first results


Advanced search

Message boards : Science : first results

AuthorMessage
Konstantin L. Metlov
Forum moderator
Project developer
Project scientist
Avatar
Send message
Joined: Jun 2 08
Posts: 104
Credit: 47,087
RAC: 0
Message 420 - Posted 24 Nov 2008 16:12:01 UTC

    I have finished processing a batch and built the first magnetic phase diagram last week, based on our computation.

    The main difficulty (why the data took so long to process and why there still might be some glitches in the dataset) was posed by "bad" data points. These points correspond to the results, returned by faulty computers. Their recalculation on a different PC produces the correct answers, but the main difficulty was finding (manually) -- which points to recalculate. By now, I hope to have found most of these points, influencing the phase diagram below. But the data set will have to be recalculated with enabled BOINC redundancy (sending the same workunit to two hosts at least and comparing the result). I've started this run of the project today.

    Another problem, solved by the new run, is that now the whole range of particle's aspect ratios is covered uniformly, something I have overlooked, submitting the first batch.

    Now, the ground states:


    This is the raw map, showing the values of aspect ratios, which have actually been calculated (non-uniform coverage and the gap in the bottom is the result of the second of the previously mentioned problems). The vertical coordinate on the plot is the cylinder's radius, divided by the material's exchange length. The horisontal is thickness, divided by the exchange length. The range of vertical and horizontal coordinates is from 0 to 5 (which is about 100nm for Permalloy).


    This is the close up of the origin of the diagram.


    This is the same diagram, but with spline interpolation across the aspect ratios. The waviness of the orange/yellow boundary at the bottom is, probably, due to another set of "bad" points I could not find yet and should be fixed by the new batch.

    The colors of the regions correspond to different magnetization ground states. You can see sketches and names of these states in the slideshow inside your BOINC client if you switch it to the "simple view".

    There are many interesting features on this diagram I'd rather spend time discussing in the paper, but if you ask the right questions I'll be glad to explain it here as well.

    The data set we have calculated contains much more than this diagram, it also contains the metastable states and the energy barriers between them, but that will be the subject of future work. The new picture of ground states (above) is surprising enough to publish it separately.

    OK. Now your questons. ;-)

    ---
    Konstantin

    PLEASE NOTE: the above pictures are shown here on the same basis as "preprint". They are part of the forthcoming and developing scientific paper. Please contact me for an updated reference to this work, if you need it.

    Profile Steven Pletsch
    Avatar
    Send message
    Joined: Jun 17 08
    Posts: 10
    Credit: 102,269
    RAC: 0
    Message 483 - Posted 6 Jan 2009 17:32:01 UTC

      Last modified: 6 Jan 2009 17:38:53 UTC

      I'd be a liar if I said I knew enough to ask an intelligent question about the science, but I am curious about how you can obtain the same results from multiple computers without homogeneous redundancy. Are variations in the floating point numbers affecting the validation of WU's or is there a "margin" you are granting to ensure that valid results are accepted given slight variations in floating point operations of different hardware ?

      Thanks :)

      BTW: Congratulations (belated) on the first results
      ____________
      "Every passing hour brings the Solar System forty-three thousand miles closer to Globular Cluster M13 in Hercules -- and still there are some misfits who insist that there is no such thing as progress." - Kurt Vonnegut

      Konstantin L. Metlov
      Forum moderator
      Project developer
      Project scientist
      Avatar
      Send message
      Joined: Jun 2 08
      Posts: 104
      Credit: 47,087
      RAC: 0
      Message 485 - Posted 10 Jan 2009 14:17:59 UTC - in response to Message 483.

        The calculation we are doing is approximate. Roughly speaking, only the leading digits in the resulting numbers are valid (their number depends mainly on the parameters of the algorithm, such as the number of finite elements, order of the multipole expansion of "far" field, etc). Validator program compares just these leading digits, which should definitely be valid on every platform. The result, however, contains more valid digits (their number may differ from platform to platform).

        ---
        Konstantin.

        Post to thread

        Message boards : Science : first results


        Home | My Account | Message Boards


        Copyright © 2014 Konstantin L. Metlov