How to continue a BSK calculation?

[gfluo]

Dear friends,

Recently I carry out a very time-consuming (over three months) BSE calculation and for machine maintaining reason I stopped yambo in the middle of the BSK part calculation. Though it is said that yambo can continue the calculation automatically, I found it is impossible in this part, at least not possible by just rerunning it.

The cause seems that yambo simply checks the existence of the db.BS_Q1 file, which is created at the very beginning of the BSK calculation: If exists, yambo thinks that the BSK calculation finished and does nothing; If not, yambo will start this calculation from scratch. I did several tests and still have not figured out how to solve this problem. Could you please enlighten me?

P.S. I am using the latest yambo-3.2.3-rev.11; This post (viewtopic.php?f=16&t=126) described similar situation, except that there many of the first-calculated fragments (like *1_1, *1_2) are missing while in my case the end of fragments (like *200_199, *200_200) are unfinished.

Sincerely,

[andrea marini]

Recently I carry out a very time-consuming (over three months) BSE calculation and for machine maintaining reason I stopped yambo in the middle of the BSK part calculation.
Though it is said that yambo can continue the calculation automatically, I found it is impossible in this part, at least not possible by just rerunning it.

THREE MONTHS! WoW Well I wrote Yambo to be restarted in reasonable cases. I cannot even immagine how many fragmentes, I/O etc. etc. you have produced in three months. So it is a miracle that you actually managed to run for THREE months.

But, for my curiosity, tell us more about the run. How many atoms, G-vectors, processoros. How big is the Hamiltonian? Can you attach the report and the log file of the calculation ? What Machine are you using ?

The cause seems that yambo simply checks the existence of the db.BS_Q1 file, which is created at the very beginning of the BSK calculation: If exists, yambo thinks that the BSK calculation finished and does nothing; If not, yambo will start this calculation from scratch. I did several tests and still have not figured out how to solve this problem. Could you please enlighten me?

It should not be like this. But I checked in the repository hosted on qe-forge and I noticed that in revisions 12 and 14 (now the latest is rev. 11) I fixed the use of fragments in an MPI calculation. Maybe this can solve your problem. Would you bne so kind to register on the qe-forge portal and download the latest revision (16) from the SVN repository ? Soon we will also release a new version.

Cheers

Andrea

[gfluo]

Dear Dr. Andrea,

Thanks very much for your reply. The system I am studying is an 18-carbon-atom low dimensional material with relatively large vacuum space. After doing some convergence tests, I found it requires a very small k-point spacing and therefore very large memory, especially in the BSE solver part by using diagonalization. I carried it out on an 8-core cluster, only which has a 64-GB memory. The log and report files are attached for your inspection.

I did tests with both yambo-3.2.2-rev5 and yambo-3.2.3-rev11, and both of them act as I described before. For your convenience, I also attached a test file of the 3D LiF system (in a separate post because of the attachment size limit), which can be finished in minutes.

I registered in qe-forge.org with a username of gfluo but cannot find the yambo revision 16 (see http://qe-forge.org/frs/?group_id=34). Could you please direct me?

Sincerely,

[gfluo]

Dear Dr. Andrea,

The attached is the test input files of Abinit and Yambo I used.

Sincerely,

[andrea marini]

I registered in qe-forge.org with a username of gfluo but cannot find the yambo revision 16 (see http://qe-forge.org/frs/?group_id=34). Could you please direct me?

You need to join the Yambo project and, then, you will get access to the SVN repository. I have made more clear the instructions on the download page. Try following them and ket us know if you have problems.

Andrea

[gfluo]

Dear Dr. Andrea,

Today, I surprisingly find that a new version, yambo-3.2.4-rev.17, is available on the website. So I install it and try the former test. The outcome is great: the BSK part can finally be continued smoothly in this version!

You help me save a lot of time. Thanks so much!

Sincerely,