p-norms
Depending on how you scale them, there are two kinds of
-lengths, which “sum” the numbers, -means, which “average” the numbers.
-length
This is the usual notion of
In particular:
is the Manhattan length, is the Euclidean length, is just the (absolute) maximum: .
The
It decreases a lot unless is sparse
Let
with equality iff all but one of the coordinates are zero: vector
This is actually robust fact: if the ratio
so you can make sure that
and then the number of nonzero entries in
See Lower-degree norms give sparsity for more intuition and details.
-mean
Let
More generally, we can consider the
In particular:
is the arithmetic mean and is the expectation, is the quadratic mean, and are just the maximum.2
The
It increases a lot unless is spread / is reasonable
Let
and thus also
The equality case is when
is constant: it’s “perfectly reasonable”,- all the
’s are equal: they are “perfectly spread”.
This is also a robust fact: if the ratio
so in particular, when the ratio is constant,
Note: I’m pretty sure you can also show that when
For
Overall picture
Bounding either side
The
And on the other hand, we can use the
But on the third hand, you can’t bound how fast the moments of a continuous random variable increase: in fact, it could be that the
So we get the following picture (for
- The
-norm is stuck between (“spread” regime) and (“sparse” regime). - The
-mean is stuck between (“spread” regime) and (“sparse” regime). - The
moment is stuck between (“reasonable/concentrated” regime) and- either
if is discrete (“sparse” regime), - or
if is continuous (“spiky” regime).
- either

TODO: change “concentrated” to “spread” in this graph #figure TODO: also make a “table” #figure
Qualities being tracked
We can use ratios between norms to track different qualitative behaviors (the ratios below are
- for a vector
, the length ratios track its sparsity; - for a sequence
, the mean ratios track its spreadness; - for a random variable
, the moment ratios track its reasonableness;- by extension, so do the norm ratios
for a function (by considering its value over a random input );
- by extension, so do the norm ratios
- for a discrete random variable
, the power entropy ratios (for ) track its flatness.
TODO: add something about Density functions (are inner products like