Sigma algebras, probability spaces, measure theory

The scaffolding of randomness

June 21, 2015 — June 21, 2015

algebra
functional analysis
networks
probability
time series

The most popular axiomatic foundation for classical commutative probability, i.e. my bread and butter.

I learned about sigma algebras and probability spaces and all the scaffolding of modern probability in the context of financial mathematics, where the emphasis is on proving things about certain pathologies of probability which obtain in certain pathological limits of uncountable what-have-yous, in order to demonstrate that you are clever enough to get good grades in order that you can ignore it for the rest of your career as a financial trader trading in discrete time where none of those theorems matter.

BUT! There is interesting stuff here. Not just pathologies of measure theory, but also the interaction between measure theory and the Kolmogorov axioms that give us actual probability, and there are real problems that can arise in actual use. 🏗 They do not often explain why we bother with this construction here, but there are practical consequences if we ignore it. See Djalil Chafaï’s explanation for a couple of great examples. Some subtle interactions arise in probability metrics.

Interesting mental exercises here: conditionalising corresponds with differentiation Ben Green points out. (See also)

Hopefully one day I will have time to return to this with some more profound insights, but… later.

1 References

Campbell, Syed, Yang, et al. 2019. arXiv:1906.09507 [Math, Stat].
Caves, Fuchs, and Schack. 2002. Journal of Mathematical Physics.
Cover, Gács, and Gray. 1989. The Annals of Probability.
Diaconis, and Freedman. 1980a. The Annals of Probability.
———. 1980b. The Annals of Probability.
Heunen, Kammar, Staton, et al. 2017. arXiv:1701.02547 [Cs, Math].
Orbanz, and Roy. 2015. IEEE Transactions on Pattern Analysis and Machine Intelligence.
Radul, and Alexeev. 2020. arXiv:2010.09647 [Cs].