Hi,
I am implementing the worm algorithm for the Bose Hubbard model. On of the "worm moves" is to insert a kink after the MASHA head this can cause a reconnection of the worm. I think that this move is possible only when there is a world line in the target site, other wise there is nothing to connect to. I tried to see were this condition is implement in the code but I failed to find it. I'll be grateful if someone can direct me to the location of this condition in the code, or tell me were my reasoning is wrong (may be there is some other way to close the world line).
Thanks in advance
Snir
On 30 Dec 2009, at 10:08, Snir Gazit wrote:
Hi,
I am implementing the worm algorithm for the Bose Hubbard model. On of the "worm moves" is to insert a kink after the MASHA head this can cause a reconnection of the worm. I think that this move is possible only when there is a world line in the target site, other wise there is nothing to connect to. I tried to see were this condition is implement in the code but I failed to find it. I'll be grateful if someone can direct me to the location of this condition in the code, or tell me were my reasoning is wrong (may be there is some other way to close the world line).
Thanks in advance
Snir
We calculate the matrix element for the proposed insertion. In the case you mention the matrix element is 0, and the proposed update is rejected immediately.
Matthias
Thanks for the quick and helpful reply!
Snir
On Wed, Dec 30, 2009 at 11:24 AM, Matthias Troyer troyer@phys.ethz.chwrote:
On 30 Dec 2009, at 10:08, Snir Gazit wrote:
Hi,
I am implementing the worm algorithm for the Bose Hubbard model. On of the "worm moves" is to insert a kink after the MASHA head this can
cause a reconnection of the worm.
I think that this move is possible only when there is a world line in the
target site, other wise there is nothing to connect to.
I tried to see were this condition is implement in the code but I failed
to find it.
I'll be grateful if someone can direct me to the location of this
condition in the code, or tell me were my reasoning is wrong (may be there is some other way to close the world line).
Thanks in advance
Snir
We calculate the matrix element for the proposed insertion. In the case you mention the matrix element is 0, and the proposed update is rejected immediately.
Matthias
Hi,
One more question regarding the implementation.
Isn't it better to check if the reconnection move is feasible? (same like the close/open move in which IRA and MASHA should be on the same site). I understand that the ALPS implementation keeps the jump and reconnection process symmetric. But essentially they are not because a regular jump (where the kink is inserted before MASHA) is always possible unlike the reconnection process. Is there is something I missing?
Thanks again for the assistance.
snir
On Wed, Dec 30, 2009 at 11:31 AM, Snir Gazit snirgaz@tx.technion.ac.ilwrote:
Thanks for the quick and helpful reply!
Snir
On Wed, Dec 30, 2009 at 11:24 AM, Matthias Troyer troyer@phys.ethz.chwrote:
On 30 Dec 2009, at 10:08, Snir Gazit wrote:
Hi,
I am implementing the worm algorithm for the Bose Hubbard model. On of the "worm moves" is to insert a kink after the MASHA head this can
cause a reconnection of the worm.
I think that this move is possible only when there is a world line in
the target site, other wise there is nothing to connect to.
I tried to see were this condition is implement in the code but I failed
to find it.
I'll be grateful if someone can direct me to the location of this
condition in the code, or tell me were my reasoning is wrong (may be there is some other way to close the world line).
Thanks in advance
Snir
We calculate the matrix element for the proposed insertion. In the case you mention the matrix element is 0, and the proposed update is rejected immediately.
Matthias
On 30 Dec 2009, at 11:50, Snir Gazit wrote:
Hi,
One more question regarding the implementation.
Isn't it better to check if the reconnection move is feasible? (same like the close/open move in which IRA and MASHA should be on the same site). I understand that the ALPS implementation keeps the jump and reconnection process symmetric. But essentially they are not because a regular jump (where the kink is inserted before MASHA) is always possible unlike the reconnection process. Is there is something I missing?
A regular jump is also not symmetric if there is a cutoff in the total occupation number, as is the case for example for spin models.
And actually we make the check for the matrix element as the first thing in the move - we thus do what you propose
Matthias
Thanks again :)
On Wed, Dec 30, 2009 at 1:00 PM, Matthias Troyer troyer@phys.ethz.chwrote:
On 30 Dec 2009, at 11:50, Snir Gazit wrote:
Hi,
One more question regarding the implementation.
Isn't it better to check if the reconnection move is feasible? (same like
the close/open move in which IRA and MASHA should be on the same site).
I understand that the ALPS implementation keeps the jump and reconnection
process symmetric. But essentially they are not because a regular jump (where the kink is inserted before MASHA) is always possible unlike the reconnection process.
Is there is something I missing?
A regular jump is also not symmetric if there is a cutoff in the total occupation number, as is the case for example for spin models.
And actually we make the check for the matrix element as the first thing in the move - we thus do what you propose
Matthias
Hi,
I have yet another question :).
The Insert jump move inserts a kink only in the time segment in which the occupation number of the original and target site don't change. This leads to the restriction on the remove jump (do to detailed balance) that a kink is removed only if in the time segment between the kink and the Masha head the occupation number remains the same. Am I right? If so, doesn't this restriction compromise ergodicy because it restricts the paths in which the worm can choose? Do you think that the computational over head in inserting a kink in an arbitrary position between the last kink (before Masha) and Masha is reasonable?
Thanks allot :)
Snir
On Wed, Dec 30, 2009 at 1:23 PM, Snir Gazit snirgaz@tx.technion.ac.ilwrote:
Thanks again :)
On Wed, Dec 30, 2009 at 1:00 PM, Matthias Troyer troyer@phys.ethz.chwrote:
On 30 Dec 2009, at 11:50, Snir Gazit wrote:
Hi,
One more question regarding the implementation.
Isn't it better to check if the reconnection move is feasible? (same
like the close/open move in which IRA and MASHA should be on the same site).
I understand that the ALPS implementation keeps the jump and
reconnection process symmetric. But essentially they are not because a regular jump (where the kink is inserted before MASHA) is always possible unlike the reconnection process.
Is there is something I missing?
A regular jump is also not symmetric if there is a cutoff in the total occupation number, as is the case for example for spin models.
And actually we make the check for the matrix element as the first thing in the move - we thus do what you propose
Matthias
No, it does not restrict ergodicity since shift_kink can move Masha to an adjacent segment
Matthias
On 2 Jan 2010, at 11:05, Snir Gazit wrote:
Hi,
I have yet another question :).
The Insert jump move inserts a kink only in the time segment in which the occupation number of the original and target site don't change. This leads to the restriction on the remove jump (do to detailed balance) that a kink is removed only if in the time segment between the kink and the Masha head the occupation number remains the same. Am I right? If so, doesn't this restriction compromise ergodicy because it restricts the paths in which the worm can choose? Do you think that the computational over head in inserting a kink in an arbitrary position between the last kink (before Masha) and Masha is reasonable?
Thanks allot :)
Snir
On Wed, Dec 30, 2009 at 1:23 PM, Snir Gazit snirgaz@tx.technion.ac.il wrote: Thanks again :)
On Wed, Dec 30, 2009 at 1:00 PM, Matthias Troyer troyer@phys.ethz.ch wrote:
On 30 Dec 2009, at 11:50, Snir Gazit wrote:
Hi,
One more question regarding the implementation.
Isn't it better to check if the reconnection move is feasible? (same like the close/open move in which IRA and MASHA should be on the same site). I understand that the ALPS implementation keeps the jump and reconnection process symmetric. But essentially they are not because a regular jump (where the kink is inserted before MASHA) is always possible unlike the reconnection process. Is there is something I missing?
A regular jump is also not symmetric if there is a cutoff in the total occupation number, as is the case for example for spin models.
And actually we make the check for the matrix element as the first thing in the move - we thus do what you propose
Matthias
Hi,
Thanks for the quick reply :).
Just to make sure I understood correctly. In the shift_kink move, the Masha head move in time is not restricted to the time segment in which the occupation number in the site remains the same? As far as I can see from the code implementation, the shift_kink moves Masha head in the time segment were the occupation number remain the same.
Thanks Again :), for clarifying me all those subtle issues...
Snir
On Sat, Jan 2, 2010 at 12:11 PM, Matthias Troyer troyer@phys.ethz.chwrote:
No, it does not restrict ergodicity since shift_kink can move Masha to an adjacent segment
Matthias
On 2 Jan 2010, at 11:05, Snir Gazit wrote:
Hi,
I have yet another question :).
The Insert jump move inserts a kink only in the time segment in which the occupation number of the original and target site don't change. This leads to the restriction on the remove jump (do to detailed balance) that a kink is removed only if in the time segment between the kink and the Masha head the occupation number remains the same. Am I right? If so, doesn't this restriction compromise ergodicy because it restricts the paths in which the worm can choose? Do you think that the computational over head in inserting a kink in an arbitrary position between the last kink (before Masha) and Masha is reasonable?
Thanks allot :)
Snir
On Wed, Dec 30, 2009 at 1:23 PM, Snir Gazit snirgaz@tx.technion.ac.ilwrote:
Thanks again :)
On Wed, Dec 30, 2009 at 1:00 PM, Matthias Troyer troyer@phys.ethz.chwrote:
On 30 Dec 2009, at 11:50, Snir Gazit wrote:
Hi,
One more question regarding the implementation.
Isn't it better to check if the reconnection move is feasible? (same
like the close/open move in which IRA and MASHA should be on the same site).
I understand that the ALPS implementation keeps the jump and
reconnection process symmetric. But essentially they are not because a regular jump (where the kink is inserted before MASHA) is always possible unlike the reconnection process.
Is there is something I missing?
A regular jump is also not symmetric if there is a cutoff in the total occupation number, as is the case for example for spin models.
And actually we make the check for the matrix element as the first thing in the move - we thus do what you propose
Matthias
shift_kink shifts within a segment on the local site, but the occupations on the neighboring sites can change. That solves your ergodicity issue.
Matthias
On 2 Jan 2010, at 11:31, Snir Gazit wrote:
Hi,
Thanks for the quick reply :).
Just to make sure I understood correctly. In the shift_kink move, the Masha head move in time is not restricted to the time segment in which the occupation number in the site remains the same? As far as I can see from the code implementation, the shift_kink moves Masha head in the time segment were the occupation number remain the same.
Thanks Again :), for clarifying me all those subtle issues...
Snir
On Sat, Jan 2, 2010 at 12:11 PM, Matthias Troyer troyer@phys.ethz.ch wrote: No, it does not restrict ergodicity since shift_kink can move Masha to an adjacent segment
Matthias
On 2 Jan 2010, at 11:05, Snir Gazit wrote:
Hi,
I have yet another question :).
The Insert jump move inserts a kink only in the time segment in which the occupation number of the original and target site don't change. This leads to the restriction on the remove jump (do to detailed balance) that a kink is removed only if in the time segment between the kink and the Masha head the occupation number remains the same. Am I right? If so, doesn't this restriction compromise ergodicy because it restricts the paths in which the worm can choose? Do you think that the computational over head in inserting a kink in an arbitrary position between the last kink (before Masha) and Masha is reasonable?
Thanks allot :)
Snir
On Wed, Dec 30, 2009 at 1:23 PM, Snir Gazit snirgaz@tx.technion.ac.il wrote: Thanks again :)
On Wed, Dec 30, 2009 at 1:00 PM, Matthias Troyer troyer@phys.ethz.ch wrote:
On 30 Dec 2009, at 11:50, Snir Gazit wrote:
Hi,
One more question regarding the implementation.
Isn't it better to check if the reconnection move is feasible? (same like the close/open move in which IRA and MASHA should be on the same site). I understand that the ALPS implementation keeps the jump and reconnection process symmetric. But essentially they are not because a regular jump (where the kink is inserted before MASHA) is always possible unlike the reconnection process. Is there is something I missing?
A regular jump is also not symmetric if there is a cutoff in the total occupation number, as is the case for example for spin models.
And actually we make the check for the matrix element as the first thing in the move - we thus do what you propose
Matthias
comp-phys-alps-users@lists.phys.ethz.ch