search

lab-cosmo / i-pi-dev_archive

Development version of i-PI
21 stars 12 forks source link

Conserved quanty with thermostat PILE #237

Closed litman90 closed 5 years ago

litman90 commented 6 years ago

The conserved quantity is not conserved when pile_l or pile_g is used.

Just for reference: system: ice (288 atoms) -nbeads=1, nve, nvt (svr) and nvt (langevin) work fine -nbeads=48 nve works fine -nbeads=48 nvt(pile_l) and nvt(pile_g) have a drift of 0.5eV/ps

The problem is still there when "fixcom=False " is set.

ceriottm commented 6 years ago

Does the problem go away when you set pile_lambda=0.1 ?

litman90 commented 6 years ago

No. The drift is reduced by ~40% tough

mahrossi commented 6 years ago

Is the problem also there for pile_l with 1 bead?

Cheers,

Mariana

On Tue, Jun 12, 2018 at 10:20 AM, litman90 notifications@github.com wrote:

No. The drift is reduced by ~40% tough

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/cosmo-epfl/i-pi-dev/issues/237#issuecomment-396507070, or mute the thread https://github.com/notifications/unsubscribe-auth/AGbkQiA0KJ0xssIkFW-fwZ7Cg-jZWVpNks5t73nfgaJpZM4UjBBN .

litman90 commented 6 years ago

No. Regards, Yair

On 12 June 2018 at 10:47, Mariana Rossi notifications@github.com wrote:

Is the problem also there for pile_l with 1 bead?

Cheers,

Mariana

On Tue, Jun 12, 2018 at 10:20 AM, litman90 notifications@github.com wrote:

No. The drift is reduced by ~40% tough

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/cosmo-epfl/i-pi-dev/issues/237# issuecomment-396507070, or mute the thread https://github.com/notifications/unsubscribe-auth/AGbkQiA0KJ0xssIkFW- fwZ7Cg-jZWVpNks5t73nfgaJpZM4UjBBN .

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/cosmo-epfl/i-pi-dev/issues/237#issuecomment-396514571, or mute the thread https://github.com/notifications/unsubscribe-auth/AXH6e8k3BeGHLHW9dS49kcJu2OGFwtYoks5t74AegaJpZM4UjBBN .

ceriottm commented 6 years ago

I was asking because in general strongly overdamped Langevin dynamics interferes with energy conservation (basically velocity verlet is second-order but assuming smooth velocities, which you don't have with strong overdamping). This is not a practical problem, because the thermostat mops up most problems. Try to see what happens if you run classical with a Langevin thermostat with a time constant of 1fs. You should see similar drift. Michele

On 12 June 2018 at 12:14, litman90 notifications@github.com wrote:

No. Regards, Yair

On 12 June 2018 at 10:47, Mariana Rossi notifications@github.com wrote:

Is the problem also there for pile_l with 1 bead?

Cheers,

Mariana

On Tue, Jun 12, 2018 at 10:20 AM, litman90 notifications@github.com wrote:

No. The drift is reduced by ~40% tough

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/cosmo-epfl/i-pi-dev/issues/237# issuecomment-396507070, or mute the thread https://github.com/notifications/unsubscribe-auth/AGbkQiA0KJ0xssIkFW- fwZ7Cg-jZWVpNks5t73nfgaJpZM4UjBBN .

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/cosmo-epfl/i-pi-dev/issues/237# issuecomment-396514571, or mute the thread https://github.com/notifications/unsubscribe-auth/ AXH6e8k3BeGHLHW9dS49kcJu2OGFwtYoks5t74AegaJpZM4UjBBN .

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/cosmo-epfl/i-pi-dev/issues/237#issuecomment-396539428, or mute the thread https://github.com/notifications/unsubscribe-auth/ABESZ7cILGh2dmdggdLUBK8nZEExGwg8ks5t75SJgaJpZM4UjBBN .

litman90 commented 6 years ago

So, nbeads=1 with a langevin thermostat (tau=1ps) has a comparable drift to the other similations that I have already mention. Of course after I rescale it w.r.t the numbers of beads. But pile_g/pile_l with lambda=0.5 is not overdamped, is it? And if the velocity verlet algorithm integrates "so bad" can we guarantee that we are sampling the ensemble we want? Regards, Yair

On 12 June 2018 at 23:35, Michele Ceriotti notifications@github.com wrote:

I was asking because in general strongly overdamped Langevin dynamics interferes with energy conservation (basically velocity verlet is second-order but assuming smooth velocities, which you don't have with strong overdamping). This is not a practical problem, because the thermostat mops up most problems. Try to see what happens if you run classical with a Langevin thermostat with a time constant of 1fs. You should see similar drift. Michele

On 12 June 2018 at 12:14, litman90 notifications@github.com wrote:

No. Regards, Yair

On 12 June 2018 at 10:47, Mariana Rossi notifications@github.com wrote:

Is the problem also there for pile_l with 1 bead?

Cheers,

Mariana

On Tue, Jun 12, 2018 at 10:20 AM, litman90 notifications@github.com wrote:

No. The drift is reduced by ~40% tough

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/cosmo-epfl/i-pi-dev/issues/237# issuecomment-396507070, or mute the thread https://github.com/notifications/unsubscribe- auth/AGbkQiA0KJ0xssIkFW- fwZ7Cg-jZWVpNks5t73nfgaJpZM4UjBBN .

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/cosmo-epfl/i-pi-dev/issues/237# issuecomment-396514571, or mute the thread https://github.com/notifications/unsubscribe-auth/ AXH6e8k3BeGHLHW9dS49kcJu2OGFwtYoks5t74AegaJpZM4UjBBN .

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/cosmo-epfl/i-pi-dev/issues/237# issuecomment-396539428, or mute the thread https://github.com/notifications/unsubscribe-auth/ ABESZ7cILGh2dmdggdLUBK8nZEExGwg8ks5t75SJgaJpZM4UjBBN

.

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/cosmo-epfl/i-pi-dev/issues/237#issuecomment-396742370, or mute the thread https://github.com/notifications/unsubscribe-auth/AXH6ey71cI7RrD4s_wqlyMi7d9SgBDP5ks5t8DQjgaJpZM4UjBBN .