Re: PDEs Involving Absolute Value


In article <d4d37c37-0516-4725-8126-9c8f5c6aea6c@xxxxxxxxxxxxxxxxxxxxxxxxxxx>,
aruzinsky <aruzinsky@xxxxxxxxxxxxxxxxxxxx> writes:

You have wrongly assumed that this is my problem, It is not my
problem, i.e., I already know how to do proper L1 norm minimization,
and I find your insuation otherwise offensive. The question that you
failed to address was, "How do so many people get away with this?" In
fact, http://www.duke.edu/~sf59/SRfinal.pdf , and many other papers
are wrong because "if you have a function minimization with a
nonsmooth function, then indeed you never should use gradient based
methods since almost for sure they will stick up at a nonoptimal
point."

You and others who are aware of this either are

1. unaware of such published mistakes.

or

2. remiss as referees by passing such mistakes in peer reviewed
papers.

or

3. failed to write letters to publishers correcting mistakes after the
papers are published.

Why, it is almost as if someone published 2 + 2 =3D 5, and then many
other authors copied and published that mistake while people who knew
better allowed it to happen. Just think how much social chaos that
would cause.

Now that you know, why don't you write a letter to the publisher?
And, seek out similar papers and write letters to their publishers?


aaaah, no I see the sense behind your original question, but you had better
wrote in your initial contribution what you wrote here.
Concerning the paper you pointed too: yes, there are errors
(the gradient of the l1-norm is indeed quite funny)
but due to the regularization terms used this seemingly had no severe influence.
(although I found the results there not quite impressive)

well, maybe I am too happy to read only journals with a high quality refereeing
process and detect errors of this level not too often.
If you feel disappointed by such, then you should not address this to this
group which cannot hinder this.
you have two possibilities:
1. write to the authors and ask for explanation.
maybe you get no answer.
2. write a paper "remark on the paper.... of ... " there you point out
the error and show that you can do better.
if this is really better, then such a paper will be published,
possibly after the editor did contact the authors of the wrong
statement giving them a chance for correction (which results in another paper,
and indeed, there are too many papers around)
you should be assured that as a referee and associate editor I do my best
to avoid the publication of nonsense, and this takes a lot of my time.
best wishes
p. spellucci

.

Quantcast