陶哲轩博客写数学问题时，通常先把对象定义清楚，再给直觉、反例和证明轮廓。把「Distinguished Lecture Series III: Charles Fefferman, “Interpolation of functions on R^n”」改写成可阅读的中文笔记，重点是：问题在问什么、已知到哪一步、下一步最容易走偏在哪。原站广告、分享条和导航已去掉。

问题在问什么

Today, Charlie wrapped up several loose ends in his lectures, including the connection with the classical Whitney extension theorem , the role of convex bodies and Whitney convexity, and a glimpse as to how one obtains the remarkably fast (almost linear time) algorithms in which one actually computes interpolation of functions from finite amounts of data.

With all the notational machinery from the preceding lecture, the Whitney extension theorem can be stated rather cleanly in the language of bundles, sections, and Glaeser-stability as follows. Consider a bundle on a compact set in which every fibre is simply a point: . The Whitney extension problem is then to see whether such bundles have sections. The theorem is then:

已知结果和反例

Whitney extension theorem . Let be a bundle in which every fibre is a point. Then this bundle has sections if and only if it is Glaeser stable.

It is then clear that the main theorem of the previous lecture generalises the Whitney extension theorem (which represents the case of exactly one stratum, consisting entirely of zero-dimensional fibres). It is then unsurprising that the main tools used to prove the latter theorem (Whitney cubes, smooth partitions of unity, etc.) also come up in the proof of the former. It is also worth noting that Whitney’s proof shows that the map from the polynomials to the section F can b

证明或构造的主线

In the proof of the Whitney extension theorem, it was necessary to smoothly partition space into cubes, perform some sort of extension on each cube, and then sum everything up. In order to guarantee convergence, it is necessary that the bounds on the extension are (a) uniform in some sense over all cubes, and (b) are stable under multiplication by smooth cutoff functions. It turns out that we can abstract these properties by means of a quantitative generalisation of the notio

We can generalise the notion of an ideal as follows. If and $A > 0$, we say that a set is Whitney convex with Whitney constant A if it is a closed convex symmetric set, and if for every we have the ideal-like property

阅读时建议盯住的点

The point if this rather technical definition is that if a jet is lying in $\sigma(x)$ and it has controlled behaviour at some small length scale , then one can smoothly localise to that scale without causing the jet to escape too much. This is the key property needed for Whitney-type arguments to work. (This is why Whitney-type extension problems are tractable, whereas general linear PDE problems are not; one can glue together local extensions to form a global extension via

There was also a slightly weaker notion of Whitney convexity relating to a modulus of continuity , called Whitney -convexity, which was the same except that the polynomial P had to now obey the bounds instead of .

阅读和落地时建议先做的 5 件事

1. 用自己的语言重写定义和结论，不看原文能不能说清对象是什么。
2. 找一个最小反例或边界情形，确认假设少一条会怎样。
3. 把证明拆成可独立检验的引理，每步只保留一个新想法。
4. 若涉及计算或形式化，先写可复现的小例子，再谈一般情形。
5. 记下尚未解决的缺口：缺估计、缺构造，还是缺正确的范畴。

和智能体、形式化工具怎么接

龙虾PRO做 OpenClaw 落地时，数学笔记最有用的部分往往是「可检验的步骤」：定义、反例、引理边界。智能体适合帮忙展开计算和检索，不适合代替你决定哪条假设能扔。

本文侧重全链路风控方法论。落地时请用自身业务单据做回放验证，不要把示例阈值直接当生产策略。 相关：风控体检 · 方案资源