Numerical Analysis · Partial Differential Equations

ISHIZAKA Hiroki

石坂 宏樹 · Independent Researcher

Mathematics makes the invisible visible.

数学は見えないものを可視化する

I study how mesh geometry and unresolved dynamics affect the accuracy and stability of computation, and how mathematical analysis can make those effects visible.

Research update · 20 November 2026 The journal article page is now available for Weighted well-posedness and kernel stability for coercive evolution equations with measure-valued delays in Evolution Equations and Control Theory (DOI: 10.3934/eect.2026.100844). Low-order CR–RT equilibrated-flux certification for semilinear problems on anisotropic meshes was also accepted for publication in the Journal of Computational and Applied Mathematics (12 November 2026).

Mathematics for Trustworthy Simulation

AIの計算能力が向上するほど「計算できること」と「保証できること」の間にある数学的問題が鮮明になる

This research vision is developed through two complementary programmes: Certified finite element analysis and Stable reduction of unresolved dynamics. The first asks when a discretised computation can be trusted; the second asks when a reduced model can be trusted after dynamically relevant information has been eliminated.

Mathematical foundations of certified finite element analysis

My research goes beyond convergence estimates to investigate the mathematical conditions under which finite element approximations remain reliable. I study how mesh geometry, approximation spaces and the structure of partial differential equations interact to determine accuracy, stability and qualitative properties of numerical solutions.

Starting from computable geometric conditions and anisotropic interpolation theory, my work extends to nonconforming and mixed finite elements, discontinuous Galerkin methods, discrete functional inequalities, pressure robustness and the preservation of mathematical properties such as conservation and monotonicity.

Aim
To develop a mathematical theory that identifies which assumptions are essential, which can be relaxed, and where numerical guarantees break down.

My long-term goal is to connect geometric analysis, structure-preserving discretisation, reconstruction techniques and rigorous error certification. Rather than relying solely on convergence rates, I seek to establish when a numerical approximation can be trusted, what information may have been lost through discretisation, and how its accuracy can be verified through computable bounds.

Anisotropic FEM · Mesh geometry · CR / RT · Morley · Nitsche / DG · Stokes / Navier–Stokes · Equilibrated flux
FEM research overview →

Stable reduction of unresolved dynamics未解像ダイナミクスの安定縮約理論

My research asks what remains of a dynamical system when unresolved degrees of freedom are eliminated. A reduced model may reproduce observations yet lose stability, dissipation or long-time behaviour. I ask when its predictions remain mathematically justified.

I analyse discarded states using operator-valued memory, delay terms and nonlocal closures, alongside coercivity, appropriate solution spaces and quantitative stability estimates. These are tools for tracking information loss, identifying mechanisms of stability or instability, and measuring the difference between full and reduced dynamics.

Aim
To establish mathematical criteria for deciding when a reduction is admissible, when it must be rejected, and when the available information is insufficient for certification.

My long-term goal is a theory beyond any particular memory kernel or evolution model, with possible applications to multiscale diffusion, fluid instabilities and other high-dimensional systems. The central question is whether a finite set of retained variables can capture the influence of eliminated degrees of freedom, and how to quantify the limits of that description through rigorous, computable bounds.

Unresolved dynamics · Information loss · Stability boundaries · Reduction criteria · Memory and delay · Coercivity · Certified approximation
Research overview →

A shared question

What remains stable, accurate and certifiable after information has been discretised, eliminated or compressed?

Goal

複雑なPDEの自由度を削減したとき,どの情報が失われ,その情報損失が安定性・不安定性・長時間予測にどう影響するかを定量化する.さらに,縮約の妥当性を数学的・数値的に判定できる理論を確立する.

AI4Math &
Mathematical Assurance

Mathematics for reliable scientific AI.

I am exploring mathematical foundations for reliable AI for Science. The aim is to quantify information lost through learning, model reduction and numerical discretisation, and to assess whether scientific predictions can be trusted.

Current directions connect certified PDE computation, structure-preserving learning of unresolved dynamics, memory and reduced models, and self-correcting scientific computation supported by computable error bounds.

AI4Math · Mathematical assurance · Certified scientific computing · Unresolved dynamics · Structure-preserving learning

Research direction →

Selected publications

Selected peer-reviewed work in finite element analysis and evolution equations.

Public manuscripts

Current public manuscripts [S2]–[S4] and [S6]–[S11], with links to arXiv. Former [S1] and [S5] are now listed under peer-reviewed publications.

[S11] · 2026 ·

Information Limits for Stable Reduction of Coupled Diffusion Systems

arXiv 2611.02977 →
[S10] · 2026 ·

Inverse monotonicity of the Crouzeix--Raviart method beyond local \(Z\)-matrix conditions

arXiv →
[S9] · 2026 · Memory / delay

Sharp Wasserstein stability and small-delay thresholds for evolution equations with measure-valued principal-part delays

arXiv 2610.24635 →
[S8] · 2026 · Spherical FEM

Anisotropic Lagrange interpolation on exact spherical triangles

arXiv 2610.09686 →
[S7] · 2026 · Anisotropic FEM

Crouzeix–Raviart–Marini realisation of computable a priori L²-error bounds on anisotropic meshes

arXiv 2609.06057 →
[S6] · 2026 · Spherical FEM

Exact characterisation of maximum-angle conditions for spherical finite element meshes

arXiv 2608.11353 →
[S4] · 2026 · Degenerate diffusion

Graph-space well-posedness for diffusion equations with degenerate instantaneous diffusion

Stable functional setting for degenerate history-dependent evolution.

arXiv 2607.12871 →
[S3] · 2026 · Memory stability

High-frequency coercivity loss for completely monotone memory on bounded time intervals

arXiv 2607.12482 →
[S2] · 2026 · Exact-curved FEM

Exact-curved Lagrange finite elements for the Poisson problem in two dimensions

arXiv 2605.26566 →

About

I am an independent researcher in numerical analysis. My work spans finite element analysis, partial differential equations and mathematical certification.

I am particularly interested in mathematical structures that computation or model reduction can leave unseen: discretisation error, mesh geometry, unresolved scales, memory and latent states.

Professional activities

Independent Researcher — Numerical Analysis, Partial Differential Equations, and Finite Element Methods
AI Evaluator — Mathematics / Reasoning / Numerical Methods
Scientific-Computing SME — PDEs & FEMs
Languages — JP / EN
Memberships — Mathematical Society of Japan · Japan Society for Industrial and Applied Mathematics
Name
Hiroki Ishizaka, Ph.D. / 石坂 宏樹
Affiliation
Independent Researcher · Team FEM · Matsuyama, Japan
Email
h.ishizaka005 [at] gmail.com