Total variation distance of probability measures explained

Total variation distance of probability measures should not be confused with Total variation.

In probability theory, the total variation distance is a distance measure for probability distributions. It is an example of a statistical distance metric, and is sometimes called the statistical distance, statistical difference or variational distance.

Definition

(\Omega, l{F}) and probability measures P and Q defined on (\Omega, l{F}) .The total variation distance between P and Q is defined as^[1] \delta(P,Q) = \sup }\left|P(A)-Q(A)\right|.This is the largest absolute difference between the probabilities that the two probability distributions assign to the same event.

Properties

The total variation distance is an f-divergence and an integral probability metric.

Relation to other distances

The total variation distance is related to the Kullback–Leibler divergence by Pinsker’s inequality:

\delta(P,Q) \le \sqrt{ (1)/(2) D _{KL} (P\parallel Q)}.

One also has the following inequality, due to Bretagnolle and Huber^[2] (see also ^[3]), which has the advantage of providing a non-vacuous bound even when

style D _{KL} (P\parallel Q)> 2 \colon \delta(P,Q) \le \sqrt{ 1- e ^{- D _{KL} (P\parallel Q)} }.

The total variation distance is half of the L¹ distance between the probability functions:on discrete domains, this is the distance between the probability mass functions^[4]

\delta(P, Q) = (12)/(\sum) _{x} |P(x) - Q(x)|, and when the distributions have standard probability density functions and,^[5] \delta(P, Q) = (12)/(\int) | p(x) - q(x) | d x (or the analogous distance between Radon-Nikodym derivatives with any common dominating measure). This result can be shown by noticing that the supremum in the definition is achieved exactly at the set where one distribution dominates the other.^[6] H(P,Q) as follows:^[7] H ^{2(P,Q)} \leq \delta(P,Q) \leq \sqrt 2 H(P,Q). These inequalities follow immediately from the inequalities between the 1-norm and the 2-norm.

The total variation distance (or half the norm) arises as the optimal transportation cost, when the cost function is

c(x,y) = { 1 }_, that is,(1)/(2) \| P - Q \| _{1} = \delta(P,Q) = inf \ { P (X ≠ Y ) : Law (X) = P , Law (Y) = Q \ } = inf _{\pi} \operatorname{E} _{\pi} [{ 1 }_],where the expectation is taken with respect to the probability measure \pi on the space where (x,y) lives, and the infimum is taken over all such \pi with marginals P and Q , respectively.^[8]

See also

• Total variation
• Kolmogorov–Smirnov test
• Wasserstein metric

Notes and References

1. Web site: Chatterjee . Sourav . Distances between probability measures . UC Berkeley . 21 June 2013 . dead . https://web.archive.org/web/20080708205758/http://www.stat.berkeley.edu/%7Esourav/Lecture2.pdf . July 8, 2008 .
2. Bretagnolle, J.; Huber, C, Estimation des densités: risque minimax, Séminaire de Probabilités, XII (Univ. Strasbourg, Strasbourg, 1976/1977), pp. 342–363, Lecture Notes in Math., 649, Springer, Berlin, 1978, Lemma 2.1 (French).
3. Tsybakov, Alexandre B., Introduction to nonparametric estimation, Revised and extended from the 2004 French original. Translated by Vladimir Zaiats. Springer Series in Statistics. Springer, New York, 2009. xii+214 pp., Equation 2.25.
4. David A. Levin, Yuval Peres, Elizabeth L. Wilmer, Markov Chains and Mixing Times, 2nd. rev. ed. (AMS, 2017), Proposition 4.2, p. 48.
5. Book: Tsybakov . Aleksandr B. . Introduction to nonparametric estimation . 2009 . Springer . New York, NY . 978-0-387-79051-0 . rev. and extended version of the French Book . Lemma 2.1.
6. Book: Devroye, Luc . A Probabilistic Theory of Pattern Recognition . Györfi . Laszlo . Lugosi . Gabor . 1996-04-04 . Springer . 978-0-387-94618-4 . Corrected . New York . en.
7. Web site: Lecture notes on communication complexity . September 23, 2011 . Prahladh . Harsha .
8. Book: Villani, Cédric. Optimal Transport, Old and New. 338. Springer-Verlag Berlin Heidelberg. 2009. 978-3-540-71049-3. 10. en. 10.1007/978-3-540-71050-9. Grundlehren der mathematischen Wissenschaften.