I think the authors have substantially improved the paper. There are still a few points that need to be improved, but the overall quality is now much more satisfactory.
General comment:
Please use floating point notation (e.g. 1.5 instead of 3/2), when you have no prediction or argument for the expoenents to take these simple rational values. Otherwise it is misleading, as usually when an exponent is *exactly*U equal to a ratinal number, there is (relatively) simple argument explaining that result. In your paper I believe most exponents are irrational numbers.
1.0
Ok, I see you added some litterature, I haven't compared in detail but reading the whole paper I found it more clear.
I see you cited Landes and Lippiello 2016, this is nice but not very necessary: please do not feel like you should cite me because I'm refereeing. Cite if you truly believe it is relevant.
I think you would gain readership by further putting things into context in a precise way, comparing your results quantitatively with other model's, but that may also be for a separate publication, it's your choice. Now the reader is not lost, I think.
1.1
Ok, good.
1.2
Thank you for this nice discussion and adding these itneresting results.
You should add one or two tentative fits and their corresponding power-law exponents in Fig 4.
About your last comment on 1.2, a remark:
you say that V and A are equally representations of the energy .... but it's like saying that velocity v and kinetic energy E_k of a system are equally valid representations of its temperature: instead, we have |v|^2 ~ T ~ E_k, not |v|~T~E_k. Exponents change if you use a variable other than the correct one (or not proportional to it).
If P(A) ~ A^-1.6 and V~A^1.5 for instance, using P(A) dA = P(V) dV , you get (I think) P(V)~V^-((1.6-1.5+1)/1.5)=V^-1.4.
I do not think it is crucial for your results that the exponents match very well: the key result is that you have bimodal statistics in both space and time distributions which appear as a result of introducing p. So even if the exponent of fig 4 is not very close to the famous 5/3=1.6666666 you wish for, it's ok, your paper is worthwhile (amyway the "true" value of b is very debated).
1.3
Ok... So you elected to call M (and sometimes m) the magnitude and m (?) its threshold... why not use m always and m_th for the threshold ?
1.4
Ok, excellent, this point is now much clearer to me and clearer in the paper.
1.5
I appreciate your work, but cannot find this discussion in the revised paper. Where did you include (part of) this discussion?
Let me add a comment for you:
what I was trying to exaplain is that because of this effect (of Fig ii of your reply), your model may be described by "count", the y-axis of Fig ii, instead of the proba p. Let me call "count" C here. Using C as parameter is completely equivalent to using p.
Using C as control parameter, it becomes obvious that as soon as C > 10^5, i.e. 1000 times its baseline value, what you are actually doing is a quasi- extremal dynamics, since you are almost always picking this site.
For lower values of C (in the range p~0.007 I guess C is much closer to its baseline value), you are not doing extremal dyanmics, but since you load all sites at random almost equally, your loading protocol is in effect quite similar to uniform loading.
I just noticed this fact while reading your paper and I think one needs to study this matter carefully in order to compare with other loading protocols.
It is not necessary to have this discussion in full in your paper, a short comment to let the reader realize this fact will be enough.
1.6
OK
1.7
OK
1.8
OK
1.9
Thanks, this is now very clear when reading the paper, and furthermore one understands why it is important to threshold (relative to interevent time statistics). I learned something new, thank you !
1.10
OK |