CSAS6 keff Inconsistency

[hargreaves]

When I use CSAS6 to build my input file the resulting keff value is inconsistent across machines. At work we have several servers to run SCALE on and across four different machines four different keff values were obtained with the exact same input file. Why is this and would this be resolved by modeling with a KENO input instead of CSAS?

Thank you!

[Willem van Rooijen]

It is impossible to answer your question.

- what OS on the various servers?

- which versions of SCALE on the various servers?

- which versions of the cross section libraries on the various servers?

- how "different" is "different"?

2025年10月29日水曜日 23:09:45 UTC+9 hargreaves:

[A. Hargreaves]

Since I posted this I"ve looked into it a bit more with folks from work. So far we have had this issue on:

- 2 Mac computers (latest OS and updates), 1 windows machine (latest OS and updates), and a server based on two different user profiles (I have no clue about any info on this)

- All machines are using SCALE 6.3.1

- All files were run using ce_v8.0 

- Differences were about 700 pcm

[Willem van Rooijen]

Next questions / options:

- are you changing the random number sequence in KENO (RND= in the KENO input)?

- do you always get the same result on the same machine (i.e. on computer X, is the result always k_X and on computer Y, is it always k_Y)?

- do you get 2 different results for the 2 user profiles on the server?

- do you get 2 different results on the Mac computers?

I would start out by comparing the output files (use something like "diff") and check for any differences.

2025年10月30日木曜日 22:53:01 UTC+9 A. Hargreaves:

[A. Hargreaves]

- no

- yes

- yes

- yes

Also of note, no similar issue of difference between these machines when using one of the pre-built SCALE sample files that came on installation. Running a pre-built file produces the same results across all platforms, it seems to just be one tricky one (that I built but is fairly simplistic)

[Willem van Rooijen]

It sounds very mysterious. But computers are deterministic (even if you are running a stochastic simulation).

- compare the input files across the various machines to make sure that, in fact, all the input files are identical. Do not only check visually, also make sure about tabulations (tabs versus spaces) and newline characters; make sure that none of the input lines are too long (I recommend to use up to 80 columns per line). Use "diff" and a proper text editor (such as Notepad++ on Windows or vi/vim (available on Mac, Windows and linux))

- compare output files; if the k-eff is different, there must be other differences in the output files

- make sure that the cross section files are the same on all machines (i.e. same file size, modification date, etc. "diff" would be the best test)

I am guessing that the "server" is running linux, so you are probably dealing with 3 different versions of SCALE (Mac, Windows, linux). That by itself __could__ be a source of the problem, but since the test cases are not problematic your issue is somewhat mysterious. 

If possible, post your input file, so that others can try to reproduce the problem.

Namizono

2025年10月31日金曜日 6:58:30 UTC+9 A. Hargreaves: