Imagine trying to simulate a block of rubber or a thin sheet of metal. Intuitively, if you push on it, it should bend or deform smoothly. However, standard computer simulations often encounter a frustrating phenomenon: the material suddenly behaves as if it is completely rigid, frozen in place. This artificial stiffening is known as numerical locking.
Locking is not a physical property of the material; it is a mathematical artifact arising from how we discretize continuous physics into finite elements or particles. When the equations governing a simulation are forced to satisfy strict constraints—such as preserving volume perfectly or preventing unphysical shearing—the available degrees of freedom in standard elements become over-constrained. The only mathematical solution that satisfies these constraints is one where the material barely moves at all, causing the simulation to stall or yield highly inaccurate, overly stiff results.
The Root Causes: Incompressibility and Extreme Aspect Ratios
Numerical locking typically manifests in two primary regimes: nearly incompressible materials and thin, high-aspect-ratio geometries.
In nearly incompressible materials (where Poisson's ratio approaches the physical limit of 0.5, such as rubber, water, or soft biological tissues), the relationship between volume preservation and material stiffness becomes highly non-linear, as noted by Chen et al.. Standard displacement-based finite elements, such as the tetrahedral meshes frequently used in soft tissue simulations, struggle to resolve these constraints. As Salehi et al. and Descamps et al. point out, enforcing volume preservation across standard elements over-constrains the displacement fields, leading directly to volumetric locking (or Poisson-locking) and severe stress oscillations. This issue extends to fluid simulations as well; Li et al. observe that enforcing incompressibility on standard computational grids can induce fluid leakage and locking artifacts.
A parallel issue occurs in structural elements with high aspect ratios, such as thin beams, shells, or engineering textiles. As Shafqat et al. explain, standard 3D solid elements exhibit artificial stiffness when subjected to bending because they cannot represent pure bending modes without introducing spurious shear strains. This structural locking leads to a massive underestimation of displacements. Furthermore, when simulating systems with extreme stiffness disparities—such as rigid bones embedded in soft muscle tissue—standard solvers frequently lock up because they cannot effectively reconcile the vastly different elastic properties across material boundaries, a challenge highlighted by Modi et al..
Unburdening Displacements via Mixed Formulations
The most prominent strategy to eliminate volumetric locking is the use of mixed or hybrid formulations. Instead of relying solely on a single displacement field to compute both shape changes and internal pressures, mixed formulations introduce auxiliary independent variables.
By treating pressure or stress as an independent variable alongside displacement—typically rooted in variational principles like the Hellinger-Reissner principle—the mathematical constraints are decoupled. As Huang et al. (2014) demonstrate, this dual-variable approach provides a flexible framework that remains completely immune to Poisson-locking, ensuring numerical stability even as the Poisson's ratio approaches 0.5. Treister et al. and Ji leverage this strategy by incorporating an auxiliary pressure variable to maintain stability and accurate stress calculations during large deformations of nearly incompressible materials like rubber.
To make these mixed elements robust, researchers employ stabilization techniques. For instance, Gsell et al. utilize stabilized P1-P1 elements and adapt the classic MINI element by enriching the displacement approximation space with higher-order polynomial components. This prevents the numerical errors and artificial stiffening that commonly plague standard linear elements or traditional multiplicative splits. However, these formulations introduce their own challenges: in the strict incompressibility limit, solvers must handle complex saddle-point problems and the null space of divergence-free functions, which requires specialized multigrid methods to maintain parameter robustness, as discussed by Adler et al..
Smoothing Constraints via Strain Modification and Averaging
Rather than adding variables, another powerful approach modifies how strains are calculated or sampled across the simulation domain to prevent local over-constraint.
In particle-based methods like the Material Point Method (MPM), Zhao et al. introduce a method that averages the deformation gradient across the simulation domain. By utilizing the background grid as a natural filter to smooth Jacobian values during standard particle-to-grid and grid-to-particle transfers, the formulation eliminates the localized volume-preservation bottlenecks. This yields smooth, physically plausible stress fields without requiring specialized geometric averaging schemes or suffering from checkerboard artifacts.
For thin-walled structures and textiles, structural locking is mitigated by combining Enhanced Assumed Strain (EAS) and Assumed Natural Strain (ANS) methods. As Schafer et al. (2021) show, integrating these techniques into solid-shell elements allows simulations to capture through-thickness compaction and complex bending or wrinkling patterns without introducing artificial stiffness.
Interestingly, while locking is usually an error to be avoided, controlled physical locking mechanisms can sometimes be beneficial. Schafer et al. (2022) demonstrate that incorporating a perpendicular compressive locking mechanism is necessary to model the physical limit of fiber compaction in non-crimp fabrics, preventing numerical element collapse under high shear forces.
Leveraging Splines and Alternative Discretization Spaces
Shifting the underlying mathematical representation of the geometry offers another path to locking-free simulation. Isogeometric Analysis (IGA), which uses smooth spline basis functions (like NURBS) instead of traditional polynomials, provides unique opportunities to bypass locking.
In cloth simulation, Zheng et al. show that sampling strain-limiting constraints precisely at the knot mesh vertices of spline-based shells effectively eliminates numerical locking and artificial stiffness. This allows the model to capture realistic, inextensible cloth deformations while maintaining grid independence and avoiding hourglassing artifacts. By carefully aligning where constraints are evaluated with the underlying spline structure, the simulation retains the flexibility needed to represent complex physical behaviors accurately.
Key Foundational Work
An adaptive hybrid stress transition quadrilateral finite element method for linear elasticity (2014) — contributed 3 concept elements to this direction.
A framework for facial surgery simulation (2002) — contributed 1 concept elements to this direction.