Re: [ALPS-users] dmrg errors
Hi Matthias, it seems to me that the segfaults occur for lattices with different types of vertices. For example, if I add fields at the beginning and the end of a Heisenberg-chain, I get segmentation faults for the following parameter file LATTICE_LIBRARY="mychain.xml" LATTICE="my open chain lattice N=16" MODEL="spin" NUMBER_EIGENVALUES=1 SWEEPS=4 COMPLEX=1 L=16 h0=1.5 h1=0.0 h2=1.5 MEASURE_LOCAL[Sz]=Sz {MAXSTATES=20} Strangely, if I change some parameters, for example set h0 and h2=1.0 or VERBOSE=1 everything works fine... Here mychain.xml just defines an open chain with edge-fields: <LATTICES> <GRAPH name = "my open chain lattice N=16" dimension="1" vertices="16" edges="15"> <VERTEX id="1" type="0"><COORDINATE>0</COORDINATE></VERTEX> <VERTEX id="2" type="1"><COORDINATE>1</COORDINATE></VERTEX> <VERTEX id="3" type="1"><COORDINATE>2</COORDINATE></VERTEX> <VERTEX id="4" type="1"><COORDINATE>3</COORDINATE></VERTEX> <VERTEX id="5" type="1"><COORDINATE>4</COORDINATE></VERTEX> <VERTEX id="6" type="1"><COORDINATE>5</COORDINATE></VERTEX> <VERTEX id="7" type="1"><COORDINATE>6</COORDINATE></VERTEX> <VERTEX id="8" type="1"><COORDINATE>7</COORDINATE></VERTEX> <VERTEX id="9" type="1"><COORDINATE>8</COORDINATE></VERTEX> <VERTEX id="10" type="1"><COORDINATE>9</COORDINATE></VERTEX> <VERTEX id="11" type="1"><COORDINATE>10</COORDINATE></VERTEX> <VERTEX id="12" type="1"><COORDINATE>11</COORDINATE></VERTEX> <VERTEX id="13" type="1"><COORDINATE>12</COORDINATE></VERTEX> <VERTEX id="14" type="1"><COORDINATE>13</COORDINATE></VERTEX> <VERTEX id="15" type="1"><COORDINATE>14</COORDINATE></VERTEX> <VERTEX id="16" type="2"><COORDINATE>15</COORDINATE></VERTEX> <EDGE source="1" target="2" id="1" type="0" vector="1"/> <EDGE source="2" target="3" id="2" type="0" vector="1"/> <EDGE source="3" target="4" id="3" type="0" vector="1"/> <EDGE source="4" target="5" id="4" type="0" vector="1"/> <EDGE source="5" target="6" id="5" type="0" vector="1"/> <EDGE source="6" target="7" id="6" type="0" vector="1"/> <EDGE source="7" target="8" id="7" type="0" vector="1"/> <EDGE source="8" target="9" id="8" type="0" vector="1"/> <EDGE source="9" target="10" id="9" type="0" vector="1"/> <EDGE source="10" target="11" id="10" type="0" vector="1"/> <EDGE source="11" target="12" id="11" type="0" vector="1"/> <EDGE source="12" target="13" id="12" type="0" vector="1"/> <EDGE source="13" target="14" id="13" type="0" vector="1"/> <EDGE source="14" target="15" id="14" type="0" vector="1"/> <EDGE source="15" target="16" id="15" type="0" vector="1"/> </GRAPH> </LATTICES> Can you reproduce this error?! Sebastian On 23/07/10 15:23, Matthias Troyer wrote:
Hi Sebastian,
can you please post an example parameter file that segfaults at the beginning?
Matthias
On 23 Jul 2010, at 03:53, Sebastian Greschner wrote:
Hi Matthias,
Thank you - now dmrg gives the correct results!
Unfortunately there is still one (but probably unrelated) problem with dmrg (I am working on a Debian Lenny system - Intel Xeon): Depending on the parameters I choose (some special model parameters, Number of MAXSTATES, etc.) in the parameter file, I sometimes get segmentation faults at the beginning or at the end. Sorry - I haven't figured out yet, when this happens exactly, but maybe somebody has observed similar errors...
Best, Sebastian
On 22/07/10 17:25, Matthias Troyer wrote:
On 22 Jul 2010, at 05:33, Sebastian Greschner wrote:
Hi,
I am having some trouble with the dmrg-program (e.g. of current ALPS 2.0b3) and "complex-valued" models. It seems to me that dmrg ignores the imaginary part of model-parameters (even with COMPLEX set to true). "sparsediag" gives the correct results.
Is this a bug or a general restriction of the dmrg-program? Or is it just a problem of my model or my local installation?
Thanks in advance and greetings, Sebastian
Hi Sebastian,
I fixed one bug and the fix will show up in the MacOS and source versions of thennext nightly snapshot (Friday morning). Can you please try that?
Matthias
Hi, this is not related to multiple vertex type but due to you not specifying conserved quantum numbers (which would also speed up the code). The bug has been fixed and the will be in Monday's daily snapshot Matthias On 24 Jul 2010, at 11:19, sebastian.greschner@itp.uni-hannover.de wrote:
Hi Matthias,
it seems to me that the segfaults occur for lattices with different types of vertices. For example, if I add fields at the beginning and the end of a Heisenberg-chain, I get segmentation faults for the following parameter file
LATTICE_LIBRARY="mychain.xml" LATTICE="my open chain lattice N=16" MODEL="spin" NUMBER_EIGENVALUES=1 SWEEPS=4 COMPLEX=1 L=16 h0=1.5 h1=0.0 h2=1.5 MEASURE_LOCAL[Sz]=Sz {MAXSTATES=20}
Strangely, if I change some parameters, for example set h0 and h2=1.0 or VERBOSE=1 everything works fine...
Here mychain.xml just defines an open chain with edge-fields:
<LATTICES> <GRAPH name = "my open chain lattice N=16" dimension="1" vertices="16" edges="15"> <VERTEX id="1" type="0"><COORDINATE>0</COORDINATE></VERTEX> <VERTEX id="2" type="1"><COORDINATE>1</COORDINATE></VERTEX> <VERTEX id="3" type="1"><COORDINATE>2</COORDINATE></VERTEX> <VERTEX id="4" type="1"><COORDINATE>3</COORDINATE></VERTEX> <VERTEX id="5" type="1"><COORDINATE>4</COORDINATE></VERTEX> <VERTEX id="6" type="1"><COORDINATE>5</COORDINATE></VERTEX> <VERTEX id="7" type="1"><COORDINATE>6</COORDINATE></VERTEX> <VERTEX id="8" type="1"><COORDINATE>7</COORDINATE></VERTEX> <VERTEX id="9" type="1"><COORDINATE>8</COORDINATE></VERTEX> <VERTEX id="10" type="1"><COORDINATE>9</COORDINATE></VERTEX> <VERTEX id="11" type="1"><COORDINATE>10</COORDINATE></VERTEX> <VERTEX id="12" type="1"><COORDINATE>11</COORDINATE></VERTEX> <VERTEX id="13" type="1"><COORDINATE>12</COORDINATE></VERTEX> <VERTEX id="14" type="1"><COORDINATE>13</COORDINATE></VERTEX> <VERTEX id="15" type="1"><COORDINATE>14</COORDINATE></VERTEX> <VERTEX id="16" type="2"><COORDINATE>15</COORDINATE></VERTEX> <EDGE source="1" target="2" id="1" type="0" vector="1"/> <EDGE source="2" target="3" id="2" type="0" vector="1"/> <EDGE source="3" target="4" id="3" type="0" vector="1"/> <EDGE source="4" target="5" id="4" type="0" vector="1"/> <EDGE source="5" target="6" id="5" type="0" vector="1"/> <EDGE source="6" target="7" id="6" type="0" vector="1"/> <EDGE source="7" target="8" id="7" type="0" vector="1"/> <EDGE source="8" target="9" id="8" type="0" vector="1"/> <EDGE source="9" target="10" id="9" type="0" vector="1"/> <EDGE source="10" target="11" id="10" type="0" vector="1"/> <EDGE source="11" target="12" id="11" type="0" vector="1"/> <EDGE source="12" target="13" id="12" type="0" vector="1"/> <EDGE source="13" target="14" id="13" type="0" vector="1"/> <EDGE source="14" target="15" id="14" type="0" vector="1"/> <EDGE source="15" target="16" id="15" type="0" vector="1"/> </GRAPH> </LATTICES>
Can you reproduce this error?!
Sebastian
On 23/07/10 15:23, Matthias Troyer wrote:
Hi Sebastian,
can you please post an example parameter file that segfaults at the beginning?
Matthias
On 23 Jul 2010, at 03:53, Sebastian Greschner wrote:
Hi Matthias,
Thank you - now dmrg gives the correct results!
Unfortunately there is still one (but probably unrelated) problem with dmrg (I am working on a Debian Lenny system - Intel Xeon): Depending on the parameters I choose (some special model parameters, Number of MAXSTATES, etc.) in the parameter file, I sometimes get segmentation faults at the beginning or at the end. Sorry - I haven't figured out yet, when this happens exactly, but maybe somebody has observed similar errors...
Best, Sebastian
On 22/07/10 17:25, Matthias Troyer wrote:
On 22 Jul 2010, at 05:33, Sebastian Greschner wrote:
Hi,
I am having some trouble with the dmrg-program (e.g. of current ALPS 2.0b3) and "complex-valued" models. It seems to me that dmrg ignores the imaginary part of model-parameters (even with COMPLEX set to true). "sparsediag" gives the correct results.
Is this a bug or a general restriction of the dmrg-program? Or is it just a problem of my model or my local installation?
Thanks in advance and greetings, Sebastian
Hi Sebastian,
I fixed one bug and the fix will show up in the MacOS and source versions of thennext nightly snapshot (Friday morning). Can you please try that?
Matthias
participants (2)
-
Matthias Troyer
-
sebastian.greschner＠itp.uni-hannover.de