Research Visions

From current results
to the next mathematical questions.

The first layer is a compact map of the research programme. Each card then opens the full detailed Vision, so the new design improves navigation without discarding the mathematical development already written.

Core programme

The work to build first.

Vision 10, 11, 17, 18, 20 and 21 form the present Core, with geometry, unresolved dynamics, exact-curved analysis and certification developed in parallel but with explicit dependencies.

Core G1 · Geometry / CR–RT

Anisotropic CR–RT, discrete maximum principles and maximum-norm analysis

Separate element geometry, metric-angle monotonicity, patch connectivity and discrete Green-function effects rather than forcing them into one mesh condition.

Vision 20
Read the anisotropic FEM entry route →
Core G2 · 3D de Rham

3D Nédélec interpolation to anisotropic de Rham, Hodge and Maxwell theory

Build quantitative Nédélec stability together with existing RT estimates, then move to commuting projections, discrete Poincaré inequalities, Hodge decomposition and Maxwell certification.

Vision 21
Core M · Unresolved dynamics

Stable reduction theory and structure-preserving memory computation

Vision 18 asks how history operators arise from unresolved dynamics; Vision 11 asks how a given memory or delay operator should be discretised without destroying its structure.

Visions 11 & 18
Read the stable-reduction entry route →
Core Q · Exact-curved geometry

Exact-curved CR–RT, shape differentiation, ALE and FSI

Keep geometry error distinct from PDE and discretisation error as exact-curved finite elements are extended to moving domains and certified interface outputs.

Vision 10
Read the Exact-curved FEM entry route →
Cross-cutting V · Certification

Certified output, formal reasoning and AI4Math-ready numerical analysis

Connect equilibrated fluxes, DWR, pointwise certificates, geometric conditions and model-reduction errors into reusable and verifiable mathematical certificates.

Vision 17
Current research order

Three lanes, not one queue.

The geometry/FEM, memory/unresolved-dynamics and exact-curved programmes remain distinct research lanes.

Geometry & FEM

CR-DMP + anisotropic L∞→3D Nédélec→de Rham / Hodge→Maxwell / certified bounds

Memory & unresolved dynamics

continuous foundations→BDF–CQ→anisotropic CR–RT→stable reduction→identification + certification

Exact-curved geometry

curved Lagrange→curved CR–RT→shape differentiation→ALE / FSI→certified interfaces
Research-line tags
G · Geometry & Anisotropic FEM TheoryR · Reliable & Goal-Oriented PDE SimulationA · Automation, Learning & New Numerical MethodsV · Verification, Formal Reasoning & AI4Math
Vision map

Scan first. Read deeply when needed.

Open “Read full vision” to load the detailed text, equations, research questions and related-Vision links.

Vision 1

Geometric conditions on spheres and manifolds

Extend finite-element geometric conditions from Euclidean simplices to spherical and manifold settings.

G
Read full vision
Vision 2

Deep-learning numerical methods and geometric conditions

Relate sampling geometry in PINNs and Deep Ritz methods to geometric conditions familiar from FEM.

AG
Read full vision
Vision 3

Polygonal elements and star-shaped conditions

Revisit polygonal and VEM geometry while allowing stronger anisotropy.

G
Read full vision
Vision 4

Automation of numerical computation with machine learning

Use geometric and error information to support adaptation, mesh generation and method selection.

AR
Read full vision
Vision 5

Abstraction and automation of mathematical structure

Extract reusable mathematical patterns behind stability and approximation across numerical schemes.

AG
Read full vision
Vision 6

Evolution problems and pressure robustness

Connect time integration, pressure robustness, anisotropy and long-time stability.

RG
Read full vision
Vision 7

Stochastic PDEs and geometric conditions

Study geometry, adaptivity and reliability when PDE data or coefficients are random.

GA
Read full vision
Vision 7b

Reliability of inference with uncertain or identified coefficients

Separate data perturbation and numerical-discretisation error in inference outputs.

GRA
Read full vision
Vision 8

Richardson extrapolation, geometry and learned solvers

Reconsider extrapolation across classical FEM and deep-learning numerical methods.

RA
Read full vision
Vision 9

Defect collection across numerical methods

Use residual and defect structures as a common language for classical and newer schemes.

A
Read full vision
Vision 10 · Core

Exact-curved geometry programme

Curved CR–RT, shape differentiation, ALE, FSI and certified interface quantities.

GRV
Read full vision
Vision 11 · Core

Structure-preserving numerical analysis for evolution equations with memory and delay

Given a history operator, preserve its graph-energy or measure structure in full discretisation.

RGV
Read full vision
Vision 12

Extreme weather and disaster models

Long-term expansion towards multiscale, nonlinear and strongly anisotropic real-world systems.

RA
Read full vision
Vision 13

Anisotropic PDEs and numerical analysis

Extend anisotropic-mesh theory to equations whose operators are themselves direction dependent.

G
Read full vision
Vision 14

Metric-geometric redesign of anisotropic FEM

Reformulate geometry and error analysis in a metric-invariant language with space–time extensions.

GR
Read full vision
Vision 15

Representation · metric · projection

Build a common conceptual map for FEM, meshless methods, deep learning and information geometry.

AG
Read full vision
Vision 16

VEM theory permitting anisotropic meshes

Move beyond conventional star-shaped and shape-regular assumptions for polygonal elements.

G
Read full vision
Vision 17 · Core

Certified FEM in the AI4Math era

Organise geometric assumptions, error bounds, flux certificates and proof structure for verifiable computation.

VRGA
Read full vision
Vision 18 · Core

Stable reduction theory for unresolved dynamics

Unresolved dynamics → history operator → stable reduced model → computation → identification → certification.

RAV
Read full vision
Vision 19 · Expansion

Dynamic progression forecasting with treatment history and latent states

Use history-dependent inference and certified spatio-temporal prediction for disease-progression applications.

RAV
Read full vision
Vision 20 · Core

Discrete maximum principles and maximum-norm error estimates for anisotropic CR–RT

Separate monotonicity, pointwise accuracy, local geometry and global connectivity.

GRV
Read full vision
Vision 21 · Core

3D Nédélec interpolation to anisotropic de Rham, Hodge and Maxwell theory

Build the low-order three-dimensional complex first, then move toward compactness and certification.

GRV
Read full vision