2D Heisenberg model with open boundary in one spatial direction
Hello all I think I should try to make my quick questions more precise (hence slightly more interesting). For a 2D Heisenberg model with open boundary in one spatial direction, can one measure the spin-spin correlations as well as the on-site magnitization as functions of distance from the open boundary using ALPS without modifying the codes dramatically? Thanks in advance! Sincerely, FJ
On Jan 21, 2009, at 5:47 AM, Fu-Jiun Jiang ITP wrote:
Hello all
I think I should try to make my quick questions more precise (hence slightly more interesting). For a 2D Heisenberg model with open boundary in one spatial direction, can one measure the spin- spin correlations as well as the on-site magnitization as functions of distance from the open boundary using ALPS without modifying the codes dramatically? Thanks in advance!
Yes, just use the ladder lattice which has just the boundary conditions you want. Matthias
Hello all Thanks for the quick response. This is exactly what I thought (and did). However to my surprise, the output file of " ladder " is so large (compared to the usual square lattices), it is impossible to " firefox " or " convert2tex " the " *.tast1.out.xml " file! Any suggestion will be appreciated! Sincerely, FJ On Wed, 21 Jan 2009, Matthias Troyer wrote:
On Jan 21, 2009, at 5:47 AM, Fu-Jiun Jiang ITP wrote:
Hello all
I think I should try to make my quick questions more precise (hence slightly more interesting). For a 2D Heisenberg model with open boundary in one spatial direction, can one measure the spin-spin correlations as well as the on-site magnitization as functions of distance from the open boundary using ALPS without modifying the codes dramatically? Thanks in advance!
Yes, just use the ladder lattice which has just the boundary conditions you want.
Matthias
From: Fu-Jiun Jiang ITP <fjjiang@itp.unibe.ch> Date: Thu, 22 Jan 2009 13:08:53 +0100 (CET)
Thanks for the quick response. This is exactly what I thought (and did). However to my surprise, the output file of " ladder " is so large (compared to the usual square lattices), it is impossible to " firefox " or " convert2tex " the " *.tast1.out.xml " file! Any suggestion will be appreciated!
Since your lattice does not have translational symmetry any more, all the combinations of two sites (N*(N-1)/2) will be measured and stored independently. If you are interested in the correlation function from a specific site (say the boundary site), you can specify parameter INITIAL_SITE to its index (e.g INITIAL_SITE = 0), which greatly reduces the output size from O(N^2) to O(N). Synge
Please forgive me for jumping into the discussion: what is the syntax precisely? Is this feature implemented in all the applications? I've tried using MEASURE_CORRELATIONS[SzSz,INITIAL_SITE=0]="Sz:Sz" with dmrg, and it gave me same N*(N-1)/2 values. Best, Zhenya Synge Todo wrote:
From: Fu-Jiun Jiang ITP <fjjiang@itp.unibe.ch> Date: Thu, 22 Jan 2009 13:08:53 +0100 (CET)
Thanks for the quick response. This is exactly what I thought (and did). However to my surprise, the output file of " ladder " is so large (compared to the usual square lattices), it is impossible to " firefox " or " convert2tex " the " *.tast1.out.xml " file! Any suggestion will be appreciated!
Since your lattice does not have translational symmetry any more, all the combinations of two sites (N*(N-1)/2) will be measured and stored independently. If you are interested in the correlation function from a specific site (say the boundary site), you can specify parameter INITIAL_SITE to its index (e.g INITIAL_SITE = 0), which greatly reduces the output size from O(N^2) to O(N).
Synge
From: Evgeny Burovskiy <burovski@alumni.umass.edu> Date: Fri, 23 Jan 2009 13:37:22 +0100
Please forgive me for jumping into the discussion: what is the syntax precisely? Is this feature implemented in all the applications? I've tried using MEASURE_CORRELATIONS[SzSz,INITIAL_SITE=0]="Sz:Sz" with dmrg, and it gave me same N*(N-1)/2 values.
The syntax is: MEASURE_CORRELATIONS[SzSz]="Sz:Sz" INITIAL_SITE=0 All the qmc applications should support this feature. But probably dmrg does not. Synge
Thanks for the clarification! And indeed, the INITIAL_SITE parameter is quietly ignored by dmrg.. I suppose there is more urgent work for you alps developers, but as a humble user, I'd dare begging you to implement this feature as soon as realistically possible. If one needs a valid use case, here it is: as soon as I try running dmrg for sensible system sizes, e.g. L=80, the dmrg sweeping takes 2--3 hours, while calculating a full correlation function takes rather 16--20 hours. As far as I'm concerned, being able to only calculate O(L) values for correlators would be very very much welcome--- at least if one needs to check the large-distance asymptotics of a certain correlator, e.g. to see whether it's a power law or an exponential. I further guess that I'm not the only one trying to use alps dmrg for calculating correlations, hence the effort of implementing the feature would be useful for at least a few current and future users. And as a side request: would it be possible to add a note to the alps wiki explaining this INITIAL_SITE trick? Best, Zhenya On Fri, Jan 23, 2009 at 4:02 PM, Synge Todo <wistaria@comp-phys.org> wrote:
From: Evgeny Burovskiy <burovski@alumni.umass.edu> Date: Fri, 23 Jan 2009 13:37:22 +0100
Please forgive me for jumping into the discussion: what is the syntax
precisely?
Is this feature implemented in all the applications? I've tried using MEASURE_CORRELATIONS[SzSz,INITIAL_SITE=0]="Sz:Sz" with dmrg, and it gave me same N*(N-1)/2 values.
The syntax is:
MEASURE_CORRELATIONS[SzSz]="Sz:Sz" INITIAL_SITE=0
All the qmc applications should support this feature. But probably dmrg does not.
Synge
Hi Zhenya, that trick is an unsupported and undocumented feature as you found out. I'll talk to Adrian about DMRG. INITIAL_SITE will not be useful there since with open boundary conditions you want it symmetric around the center, but we'll find a solution. Matthias On 26 Jan 2009, at 03:00, Evgeny Burovskiy wrote:
Thanks for the clarification!
And indeed, the INITIAL_SITE parameter is quietly ignored by dmrg..
I suppose there is more urgent work for you alps developers, but as a humble user, I'd dare begging you to implement this feature as soon as realistically possible. If one needs a valid use case, here it is: as soon as I try running dmrg for sensible system sizes, e.g. L=80, the dmrg sweeping takes 2--3 hours, while calculating a full correlation function takes rather 16--20 hours. As far as I'm concerned, being able to only calculate O(L) values for correlators would be very very much welcome--- at least if one needs to check the large-distance asymptotics of a certain correlator, e.g. to see whether it's a power law or an exponential. I further guess that I'm not the only one trying to use alps dmrg for calculating correlations, hence the effort of implementing the feature would be useful for at least a few current and future users.
And as a side request: would it be possible to add a note to the alps wiki explaining this INITIAL_SITE trick?
Best, Zhenya
On Fri, Jan 23, 2009 at 4:02 PM, Synge Todo <wistaria@comp-phys.org> wrote: From: Evgeny Burovskiy <burovski@alumni.umass.edu> Date: Fri, 23 Jan 2009 13:37:22 +0100
Please forgive me for jumping into the discussion: what is the
syntax precisely?
Is this feature implemented in all the applications? I've tried using MEASURE_CORRELATIONS[SzSz,INITIAL_SITE=0]="Sz:Sz" with dmrg, and it gave me same N*(N-1)/2 values.
The syntax is:
MEASURE_CORRELATIONS[SzSz]="Sz:Sz" INITIAL_SITE=0
All the qmc applications should support this feature. But probably dmrg does not.
Synge
Hello Matthias, I still believe this might come handy for DMRG, even though open boundary conditions complicate things. Maybe I'm just much too naive, but I can't see why having INITIAL_SITE=0 isn't enough. If a user has the correlations for all the pairs of sites 0--0 0--1 ... 0--L-1 she can check whether there is mirror symmetry or not. Best, Zhenya On Mon, Jan 26, 2009 at 5:03 PM, Matthias Troyer <troyer@phys.ethz.ch>wrote:
Hi Zhenya, that trick is an unsupported and undocumented feature as you found out. I'll talk to Adrian about DMRG. INITIAL_SITE will not be useful there since with open boundary conditions you want it symmetric around the center, but we'll find a solution.
Matthias
On 26 Jan 2009, at 03:00, Evgeny Burovskiy wrote:
Thanks for the clarification!
And indeed, the INITIAL_SITE parameter is quietly ignored by dmrg..
I suppose there is more urgent work for you alps developers, but as a humble user, I'd dare begging you to implement this feature as soon as realistically possible. If one needs a valid use case, here it is: as soon as I try running dmrg for sensible system sizes, e.g. L=80, the dmrg sweeping takes 2--3 hours, while calculating a full correlation function takes rather 16--20 hours. As far as I'm concerned, being able to only calculate O(L) values for correlators would be very very much welcome--- at least if one needs to check the large-distance asymptotics of a certain correlator, e.g. to see whether it's a power law or an exponential. I further guess that I'm not the only one trying to use alps dmrg for calculating correlations, hence the effort of implementing the feature would be useful for at least a few current and future users.
And as a side request: would it be possible to add a note to the alps wiki explaining this INITIAL_SITE trick?
Best, Zhenya
On Fri, Jan 23, 2009 at 4:02 PM, Synge Todo <wistaria@comp-phys.org>wrote:
From: Evgeny Burovskiy <burovski@alumni.umass.edu> Date: Fri, 23 Jan 2009 13:37:22 +0100
Please forgive me for jumping into the discussion: what is the syntax
precisely?
Is this feature implemented in all the applications? I've tried using MEASURE_CORRELATIONS[SzSz,INITIAL_SITE=0]="Sz:Sz" with dmrg, and it gave me same N*(N-1)/2 values.
The syntax is:
MEASURE_CORRELATIONS[SzSz]="Sz:Sz" INITIAL_SITE=0
All the qmc applications should support this feature. But probably dmrg does not.
Synge
participants (4)
-
Evgeny Burovskiy
-
Fu-Jiun Jiang ITP
-
Matthias Troyer
-
Synge Todo