Re: THRESHOLD in SEWARD

Posted by Victor on September 08, 2014 at 13:04:15:

In Reply to: Re: THRESHOLD in SEWARD posted by Jesus Gonzalez Vazquez on September 04, 2014 at 16:53:38:

Hi Jesus,

the 'THRESHOLD' is relevant only without packing of two-electron integrals (ERIs). By default, packing of ERIs is activated. So, if you want to discard integrals, you should specify an additional option - "NoPack". However, that is very brutal and extremely inaccurate way for packing ERIs and reducing the disk storage demands. I suggest you to consider using the Cholesky Decomposition method. The CD method is consistent and robust/accurate packing scheme. Please check the following SEWARD's option:

"Cholesky","THRCholesky", "Low Cholesky", "Medium Cholesky", "High Cholesky"

With best regards,
Victor.

: Hi,

: Thank you very much for your answer, I checked it with a diatomic molecule (but a big basis set). The input:

: &SEWARD
: Basis set
: H.aug-cc-pV5Z...
: H -1. 0. 0.
: End of Basis
: Basis set
: Cl.aug-cc-pV5Z...
: Cl +1. 0. 0.
: End of Basis
: THRESHOLD=...
: End of Input

: In case of 1e-14 (default), the output is saving:
: Integrals are written in MOLCAS2 format
: Number of integrals written on Disk = 500237956
: Number of nonzero integrals = 134879021
: Packing accuracy = 0.1000E-13
: Highest disk address written 2230852
: Diagonal and subdiagonal, symmetry allowed 2-el integral blocks are stored on Disk
: The size of the OrdInt: 1115432KB

: If one uses 100:
: Integrals are written in MOLCAS2 format
: Number of integrals written on Disk = 500237956
: Number of nonzero integrals = 0
: Packing accuracy = 0.1000E-13
: Highest disk address written 1955396
: Diagonal and subdiagonal, symmetry allowed 2-el integral blocks are stored on Disk
: The size of OrdInt: 977704KB

: There is a very tiny reduction of the file... so I cannot understand how it works, how big are the blocks?

: Thank you again,

: Jesus

: : Hi,
: : this is a poor example since it is atomic. The integrals are written to the disk in blocks. In this atomic case the number of integrals does not exceed a singel block and reducing the number of nonzero integrals will not reduce the number of blocks in the disk. Try a larger molecular system to see the effect.

Follow Ups:

Post a Followup

Name:
E-Mail:

Subject:

if B is 1s22s22p1, what is Li?

Passfield:

Comments:
: Hi Jesus, : the 'THRESHOLD' is relevant only without packing of two-electron integrals (ERIs). By default, packing of ERIs is activated. So, if you want to discard integrals, you should specify an additional option - "NoPack". However, that is very brutal and extremely inaccurate way for packing ERIs and reducing the disk storage demands. I suggest you to consider using the Cholesky Decomposition method. The CD method is consistent and robust/accurate packing scheme. Please check the following SEWARD's option: : "Cholesky","THRCholesky", "Low Cholesky", "Medium Cholesky", "High Cholesky" : : With best regards, : Victor. : : Hi, : : Thank you very much for your answer, I checked it with a diatomic molecule (but a big basis set). The input: : : &SEWARD : : Basis set : : H.aug-cc-pV5Z... : : H -1. 0. 0. : : End of Basis : : Basis set : : Cl.aug-cc-pV5Z... : : Cl +1. 0. 0. : : End of Basis : : THRESHOLD=... : : End of Input : : In case of 1e-14 (default), the output is saving: : : Integrals are written in MOLCAS2 format : : Number of integrals written on Disk = 500237956 : : Number of nonzero integrals = 134879021 : : Packing accuracy = 0.1000E-13 : : Highest disk address written 2230852 : : Diagonal and subdiagonal, symmetry allowed 2-el integral blocks are stored on Disk : : The size of the OrdInt: 1115432KB : : If one uses 100: : : Integrals are written in MOLCAS2 format : : Number of integrals written on Disk = 500237956 : : Number of nonzero integrals = 0 : : Packing accuracy = 0.1000E-13 : : Highest disk address written 1955396 : : Diagonal and subdiagonal, symmetry allowed 2-el integral blocks are stored on Disk : : The size of OrdInt: 977704KB : : There is a very tiny reduction of the file... so I cannot understand how it works, how big are the blocks? : : Thank you again, : : Jesus : : : Hi, : : : this is a poor example since it is atomic. The integrals are written to the disk in blocks. In this atomic case the number of integrals does not exceed a singel block and reducing the number of nonzero integrals will not reduce the number of blocks in the disk. Try a larger molecular system to see the effect.

[ Follow Ups ] [ Post Followup ] [ Molcas user's WWWBoard ]