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 I: Avi Wigderson, “The power and weakness of randomness in co」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

The Distinguished Lecture Series at UCLA for this winter quarter is given by Avi Wigderson , who is lecturing on “ some topics in computational complexity “. In his first lecture on Wednesday, Avi gave a wonderful talk (in his inimitably entertaining style) on “ The power and weakness of randomness in computation “. The talk was based on these slides . He also gave a sort of “encore” on zero-knowledge proofs in more informal discussions after the main talk.

As always, any errors here are due to my transcription and interpretation.

已知结果和反例

The theme of this talk was the close relationship between hardness (computational complexity) and randomness (and specifically, the use of randomness to save on resources such as time or memory).

Avi began with some examples to illustrate “easy” and “hard” computational problems. For instance:

证明或构造的主线

It is not known whether factoring is truly hard, or whether theorem-proving is truly hard. But there is a known reduction: if theorem-proving is easy (in the sense that it is a polynomial time algorithm ), then factoring is also easy. This is because theorem-proving is an NP-complete problem (this follows from the Cook-Levin theorem ).

The famous problem is thus formally equivalent to the statement that theorem proving is not easy. At a more philosophical level, one can view as an assertion that “creativity cannot be automated”.

阅读时建议盯住的点

There are many other NP-complete problems known in mathematics and science (indeed, Avi noted that such problems appear to be “uniformly distributed” throughout these disciplines). We thus have the following informal

Belief 1 : “natural” problems such as factoring, theorem-proving, 3-colourability of graphs , computation of the permanent , etc. are hard.

值得单独记下的条目

• Multiplication of two n-digit numbers is easy; the long multiplication algorithm one learns in primary school lets one do this in steps (and faster algorithms are also known ); but
• Completeness . If the prover does indeed have a proof of S, then V should return “true”.
• Soundness . If the prover does not have a proof of S, then V should return “false”.
• Otherwise, return to Step 1 and start over. If the algorithm does not return false in (say) iterations of this process, return “true”.

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