-
Notifications
You must be signed in to change notification settings - Fork 14
Known issues
-
Thermalization time of worm code
As described in the paper, in the thermalization process, we estimate the relative volumes of the worm spaces iteratively using the Wang-Landau algorithm. The estimates may converge slowly, depending on the details of the impurity model. Once we fail to obtain the converged results, we will some weird behavior. For example, we may see some error message like "Configuration space for **** was not visited" or "No observable found with the name: ***". There are three possible solutions:
(1) Run the code longer so that the estimates converge well.
(2) Run the code with multiple CPU cores so that the estimates are averaged over multiple processes at the end of the thermalization process.
(3) Turn off the measurement of unnecessary observables to improve the speed. They will help. -
Instability of worm measurement of equal-time single-particle Green's function
When the mean number of electrons is very close to 0, the measurement of equal-time single-particle/two-particle Green's functions becomes unstable. You may see error messages something like "Configuration space for **** was not visited in measurement steps. Thermalization time may be too short." This means that you failed to obtain converged estimates of the volumes of their configuration spaces in the given thermalization time. This is because the volumes of their configuration spaces vanish.
In practical, one may increase the thermalization time, or (if it does not help) change the chemical potential. This is a technical issue in the current implementation. We will solve this issue in the next minor update. -
Some libraries are not found at runtime. When you install the executalbe to your installation path by "make install", CMake removes the paths of dynamic libraries from the binary. When you launch "/path/to/install/dir/hybmat", some dynamic libraries which were visible in the build may not be found. In this case, please set your environment variables correctly (e.g., LD_LIBRARY_PATH) so that the system can find these libraries at runtime. More information is found [here] (https://cmake.org/Wiki/CMake_RPATH_handling).
-
"make test" fails in unittest_fu for Intel C++ compiler 2017. This may be related to a bug in the optimizer of the compiler. You can lower the optimization level to O1 by setting "cmake -DCMAKE_BUILD_TYPE=RelWithDebInfo".