Quasi-Graded Temporal Meshes: Analysis & Applications

• Quasi-graded temporal meshes are defined by variable time-step sizes governed by prescribed grading functions, effectively resolving singularities in time-dependent problems.
• They enhance numerical accuracy in fractional PDEs and two-phase flow by adaptively controlling step growth and ensuring mesh regularity.
• Construction methods combine global grading with localized space–time refinement, supporting high-dimensional applications like dynamic geometry and mesh extraction.

A quasi-graded temporal mesh is a discretization of the time domain in which the step sizes vary according to a prescribed grading function, locally or globally, and may differ across spatial regions, subject to mesh-ratio or minimum-step constraints. This approach is motivated by the need to resolve temporal singularities or localized time-variation in solutions to time-dependent partial differential equations, as well as to provide temporally coherent mesh extraction in high-dimensional applications such as dynamic geometry or two-phase flow. In both mathematical analysis and computational geometry, quasi-graded meshes generalize strictly graded temporal meshes by allowing for flexible, problem-adapted grading that is not globally uniform but retains controlled step growth and regularity properties.

1. Definitions and Core Structure

Let $[0, T]$ be the time interval under consideration, and let $N$ denote the total number of temporal sections. A prototypical graded temporal mesh is defined by

$$t_k = T\left( \frac{k}{N} \right)^r,\qquad k=0,1,\dots,N,$$

where $r \geq 1$ is the grading exponent, and $t_k$ are the mesh nodes. The corresponding local time increments are

$$\Delta t_k = t_k - t_{k-1} \approx r T N^{-r} k^{r-1} \quad \text{for } N \gg 1.$$

A quasi-graded temporal mesh relaxes the strict formula of graded meshes, admitting any partition $\{t_k\}$ such that: $$\Delta t_1 \simeq N^{-r}, \quad t_k \simeq N^{-r}k^r, \quad \Delta t_k \simeq \Delta t_1^{1/r} t_k^{1-1/r}$$ and the step-ratio satisfies

$$\frac{\Delta t_k}{\Delta t_{k-1}} \leq C\Bigl(\frac{k}{k-1}\Bigr)^{r-1} \leq C \quad \text{for } k\ge 2,$$

where $C$ is a constant. In practice, quasi-grading requires control over the growth of time steps and guarantees small initial increments to resolve early-time solution singularities (Kopteva et al., 2019).

2. Mathematical Motivation: Singularities and Time-Fractional Diffusion

For differential equations with weakly regular solutions, especially those involving Caputo fractional derivatives of order $\alpha \in (0,1)$, the solution often satisfies

$$|\partial_t^m u(t)| \leq C t^{\alpha-m}, \quad m=0,1,2,\quad t \in (0,T],$$

indicating singular time behavior at $t=0$ (Kopteva et al., 2019, Kumari et al., 2023). Uniform time steps yield poor resolution of this singular region. Graded or quasi-graded meshes, with $\Delta t_1 \sim N^{-r} \ll N^{-1}$, achieve higher accuracy near $t=0$ while allowing coarser steps away from the singularity.

For high-order approximations to the Caputo derivative, the truncation error over such meshes is governed by the minimum of the interpolation and grading errors, with the optimal exponent given by $r^* = (4-\alpha)/\alpha$ in fourth-order schemes (Kumari et al., 2023).

3. Construction Algorithms and Mesh Generation

The construction of quasi-graded temporal meshes depends on the context:

a) Fractional-Derivative Evolution Equations

Meshes are constructed globally based on the grading parameter $r$, leading to temporal nodes as indicated before. Modified difference schemes (e.g., L1-type or Alikhanov-type) are then discretized on these nodes: $$t_n = T \left( \frac{n}{N} \right)^r, \quad n = 0,1,\dots,N.$$ The local increments are used for discretizing convolution-type quadratures arising from Caputo derivatives, with weights that depend on the mesh geometry (Kumari et al., 2023).

b) Localized Space–Time Refinement (Two-Phase Flow, DG-FEM)

Quasi-graded temporal discretization may be applied locally in space, resulting in locally refined slabwise temporal partitions. For each spatial node $X_i$, subdivision counts $m_i$ and grading functions $\gamma_i$ are assigned adaptively according to problem indicators (e.g., level-set function, interface curvature): $$\tau_i^j = t_n + \Delta t \cdot \gamma_i(j / m_i), \quad j = 0,1,\dots, m_i.$$ Subsequent simplex construction applies a Freudenthal subdivision to the extruded space–time polytope, ensuring the compatibility of different temporal subdivisions across the spatial mesh (Karyofylli et al., 2019).

c) High-Dimensional Spatio-Temporal Meshes (Dynamic Geometry)

For applications such as temporally smooth mesh extraction in 4D (space-time), a spacetime binary-octree alternates between spatial and temporal splits according to screen-space error and temporal coherence thresholds. Leaves of the tree correspond to spacetime blocks over which dual contouring (4D "Marching Cubes") is applied, with quasi-grading enforced by minimum temporal interval constraints $\tau_0$ (Ma et al., 16 Sep 2025).

4. Stability, Convergence, and Error Analysis

Stability and convergence of discretization schemes on quasi-graded meshes are established through barrier function techniques and discrete maximum principles. For L1-type and Alikhanov-type discretizations, convergence rates depend explicitly on the grading parameter $r$ and the order $\alpha$ of the fractional derivative: $$\max_k |u(t_k) - U^k| \leq C N^{-\min\{\alpha r, 1\}}$$ for the L1 scheme, and

$$\max_k |u(t_k) - U^k| \leq C N^{-\min\{2\alpha r, 2\}}$$

for Alikhanov, valid for weakly regular solutions (Kopteva et al., 2019). To recover full first-order accuracy (L1) or second-order accuracy (Alikhanov) at positive times, it suffices to choose $r \geq 1/\alpha$. For optimal global-in-time accuracy, stricter criteria, such as $r \geq (2-\alpha)/\alpha$, must be met.

In high-order Caputo schemes, optimal convergence $O(N^{-(4-\alpha)})$ is attained when $r = (4-\alpha)/\alpha$, compensating both the singularity and polynomial approximation errors (Kumari et al., 2023).

Empirical results confirm the necessity of the optimal grading parameter. For instance, with $\alpha=0.6$, uniform grading ($r=1$) achieves only $O(\tau^\alpha)$, while optimal $r$ restores the full expected rate $O(\tau^{4-\alpha})$ (Kumari et al., 2023).

5. Applications in Numerical Simulation

Quasi-graded temporal meshes are fundamental in several computational contexts:

• Fractional PDEs with Singularities: In subdiffusion and related fractional equations, quasi-graded meshes enable accurate time discretization despite initial singularities, yielding optimal rates of convergence for both L1 and high-order schemes (Kumari et al., 2023, Kopteva et al., 2019).
• Space–Time Finite Elements in Two-Phase Flow: Adaptive, locally refined, quasi-graded meshes track moving interfaces, achieve high spatial and temporal resolution near features, and reduce computational cost by allowing coarser meshing elsewhere. The narrow-band strategy and Freudenthal simplex construction ensure local grading around interfaces while preserving global mesh compatibility (Karyofylli et al., 2019).
• Procedural and Dynamic Geometry (4D Mesh Extraction): In temporally smooth mesh extraction for camera trajectories, quasi-graded time discretization is enforced by limiting temporal splits in the underlying spacetime tree, controlling the minimal lifetime $\tau_0$ of nodes and thus the temporal coherence of mesh transitions. This prevents mesh "popping" and ensures connectivity remains stable over user-defined intervals. Smoothness metrics include maximum vertex displacement, normal-continuity error, and SSIM photometric consistency; worst-case discontinuities are dramatically reduced compared to per-frame or static mesh approaches (Ma et al., 16 Sep 2025).

6. Construction Constraints, Control, and Practical Guidelines

A principal feature of quasi-graded temporal meshes is the user-controlled adaptivity in step sizes, subject to constraints that preserve mesh quality and scheme stability:

• Step-Ratio Bound: Growth of $\Delta t_k$ is bounded to avoid rapid changes that would destabilize numerical schemes (Kopteva et al., 2019).
• Minimum Time-Interval ($\tau_0$): In 4D and other dynamic applications, this enforces that combinatorial mesh updates can occur no more rapidly than every $\tau_0$ seconds, giving precise control over the temporal granularity of LOD and mesh transitions (Ma et al., 16 Sep 2025).
• Spatial Adaptivity: Subdivision counts and local grading exponents can be assigned based on error indicators or physical features (e.g., interface geometry in multiphase flow), enforced within each slab or element (Karyofylli et al., 2019).
• Optimal Grading Parameter Selection: Analysis provides explicit criteria for $r$ as a function of equation order and desired convergence (Kumari et al., 2023, Kopteva et al., 2019).

7. Summary Table: Quasi-Graded Temporal Meshes Across Domains

Domain/Application	Grading Principle	Key Constraints/Parameters
Fractional PDEs (global)	$t_n = T(n/N)^r$	$r^*=(4-\alpha)/\alpha$, step-ratio bound
Two-phase flow FE (local)	Band-based local grading	Interface proximity, $m_i$, $\gamma_i(s)$
4D mesh extraction (geometry/vision)	Min interval per spacetime cell, tree splits	$\tau_0$, projected diameter $\hat D_2$

Quasi-graded temporal meshes provide a unifying abstraction for time-adaptive discretizations across differential equations, finite element methods, and high-dimensional geometric applications, achieving spatially and temporally optimized resolution, mathematical rigor, and computational tractability (Kumari et al., 2023, Karyofylli et al., 2019, Kopteva et al., 2019, Ma et al., 16 Sep 2025).