Exact objective value.

[mScid]

Hello everyone.

First of all thank you for this wonderful package.
I just want to ask if there is a way to get the exact number for the objective value. When I run sqsgen output -o [NAME] list, I get objective values of rounded off numbers like 0.00000, 0.33333, 0.50000, 0.66667.

Thank you!

[dgehringer]

Hi @mScid!

First, I glad to hear that you enjoy the package. Please feel free to open feature requests at any time!

Basically, there is certainly the possibility to extract that number. But I want to give some background here

• To extract the "exact" number I'd suggest you use the python API. Your exact use case is documented with a code example here.

But let me give you some context. the underlying C++ code is templated, that means it can run using float or double numbers.
In Python there is no distinction between this data-types since there exists only a float with arbitrary precision.

By default sqsgen uses single = float precision. That means you have 32bit number. Translating that to a decimal scale you have only a precision of 6-7 digits. This is the reason why you see 5 places after the decimal point.

The reason is that storing the results halves the amount of memory used.

In case you really want higher precision (I cannot judge you use-case) I strongly encourage you to set:

{
  "prec": "double",
}

This will make sqsgen to use double numbers under the hood resulting in approximately twice (13 digits) of accuracy.

Hope that help

Regards
Dominik

[mScid]

Thank you for the reply.

The one installed on our cluster which my adviser uses is version 0.3. The values he gets for the objective numbers are like this 0.20408163265305967, which we see as "more accurate"? than let's say 0.833333313465118 (this was ran with double precision.

I'm currently just using it for convergence test running from 1e5 to 1e11 iterations. Thank you for the reply.

Best,
Dinan

[dgehringer]

Well we have rewritten from scratch the whole package at version 0.4, as highlighted on the repositories main page.
Well the actual number depends highly on you input configurations and chosen shell_weights, as well as the number of iterations.

In case you share your config, we can discuss further,

Best
Dominik

[mScid]

Thank you.

I've discovered that even the older version outputs the same "rounded off" numbers, which is due to the input. My input is a 2D HEA of 5 metals with shell weight of 1: 1.0, 2: 0.5. This is now considered answered. Thank you again.

Best Regards,
Dinan