Monthly PR Calculation

What is the PR of a page prior to a PR calculation?

[speda1]

After the deep crawl has captured a 'snapshot' of the web, the calculation of PR can begin. We know that the calculation takes many iterations. Any thoughts on what PR is assigned each page prior to the first iteration? They could start all pages at 1 (or equivelant) or at the previously calculated PR (from last month).

My thinking is that if Google uses the PR calculated from the previous dance, less iterations would be necessary. This might also explain older sites having a PR advantage over newer sites (at least for a couple months).

To start at 1 each month would require more iterations, but would probably be more accurate.

Highly speculative, but anyone have any other ideas?

[Oaf357]

What if it's a top to bottom thing?

What if all sites start at 10 and get dinged for certain things? Not enough keyword density, inaccurate title tags, etc. could all be things that get factored in.

[speda1]

Starting at 10 would require more iterations than starting at 1. Also, they would have to perform a calculation when a page did'nt have an inbound link. It does'nt make sense to do it that way.

[BigDave]

Enough of the web changes drastically each month that I don't think there would be much advantage to starting with the previous month's PR.

With more of the web becoming dynamic, this increases the change even more.

If we take Webmaster World as an example, the high ranking pages this month, will be pretty deeply buried next month. Forums, blogs and news sites with high PR give a huge boost with links off their front pages when they are hit by the bot, but a month later they are on totally different topics.

[doc_z]

> My thinking is that if Google uses the PR calculated from the previous dance, less iterations would be necessary.

Yes, I agree. Starting from the 'old solution' is computationally less expensive. (Although, the solution is independent from the initial values for any reasonable configuration.)

> This might also explain older sites having a PR advantage over newer sites (at least for a couple months).

If this would be true, Google has chosen the criterion to stop the iterations (e.g. norm of PR_n+1 - PR_n) too high, i.e. they are performing too few iterations.

> To start at 1 each month would require more iterations, but would probably be more accurate.

No.

> Starting at 10 would require more iterations than starting at 1.

Not necessarily.

> Also, they would have to perform a calculation when a page did'nt have an inbound link. It does'nt make sense to do it that way.

Yes, it doesn't make sence. However, the solution is the same.
(I know that Google doesn't consider pages which have no inbound links. However, as long as the damping factor d is smaller than 1, no problem exists since there is a unique solution. In practice, a value very close to 1 would take more effort. For the value of d currently used by Google there should be no problem.)

By the way, I would guess that Google changed the algorithm for the calculation from the original (simple) Jacobi-iteration to a more state of the art one. Of course, this has no influence to the solution but for the CPU time needed for the calculation.