problem with worm algorithm
Hi, I have encountered problem with worm algorithm, and I would be very grateful for help. I'm using ALPS 1.3.4, and the problem is as follows. In low temperatures (around 0.05 with U=1) and hopping parameter around 0.2 simulation runs normally and then just stops - but there is no error. So it seems to be everything ok, simulation status is "running" - but there is no progress. If I rerun the simulation on the same parameters few times, it eventually works to the end. Problem intesifies with growing hopping parameter J - for J=0.1 this problem happens very rarely to me. I would really appreciate any help. Best regards Jakub Kolasinski
Can you please post an input file that shows the problem? On 11 Oct 2010, at 21:26, Jakub Kolasinski wrote:
Hi, I have encountered problem with worm algorithm, and I would be very grateful for help. I'm using ALPS 1.3.4, and the problem is as follows. In low temperatures (around 0.05 with U=1) and hopping parameter around 0.2 simulation runs normally and then just stops - but there is no error. So it seems to be everything ok, simulation status is "running" - but there is no progress. If I rerun the simulation on the same parameters few times, it eventually works to the end. Problem intesifies with growing hopping parameter J - for J=0.1 this problem happens very rarely to me. I would really appreciate any help.
Best regards Jakub Kolasinski
Hi I ran some simulations using the worm-algorithm for softcore bosons (Nmax = 2) on the kagome for determining the MI-SF transition. I get convergent results (3x3, 6x6, 9x9 lattices) and a just-ok scaling law (with 600000 sweeps and 150000 thermalization); my understanding is that there should be no sign-problem for this algorithm on this model. Can someone confirm this, or must I experiment with the sweeps and thermalization? Thanks, Vipin
On 15 Oct 2010, at 02:37, Vipin Varma wrote:
Hi
I ran some simulations using the worm-algorithm for softcore bosons (Nmax = 2) on the kagome for determining the MI-SF transition. I get convergent results (3x3, 6x6, 9x9 lattices) and a just-ok scaling law (with 600000 sweeps and 150000 thermalization); my understanding is that there should be no sign-problem for this algorithm on this model. Can someone confirm this, or must I experiment with the sweeps and thermalization?
Thanks, Vipin
Hi Vipin, if the hopping terms all have negative amplitude (positive t) then there will be no sign problem. 3x3 seems to be too small to get any reasonable scaling 150000 thermalization might be a bit excessive. What do you get for the autocorrelation times of the superfluid density (winding)? Matthias
Hi Matthias. I get autocorrelation times ranging from 1.96 - 51.5. I'm guessing that they are in the same units as the thermalization? However, I do get 5 infinities for Tau out of 51 runs: 4 "nan" for 8x8 (t/U = 0.025, 0.03, 0.035, 0.04), and 1 "nan" for 3x3 (for t/U = 0.025) which can be ignored for scaling probably. Vipin On Sun, 17 Oct 2010, Matthias Troyer wrote:
On 15 Oct 2010, at 02:37, Vipin Varma wrote:
Hi
I ran some simulations using the worm-algorithm for softcore bosons (Nmax = 2) on the kagome for determining the MI-SF transition. I get convergent results (3x3, 6x6, 9x9 lattices) and a just-ok scaling law (with 600000 sweeps and 150000 thermalization); my understanding is that there should be no sign-problem for this algorithm on this model. Can someone confirm this, or must I experiment with the sweeps and thermalization?
Thanks, Vipin
Hi Vipin,
if the hopping terms all have negative amplitude (positive t) then there will be no sign problem. 3x3 seems to be too small to get any reasonable scaling 150000 thermalization might be a bit excessive. What do you get for the autocorrelation times of the superfluid density (winding)?
Matthias
On 18 Oct 2010, at 05:51, Vipin Varma wrote:
Hi Matthias.
I get autocorrelation times ranging from 1.96 - 51.5. I'm guessing that they are in the same units as the thermalization?
Yes, that means that something like 10000 sweeps should be plenty here, since you are still close to the SF phase.
However, I do get 5 infinities for Tau out of 51 runs: 4 "nan" for 8x8 (t/U = 0.025, 0.03, 0.035, 0.04), and 1 "nan" for 3x3 (for t/U = 0.025) which can be ignored for scaling probably.
Is the error at these quantities exactly 0? I don't fully understand what your question is, though. Matthias
Vipin
On Sun, 17 Oct 2010, Matthias Troyer wrote:
On 15 Oct 2010, at 02:37, Vipin Varma wrote:
Hi
I ran some simulations using the worm-algorithm for softcore bosons (Nmax = 2) on the kagome for determining the MI-SF transition. I get convergent results (3x3, 6x6, 9x9 lattices) and a just-ok scaling law (with 600000 sweeps and 150000 thermalization); my understanding is that there should be no sign-problem for this algorithm on this model. Can someone confirm this, or must I experiment with the sweeps and thermalization?
Thanks, Vipin
Hi Vipin,
if the hopping terms all have negative amplitude (positive t) then there will be no sign problem. 3x3 seems to be too small to get any reasonable scaling 150000 thermalization might be a bit excessive. What do you get for the autocorrelation times of the superfluid density (winding)?
Matthias
Yes, the errors and the mean values are zero. The question was: this probably occurs because the system is in the MI phase at these hopping values? Vipin On Mon, 18 Oct 2010, Matthias Troyer wrote:
On 18 Oct 2010, at 05:51, Vipin Varma wrote:
Hi Matthias.
I get autocorrelation times ranging from 1.96 - 51.5. I'm guessing that they are in the same units as the thermalization?
Yes, that means that something like 10000 sweeps should be plenty here, since you are still close to the SF phase.
However, I do get 5 infinities for Tau out of 51 runs: 4 "nan" for 8x8 (t/U = 0.025, 0.03, 0.035, 0.04), and 1 "nan" for 3x3 (for t/U = 0.025) which can be ignored for scaling probably.
Is the error at these quantities exactly 0?
I don't fully understand what your question is, though.
Matthias
Vipin
On Sun, 17 Oct 2010, Matthias Troyer wrote:
On 15 Oct 2010, at 02:37, Vipin Varma wrote:
Hi
I ran some simulations using the worm-algorithm for softcore bosons (Nmax = 2) on the kagome for determining the MI-SF transition. I get convergent results (3x3, 6x6, 9x9 lattices) and a just-ok scaling law (with 600000 sweeps and 150000 thermalization); my understanding is that there should be no sign-problem for this algorithm on this model. Can someone confirm this, or must I experiment with the sweeps and thermalization?
Thanks, Vipin
Hi Vipin,
if the hopping terms all have negative amplitude (positive t) then there will be no sign problem. 3x3 seems to be too small to get any reasonable scaling 150000 thermalization might be a bit excessive. What do you get for the autocorrelation times of the superfluid density (winding)?
Matthias
On 18 Oct 2010, at 06:16, Vipin Varma wrote:
Yes, the errors and the mean values are zero. The question was: this probably occurs because the system is in the MI phase at these hopping values?
Yes, deep inside so that there is no winding at all throughout the simulation Matthias
participants (3)
-
Jakub Kolasinski
-
Matthias Troyer
-
Vipin Varma