Soft Mobility Theory

arXiv:2605.23869 · physics.flu-dyn · Submitted 2026-05-22 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: We're looking at the paper: "Soft Mobility Theory".

Title and authors: Mira: We’ve seen that the title is "Soft Mobility Theory," but what does that actually mean for the core of the paper? It suggests a new way to describe motion when things are squishy rather than just rigid particles.

Kai: Basically, it gives you an explicit, shape-dependent linear relation between velocities and loads for a single body, which is really helpful for figuring out how to design something.

Lev: I see that explicit relation being the core appeal here. Because in simulations, especially when we try to do inverse design—figuring out what shape or stiffness we need—you usually end up with some kind of complex saddle-point system you have to solve iteratively.

Kai: Exactly. The paper suggests they invert that dense grand mobility tensor directly, which is computationally expensive but differentiable, meaning the gradient for optimization comes almost for free through reverse mode automatic differentiation.

Mira: That’s a key point because it bypasses the need to store all those intermediate states that would normally make those iterative solvers incredibly memory intensive.

Lev: For real hardware running on actual chips, if you try to solve that grand mobility tensor directly, the sheer size of that matrix is a problem, but if you only have a few degrees of freedom, like this paper suggests—say N equals small—it becomes feasible for gradient-based optimization.

Kai: So what they’re doing with these sphere assemblies is using the Rotne–Prager–Yamakawa approximation to get a closed-form mobility that works for those many bodies interacting in a flow.

Mira: It’s an established tool for rigid suspensions, and they are showing how you can use it as a starting point to treat deformable bodies as assemblies of rigid sub-bodies linked by elastic constraints.

Lev: That’s where the complexity shifts from simulating millions of degrees of freedom down to managing the constraints between these few interacting components.

Kai: So, if we think about what this means for the world right now, it means we can move away from just modeling rigid particles and start designing systems that actually deform intelligently in a fluid.

Mira: It opens up inverse design for soft microrobotics because you aren't stuck with pre-set shapes; you can optimize the shape itself to get better performance.

Lev: I’m interested in the fidelity here. If we were trying to run this on real hardware, we’d have to worry about how much error comes from approximating that background flow field when treating every body as a deformable entity.

The paper's summary: Kai: Now, let’s look at what the paper actually summarizes in the middle section of the text. They describe this configuration-dependent ordinary differential equation that governs how the generalized coordinates of the body move over time.

Mira: That equation is built from decomposing a virtual velocity into a rigid-body part and a strain-producing deformation part, and showing how those parts interact with the hyperelastic terms in that main mobility equation.

Lev: It’s interesting that they break down this motion description into components: the elastic mobility carries the response to internal elastic forces, and there’s another one, the body-force mobility, which handles external field forces.

Kai: And then there's this flow-coupling tensor that they call C E, which is basically the soft-body version of the Bretherton tensor that links your body directly to the background rate of strain in the fluid.

Mira: So, what’s important here is that all these terms—mobility, advection, and flow-gradient coupling—are explicitly functions of the instantaneous deformation. That’s what makes it configuration-dependent.

Lev: If those tensors aren't just constants defined beforehand, but they change based on the current shape and strain, then you can’t use a simple pre-calculated model for motion; you have to solve this evolving equation at every step.

Kai: That dependence is what allows the inverse design to work because we can make those tensors differentiable with respect to the design parameters—like spring stiffness or sphere radii.

Mira: And they show that by using this framework, you can actually optimize a three-sphere swimmer, finding that the optimal shape is about seven point four eight times more efficient than a non-optimized one in shear flow conditions.

The paper's improvements: Lev: The authors mention several things they did to make this framework better than what was already out there, focusing on making it usable for optimization.

Kai: One big improvement is the implementation itself: they built a Python library called softmobility that uses JAX, which makes every tensor differentiable end-to-end.

Mira: That differentiability allows the Optax library to do gradient-based optimization of shape and stiffness parameters directly, using reverse mode automatic differentiation to build the training signal.

Lev: That’s powerful because it means we don't need to manually derive gradients for every single design parameter; the system does that for us based on how much an objective function changes with those parameters.

Kai: And they are specifically targeting systems with very few degrees of freedom, typically N equals O(one), which is why they focus on optimizing a single deformable body rather than simulating huge suspensions <ref:2605.23869#pg1>.

Mira: They also point out that while the Rotne–Prager–Yamakawa approximation is convenient, it’s a pairwise far-field expansion and breaks down when spheres get close enough to touch, which they handle by allowing you to plug in different mobility models if you need lubrication corrections.

Lev: So, the limitation they admit is that this framework is tailored for systems with few degrees of freedom, meaning it’s less competitive than standard Stokesian dynamics when you’re trying to simulate a huge suspension with millions of rigid bodies.

Kai: But for optimizing a single swimmer or microdevice, that trade-off seems worthwhile because the gradient calculation is essentially free.

Conclusion: Mira: So, wrapping up on "Soft Mobility Theory," the main implication is that we have a way to design soft bodies where the mobility and coupling effects are inherently linked to their current shape during fluid flow.

Kai: It provides an explicit ordinary differential equation for motion, which is much more transparent than relying on black-box simulations when trying to optimize a physical device.

Lev: From my side, the engineering takeaway is that if you want to do this practically with real hardware, you need to make sure your chosen mobility model handles those near-contact situations where the approximation breaks down.

Kai: And for the immediate future, they suggest extensions like adding Brownian fluctuations or incorporating far-field hydrodynamic interactions between soft bodies as deformable multipoles.

Mira: The ultimate goal seems to be using this differentiable rollout for systematic inverse design campaigns, especially for optimizing flexible microswimmers that need to navigate complex flows like turbulent ones.

Lev: For me, the future of this is seeing if we can use these tools to learn soft body controllers, applying physical intelligence ideas from morphological computation and physical intelligence.

Kai: So we’ve seen how they used this framework to optimize a three-sphere swimmer and design a soft gyrotactic surfer that swims nineteen percent faster than its rigid counterpart in Taylor–Green flow.

Mira: That result is compelling because it shows that flexibility can actually turn a slowdown into a speedup by exploiting passive deformation in the flow.

Lev: It’s about finding those optimal design parameters, which they found to be spring stiffness k and sphere radii a1 and l1, at values like zero point one six nine for the ratio of sphere radii to rest lengths and eighteen point two for the rigidity bar k.

Kai: Those specific numbers show where the sweet spot is between having floppy springs where k goes to zero, and very stiff springs where k goes to infinity.

Mira: It’s a neat demonstration that this framework can give you a concrete design target rather than just saying "it works."

Lev: In summary, the "Soft Mobility Theory" gives us a differentiable path for designing flexible swimmers by leveraging explicit dependencies on deformation and providing gradient information efficiently.

Aix-Marseille University · CNRS · Centrale Med

physics.flu-dyn

Submitted: 2026-05-22

Updated: 2026-11-17

Comments: 24 pages, 11 figures

Code: https://github.com/C0PEP0D/SoftMobility

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 83/100

The gist: Predicting how a deformable body moves and deforms in a viscous flow underlies problems ranging from microorganism locomotion to soft microrobotics…

Key concepts

Soft Mobility Theory
This theory derives a configuration-dependent ordinary differential equation that describes how a deformable body moves in a viscous fluid. It extends classical rigid-body mobility by making the mobility, elastic, and flow-coupling properties explicitly dependent on the body's current shape and deformation state.
Virtual Power and Lorentz Reciprocal Theorem
These mathematical principles are used to obtain the weak formulation of equilibrium equations for a hyperelastic body in Stokes flow. They lead to an equation that governs the body's motion based on its generalized coordinates, simplifying complex continuum mechanics into a tractable differential equation.
JAX Implementation and Inverse Design
The framework is coded using JAX, which allows the entire simulation to be end-to-end differentiable. This differentiability enables gradient-based optimization of design parameters (like shape or stiffness) by calculating gradients for objectives like maximizing swimming speed, without needing complex iterative solvers.

Terminology

Summary

Predicting how a deformable body moves and deforms in a viscous flow underlies problems ranging from microorganism locomotion to soft microrobotics, yet existing frameworks are either problemspecific or ill-suited to inverse design The authors propose the soft mobility theory: applying the principle of virtual power and the Lorentz reciprocal theorem to a hyperelastic body in a background Stokes flow yields a configuration-dependent ordinary differential equation for the generalized coordinates of the body This soft mobility equation extends classical rigid-body mobility theory in that the mobility, elastic, body-force, and flow-coupling tensors all depend explicitly on the instantaneous deformation They specialize the framework to assemblies of hydrodynamically interacting spheres connected by elastic springs, using the Rotne–Prager–Yamakawa approximation to compute the mobility, and validate it on canonical problems spanning rigid and flexible bodies in quiescent and shear flows An open-source JAX implementation makes entire simulations end-to-end differentiable, allowing for efficient gradient-based inverse design

Theoretical Foundation of Soft Mobility

The soft mobility theory is derived from the continuum elastohydrodynamic problem of a hyperelastic body in a background Stokes flow The weak formulation of the equilibrium equations is obtained by applying the principle of virtual power and the Lorentz reciprocal theorem This leads to an equation (14) that governs the body's motion in terms of generalized coordinates The structure of Eq. (14) becomes transparent through the decomposition of χ˙ into a rigid-body part and a strain-producing deformation part For any rigid-body virtual velocity V∗ = u∗0 + ω∗0 × (χ(x, t) − r0), the hyperelastic term in Eq. p. 4 (14) vanishes because F⊤ · ∇V∗ is then skew-symmetric in Eq. p. 4 (5) when ∇V∗ = [ω∗0]× · F

Modeling Sphere Assemblies

The framework specializes the theory to assemblies of spheres connected by springs, using a six-component position Xi(Q, t), a function of time and the NQ degrees of freedom Q representing the deformation The grand velocity V = [V1,...,VN] in the body frame satisfies V = Jsph · Q˙ + Vact The full state of the assembly in the laboratory frame is therefore given by the generalized coordinates (of size Nq = 6 + NQ) q = [x0, Q]

Numerical Implementation and Optimization

The framework is implemented in softmobility, a Python library that uses JAX for end-to-end differentiable simulations This allows for gradient-based optimization of shape and stiffness parameters via the Optax library The framework is most useful when one wants to optimize the design of a single deformable body, as it inverts the dense 6N × 6N grand mobility matrix directly

Application and Validation

The framework was validated on five canonical problems spanning rigid and flexible bodies in quiescent and shear flows These include the gravity-driven sinking of a chiral rigid body, the wrapping of a rotating clamped fiber, soft Jeffery orbits in shear, optimization of a three-sphere swimmer, and a soft gyrotactic surfer The soft surfer demonstrates that flexibility can turn a slowdown into a speedup by exploiting passive deformation in Taylor–Green flow

Future Extensions

The soft mobility theory could be extended in three directions, including adding Brownian fluctuations to connect it to recent work on Brownian dynamics Incorporating far-field hydrodynamic interactions between soft bodies by treating each as a deformable multipole Finally, the differentiable rollout invites systematic inverse-design campaigns for optimizing flexible microswimmers for magnetic actuation or navigation in turbulent flows The soft surfer of §IV E is a first proof of concept in this direction This framework combines the projection idea of Ref. [59] with three additions: a closed-form differentiable mobility (here Rotne–Prager–Yamakawa, though any differentiable mobility model is compatible), explicit elastic and flow-coupling tensors, and a JAX implementation that makes every tensor differentiable with respect to the design parameters The cost of one forward simulation exceeds that of an explicit bead-spring or multiblob simulation, but the gradient of any scalar objective with respect to the design parameters comes essentially for free through reverse-mode automatic differentiation The framework is tailored for systems with few degrees of freedom, typically NQ = O(1), in contrast to standard Stokesian dynamics approaches that handle suspensions with NQ = O(N), where N = O(103) rigid bodies or more The authors also note that the Rotne–Prager–Yamakawa approximation is a pairwise far-field expansion that breaks down at lubrication contact between spheres Problems with nearly touching elements would require lubrication corrections or a different mobility model, which the framework accommodates as long as that model remains differentiable The soft mobility theory is tailored for systems with few degrees of freedom, typically NQ = O(1), in contrast to standard Stokesian dynamics approaches that handle suspensions with NQ = O(N), where N = O(103) rigid bodies or more The framework is therefore most useful when one wants to optimize the design of a single deformable body, and least competitive when one wants to simulate large suspensions for which Stokesian dynamics remains the right tool

Where the gains come from

The primary advantage of this framework is that while one forward simulation costs more than explicit bead-spring or multiblob simulations, the gradient of any scalar objective with respect to the design parameters comes essentially for free through reverse-mode automatic differentiation This contrasts with methods that invert the saddle-point system arising from constraints using preconditioned iterative solvers, which are difficult to differentiate by reverse mode because storing intermediate states incurs a prohibitive memory cost The authors instead invert the dense 6N × 6N grand mobility matrix directly in Eq. (29), which is computationally costly but differentiable

Pruning redundant reasoning tokens

The paper develops a general “soft mobility” framework that simultaneously is derived from the continuum elastohydrodynamic problem of a hyperelastic body in a background Stokes flow; accommodates arbitrary generalized coordinates and yields an explicit, configuration-dependent ordinary differential equation in which mobility, advection, and flow-gradient coupling tensors all appear This framework is constructed so that all tensors are smooth, differentiable functions of the design parameters, enabling end-to-end gradient-based inverse design The soft mobility equation (33) sits between two established families, bead and multiblob methods which discretize a deformable body into rigid sub-bodies and enforce inter-bead constraints by solving a saddle-point linear system at every timestep

How the training signal is built

The training signal for optimization is built through reverse-mode automatic differentiation, which allows the gradient of any scalar objective with respect to the design parameters to be computed essentially for free The authors use this to recover the asymptotic optimum of a three-sphere swimmer and design a soft gyrotactic “surfer” that exploits passive deformation to ascend faster than its rigid counterpart in a Taylor–Green flow

Summary of Key Findings

The soft mobility theory provides an explicit, configuration-dependent ordinary differential equation (33) for the generalized velocities, where the four right-hand-side terms admit a clean physical reading The elastic mobility MK carries the response to internal elastic forces, while the body-force mobility MH carries the response to external field H The flow-coupling tensor CE is the soft-body analogue of the Bretherton tensor and couples the body to the background rate-of-strain The optimization of a three-sphere swimmer shows that for optimal parameters, the final shape is 7.48 times more efficient than a non-optimized shape The soft surfer achieves an effective vertical velocity of Veff = 1. p. 0193Vswim, which means it swims 19% faster than without background flow The optimal parameters found for the soft swimmer are a1/a0 = 0. p. 0169 and κ = 18.2 This optimization lies in between the limits of stiff springs (κ → ∞) and floppy springs (κ → 0) The optimal design parameters are a1/a0 = 0. p. 7169 and κ = 18.2 The effective velocity is Veff = 1. p. 0193Vswim This means that this soft swimmer swims 19% faster than without background flow

References

[1] A. W. Visser, Hydromechanical signals in the plankton, Mar. Ecol. Prog. Ser. 222, 1 (2001)

[78] G. B Jeffery, The motion of ellipsoidal particles immersed in a viscous fluid, Proc. p. 1 R. Soc. Lond. A 102, 161 (1922)

[83] A. Montino and A. DeSimone, Three-sphere low-reynolds-number swimmer with a passive elastic arm, Eur. p. 0 Phys. J. E 38, 1 (2015)

[86] R. Monthiller, A Loisy, M Koehl, B Favier and C Eloy, Surfing on turbulence: a strategy for planktonic navigation

[91] W. T. Funkenbusch, K. S Silmore and J. W Swan, Approaches for fast brownian dynamics simulation with constraints

[74] N. Coq, O Du Roure, J Marthelot, D Bartolo and M Fermigier, Rotational dynamics of a soft filament: Wrapping transition and propulsive forces

[87] S. Mecanna, A Loisy and C Eloy, A critical assessment of reinforcement learning methods for microswimmer navigation in complex flows

[90] A.-K Tornberg and L Greengard, A fast multipole method for the three-dimensional stokes equations

[73] L. Li, H Manikantan, D Saintillan and S E Spagnolie, The sedimentation of flexible filaments

[74] N Coq, O Du Roure, M Fermigier and D Bartolo, Rotational dynamics of a soft filament: Wrapping transition and propulsive forces

[75] N Coq, O d Roure, M Fermigier and D Bartolo, Helical beating of an actuated elastic filament

[78] G B Jeffery, The motion of ellipsoidal particles immersed in a viscous fluid, Proc. R. Soc. Lond. A 102, 161 (1922)

[79] F P Bretherton, The motion of rigid particles in a shear flow at low Reynolds number, J. Fluid Mech. p. 0 p. 0 p. 0 p. 0 14, 284 (1962)

[81] K Ishimoto, Jeffery’s orbits and microswimmers in flows: A theoretical review, J. Phys. Soc. Jpn. 92, 062001 (2023)

[84] S Colabrese, K Gustavsson, A Celani and L Biferale, Flow navigation by smart microswimmers via reinforcement learning

[85] T Pedley and J O Kessler, Hydrodynamic phenomena in suspensions of swimming microorganisms, Annu. Rev. Fluid Mech. 24, 313 (1992)

[77] C F Schmid, L H Switzer and D J Klingenberg, Simulations of fiber flocculation: Effects of fiber properties and interfiber friction

[81] K Ishimoto, Jeffery’s orbits and microswimmers in flows: A theoretical review, J. Phys. Soc. Jpn. 92, 062001 (2023)

[74] N Coq, O Du Roure, M Fermigier and D Bartolo, Rotational dynamics of a soft filament: Wrapping transition and propulsive forces

[75] N Coq, O d Roure, M Fermigier and D Bartolo, Helical beating of an actuated elastic filament

[78] G B Jeffery, The motion of ellipsoidal particles immersed in a viscous fluid, Proc. R. Soc. Lond. A 102, 161 (1922)

[79] F P Bretherton, The motion of rigid particles in a shear flow at low Reynolds number, J. Fluid Mech. 14, 284 (1962)

[81] K Ishimoto, Jeffery’s orbits and microswimmers in flows: A theoretical review, J. Phys. Soc. Jpn. 92, 062001 (2023)

[84] S Colabrese, K Gustavsson, A Celani and L Biferale, Flow navigation by smart microswimmers via reinforcement learning

[85] T Pedley and J O Kessler, Hydrodynamic phenomena in suspensions of swimming microorganisms, Annu. Rev. Fluid Mech. 24, 313 (1992)

[77] C F Schmid, L H Switzer and D J Klingenberg, Simulations of fiber flocculation: Effects of fiber properties and interfiber friction

[81] K Ishimoto, Jeffery’s orbits and microswimmers in flows: A theoretical review, J. Phys. Soc. Jpn. 92, 062001 (2605. p. 023869IV]

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. End-to-end differentiable gradient-based inverse design for soft bodies can be achieved by utilizing a JAX implementation that makes every tensor differentiable with respect to design parameters, allowing for efficient optimization of shape and stiffness parameters via reverse mode automatic differentiation.

  2. The improved system can perform gradient-based inverse design to recover the asymptotic optimum of a three-sphere swimmer or design soft gyrotactic “surfers” that exploit passive deformation to ascend faster than their rigid counterparts in Taylor–Green flow.

  3. The framework allows for the optimization of three design parameters, such as spring stiffness k, sphere radii a1, and rest lengths l1 for a three-sphere swimmer to maximize the average effective vertical velocity Veff.

  4. The system can be used to design soft bodies that exhibit enhanced vertical migration in Taylor–Green flow by optimizing parameters like a1/a0 and the dimensionless rigidity bar k.

  5. The framework enables the learning of soft body controllers in the spirit of morphological computation and physical intelligence by applying differentiable rollout to optimize flexible microswimmers for magnetic actuation or navigation in turbulent flows.

Sources

Related papers