New lattice files - 7Xc2 & 9Xc2

Those of you paying attention to your clients on Friday would have notices that they were downloading new lattices, and finishing simulations more quickly. That’s because new lattices have been introduced. Also, the v4.44d client has also been finally put to rest. There will be no more lattices that allow it to be used, so after almost a year, the jitter-free 4.45 will finally be the ‘required’ client.

With the retirement of both Linac900Ext10tc2 and Linac900Ext9X we’re moving on to Linac900Ext7Xc2 and Linac900Ext9Xc2. Both are development improvements of previous lattices, and attempt to work on phase handling, as well as investigate new designs without being side-tracked by jitter.

Just how big a problem jitter (the random noise that can mean almost identical designs can have a significantly different result, and add an element of fuzziness) can be shown in this graph from the Linac900Ext10d2_zoom lattice trial tested in late 2010.
v4.45 versus v4.44d
click to enlarge. Credit: SJ Brooks
How that impacts on the yield and it’s design path is better illustrated by the following graph, showing yield against mpts for various client versions. As you can see, the rightmost edge is all 4.44d clients, meaning the 4.45 users are trying to beat a design that may be inferior.
Linac900Ext10tc2's yield/mpts graph showing results by client version
As to the new lattices themselves, they are derivatives. The final 4 (or sometimes 5) characters are the key to the differences. In this case, it’s 7Xc2 and 9Xc2. These can be broken down as follows:

  • The 7 and 9 represent the number of blocks in the design.
  • X shows that it’s got an extended maximum simulated time, just as the earlier “t” did. However, X *also* means that it requires 4.45 or later.
  • The c2 refers to the method for generating the RF phase. It is in fact the inverse wave of the d2 attribute. Without either, it’s a single wave that starts with the c2 designs. However, as the 2 indicates, it contains TWO waves, rather than one.

RF Phase genome value in lattice design
Credit S. Brooks
The two waves are for cycling. With only a single wave, the client sometimes has problems going from 999 (one end of the wave) to 000 (the other end), going to two waves enabled this to be less of a problem. However, it’s important to point out that c2 and d2 variants describe what would otherwise be IDENTICAL designs in different ways. The massive yield difference between 10td2 (3.774999%) and 10tc2 (2.885679%) point to optimizer failures.

Meanwhile, we’ll see how well these new lattices do. Already they’ve gone positive, and now that there’s clarity on the RF phase factor, we can expect things to be performing better than ever.

The official discussion thread on the new lattices is here

No comments:

Post a Comment