Usage of VSL_BRNG_SOBOL for dimension higher than 40

[CaseySt]

Hello devs and users, 

Apologize in advance if this question has been already treated but couldn't find anything similar.

My task is to generate sobol sequence of dimension higher than 40 within VSL MKL library (version after 2022). 

According official documentation, using the method vslNewStream should allow me to have a maximum dimension of 40 and in case i want to go beyond i should use vslNewStreamEx with customer polyomials and direction numbers.

The latter is what i did, using the file produced in the notorious joe e kuo notes, which works without problem.

Now my point is, when i use vslNewStream setting dimesion higher than 40, i don't get any error and i get exactly the same ouput as for vslNewStreamEx with customer parameters, as if vslNewStream could already handle dimension grater than 40 (despite what is documented).

Could someone please clarify if is really the case and i can trust the vslNewStream output without using my custom parameters?

Many thanks in advance and best Regards

[Ruqiu_C_Intel]

Hi,

May I ask where the statement of maximum dimension of 40 comes from from? From the oneMKL guidance, here is a wording for vslNewStream vs vslNewStreamEx: "Whenever possible, use vslNewStream, which is analogous to vslNewStreamEx except that it takes only one 32-bit initial condition."

Regards,

Ruqiu

[CaseySt]

Hi Ruqiu, Many thanks for your reply.

The information on vslNewStream on maximum dimension (within VSL_BRNG_SOBOL generator) i found it here SOBOL 

Moreover i can confirm that the maximum dimension (again when SL_BRNG_SOBOL generator is used) is 40 since above this number the method returns by default the sobol sequence of dimension 1 (tested in several way i can provide the test if needed).

I observed the same behavior for vslNewStreamEx, even if i provide custom parameters such sobol direction numbers and polynomial, this is why i get the same results for dimension higher than 40 (my initial topic).

Now vslNewStreamEx ignoring the custom parameter seems to be a well known issue of the mkl library, i found a post in this forum dated 2008 when an intel developer clamed they were working on fix it but apparently did not.