Numerical Computation and Mathematical Software

Numerical computation and mathematical software form the backbone of modern science and engineering by providing robust tools to approximate, simulate and analyse complex systems. Core components include algorithms for solving linear and nonlinear equations, eigenvalue problems and optimisation tasks, together with discretisation methods for differential and integral equations such as finite differences, finite volumes and finite elements. These numerical kernels are encapsulated in mature software libraries and frameworks that exploit parallelism—from multicore processors to graphics accelerators and large clusters—while managing issues of accuracy, stability and performance portability. Computer algebra systems extend these capabilities by supporting symbolic manipulation, exact arithmetic and analytic simplification. High-order methods, adaptive mesh refinement and multigrid techniques accelerate convergence, and the integration of machine-learning components is widening the scope of algorithmic adaptivity. Meanwhile, software engineering advances—domain-specific languages, performance-portable programming models and automated code generation—ensure that mathematical software remains efficient across ever-evolving hardware architectures. Collectively, these developments empower researchers to tackle large-scale simulations, real-time data assimilation and optimisation problems that were previously infeasible.

Research from Nature Portfolio

A new time-consistent stabilisation in finite element methods addresses the loss of accuracy at small time steps in incompressible flow simulations. By replacing the conventional time-step-dependent stabilisation parameter with a physics-based time scale derived from the ratio of acceleration and velocity norms, the method maintains optimal accuracy in cardiovascular flow and fluid–structure interaction problems, even under highly refined temporal discretisation.

Researchers have demonstrated that low-precision fixed-point processors, when paired with residual iteration in a simple Richardson scheme, can achieve solver convergence and solution precision beyond the native format. This approach opens the door to energy-efficient analog and digital hardware for large-scale linear inverse problems without compromising numerical fidelity.

A spurious oscillation reduction strategy for transient diffusion and wave equations uses analytical solutions of simple test problems to precondition finite element discretisations. This method eliminates non-physical oscillations in problems with impulsive point sources, enabling accurate wave and heat propagation simulations without resorting to regularisation of singular source terms.

Numerical Computation and Mathematical Software publication trend

The graph below shows the total number of articles in numerical computation and mathematical software across all publications each year (not limited to Nature Index journals).

Technical terms

Finite Element Method (FEM): A variational discretisation that subdivides a domain into elements and approximates solutions by piecewise polynomial basis functions.

Residual iteration: A corrective scheme that refines an approximate solution by iteratively solving for the error in the residual, often used to regain precision lost in low-order arithmetic.

Adaptive Mesh Refinement (AMR): A technique that dynamically refines or coarsens a computational mesh in regions indicated by error estimators to concentrate resolution where it is most needed.

Performance portability: The capacity of software to deliver efficient execution across multiple hardware architectures from a single code base, typically via abstract parallelism models.

Unit round-off (machine epsilon): The smallest relative difference between two distinct floating-point numbers in a given format, governing the bound on rounding errors.

References

1. A time-consistent stabilized finite element method for fluids with applications to hemodynamics. Scientific Reports (2023).
2. Fixed-point iterative linear inverse solver with extended precision. Scientific Reports (2023).
3. Spurious oscillations reduction in transient diffusion and wave propagation problems discretized with the Finite Element Method. Scientific Reports (2022).
4. Kokkos 3: Programming Model Extensions for the Exascale Era. IEEE Transactions on Parallel and Distributed Systems (2021).
5. Floating-point arithmetic. Acta Numerica (2023).