Posts ²⁷

Navier–Stokes Existence and Smoothness

The motion of a viscous incompressible fluid is described by the Navier–Stokes equations, first written down by Claude-Louis Navier in 1822 and given their modern form by George Gabriel Stokes. Whether smooth solutions to these equations can always be continued for all time (or whether they can spontaneously develop a singularity at some finite time) is one of the deepest open problems in mathematics, and one of the seven Clay Millennium Prize Problems, carrying a 1,000,000$ prize for a solution.

Navier–Stokes Regularity: The Uniqueness of Weak Solutions

The companion post on Navier–Stokes existence and smoothness asked whether smooth solutions can break down in finite time. This post asks the opposite question: when a solution is only weakly defined, satisfying the equations in an integral sense rather than pointwise, is it uniquely determined by its initial data? The answer, developed over the last two decades through a dramatic series of results, is a resounding no in many regimes. The frontier is now whether the physically natural class of Leray–Hopf weak solutions retains uniqueness.

The Regularity Problem for the 3D Euler Equations

Leonhard Euler wrote down the equations governing the motion of an ideal incompressible fluid in 1757. Whether smooth solutions to these equations can develop a singularity in finite time, a point at which derivatives of the velocity blow up, has been an open problem ever since, and remains one of the central questions in mathematical fluid dynamics. Problem (Euler Regularity) Let $u_0 : \mathbb{R}^3 \to \mathbb{R}^3$ be a smooth, divergence-free initial velocity field with sufficient decay at infinity. Does the unique local smooth solution $u(x,t)$ to the 3D incompressible Euler equations $$\partial_t u + (u \cdot \nabla)u + \nabla p = 0, \qquad \nabla \cdot u = 0, \qquad u(\cdot,0)=u_0$$ remain smooth for all time $t > 0$?

$C^r$ Stability Conjecture

Structural stability is a global topological property: a dynamical system is structurally stable if all nearby systems have the same orbit structure, up to continuous reparametrisation. Hyperbolicity is a local differential property: the tangent bundle over the recurrent set splits into uniformly contracting and expanding directions. That these two conditions should be equivalent is one of the deepest principles in smooth dynamics. Conjecture ($C^r$ Stability Conjecture, Palis–Smale, ~1970) Let $M$ be a closed smooth manifold and $r \geq 1$. If $f \in \mathrm{Diff}^r(M)$ is $C^r$-structurally stable, then $f$ is hyperbolic, i.e., it satisfies Axiom A and the Strong Transversality Condition.

Inequality for Square-Summable Complex Series

Some inequalities look formidable until the right decomposition makes them transparent. The conjecture below, posed by Zoltan Retkes on the Open Problem Garden in 2012 with a £10 prize attached, is one such case: once the dyadic structure of the positive integers is made explicit, the proof reduces to two classical facts. Conjecture (Retkes, 2012), now proved For all $\alpha = (\alpha_1, \alpha_2, \ldots) \in \ell^2(\mathbb{C})$, $$\sum_{n \geq 1} |\alpha_n|^2 \geq \frac{6}{\pi^2} \sum_{k \geq 0} \left|, \sum_{l \geq 0} \frac{\alpha_{2^k(2l+1)}}{l+1} ,\right|^2.$$

Recent Advances in Neural Network Optimization for LLM Training

The optimization landscape for LLM training looks very different from two years ago. AdamW still dominates production runs, but a wave of research is eroding that dominance from multiple angles simultaneously: matrix-aware optimizers, horizon-free schedulers, a sharply revised understanding of µP, and communication-efficient distributed methods. This post synthesizes 18 recent papers across five interconnected fronts. The unifying thread is an active re-examination of long-held assumptions, from whether gradient geometry matters, to what µP is actually doing, to whether weight decay is a regularizer at all.

The Invariant Subspace Problem

Few questions in functional analysis have attracted sustained attention across as many decades as this one. It sits at the confluence of operator theory, spectral theory, and complex analysis, and every partial result has opened new territory rather than narrowing the problem to a routine case. Problem (Invariant Subspace Problem) Does every bounded linear operator $T$ on an infinite-dimensional separable complex Hilbert space $\mathcal{H}$ have a non-trivial closed invariant subspace?

Something Like Picard for 1-Forms

Picard’s great theorem is a statement about how wildly a holomorphic function can behave near an essential singularity. The conjecture below asks whether injectivity of local primitives of a 1-form is enough to rule out such wild behaviour at the origin, forcing the 1-form to extend meromorphically across the puncture. Conjecture (Elsner, 2010) Let $D$ be the open unit disk and let $U_1,\dots,U_n$ be open sets with $\bigcup_{j=1}^n U_j = D\setminus{0}$. Suppose there are injective holomorphic functions $f_j : U_j \to \mathbb{C}$ such that $$\mathrm{d}f_j = \mathrm{d}f_k \quad \text{on every connected component of } U_j \cap U_k.$$ Then the $\mathrm{d}f_j$ glue together to a meromorphic 1-form on $D$.

Criterion for Boundedness of Power Series

Introduction & Problem Statement # Power series constitute one of the most ubiquitous objects in analysis. A power series $\sum_{n=0}^{\infty}a_n x^n$ with infinite radius of convergence defines a real-entire function $f:\mathbb{R}\to\mathbb{R}$. Whereas the question of convergence is completely settled by Cauchy–Hadamard theory, the question of boundedness of the sum function is far subtler and, as of this writing, remains open. Question 1 (Rüdinger, 2009) Let $(a_n) _{n\ge 0}$ be a sequence of real numbers such that the power series $\sum _{n=0}^{\infty}a_n x^n$ converges for every $x\in\mathbb{R}$, thereby defining a smooth function $f:\mathbb{R}\to\mathbb{R}$. Give a necessary and sufficient criterion on $(a_n)$ for $f$ to be bounded on $\mathbb{R}$.

Brezis' first open problem - An elliptic equation involving the critical exponent in 3D

Yamabe problem # Yamabe problem: Suppose $(\mathcal{M}, g_0)$ is a compact closed Riemannian manifold with dimension $N \geq 3$, does there exist a conformal metric $g = u^{\frac{4}{N-2}}g_0$ which has constant scalar curvature $R_g \equiv C$? Find $u > 0$ on $\mathcal{M}$ such that $$ -\frac{4(N-1)}{N-2}\Delta_{g_0}u + R_{g_0}u = Cu^{\frac{N+2}{N-2}}\qquad\text{on }\mathcal{M}. $$ Some results: Trudinger [1968]: if $g$ has non-positive scalar curvature. Aubin [1976]: $N \geq 6$ and $(\mathcal{M}, g)$ not locally conformally flat. Schoen [1984]: any dimension, the remaining cases, assuming the Positive Mass Theorem by Schoen-Yau [1979]. A special case # Consider the special case where $\mathcal{M}$ is a bounded domain $\Omega$ in $\mathbb{R}^{N}$: $$ \begin{cases} -\Delta u = u^{\frac{N+2}{N-2}}\qquad\text{in }\Omega, \\ u > 0\qquad\text{in }\Omega, \\ u = 0\qquad\text{on }\partial\Omega. \end{cases} $$

Recent Advances in KAN-Based Numerical PDE Solvers

Kolmogorov-Arnold Networks (KANs), introduced in 2024, have rapidly become one of the most active frontiers in scientific machine learning for solving partial differential equations (PDEs) (Liu et al., 2024). Unlike Multi-Layer Perceptrons (MLPs), which apply fixed activation functions at nodes, KANs place learnable univariate activation functions on edges, grounded in the Kolmogorov-Arnold representation theorem: every continuous multivariate function can be expressed as a composition of univariate functions and summations. This structural difference gives KANs two key properties relevant to PDE numerics — higher interpretability and parameter efficiency — making them an appealing successor to MLP-based Physics-Informed Neural Networks (PINNs).

Recent Advances in Numerical PDEs

Numerical methods for partial differential equations (PDEs) have entered a period of rapid transformation, driven by two converging forces: deep learning’s maturation as a tool for high-dimensional function approximation, and the resurgence of classical methods augmented by machine learning. The field broadly divides into physics-informed machine learning, neural operator learning, foundation models for PDEs, and the continuing evolution of classical high-order, structure-preserving, and data-driven discovery methods. Quantum computing and laser-based hardware solvers are also beginning to enter the landscape. This survey organises the most active research fronts, highlights landmark and recent key papers, and identifies open problems as of early 2026.

Recent Advances in Steady States of Navier-Stokes Equations

The study of steady-state and self-similar solutions of the incompressible Navier-Stokes equations (NSE) has undergone remarkable progress in the 2020s. This post surveys landmark results from 2024–2026 touching on existence, uniqueness, classification, and stability of such solutions. The stationary (steady) NSE in $\mathbb{R}^3$ reads: $$-\nu \Delta u + (u \cdot \nabla) u + \nabla p = 0, \quad \operatorname{div} u = 0.$$ A central object of the self-similar theory is the class of $(-1)$-homogeneous (scale-invariant) solutions: a function $u$ is $(-1)$-homogeneous if $u(\lambda x) = \lambda^{-1} u(x)$ for all $\lambda > 0$. These are precisely the profiles of forward self-similar solutions $u(x,t) = t^{-1/2} U(x/\sqrt{t})$ of the time-dependent NSE.

Recent Research Directions in Analysis of PDEs 2021–2026

The arXiv section of Analysis of Partial Differential Equations is one of the most prolific areas of pure mathematics, producing over 400 preprints per month as of early 2026. The period 2021–2026 has witnessed landmark breakthroughs — including a computer-assisted proof of finite-time singularity in the 3D Euler equations, the resolution of Hilbert’s Sixth Problem via kinetic theory, and the emergence of probabilistic and nonlocal operator methods as dominant paradigms. This survey identifies, categorises, and profiles the key research directions and landmark papers in math.AP during this era.

Paper Reading - Optimization problems for elliptic PDEs (2601.01591)

This paper is a panoramic tour of three families of optimal control problems for elliptic PDEs: where the control is the coefficient, the potential, or the source term, unifying and sharpening results from the authors’ previous works. Three ways to control an elliptic PDE # The authors always consider a Dirichlet problem on a bounded domain $\Omega \subset \mathbb{R}^d$, with the solution $u$ as the state and a function (or measure) as the control. They study three settings:

Paper Reading - Optimal coefficients for elliptic PDEs (2512.08431)

This paper gives a clear, fairly complete picture of how to optimally choose the coefficient $a(x)$ (think “material quality”) in an elliptic PDE, with compliance as the main model and then a general optimal control formulation. Problem Setup # Considering the boundary value problem: $$ -{\rm div}(a(x)\nabla u) = f \quad\text{in } \Omega,\qquad u=0 \text{ on } \partial\Omega, $$ where $\Omega$ is a bounded domain, $f$ is a given load, and $a(x)$ is the design variable.

Paper Reading - Optimal sources for elliptic PDEs (2509.01521)

Introduction # The authors study how to “best choose” a source term $f$ in a Poisson-type equation $$ -\Delta u = f \quad\quad\text{in }\Omega,\quad u = 0\text{ on }\partial\Omega, $$ so that a given performance measure (a cost functional) is optimized. The twist is that the source itself is the control, and it can be subject to various constraints (size, bounds, sign, etc.). This makes the problem sit at the intersection of optimal control, shape optimization, and regularity theory.

Restriction and extension

Considering a smooth compact hyper-surface $\mathcal{S}$ in $\mathbb{R}^d$ with surface measure $d\sigma$. Given $f \in L^1(\mathbb{R}^d)$, the Fourier transform defined as follow: $$ \begin{equation} \hat{f}(x) = \int_{\mathbb{R}^d}e^{-2\pi i x \xi}f(x)dx \end{equation} $$ which by Riemann-Lebesgue is a bounded, continuous function vanishing at infinity. Since $\hat{f}$ is continuous on $\mathbb{R}^d$, by the Rimann-Lesbegue lemma its restriction to the compact hyper-surface $S \subset \mathbb{R}^d$ is is well-defined pointwise. Specifically, the restriction $\hat{f}\mid_{S}: S \rightarrow \mathbb{C}$ is the continuous function given by $$ \begin{equation} \hat{f}\mid_{S}(\sigma) = \hat{f}(\sigma) = \int_{\mathbb{R}^d}e^{-2\pi i x \xi}f(x)dx \end{equation} $$ for each $\sigma \in S$. This is bounded (as $\hat{f}$ is bounded) and can be integrated against the surface measure $d\sigma$ on $S$.

Proof of Theorem of solution of wave equation in the case $n = 1$

Solution of Brezis Problem 8.24 (1) and (2)

Solution of Evans PDE Problem 13

A lemma of J. L. Lions

This post explores J. L. Lions’ lemma about Banach spaces with compact injection, including applications to functional analysis. Lemma statement: Let $X$, $Y$, and $Z$ be three Banach spaces with norms $|| \cdot ||_X$, $|| \cdot ||_Y$, and $|| \cdot ||_Z$. Assume that $X \subset Y$ with compact injection and that $Y \subset Z$ with continuous injection. Prove that $$ \forall \varepsilon > 0, \exists C_\varepsilon > 0 \text{ satisfying } || u ||_Y \leq \varepsilon || u ||_X + C _{\varepsilon}|| u ||_Z,\quad \forall u \in X $$

Complex Hahn-Banach Theorem

Let $X$ be a complex vector space, $X_0$ one of its subspaces, $p: X \to \mathbb{R}_+$ such that $$ p(\lambda x) = |\lambda| p(x), \quad \forall \lambda \in \mathbb{C}, x \in X \text{ and } p(x + y) \leq p(x) + p(y), \quad \forall x, y \in X, $$ satisfying $|f(x)| \leq p(x)$, $\forall x \in X_0$, where $f: X_0 \to \mathbb{C}$ is linear. Under these conditions, there exists a linear functional $F: X \to \mathbb{C}$ such that $F|_{X_0} = f$ and

Real Hahn-Banach Theorem

Suppose $X$ is a vector space over $\mathbb{R}$, $p: X \to \mathbb{R}$ has the following properties: $p(X) = \lambda p(x)$, $\forall x \in X$, $\lambda \in \mathbb{R}_+$ and $p(x + y) \leq p(x) + p(y)$, $\forall x, y \in X$. Let $X_0$ be a subspace of $X$ and $u: X_0 \to \mathbb{R}$ a linear functional such that $u(x) \leq p(x)$, $\forall x \in X_0$. Then we can find $f: X \to \mathbb{R}$ a linear functional such that $f|_{X_0} = u$ and $f(x) \leq u(x)$, $\forall x \in X$.

Riesz Representation Theorem

1. Riesz Representation Theorem # Let $H$ be a Hilbert space over $\mathbb{R}$ or $\mathbb{C}$, and $T$ be a bounded linear functional on $H$ (a bounded operator from $H$ to the field $\mathbb{R}$ or $\mathbb{C}$, where $H$ is defined over that field). The following is known as the Riesz Representation Theorem: Theorem 1: If $T$ is a bounded linear functional on the Hilbert space $H$, then there exists $g \in H$ such that for every $f \in H$, we have: $$ T(f) = \langle f, g \rangle. $$

The application of Hahn-Banach Theorem 01

Suppose $X$ is a normed space and $X_0$ is a closed subspace of $X$ and $x_0 \in X \setminus X_0$. Then we can find $f \in X’$ such that $f(x_0) = 1$ and $f(x) = 0$, $\forall x \in X_0$. Proof: Since $x_0 \notin X_0$, we can find $\delta > 0$ such that $|x_0 - x| \geq \delta$, $\forall x \in X_0$, which is equivalent to $1 \leq \dfrac{|x_0 - x|}{\delta}$, $\forall x \in X_0$.

The application of Hahn-Banach Theorem 02

$X'$ = $\{ f: X \to \mathbb{K} \}$ where $f$ is is linear and continuous and $X$ is a Banach space over $\mathbb{K}$. Prove that $X' \neq {0}$, in fact, for every $x \neq 0 \in X$, we can find $f \in X’$ such that $f(x) = |x|$ and $|f| = 1$. Proof: Pick $x_0 \in X$. Define $X_0 = x_0 \cdot \mathbb{K}$, a subspace of $X$, and $g: X_0 \to \mathbb{K}$, $g(x) = x$, which is linear. Since $g$ and $|\cdot|$ satisfy the conditions of the Hahn-Banach theorem, we can find $f: X \to \mathbb{K}$ such that $f|_{X_0} = g$, $f$ is linear and $f(x) \leq |x|$, $\forall x \in X$. Therefore $f(x_0) = g(x_0) = |x_0|$ and $|f| \leq 1$. The equality $f(x_0) = |x_0|$ guarantees that $|f| = 1$.