Trace distance between two probability distributions - prove

In summary, the trace distance between two probability distributions is the sum of the absolute values of the differences between the probabilities of each element in the subset.
  • #1
Emil_M
46
2

Homework Statement


Let ##\{p_x\}## and ##\{q_x\}## be two probability distributions over the same index set ##\{x\}={1,2,...,N}##. Then, the trace distance between them is given by ##D(p_x,q_x):=\frac{1}{2} \sum_x |p_x-q_x|##.

Prove that ##D(p_x,q,_x)=max_S |p(S)-q(S)|=max_S | \sum_{x \in S} p_x - \sum_{x \in S} q_x|##, where the maximization is over all subsets ##S## of the index set ##\{x\}##.

Homework Equations


See above

The Attempt at a Solution


[itex]
\begin{align*}
\frac{1}{2} \sum_x |p_x-q_x| &\geq | \sum_{x \in S} p_x - \sum_{x \in S} q_x|\\
&= |\sum_{x \in S} p_x-q_x|
\end{align*}
[/itex]

Then, I have tried playing around with the triangle inequality, but that didn't go anywhere...

Thanks for you help!
 
Last edited:
Physics news on Phys.org
  • #2
The key piece of information here is that all the probabilities will sum to 1.
If p(1) - q(1) = z, then the sum of q(2) to q(N) minus the sum of p(2) to p(N) will also be z.
 
  • #3
Emil_M said:

Homework Statement


Let ##\{p_x\}## and ##\{q_x\}## be two probability distributions over the same index set ##\{x\}={1,2,...,N}##. Then, the trace distance between them is given by ##D(p_x,q_x):=\frac{1}{2} \sum_x |p_x-q_x|##.

Prove that ##D(p_x,q,_x)=max_S |p(S)-q(S)|=max_S | \sum_{x \in S} p_x - \sum_{x \in S} q_x|##, where the maximization is over all subsets ##S## of the index set ##\{x\}##.

Homework Equations


See above

The Attempt at a Solution


[itex]
\begin{align*}
\frac{1}{2} \sum_x |p_x-q_x| &\geq | \sum_{x \in S} p_x - \sum_{x \in S} q_x|\\
&= |\sum_{x \in S} p_x-q_x|
\end{align*}
[/itex]

Then, I have tried playing around with the triangle inequality, but that didn't go anywhere...

Thanks for you help!

If you let ##p_x - q_x = r_x##, the ##r_x## sum to zero. We can re-write ##D(p_x,q_x)## as ##(1/2) \sum_x |r_x| ## and ##p(S) - q(S) = \sum_{x \in S} r_x##.

We can write
[tex] \sum_{x \in S} r_x = \underbrace{\sum_{x \in S, r_x > 0} r_x }_ {\geq 0} +
\underbrace{\sum_{x \in S, r_x < 0} r_x }_ {\leq 0} [/tex]
Think about what properties ##S## must have in order that the absolute value of the above sum be maximal, say at the subset ##S = S_0##.
 
Last edited:
  • #4
Thanks for your help! I get it now
 

FAQ: Trace distance between two probability distributions - prove

What is the trace distance between two probability distributions?

The trace distance between two probability distributions is a measure of the difference between them. It is defined as the sum of the absolute differences between the eigenvalues of the two distributions.

How is the trace distance calculated?

The trace distance is calculated by first finding the eigenvalues of each distribution. Then, the absolute differences between the corresponding eigenvalues are calculated and summed to get the final value.

What does it mean to prove the trace distance between two probability distributions?

Proving the trace distance between two probability distributions means showing that the calculated value accurately represents the difference between the two distributions. This can be done by comparing the trace distance to other measures of difference, or by using the trace distance in other calculations or analyses.

Why is the trace distance important in probability theory?

The trace distance is important in probability theory because it provides a quantitative measure of the difference between two distributions. This can be useful in various applications, such as comparing the performance of different models or evaluating the accuracy of predictions.

Can the trace distance be used for any type of probability distribution?

Yes, the trace distance can be used for any type of probability distribution, including discrete, continuous, and mixed distributions. As long as the distributions have well-defined eigenvalues, the trace distance can be calculated and used to measure their difference.

Back
Top