiments and computations on internal waves generated by translating sources undergoing forced oscillation (Stevenson & Thomas Reference Stevenson and Thomas1969; Voisin Reference Voisin1994). Accordingly, we also employ this classical theory to assess the present simulations. For a travelling and oscillating source, the oscillation-induced wave system is superimposed on the steady wave system that remains stationary relative to the source. Specifically, the wave system resulting from the steady motion consists of a series of semicircles centred on the source, whereas the oscillatory wave system displays different patterns governed by the ratio of the forcing frequency to the buoyancy frequency
$\varOmega = {\omega _f}/N$
(Lighthill Reference Lighthill1967). Here, we consider an undulatory swimmer with fixed kinematics of
${A^*} = 0.1$
and
${f^*} = 1.0$
. Varying the buoyancy frequency
$N$
, we examine two typical scenarios, namely
$\varOmega \lt 1$
(
$\varOmega = 0.55$
) and
$\varOmega \gt 1$
(
$\varOmega = 1.75$
).
Instantaneous contours of
$z$
-component of baroclinic vorticity generation,
$\omega _b$
, arising from the misalignment of the density gradient with the pressure gradient, for a self-propelled undulatory swimmer at (
$a$
)
$\textit{Fr} = 3.0$
, (
$b$
)
$\textit{Fr} = 1.0$
and (
$c$
)
$\textit{Fr} = 0.5$
. The results are normalised by
${g_0}{\rho _0}/L$
, where
${g_0} = 10\,\mathrm{m}\,{\mathrm{s}^{ - 2}}$
denotes the gravitational acceleration under standard conditions. The swimmer kinematics are
${A^*} = 0.1$
and
${f^*} = 1.0$
.

The comparison of our simulations with the results predicted by the theory of Lighthill (Reference Lighthill1967) is presented in figure 4. Figure 4(a,c) shows the curves of constant phase of the internal waves, such as wave crests, in the theoretical predictions. Figure 4(b,d) exhibits the schlieren contours characterising the magnitude of density gradient, obtained from the present numerical simulations. According to Lighthill (Reference Lighthill1967), the oscillation-induced wave system at
$\varOmega \lt 1$
exhibits a radial pattern that spreads both upstream and downstream (see figure 4
$a$
), which transitions to a wedge-shaped configuration trailing behind the source when
$\varOmega \gt 1$
(see figure 4
$c$
). From figure 4(
$b$
,
$d$
) we see that these dynamical behaviours are accurately captured by the present numerical simulations. In particular, figure 4 provides quantitative evidence supporting the earlier discussion on the interplay between inertia and buoyancy. Overall, our numerical simulations show excellent agreement with the theoretical predictions for the final superposed wave patterns, with remarkably consistent wave spacing. The major discrepancy between the two results lies in the flow structures in the central wake, due to the fact that vortex shedding and nonlinear behaviours (such as wave interactions and wave breaking), which are inherently resolved in our numerical simulations, are absent in the theoretical calculations based on linear assumptions. In addition, we find that the internal wave patterns induced by the self-propelled swimmer show better agreement with the theoretical predictions than those generated by a tethered oscillating foil (Wang et al. Reference Wang, Kandel and Deng2025), highlighting the necessity of accounting for self-propelled motion when studying swimming in a complex fluid environment.
We pay further attention to the underlying physics that give rise to the main differences in vorticity configurations between stratified and homogeneous fluids. Taking the curl of the momentum equation (2.6) yields the following vorticity equation for stratified fluids under the Boussinesq approximation:
The first term on the right-hand side of (3.1), which represents vorticity stretching, vanishes identically in two-dimensional flows. The cross term on the right-hand side of (3.1) is called the baroclinic torque (Sutherland Reference Sutherland2010), arising from the misalignment between the density and pressure gradients. This baroclinic torque, also known as baroclinic vorticity generation, can contribute to the generation of vorticity throughout the fluid interior. Figure 5 shows the
$z$
-component of the generation of baroclinic vorticity,
${\boldsymbol{\omega } _b} = \boldsymbol{g} \times \boldsymbol{\nabla }\!\rho$
, for three stratification levels. We find that the spatial and temporal distributions of baroclinic torque (figure 5) coincide with those of vorticity structures in regions far from the swimmer (see figure 3
$b$
). This suggests that the generation of baroclinic vorticity accounts for the broadly distributed vorticity patterns in stratified fluids, which contrasts sharply with the homogeneous observation that vorticity structures are created only along the body surface (see figures 2 and 3). As stratification strength increases, baroclinic vorticity generation plays an increasingly important role in driving the flow configurations. At relatively weak stratification, the baroclinic torque mainly acts in the wake region, as shown in figure 5(
$a$
). With increasing stratification, baroclinic vorticity generation first becomes pronounced around the posterior body of the swimmer (see figure 5
$b$
), eventually extending throughout the body (see figure 5
$c$
). Clearly, this baroclinic contribution to vorticity dynamics can significantly affect the stress distribution throughout the body of the swimmer, thereby leading to new swimming behaviours in stratified fluids.
Effect of density stratification on the velocity evolution of the swimmer with
$f^* = 1.0$
for (a)
$A^* = 0.1$
and (b)
$A^* = 0.2$
. The results presented here span 15 undulation cycles during which the swimmer accelerates from rest and eventually reaches the steady state.

Figure 6. Long description
Two line graphs depict the velocity evolution of a swimmer under different density stratifications. Panel A: The line graph shows the velocity (U) over time (t/T) for various Froude numbers (\textit{Fr}). The x-axis represents time normalized by period (t/T) ranging from 0 to 15, and the y-axis represents velocity (U) ranging from 0 to 1.0. The inset zooms in on the time range from 14.0 to 15.0. Panel B: The line graph shows the velocity (U) over time (t/T) for various Froude numbers (\textit{Fr}). The x-axis represents time normalized by period (t/T) ranging from 0 to 15, and the y-axis represents velocity (U) ranging from 0 to 1.0. The inset zooms in on the time range from 14.0 to 15.0. Each line represents a different Froude number, with a legend indicating the specific values.
3.2. Effects of stratification on swimming performance
We now turn to the effects of stratification on swimming behaviours. The swimmer that undulates its body with different amplitudes while keeping the frequency fixed at
$f^* = 1.0$
is first considered. Figure 6 displays the evolution of the swimming velocity over time in different
$\textit{Fr}$
. Since the swimmer starts from rest, its instantaneous swimming velocity exhibits an initial acceleration phase as the thrust dominates the drag, followed by a steady swimming phase with an asymptotic mean forward velocity when the thrust is balanced by the drag. The results show that the steady state is reached after approximately 12 undulation cycles for all stratifications considered here, with a variation in the mean velocity between two consecutive cycles of less than 0.5 %. The steady state quantities are thus calculated by averaging the instantaneous results over one undulation cycle once the system has reached steady state. The swimmer undergoes a similar acceleration process in stratified fluids compared with homogeneous swimming, although the final steady-state velocity shows a strong dependence on
$\textit{Fr}$
. As expected, a higher undulation amplitude leads to a larger swimming velocity and more pronounced velocity fluctuations during the steady swimming phase, which holds for both stratified and homogeneous fluids.
A notable observation in figure 6 is that the swimmer in stratified fluids takes a longer time to attain the final steady-state speed than in the homogeneous case. Considering the presence of internal waves, the delayed asymptotic approach of swimming speed in a stratified fluid is likely associated with the history effects arising from the continuous generation, propagation and interaction of waves. As shown in figure 3, unlike the flow structures that are convected exclusively downstream in homogeneous fluids, the internal waves induced in a stratified fluid tend to persist around the swimmer. Consequently, the evolution of these internal waves generated during previous swimming cycles and their interactions with newly generated waves can exert a persistent influence on the swimmer’s locomotion. This prolongs the adjustment process of the surrounding flow towards the final periodic state, presumably contributing to the slower asymptotic approach of the swimming speed in stratified fluids compared with the homogeneous case.
Effect of density stratification on the steady-state swimming performance of the swimmer at different undulation amplitudes: (a) swimming velocity, (b) input power and (c)
$\textit{COT}$
. The different colours in (a) distinguish three regimes denoted by R-I, R-II and R-III, in which the swimming velocity exhibits different
$\textit{Fr}$
dependence. The zoomed-in plots (d)–(f) show a range of
$\textit{Fr}$
corresponding to the regime marked by the dashed rectangle in (c), where the swimmer in stratified fluids swims more efficiently than in a homogeneous fluid. The dashed lines in each plot indicate the homogeneous results. Error bars in (d)–(f) represent the standard deviation of the cycle-averaged
$\textit{COT}$
over four successive undulation cycles after the swimmer reaches the steady state.

The steady-state swimming performance of the swimmer with three undulation amplitudes in stratified fluids is summarised in figure 7. The corresponding homogeneous results are also plotted for comparison. We find that stratification significantly improves swimming velocity, as shown in figure 7(
$a$
). In particular, an undulatory swimmer consistently achieves greater speeds in stratified fluids compared with those in a homogeneous fluid. The maximum increases in swimming velocity relative to their homogeneous counterparts are 40.28 %, 24.74 % and 21.78 % for
$A^* = 0.10$
,
$A^* = 0.15$
and
$A^* = 0.20$
, respectively. According to the dependence of swimming velocity on
$\textit{Fr}$
, the range of stratification strength considered here can be divided into three regimes, denoted as R-I, R-II and R-III. The R-III regime with a range of
$3 \lt \textit{Fr} \le 20$
represents weakly stratified fluids, in which the stratification effects are less pronounced and, therefore, the results are close to those under homogeneous conditions. As
$\textit{Fr}$
increases towards infinity, the stratified case gradually approaches its homogeneous counterpart. Moderate stratifications are characterised by the R-II regime, which corresponds to a range of
$0.8 \lt \textit{Fr} \le 3$
. In this regime, the effects of stratification on undulatory swimming become dominant, and the swimmer’s speed increases rapidly and almost linearly with increasing stratification strength. Further increase in stratification yields the regime R-I spanning a range of
$0.5 \le \textit{Fr} \le 0.8$
. We see that the swimming velocity of the swimmer begins to decrease in such strongly stratified fluids, suggesting a transition in the underlying physics.
The input power of the swimmer shown in figure 7(
$b$
) displays a monotonic increase with increasing stratification strength. Similarly, the difference in power expenditure between stratified and homogeneous fluids is negligible in regime R-III, whereas the stratification-induced increase in power becomes significant in regimes R-I and R-II. In general, an undulatory swimmer expends more power in stratified fluids than in a homogeneous fluid. Figure 7(
$c$
) illustrates the effects of stratification on the efficiency of swimming characterised by
$\textit{COT}$
. We see that the variation of
$\textit{COT}$
with
$\textit{Fr}$
follows a trend similar to that of the input power. As mentioned above, a higher
$\textit{COT}$
indicates a lower swimming efficiency. In general, stratification has been found to reduce swimming efficiency, as evidenced by the increase in
$\textit{COT}$
with decreasing
$\textit{Fr}$
. Nevertheless, we also observe a range of
$\textit{Fr}$
in which
$\textit{COT}$
in stratified fluids is slightly lower than that in a homogeneous fluid, as shown by the zoomed-in plots in figure 7(
$d$
–
$f$
). This suggests that stratification may also lead to more efficient swimming under certain conditions. Given the small variations of
$\textit{COT}$
in figure 7(
$d$
–
$f$
), we perform an uncertainty analysis to assess whether these changes exceed the estimated numerical and statistical uncertainties. The sensitivity analysis of temporal and spatial resolutions for the most challenging case in Appendix A shows that the differences in the cycle-averaged results between the two finest resolutions for time step, grid and domain size are 0.33 %, 0.08 % and 0.02 %, respectively. These values are all smaller than the stratification-induced variations (around 1 %), suggesting that numerical uncertainty is not the dominant contributor to these small changes. The standard deviation of the cycle-averaged
$\textit{COT}$
over four successive undulation cycles after the swimmer reaches the steady state is further calculated, as shown by the error bars in figure 7(
$d$
–
$f$
). The relative standard deviation is below 0.21 % for all considered cases, indicating that the cycle-to-cycle variability is negligible. Hence, the stratification-induced variations of
$\textit{COT}$
in figure 7(
$d$
–
$f$
) can be considered to be resolved. However, it should be noted that such efficiency benefits due to stratification effects are marginal, as the stratification-induced decrease in
$\textit{COT}$
is less than 5 %.
For a self-propelled swimmer, the drag force acting on the body increases monotonically with the swimming velocity. The increase in swimming velocity with increasing stratification strength thus indicates a higher thrust production, as thrust must balance drag during steady-state swimming at a constant speed. Therefore, the present observations of a freely swimming swimmer are consistent with our previous study on a tethered oscillating foil in an incoming stratified flow (Wang et al. Reference Wang, Kandel and Deng2025), which showed that stratification can markedly improve thrust production, accompanied by a minor increase (
$\lt 5\,\%$
) in propulsive efficiency.
The effects of stratification have been widely reported to lead to increased power expenditure for vertically swimming organisms that employ different propulsion mechanisms (Doostmohammadi et al. Reference Doostmohammadi, Stocker and Ardekani2012; Dandekar et al. Reference Dandekar, Shaik and Ardekani2019). This behaviour is qualitatively consistent with our observations of a horizontally swimming swimmer. However, our findings that stratification increases swimming velocity when the swimmer moves in a direction perpendicular to gravity are in striking contrast to earlier studies (Dandekar et al. Reference Dandekar, Shaik and Ardekani2019; More & Ardekani Reference More and Ardekani2020), which reported that the vertical motion of the swimmer is mostly hindered in stratified fluids. More & Ardekani (Reference More and Ardekani2020) attributed the reduction in the swimming speed of a vertically orientated squirmer to twofold reasons. First, the lighter fluid is trapped into the swimmer’s recirculatory region as it moves towards the heavier fluid, resulting in an increased buoyancy force decreasing the swimming speed. Second, the isopycnals become less deformable as the stratification increases, providing the resistance to the flow generated by the swimmer and thus hampering its self-propulsion. However, this is not the case for the current undulatory swimmer undergoing horizontal motion. On the one hand, the entrained fluids resulting from swimmer’s motion are not trapped around its body when it swims horizontally, suggesting that this buoyancy force due to fluid trapping in vertical motion is absent in horizontal swimming; on the other hand, the current swimmer moves along a direction parallel to isopycnals, indicating that the isopycnal deformation has a limited effect. Hence, the marked enhancement in the horizontal motion of the undulatory swimmer observed in this study is driven by other underlying mechanisms.
From the perspective of classical theory, the propulsion of a fish-like swimmer at relatively high Reynolds numbers arises from the combined contributions of circulatory forces and inertial reaction forces (Smits Reference Smits2019). The circulatory forces originate from the lift force created by the circulation around the swimmer and therefore depends on the angle of attack of the body relative to the incoming flow. The inertial reaction forces represent the contribution associated with the acceleration of the surrounding fluid induced by the body’s undulation or oscillation. In general, inertial reaction forces dominate undulatory swimming, whereas the swimming of oscillatory swimmers is driven by both circulatory forces and inertial reaction forces (Smits Reference Smits2019; Daghooghi et al. Reference Daghooghi, Mallepaddi and Borazjani2025). Thus, it is reasonable to neglect the circulatory forces when modelling an undulatory swimmer with the long slender body (Wu Reference Wu2011; Gazzola et al. Reference Gazzola, Argentina and Mahadevan2014), such as the swimmer studied here.
Swimming in density-stratified fluids further complicates the problem by introducing two effects that differentiate the fluid dynamics from those in a homogeneous fluid, namely, the diffusion of the stratifying agent and the buoyancy variation due to the change in fluid density (Doostmohammadi et al. Reference Doostmohammadi, Stocker and Ardekani2012). The diffusion effect is characterised by
$Pr$
, which represents the ratio of momentum to stratifying-agent diffusivity. Although diffusion-driven flow has been reported to generate propulsion for a neutrally buoyant object floating in a density-stratified fluid (Allshouse et al. Reference Allshouse, Barad and Peacock2010), the velocity induced by this effect is extremely small and manifest only over sufficiently long time scales. In particular, we fix the Prandtl number at
$Pr = 7$
for all simulations so that the swimming behaviours observed in this study are independent of the diffusion of the stratifying agent. By contrast, the effects of buoyancy are generally quantified using
$\textit{Fr}$
. The strong dependence of the present results on
$\textit{Fr}$
indicates that undulatory swimming in a stratified fluid is primarily governed by buoyancy effects.
We find that the variation in inertial reaction force due to buoyancy effects is the leading mechanism responsible for the distinctive swimming behaviours of an undulatory swimmer in stratified fluids. Figure 8 schematically illustrates the physical process underlying this mechanism. As the swimmer undulates its body in a stratified fluid, the adjacent fluid parcel is displaced to a new position where its density differs from that of the ambient fluid. Hence, in addition to the force
$\boldsymbol{F}_{\!S}$
exerted by the swimmer, this fluid parcel experiences an additional buoyancy force
$\boldsymbol{F}_{\!B}$
, which is absent in a homogeneous fluid and becomes stronger with increasing stratification. The added buoyancy force
$\boldsymbol{F}_{\!B}$
, which tends to restore the fluid parcel to its neutrally buoyant position, increases the effective inertia of the fluid parcel accelerated by swimmer’s undulation. As a result, the swimmer equivalently displaces a heavier fluid, thereby experiencing a higher reaction force
$\boldsymbol{F}_{\!R}$
. The increased reaction force
$\boldsymbol{F}_{\!R}$
leads to increases in both the horizontal component
$\boldsymbol{F}_{\!x}$
that enhances the swimming velocity and the vertical component
$\boldsymbol{F}_{\!y}$
that augments the power expenditure, as observed in figure 7. This physical process occurs regardless of whether the fluid parcel is displaced upwards or downwards.
Schematic representation of buoyancy effects on the inertial reaction force for an undulatory swimmer in stratified fluids. A fluid parcel is displaced from its initial position indicated by the dashed ellipse to a new position indicated by the solid ellipse due to the swimmer’s undulation, thereby experiencing an added buoyancy
$\boldsymbol{F}_{\!B}$
. The swimmer-induced forcing acting on the fluid parcel is denoted by
$\boldsymbol{F}_{\!S}$
. The added buoyancy
$\boldsymbol{F}_{\!B}$
increases the effective inertia of the accelerated fluid parcel, resulting in a higher reaction force
$\boldsymbol{F}_{\!R}$
, and correspondingly larger horizontal and vertical components,
$\boldsymbol{F}_{\!x}$
and
$\boldsymbol{F}_{\!y}$
.

Figure 8. Long description
A diagram of the forces acting on an undulatory swimmer in stratified fluids. The swimmer is represented by a wavy black line moving through the fluid. The diagram shows various forces including thrust, reaction force, lift, added buoyancy, and the force exerted by the swimmer. The fluid parcel is displaced from its initial position to a new position due to the swimmer’s undulation, experiencing added buoyancy. The swimmer-induced forcing acting on the fluid parcel is denoted by F_S. The added buoyancy increases the effective inertia of the accelerated fluid parcel, resulting in a higher reaction force F_R, and correspondingly larger horizontal and vertical components, F_x and F_y.
3.3. Interaction between undulation and buoyancy frequencies
Fluid parcels in a stratified fluid exhibit a natural oscillation frequency, the Brunt–Väisälä frequency
$N$
, which is also known as the buoyancy frequency. Hence, the fundamental interaction between undulation and buoyancy frequencies is of particular interest in this study. Since the effects of stratification on swimming behaviours only become significant in regimes R-I and R-II (see figure 7), we focus on the Froude number range of
$0.5 \le \textit{Fr} \le 3$
in this section. Specifically, a parametric study of undulatory swimming is carried out within the parameter space
$({f^*},\textit{Fr})$
at a constant amplitude
$A^* = 0.1$
.
Effect of density stratification on the velocity evolution of the swimmer with
$A^* = 0.1$
for (a)
$f^* = 0.5$
, (b)
$f^* = 1.0$
, (c)
$f^* = 1.5$
and (d)
$f^* = 2.0$
. The results are presented as a function of dimensionless swimming distance
$d/L$
, measured from the starting point.

Figure 9 shows the evolution of the swimming velocity as a function of the cruising distance at four different undulation frequencies. The swimmer is free to swim over 12 body lengths in the central region of the computational domain, where the mesh is carefully designed and refined locally. The results indicate that, for the parameter ranges considered, the swimmer generally reaches a steady state after travelling approximately eight body lengths. The mean swimming speed in steady state is presented in figure 10. We see that the dependence of the swimming velocity on
$\textit{Fr}$
is closely correlated with
$f^*$
, confirming the anticipation that the undulation frequency is a more critical kinematic parameter than the undulation amplitude for swimming in stratified fluids. Several notable features can be observed in figure 10. First, the variation in swimming velocity with
$\textit{Fr}$
experiences a clear transition from an increasing-then-decreasing trend at low
$f^*$
to a decreasing-then-increasing trend at high
$f^*$
. As a result, we can identify a maximum value in
$\bar U$
for
${f^*} \le 1.2$
, whereas there exists a minimum
$\bar U$
when
${f^*} \ge 1.3$
. Second, we define a critical Froude number
$F{r_c}$
, which corresponds to the maximum steady-state swimming velocity for each
$f^*$
. The results show that
$F{r_c}$
tends to decrease with increasing undulation frequency within the range of
$0.5 \le {f^*} \le 1.2$
. However, the critical Froude number remains constant at
$F{r_c} = 0.5$
for
$1.3 \le {f^*} \le 2.0$
. Third, the stratification-induced change in swimming speed in relation to the homogeneous case exhibits an obvious sensitivity to
$f^*$
. The presence of stratification consistently improves swimmer motility at low
$f^*$
. By contrast, at relatively high
$f^*$
, a broad range of
$\textit{Fr}$
can be seen over which stratification suppresses swimming velocity.
Variation of the steady-state swimming velocity with stratification strength at different undulation frequencies. The dashed lines in each plot correspond to the homogeneous results. The dependence of swimming velocity on
$\textit{Fr}$
transitions from an increasing-then-decreasing trend at low
$f^*$
to a decreasing-then-increasing trend at high
$f^*$
.

Figure 10. Long description
Sixteen line graphs show the variation of steady-state swimming velocity with stratification strength at different undulation frequencies. Each graph has the x-axis labeled ‘Fr’ and the y-axis labeled ‘U’. The dashed lines in each plot correspond to the homogeneous results. The graphs show the dependence of swimming velocity on transitions from an increasing-then-decreasing trend at low to a decreasing-then-increasing trend at high. Panel A: f* = 0.5. Panel B: f* = 0.6. Panel C: f* = 0.7. Panel D: f* = 0.8. Panel E: f* = 0.9. Panel F: f* = 1.0. Panel G: f* = 1.1. Panel H: f* = 1.2. Panel I: f* = 1.3. Panel J: f* = 1.4. Panel K: f* = 1.5. Panel L: f* = 1.6. Panel M: f* = 1.7. Panel N: f* = 1.8. Panel O: f* = 1.9. Panel P: f* = 2.0.
Contour plots of swimming performance in the
$({f^*},\textit{Fr})$
parameter space: (a) steady-state swimming velocity
$\bar U$
, (b) critical Froude number
$F{r_c}$
, (c) velocity increment relative to the homogeneous case
$\Delta \bar U$
, where
$\Delta \bar U = ({{\bar U}_s} - {{\bar U}_h})/{{\bar U}_h}$
and the subscripts
$s$
and
$h$
denote stratified and homogeneous conditions, respectively, and (d) Strouhal number
$St$
. The different colours in (b) divide the
$({f^*},\textit{Fr})$
parameter space into two distinct zones. The resonant zone is characterised by a scaling relation arising from resonance between swimmer’s undulation and fluid oscillation, whereas the monotonic zone features a constant
$F{r_c}$
due to
${\omega _f} \gt N$
, where
${\omega _f} = 2\pi f$
is the angular frequency of the swimmer in radians per second.

The above swimming behaviours demonstrate a strong coupling between the undulation and buoyancy frequencies. To gain further insight, figure 11 presents the steady-state swimming performance in the parameter space
$({f^*},\textit{Fr})$
. From figure 11(
$a$
) we see that the dependence
$\textit{Fr}$
of the swimming velocity displays two distinct features, governed by the range of
$f^*$
. In the high-
$f^*$
regime, the maximum
$\bar U$
is consistently observed at the lower limit of the
$({f^*},\textit{Fr})$
plane, manifesting as a constant critical Froude number
$F{r_c} = 0.5$
. In particular, within the low-
$f^*$
regime, we can identify a peak in
$\bar U$
at a given undulation frequency, characterised by an apparent correspondence between
$F{r_c}$
and
$f^*$
. This can be seen more clearly in figure 11(
$b$
), which summarises all critical Froude numbers from our simulations in the
$({f^*},\textit{Fr})$
plane, as indicated by the orange dots. Given that the system considered here involves two characteristic frequencies, namely,
$\omega _f$
(where
${\omega _f} = 2\pi f$
) for swimmer’s undulation and
$N$
associated with the fluid oscillation, we naturally relate the local peaks in swimming velocity to the notion of resonance. Assuming a resonance between the swimmer and fluid dynamics implies that two characteristic frequencies are comparable, such that
${\omega _f} /N = O(1)$
. Thus, we infer that
$F{r_c} \sim 1/{f^*}$
. Based on this analysis, we propose the following scaling relation to fit our numerical results:
where
$\alpha$
and
$\beta$
are fitting coefficients. Since the peaks in
$\bar U$
are only observed when
${f^*} \lt 1.3$
, we fit the data in the
$f^*$
range of
$0.5 - 1.2$
, as shown by the solid green line in figure 11(
$b$
). The excellent agreement between numerical results and fitted predictions suggests that resonance is a dominant mechanism governing swimming behaviours in stratified fluids. The scaling relation (3.2) with
$\alpha = 0.37$
and
$\beta = 0.27$
accurately captures the underlying interplay between the swimmer’s undulation and fluid oscillation and can therefore be employed to predict the optimal undulation condition in a stratified environment. Here, the coefficient
$\alpha$
depends on the swimmer’s kinematics, while the constant offset
$\beta$
presumably accounts for the existence of an optimal undulation frequency in a homogeneous fluid as
$\textit{Fr}$
approaches infinity.
According to (3.2), a further increase in
$f^*$
requires a lower
$\textit{Fr}$
(i.e. stronger stratification) to trigger resonance. However, in this study, we restrict the Froude number to
$\textit{Fr} \ge 0.5$
because extremely strong stratification is numerically challenging, and stratified conditions with
$\textit{Fr} \lt 0.5$
are rarely observed in real environments. As a result, when
${f^*} \ge 1.3$
, the variation of
$\bar U$
with
$\textit{Fr}$
lies within the monotonic growth phase before reaching the resonance-induced peak, as shown in figure 10(
$i$
–
$p$
). This leads to a constant critical Froude number
$F{r_c} = 0.5$
within the range of
$1.3 \le {f^*} \le 2.0$
.
Therefore, the parameter space
$({f^*},\textit{Fr})$
examined here can be divided into two distinct zones, as shown in figure 11(
$b$
). In the resonant zone, the peaks in swimming velocity are induced by the resonance between the swimmer’s undulation and the fluid oscillation, due to the fact that
$\omega _f$
is comparable to
$N$
. A persistent increase in the swimmer’s undulation frequency yields the monotonic zone, where the largest
$N$
considered here (
$\textit{Fr} = 0.5$
) can no longer match
$\omega _f$
. In this case, the swimming velocity increases monotonically with
$\textit{Fr}$
preceding the peak, giving rise to that the maximum
$\bar U$
occurs consistently at
$\textit{Fr} = 0.5$
. It should be noted that the present findings align with our previous study (Wang et al. Reference Wang, Kandel and Deng2025), in which we found that the propulsive efficiency of a tethered oscillating foil tends to be maximised when the foil’s oscillation frequency matches the buoyancy frequency of stratified fluids. This shows that resonance is a prevalent mechanism that governs motion in stratified fluids and, more importantly, plays a dominant role in determining the optimal conditions for biological swimming. In natural environments, fish species typically exhibit undulation frequencies higher than the buoyancy frequency, suggesting that their interactions with stratified fluids mainly fall within the monotonic regime. Nevertheless, for certain aquatic swimmers with tail-beat frequencies around
$1\,\mathrm{Hz}$
or below,
$\omega _f$
becomes comparable to
$N$
in magnitude, which triggers the dynamic condition for resonance and thus implies the potential resonant interaction.
Resonant interactions between fish-like swimmers and the surrounding fluid have also been reported in the homogeneous regime. It has been shown that inertial swimmers can exploit resonant interactions with the fluid environment to select the optimal gait, thereby enhancing swimming speed and efficiency (Gazzola et al. Reference Gazzola, Argentina and Mahadevan2015). More intriguingly, aquatic swimmers were found to tune their kinematics into a narrow
$St$
interval of
$0.2 - 0.4$
, corresponding to peak propulsive efficiency (Taylor et al. Reference Taylor, Nudds and Thomas2003). Early studies (Triantafyllou et al. Reference Triantafyllou, Triantafyllou and Gopalkrishnan1991, Reference Triantafyllou, Triantafyllou and Grosenbaugh1993) attributed this phenomenon to the maximum spatial growth of wake jet instabilities, which was subsequently termed as wake resonance (Moored et al. Reference Moored, Dewey, Smits and Haj-Hariri2012). Wang (Reference Wang2000) argues that this preferred
$St$
range is connected with maximising the allowable angle of attack. Recently, Floryan et al. (Reference Floryan, Van Buren and Smits2018) performed a scaling analysis based on the thrust–drag balance and showed that fluid drag plays a crucial role in determining the optimal range of Strouhal numbers.
Despite the fact that the resonance-induced optima is also observed in our stratified simulations, the key mechanisms underlying resonance in stratified fluids differ fundamentally from those in a homogeneous fluid. The resonance in a stratified fluid originates from the interaction between the swimmer’s undulation and the oscillation of fluid parcels, inherently governed by the interplay between driving frequency
$\omega _f$
and the buoyancy frequency
$N$
, as illustrated by figure 11(
$b$
). This resonance associated with
${\omega _f}-N$
coupling has been reported in studies of the oscillations of a Cartesian diver immersed in stratified fluids (Le Gal et al. Reference Le Gal, Castillo Morales, Hernandez-Zapata and Ruiz Chavarria2022; Voisin Reference Voisin2024b
). In these works, the Cartesian diver, i.e. a neutrally buoyant body, is set into oscillation by modulating the hydrostatic pressure, and the system exhibits a resonant response when the driving frequency matches the buoyancy frequency. These observations and the present results jointly indicate that, in a stratified regime, the buoyancy frequency of fluid parcels dominates over the wake resonant frequency in governing the resonant interaction. Moreover, buoyancy frequencies in natural environments vary by several orders of magnitude (Thorpe Reference Thorpe2007), implying that the driving frequency that induces resonance can also span a wide range. Under a given stratified condition, the one-to-one correspondence between driving frequency and buoyancy frequency can yield only a single peak in the structure’s dynamic response, as demonstrated by the present study (see figure 10
$a$
–
$h$
), as well as the observations from Le Gal et al. (Reference Le Gal, Castillo Morales, Hernandez-Zapata and Ruiz Chavarria2022) and Voisin (Reference Voisin2024b). These behaviours differ markedly from the characteristics in a homogeneous regime, characterised by a narrow range of driving frequencies (Taylor et al. Reference Taylor, Nudds and Thomas2003) and multiple resonant frequencies corresponding to multiple local peak efficiencies (Moored et al. Reference Moored, Dewey, Smits and Haj-Hariri2012).
Inspired by Gazzola et al. (Reference Gazzola, Argentina and Mahadevan2015), we further provide a general physical interpretation of the enhancement of swimming velocity through resonant interactions. The buoyancy frequency of fluid parcels in a stratified regime acts as a natural frequency of the system, independent of external forcing. When the swimmer undulates its body with a driving frequency comparable to the buoyancy frequency, the phase synchronisation between the swimmer’s undulation and fluid oscillation can effectively convert the stratification-induced buoyancy into the inertial reaction force, thereby enhancing the reaction force acting on the swimmer. This case corresponds to the maximum swimming velocities observed in figure 10(
$a$
–
$h$
). Conversely, when the driving frequency does not match the buoyancy frequency, the system cannot attain a stable and beneficial phase difference between swimmer’s undulation and fluid oscillation. As a result, the buoyancy force due to stratification can either increase or reduce the inertial reaction force, because the stratification effect can increase and decrease the speed of swimming, as shown in figure 10. In the worst-case scenario, the fluid parcels oscillate in the most detrimental phase shift such that the surrounding fluid tends to drift away from the swimmer, which reduces the contribution of the swimmer’s dynamics to fluid acceleration, and thus diminishes the reaction force exerted by the fluid on the swimmer. This case corresponds to the minimum swimming velocities shown in figure 10(
$m$
–
$p$
).
Contour plots of swimming energetics in the
$({f^*},\textit{Fr})$
parameter space: (a) mean input power
$\bar P$
and (b) cost of transport
$\textit{COT}$
. The blue dots in (b) mark the region where
$CO{T_s} \lt CO{T_h}$
, implying that the swimmer achieves a higher swimming efficiency in stratified fluids than in a homogeneous fluid.

Figure 11(
$c$
) shows the contour plot of the stratification-induced variation in swimming velocity, defined as
$\Delta \bar U = ({{\bar U}_s} - {{\bar U}_h})/{{\bar U}_h}$
, where subscripts
$s$
and
$h$
denote the stratified and homogeneous results, respectively. The green solid line marks the border at which
$\bar U_s$
is equal to
$\bar U_h$
. Within the parameter space considered,
$\Delta \bar U$
attains a maximum value of 59.3 % at (
${f^*} = 0.5$
,
$\textit{Fr} = 1.0$
), while the minimum
$\Delta \bar U$
of
$- 3.0\,\%$
occurs at (
${f^*} = 2.0$
,
$\textit{Fr} = 1.2$
). We note that high values of
$\Delta \bar U$
fall mainly within the lower left-hand region of the
$({f^*},\textit{Fr})$
plane, largely overlapping with the resonant regime identified in figure 11(
$b$
). This observation supports the preceding analysis that high swimming performance in stratified fluids is closely correlated with resonant interactions. In contrast, the cases with
$\Delta \bar U \lt 0$
focus on the upper right-hand region, where the swimmer’s driving frequency deviates significantly from the buoyancy frequency of the fluid. Our study indicates that, by tuning the undulation frequency, swimmers can achieve substantial enhancement of motility when swimming in a stratified environment. The distribution of Strouhal numbers in the
$({f^*},\textit{Fr})$
plane is shown in figure 11(
$d$
). For the parameter ranges examined here, the Strouhal number of the undulatory swimmer lies between 0.25 and 0.35, aligning with the generally reported
$St$
interval in nature. In comparison, the homogeneous Strouhal numbers in the present simulations range from 0.28 at
${f^*} = 2.0$
to 0.42 at
${f^*} = 0.5$
. Overall, stratification is found to reduce the Strouhal number of undulatory swimming, resulting from the increased swimming velocity in stratified fluids, as discussed previously.
We further investigate the effects of stratification on the energetics of undulatory swimming. Self-propelled swimmers in natural environments appear to adjust their kinematics to minimise the cost of transport rather than maximise the Froude efficiency defined as the ratio of useful power to total input power (Li et al. Reference Li, Liu, Müller, Voesenek, Van and Johan2021), justifying the use of
$\textit{COT}$
as a metric to evaluate energetics in this study. Figure 12 presents the contour plots of the mean input power and
$\textit{COT}$
in the
$({f^*},\textit{Fr})$
plane. The results show that, at a given stratification, both
$\bar P$
and
$\textit{COT}$
increase monotonically with
$f^*$
. The monotonic increase of
$\textit{COT}$
with undulation frequency in our homogeneous simulations coincides with the computations of Li et al. (Reference Li, Liu, Müller, Voesenek, Van and Johan2021) and the scaling estimate
$COT \sim L{f^2}$
proposed by Sánchez-Rodríguez et al. (Reference Sánchez-Rodríguez, Raufaste and Argentina2023). This behaviour persists in stratified fluids, leading to the absence of any region of optimal swimming efficiency within the parameter space
$({f^*},\textit{Fr})$
. In a fixed kinematic condition, the decrease in
$\textit{Fr}$
results in a general increase in both
$\bar P$
and
$\textit{COT}$
, which becomes more pronounced when
$\textit{Fr} \lt 1.5$
. For an undulatory swimmer with
${f^*} = 2.0$
, the stratification level of
$\textit{Fr} = 1.0$
increases
$\textit{COT}$
by 9.2 % relative to the homogeneous case, rising sharply to 66.8 % at
$\textit{Fr} = 0.5$
. Although a small region with
$CO{T_s} \lt CO{T_h}$
is identified in the
$({f^*},\textit{Fr})$
plane, as marked by the blue dots in figure 12(
$b$
), the maximum difference between stratified and homogeneous cases remains below 5 %. Hence, our study indicates that swimmers spend more energy to sustain self-propulsion and attain a lower efficiency when swimming in a stratified environment. Energy expenditure plays a key role in driving the selection of kinematic parameters for either fish swimming or propulsion of bioinspired robotics. A typical operating scenario is characterised by selecting the appropriate combination of undulation frequency and amplitude to achieve a target swimming speed while minimising the energy expenditure. The present study implies that the stratification-induced energetic penalty may have a significant impact on the optimisation strategies used by natural swimmers and biomimetic underwater robots.
It is worth noting that the monotonic trend of a self-propelled swimmer, as shown in figure 12(
$b$
), is markedly different from the feature of Froude efficiency for a tethered oscillating foil, which generally exhibits an optimum performance region in the considered parameter space (Wang et al. Reference Wang, Kandel and Deng2024a
, Reference Wang, Kandel and Deng2025). In addition, the mild energetic benefits resulting from stratification for free swimming are confined to weak stratifications and high undulatory frequencies (see figure 12
$b$
), whereas the increase in Froude efficiency due to stratification effects in tethered cases is observed across all conditions (Wang et al. Reference Wang, Kandel and Deng2025). These behaviours suggest that stratification-induced energetic responses are governed by fundamentally different mechanisms in the two configurations.
3.4. Scaling undulatory swimming in a stratified fluid
Establishing a unifying framework that quantitatively links the locomotion of inertial aquatic swimmers spanning a wide range of characteristic length to their body kinematics and the surrounding fluid properties remains a long-standing impetus for the study of biological swimming. The pioneering work can be traced to the early empirical linear relation
$\bar U/L = (3/4)f - 1$
by Bainbridge (Reference Bainbridge1958). After a considerable interval, Gazzola et al. (Reference Gazzola, Argentina and Mahadevan2014) derived the first scaling relation for aquatic locomotion based on fundamental physical principles, which is supported by a wide range of biological datasets. According to Gazzola et al. (Reference Gazzola, Argentina and Mahadevan2014), the thrust generation results from the body-induced fluid acceleration, such that it can be scaled as
$(fA)^2$
. The hydrodynamic resistance scales as
${\bar U}^{3/2}$
in the laminar regime due to the dominance of viscous drag, while it shifts to
${\bar U}^2$
for the turbulent regime where pressure drag dominates skin drag. Balancing thrust and hydrodynamic resistance yields the scaling laws
$\textit{Re}_u \sim \textit{Re}_{\!f}^{4/3}$
for laminar flow and
$\textit{Re}_u \sim R{e_f}$
for turbulent flow, where
$\textit{Re}_u = \bar{U}\!{L}/\nu$
is the cruising Reynolds number characterising the swimmer’s velocity, and
$R{e_f} = 2\pi f\textit{AL}/\nu$
is the undulating Reynolds number characterising the undulatory motion of the swimmer. Note that
$\textit{Re}_u$
is equivalent to the Reynolds number defined in (2.16), and
$R{e_f}$
is termed the swimming number in Gazzola et al. (Reference Gazzola, Argentina and Mahadevan2014). More recently, Lin et al. (Reference Lin, Wu and Zhang2021) reported that the locomotion of a self-propelled pitching hydrofoil satisfies the scaling law
$\textit{Re}_u \sim \textit{Re}_{\!f}^{5/3}$
based on the thrust scaling
$(fA)^{5/2}$
. In contrast, Liu et al. (Reference Liu, Liu and Huang2022) indicated that the self-propulsion of a flapping flexible plate can be scaled as
$\textit{Re}_u \sim \textit{Re}_{\!f}^{3/2}$
, derived from drag scaling
${\bar U}^{4/3}$
. This inconsistency motivates us to explore the influence of stratification on the scaling of undulatory swimming.
The cruising Reynolds number
$\textit{Re}_u$
as a function of the undulating Reynolds number
$R{e_f}$
for an undulatory swimmer in a homogeneous fluid. The blue symbols represent our numerical results. The orange line indicates the scaling law derived by Gazzola et al. (Reference Gazzola, Argentina and Mahadevan2014).

The cruising Reynolds number
$\textit{Re}_u$
as a function of the undulating Reynolds number
$R{e_f}$
for an undulatory swimmer in stratified fluids. The homogeneous results, denoted by the green dashed line, serves as the reference. The black dashed line represents the turbulent scaling
$\textit{Re}_u \sim R{e_f}$
proposed by Gazzola et al. (Reference Gazzola, Argentina and Mahadevan2014). The insets (
$a$
)–(
$d$
) show the vorticity contours for the highlighted cases. Contour colours are the same as in figure 2.

Figure 14. Long description
Panel A: A line graph showing the relationship between the cruising Reynolds number (R e u) and the undulating Reynolds number (R e f) for an undulatory swimmer in stratified fluids. The horizontal axis represents R e f ranging from 10^3 to 5 x 10^3, and the vertical axis represents R e u ranging from 10^3 to 6 x 10^3. Multiple lines are plotted, each representing different Froude numbers (F r) ranging from 0.5 to 3.0. The green dashed line represents the homogeneous results, and the black dashed line represents the turbulent scaling proposed by Gazzola et al. (2014). Panel B: An inset showing vorticity contours for a case with f* = 0.5 and F r = ∞. Panel C: An inset showing vorticity contours for a case with f* = 2.0 and F r = 0.5. Panel D: An inset showing vorticity contours for a case with f* = 0.5 and F r = 0.5.
Since the filament considered in this study resembles the slender swimmer of Gazzola et al. (Reference Gazzola, Argentina and Mahadevan2014) in both kinematics and geometry, and the present simulations remain largely in the laminar regime (
$\textit{Re}_u \lt 6000$
), we employ the scaling
$\textit{Re}_u \sim \textit{Re}_{\!f}^{4/3}$
to validate our homogeneous results. Figure 13 shows the variation of
$\textit{Re}_u$
with
$R{e_f}$
in a homogeneous fluid. The excellent agreement between the scaling law of Gazzola et al. (Reference Gazzola, Argentina and Mahadevan2014) and our numerical results indicates that the undulatory filament used in this study can capture the key features of aquatic swimming.
Figure 14 presents the cruising Reynolds number
$\textit{Re}_u$
as a function of the undulating Reynolds number
$R{e_f}$
under different stratification levels. In general, the trend of
$\textit{Re}_u$
with respect to
$R{e_f}$
at varying
$\textit{Fr}$
manifests itself approximately as a family of mutually parallel straight lines, accompanied by evident shifting and overlap in the upper left-hand (low
$R{e_f}$
and low
$\textit{Fr}$
) and lower right-hand (high-
$R{e_f}$
and high-
$\textit{Fr}$
) regions. At relatively high-
$\textit{Fr}$
, undulatory swimming in stratified fluids still follows the homogeneous scaling
$\textit{Re}_u \sim \textit{Re}_{\!f}^{4/3}$
for large
$R{e_f}$
, while deviations from this power law occur as
$\textit{Re}_{\!f}$
decreases. In contrast, the evolution of
$\textit{Re}_u$
with
$\textit{Re}_{\!f}$
at low
$\textit{Fr}$
exhibits an opposite trend and differs markedly from homogeneous results. Intriguingly, the low-
$\textit{Fr}$
curves experience a curvature transition from
$m\lt 1$
at large
$\textit{Re}_{\!f}$
to
$m\gt 1$
at small
$\textit{Re}_{\!f}$
, leading to an intermediate region where undulatory locomotion in stratified fluids closely agrees with the turbulent scaling
$\textit{Re}_u \sim \textit{Re}_{\!f}$
observed in the homogeneous regime.
To gain insights into the above observations, we further check the vorticity fields corresponding to several highlighted cases, as shown by the insets in figure 14. From the inset (figure 14
$a$
) we see that the close agreement between swimming behaviours at
$\textit{Fr} = 3.0$
and
$\textit{Fr} = \infty$
for large
$R{e_f}$
results from similar flow patterns in the near wake. This flow similarity is due to the fact that when the driving frequency
$\omega _f$
greatly exceeds the buoyancy frequency
$N$
of fluid parcels, inertial effects tend to dominate the fluid dynamics and the buoyancy force is incapable of restoring the displaced fluid parcels to their neutrally buoyant positions. Under such conditions, the disturbances around the swimmer become evanescent and the amplitude of the waves exhibits an exponential attenuation, as indicated in figure 3. As a consequence, noticeable differences between the cases
$\textit{Fr} = 3.0$
and
$\textit{Fr} = \infty$
at
${f^*} = 2.0$
appear only in the far wake, whereas the near-body flow field remains nearly identical, yielding essentially the same force and locomotion characteristics. As
$\omega _f$
decreases, buoyancy effects start to play a comparable role with inertial effects in fluid motion, inducing periodic oscillations of fluid parcels. We therefore see that for
${f^*} = 0.5$
, the flow pattern around the swimmer in
$\textit{Fr} = 3.0$
is markedly different from its homogeneous counterpart, as illustrated in the inset (figure 14
$b$
). These stratification-induced flow structures can alter the hydrodynamic stress distribution along the swimmer’s body, giving rise to the deviation of locomotion from the homogeneous case at low
$R{e_f}$
. In contrast, the stratification of
$\textit{Fr} = 0.5$
corresponds to a considerably large
$N$
, such that the buoyancy force is sufficiently strong to drive the disturbed fluid parcels back to their original positions over the entire
$\omega _f$
range considered here. Consequently, we can observe that wave-like disturbances persist around the swimmer at
$\textit{Fr} = 0.5$
regardless of
$f^*$
, as shown in the insets (figures 14
$c$
and 14
$d$
). This results in distinct hydrodynamic forces on the body of the swimmer, which account for the significant departure of the locomotive behaviour from that of a homogeneous fluid across the whole range of
$\textit{Re}_{\!f}$
.
It is evident from figure 14 that the typical scaling
$\textit{Re}_u \sim \textit{Re}_{\!f}^m$
with a constant exponent
$m$
no longer holds in stratified fluids, indicating that aquatic locomotion in such environments cannot be characterised by
$R{e_f}$
alone. The pronounced dependence of
$\textit{Re}_u$
on
$\textit{Fr}$
further suggests that the Froude number, which quantifies the strength of stratification, serves as an additional key parameter governing undulatory swimming in stratified fluids. Thus, we seek new scaling relations for undulatory swimming in stratified environments taking
$R{e_f}$
and
$\textit{Fr}$
as input parameters.
Variation of the cruising Reynolds number
$\textit{Re}_u$
as a function of the scaling parameter
$\textit{Re}_{\!f}^{4/3}(1 + F{r^{ - \beta }})$
. The coefficients
$c_1$
and
$\beta$
are determined by minimising the squared residuals. The coefficient of determination for the data fit is
${R^2} = 0.918$
.

The general trend in figure 14 indicates that a simple power law can still largely represent the contribution of
$R{e_f}$
. Given the vanishing of stratification as
$\textit{Fr} \to \infty$
, we additionally introduce the factor (
$1 + F{r^{ - \beta }}$
) to account for the presence of stratification. Therefore, by retaining
$\textit{Re}_{\!f}^{4/3}$
term which holds in a homogeneous fluid and incorporating an explicit
$\textit{Fr}$
-dependent correction, we first propose the following scaling relation to characterise undulatory swimming in stratified fluids:
Figure 15 plots the variation of
$\textit{Re}_u$
as a function of the above scaling parameter, where
$\beta = 0.24$
is determined by minimising the squared residuals. Overall, the data exhibits an acceptable collapse with a coefficient of determination
${R^2} = 0.918$
. However, we note that the scaling results show a poor agreement with the dashed line of equality. This discrepancy is attributed to the fixed exponent 4/3 for
$R{e_f}$
in (3.3). Indeed, the majority of data in figure 14 demonstrates a
$R{e_f}$
dependence deviating from
$\textit{Re}_{\!f}^{4/3}$
scaling. Hence, we further examine the following scaling relations:
Here, we also provide a scaling for the dimensionless input power
$P^*$
, motivated by the finding that the input power of a self-propelled swimmer in a homogeneous fluid can similarly be scaled with
$\textit{Re}_{\!f}^m$
(Lin et al. Reference Lin, Wu and Zhang2021; Liu et al. Reference Liu, Liu and Huang2022). Figure 16 shows the data of
$\textit{Re}_u$
and
$P^*$
in the form of scaling (3.4). The best fit yields the scaling coefficients
$\alpha = 1.0$
and
$\beta = 0.25$
for
$R{e_f}$
, and
$\alpha = 2.18$
and
$\beta = 1.12$
for
$P^*$
, respectively. The datasets of both swimming velocity (figure 16
$a$
) and input power (figure 16
$b$
) are seen to collapse well onto the dashed line of equality, indicating that the established scaling relation (3.4) can serve as a valid model to explain the swimming behaviours in stratified environments. Notably, instead of the laminar scaling
$\textit{Re}_u \sim \textit{Re}_{\!f}^{4/3}$
, the
$R{e_f}$
component in (3.4) is consistent with the turbulent scaling
$\textit{Re}_u \sim R{e_f}$
observed in a homogeneous fluid (Gazzola et al. Reference Gazzola, Argentina and Mahadevan2014). This behaviour aligns with the transition trend of
$\textit{Re}_u$
with respect to
$R{e_f}$
in figure 14. Moreover, the contribution of
$R{e_f}$
to the power scaling is found to be weaker than that reported in a homogeneous fluid, where the input power has been shown to scale as
$\textit{Re}_{\!f}^3$
(Lin et al. Reference Lin, Wu and Zhang2021) or
$\textit{Re}_{\!f}^{2.7}$
(Liu et al. Reference Liu, Liu and Huang2022). This suggests that stratification tends to mitigate the role of undulation kinematics in determining power expenditure.
Variation of (
$a$
) the cruising Reynolds number
$\textit{Re}_u$
and (
$b$
) the dimensionless input power
$P^*$
as a function of the scaling parameter
$\textit{Re}_{\!f}^\alpha (1 + F{r^{ - \beta }})$
. The coefficients
$c_2$
,
$c_3$
,
$\alpha$
and
$\beta$
are determined by minimising the squared residuals. The coefficients of determination for the data fits in (
$a$
) and (
$b$
) are
${R^2} = 0.986$
and 0.976, respectively.

Figure 16. Long description
Panel A: A scatter plot depicting the cruising Reynolds number (Re_u) against the scaling parameter (c2Re_f^1.0(1 + Fr^-0.25)). The data points are represented by green circles, and a dashed orange line indicates a linear fit. The x-axis ranges from 1000 to 5000, and the y-axis ranges from 1000 to 5000. Panel B: A scatter plot showing the dimensionless input power (P*) against the scaling parameter (c3Re_f^2.18(1 + Fr^-1.12)). The data points are represented by green circles, and a dashed orange line indicates a linear fit. The x-axis ranges from 0 to 1.2, and the y-axis ranges from 0 to 1.2.
Variation of (
$a$
) the cruising Reynolds number
$\textit{Re}_u$
and (
$b$
) the dimensionless input power
$P^*$
as a function of the scaling parameter
$\textit{Re}_{\!f}^\alpha F{r^\beta }$
. The coefficients
$c_4$
,
$c_5$
,
$\alpha$
and
$\beta$
are determined by minimising the squared residuals. The coefficients of determination for the data fits in (
$a$
) and (
$b$
) are
${R^2} = 0.986$
and 0.963, respectively.

Furthermore, the features displayed in figures 11(
$a$
) and 12(
$a$
) demonstrate that the key swimming behaviours in stratified fluids can also be characterised by products of power laws of the input parameters
$R{e_f}$
and
$\textit{Fr}$
, which motivates us to propose the following alternative scaling relations:
Figure 17 presents the datasets using above scaling relations, with
$\alpha = 1.0$
and
$\beta = - 0.12$
for
$R{e_f}$
, and
$\alpha = 2.19$
and
$\beta = - 0.48$
for
$P^*$
, as determined by power law regression. As expected, the scaling (3.5) also yields an excellent collapse of the data for both
$\textit{Re}_u$
and
$P^*$
. More intriguingly, we find that the fitting features obtained from scaling (3.4) and scaling (3.5) are in close agreement, accompanied by highly consistent coefficients of determination for the two fits (see figures 16 and 17). We note that the scaling (3.4) can be rewritten as
$\textit{Re}_u \sim \textit{Re}_{\!f}^\alpha + \textit{Re}_{\!f}^\alpha F{r^{ - \beta }}$
and
${P^*} \sim \textit{Re}_{\!f}^\alpha + \textit{Re}_{\!f}^\alpha F{r^{ - \beta }}$
. The excellent agreement between the data fits using scaling (3.4) and scaling (3.5) indicates that
$\textit{Re}_{\!f}^\alpha F{r^{ - \beta }}$
plays a leading role in governing the undulatory swimming in stratified fluids, while the effect of
$\textit{Re}_{\!f}^\alpha$
is negligible. Therefore, at least within the parameter space considered here, the power law scaling (3.5) serves as a more robust theoretical model to understand and predict the swimming performance of a self-propelled swimmer in stratified environments.
Variation of (
$a$
,
$b$
) the cruising Reynolds number
$\textit{Re}_u$
and (
$c$
,
$d$
) the dimensionless input power
$P^*$
as a function of the scaling parameter
$\textit{Re}_{\!f}^\alpha F{r^\beta }$
. The data in (
$a$
) and (
$c$
) show the scaling in the strong-stratification regime with
$\textit{Fr} \le 1.2$
, whereas the results of (
$b$
) and (
$d$
) correspond to the scaling in the weak-stratification regime with
$\textit{Fr} \gt 1.2$
. The coefficients
$c_6$
–
$c_9$
,
$\alpha$
and
$\beta$
are determined by minimising the squared residuals. The coefficients of determination for the data fits in (
$a$
)–(
$d$
) are
${R^2} = 0.996$
, 0.989, 0.998 and 0.994, respectively.

Figure 18. Long description
Panel A: A scatter plot showing the relationship between the cruising Reynolds number (Re_u) and a scaling parameter (c6Re_f^0.91Fr^-0.14). The x-axis represents the scaling parameter, and the y-axis represents the cruising Reynolds number. Data points are plotted with a trend line in orange. Panel B: A scatter plot showing the relationship between the cruising Reynolds number (Re_u) and another scaling parameter (c7Re_f^1.09Fr^-0.05). The x-axis represents the scaling parameter, and the y-axis represents the cruising Reynolds number. Data points are plotted with a trend line in orange. Panel C: A scatter plot showing the relationship between the dimensionless input power (P*) and a scaling parameter (c8Re_f^1.92Fr^-0.80). The x-axis represents the scaling parameter, and the y-axis represents the dimensionless input power. Data points are plotted with a trend line in orange. Panel D: A scatter plot showing the relationship between the dimensionless input power (P*) and another scaling parameter (c9Re_f^2.83Fr^-0.07). The x-axis represents the scaling parameter, and the y-axis represents the dimensionless input power. Data points are plotted with a trend line in orange.
In view of the previously observed divergence in swimming behaviours between weakly and strongly stratified conditions, we further apply the scaling relation (3.5) to distinct stratification regimes. Specifically, the Froude number range
$0.5 \le \textit{Fr} \le 3.0$
is divided into a strong-stratification regime with
$\textit{Fr} \le 1.2$
and a weak-stratification regime with
$\textit{Fr} \gt 1.2$
, based on the sensitivity of swimming velocity to stratification strength shown in figure 11(
$a$
). The data for the cruising Reynolds number and the input power in two considered stratification regimes, expressed in terms of scaling (3.5), are presented in figure 18. Scaling the data separately in two stratification regimes leads to a noticeable improvement in the data collapse, as evidenced by the increased values of
$R^2$
when comparing figures 17 and 18. For the cruising Reynolds number
$\textit{Re}_u$
, the
$R{e_f}$
component in the scaling parameter
$\textit{Re}_{\!f}^\alpha F{r^\beta }$
remains nearly constant at
$\textit{Re}_{\!f}^{1.0}$
. In contrast, the exponent of
$R{e_f}$
in the scaling of the input power
$P^*$
undergoes a significant transition from approximately
$\textit{Re}_{\!f}^2$
in the strong-stratification regime to approximately
$\textit{Re}_{\!f}^3$
in the weak-stratification regime. In particular, the term
$\textit{Re}_{\!f}^{2.83}$
in weakly stratified fluids agrees closely with the homogeneous observations (Lin et al. Reference Lin, Wu and Zhang2021; Liu et al. Reference Liu, Liu and Huang2022). The
$\textit{Fr}$
contribution to the scaling, however, exhibits a marked difference for both
$\textit{Re}_u$
and
$P^*$
datasets when transitioning from weak to strong stratifications, rationalising the departure of swimming behaviours in strongly stratified fluids from those observed in weakly stratified and homogeneous environments. Our study reveals that the swimming performance of an undulatory swimmer in density-stratified environments can be accurately described by a multivariate power law that quantitatively relates swimming velocity and input power to body kinematics and fluid properties.
4. Conclusions
The swimming behaviour of a self-propelled undulatory swimmer in a linearly density-stratified fluid is investigated numerically. The swimmer propels itself by generating a prescribed travelling wave along its body and moves freely in the direction perpendicular to gravity, representing a canonical scenario of fish-like cruising at a constant depth. Simulations are conducted at Reynolds numbers of
$O(10^3)$
. To our knowledge, this work constitutes the first systematic study of the self-propulsion of inertial aquatic swimmers in stratified fluids. The results demonstrate that stratification levels relevant to natural aquatic environments can significantly alter fish-like swimming dynamics, giving rise to behaviours that differ fundamentally from those observed in homogeneous fluids.
We first examine the influence of stratification on wake dynamics. In the stratified regime, buoyancy restoring forces suppress the vertical motion of flow structures while promoting horizontal spreading, resulting in a highly disordered near wake. The competition between inertial and buoyancy effects further generates internal-wave-like flow patterns. The spatial organisation of these wave structures is governed by the ratio of the swimmer’s forcing frequency to the buoyancy frequency, in good agreement with theoretical predictions. We then assess swimming performance in stratified fluids. Most notably, stratification is found to substantially enhance swimming speed for horizontal locomotion, in marked contrast to previous observations that stratification inhibits vertical swimming. This speed enhancement, however, is accompanied by increased power expenditure and reduced swimming efficiency. We show that these distinctive behaviours primarily arise from an increase in inertial reaction force on the swimmer induced by buoyancy restoring effects.
A parametric study spanning the undulation frequency and the Froude number further reveals an intrinsic coupling between swimmer kinematics and fluid oscillations. As a consequence, peak swimming speed is attained when resonance occurs between the undulation frequency and the buoyancy frequency of the surrounding fluid. On the basis of this resonance mechanism, we develop a reduced theoretical model capable of predicting the optimal swimming conditions in stratified environments. Furthermore, we demonstrate that classical scaling relations derived for homogeneous fluids are no longer applicable. Instead, we establish new scaling laws that quantitatively relate swimming speed and power expenditure to body kinematics and fluid properties. In particular, a multivariate power-law formulation involving the undulatory Reynolds number and the Froude number accurately captures the swimming performance across a wide range of stratification strengths.
Overall, our results indicate that undulatory swimmers can achieve substantial speed gains during horizontal locomotion in stratified fluids, albeit at the cost of increased energetic expenditure. These findings provide new physical insights relevant to the design of bioinspired underwater vehicles and to the understanding of biological transport and aggregation near pycnoclines in lakes and oceans, where horizontal swimming has received comparatively little attention. Future work will extend the present framework to three-dimensional swimmers and vertical manoeuvres. In particular, three-dimensional lateral undulation, which characterises the dominant swimming mode of most fish species, is expected to introduce additional complexity in both internal-wave generation and swimming dynamics, and remains an open and intriguing problem.
Funding
This research has been supported by the Zhejiang Provincial Natural Science Foundation of China under grant no. LZ25A020003.
Declaration of interests
The authors report no conflict of interest.
Appendix A. Time step, grid and domain-size independence studies
We choose the most challenging case within the parameter space for independence tests, which is characterised by the strongest stratification (
$\textit{Fr} = 0.5$
) and largest undulation (
${A^*} = 0.2$
and
${f^*} = 2.0$
).
Time histories of the instantaneous swimming velocity over one undulation cycle at the steady state using different time steps. The undulation parameters are taken as
${A^*} = 0.2$
and
${f^*} = 2.0$
, combined with a stratification strength of
$\textit{Fr} = 0.5$
.

Time histories of the instantaneous swimming velocity over one undulation cycle at the steady state using different grids. The undulation parameters are taken as
${A^*} = 0.2$
and
${f^*} = 2.0$
, combined with a stratification strength of
$\textit{Fr} = 0.5$
. Here,
$n$
denotes the number of grid points along the swimmer body.

Time histories of the instantaneous swimming velocity over one undulation cycle at the steady state using different computational domains. The undulation parameters are taken as
${A^*} = 0.2$
and
${f^*} = 2.0$
, combined with a stratification strength of
$\textit{Fr} = 0.5$
.

The time independence is first demonstrated by testing three time steps, including
$\Delta {t_1} = T/500$
,
$\Delta {t_2} = T/1000$
and
$\Delta {t_3} = T/2000$
. Figure 19 shows the effect of different temporal resolutions on the evolution of the swimmer’s instantaneous velocity over one undulation cycle after the periodically steady state is reached. The maximum difference in instantaneous velocity between
$\Delta {t_2}$
and
$\Delta {t_3}$
is 0.51 %, accompanied by a 0.32 % discrepancy in average value. We thus use
$\Delta {t_2} = T/1000$
to run all the simulations.
We then assess the grid independence by examining four spatial resolutions: Grid 1 with 120 grid points along the swimmer body (2 977 836 elements), Grid 2 with 160 grid points along the swimmer body (4 521 600 elements), Grid 3 with 200 grid points along the swimmer body (6 372 256 elements) and Grid 4 with 240 grid points along the swimmer body (8 537 184 elements). The time histories of the instantaneous swimming velocity in figure 20 show an excellent agreement between Grid 3 and Grid 4 with the deviation being less than 0.1 %. Therefore, the spatial resolution Grid 3 is employed in all cases.
To ensure that the waves reflected from the side boundaries do not have a significant effect on the swimmer’s locomotion, we further carry out a domain independence test by considering three domain sizes
$40L \times 20L$
,
$70L \times 50L$
and
$100L \times 80L$
. As shown in figure 21, the time series of swimming velocity resulting from different domain sizes are found to match very closely, suggesting that the interaction between the swimmer and reflected waves is negligible for the current problem. In our simulations, the moderate domain size
$70L \times 50L$
is used to maintain accuracy and keep computational cost reasonable.



T
0.1T
t=T
A∗=0.2
f∗=1.0
ωz
Uref/L
Fr=∞
A∗=0.1
a
b
Ω=0.55
c
d
Ω=1.75
Ω=ωf/N
a
c
b
d
A∗=0.1
f∗=1.0
ρ0/L
z
ωb
a
Fr=3.0
b
Fr=1.0
c
Fr=0.5
g0ρ0/L
g0=10ms−2
A∗=0.1
f∗=1.0
f∗=1.0
A∗=0.1
A∗=0.2
COT
Fr
Fr
COT
FB
FS
FB
FR
Fx
Fy
A∗=0.1
f∗=0.5
f∗=1.0
f∗=1.5
f∗=2.0
d/L
Fr
f∗
f∗
(f∗,Fr)
U¯
Frc
ΔU¯
ΔU¯=(U¯s−U¯h)/U¯h
s
h
St
(f∗,Fr)
Frc
ωf>N
ωf=2πf
(f∗,Fr)
P¯
COT
COTs
Reu
Ref
Reu
Ref
Reu∼Ref
a
d
Reu
Ref4/3(1+Fr−β)
c1
β
R2=0.918
a
Reu
b
P∗
Refα(1+Fr−β)
c2
c3
α
β
a
b
R2=0.986
a
Reu
b
P∗
RefαFrβ
c4
c5
α
β
a
b
R2=0.986
a
b
Reu
c
d
P∗
RefαFrβ
a
c
Fr≤1.2
b
d
Fr>1.2
c6
c9
α
β
a
d
R2=0.996
A∗=0.2
f∗=2.0
Fr=0.5
A∗=0.2
f∗=2.0
Fr=0.5
n
A∗=0.2
f∗=2.0
Fr=0.5