How to compare the accuracy of simulation results?

22 views

Skip to first unread message

xing cong

unread,

Jun 5, 2024, 2:16:20 AMJun 5

to HYCOM.org Forum

Dear Professor,

I hope this message finds you well.

I am currently using HYCOM for simulations and have obtained results with different numbers of computational cores. To assess the accuracy of these results, I would like to seek your advice on a few related questions.

Specifically, I have obtained a simulation result using 5000 cores and another using 100000 cores. I aim to compare the accuracy of these two sets of results but am unsure about the methods and standards that should be employed for such a comparison. Could you please guide me on how to conduct this comparison?

Thank you very much for your assistance!

Best regards,

CongXing

Alan Wallcraft

unread,

Jun 5, 2024, 10:12:37 AMJun 5

to HYCOM.org Forum, xing cong

HYCOM is designed to be bit for bit reproducible on any number of cores. So if you used the same executable., and the recommended compiler options, the results of your two runs should be identical.

An easy way to check this is to use the cmp command on twin archive .a files from the two runs - which should be the identical.

If the two runs are not exact twins, e.g. different forcing or even a different blkdat variable like veldf4 (say), the solutions will drift apart. Quickly differing at the bit level but with noticeable differences over 30 to 90 days.