update docs

Author: hczhai

docs/source/tutorial/hubbard.ipynb

@@ -78,7 +78,7 @@
    "source": [
     "### Initialization\n",
     "\n",
-    "The SZ mode should be the most convenient for the Hubbard model. We first set a driver using this mode. ``scratch`` sets a folder for writing temparary scratch files. ``symm_type`` sets the symmetry mode (and whether the complex number should be used. Here we allow only real numbers). ``n_threads`` sets the number of threads for shared-memory parallelization.\n",
+    "The SZ mode should be the most convenient for the Hubbard model. We first set a driver using this mode. ``scratch`` sets a folder for writing temporary scratch files. When running multiple independent DMRG jobs simultaneously, each job should have a distinct scratch folder name to avoid file I/O conflicts. ``symm_type`` sets the symmetry mode (and whether the complex number should be used. Here we allow only real numbers). ``n_threads`` sets the number of threads for shared-memory parallelization.\n",
     "\n",
     "Next, we use the ``initialize_system`` method to set the property of the target wavefunction (namely, which symmetry sector is targeted).\n",
     "\n",

docs/source/tutorial/qc-hamiltonians.ipynb

@@ -75,7 +75,7 @@
         "\n",
         "Here we use ``get_rhf_integrals`` function to get the integrals. Note that in order to do DMRG in a CASCI space, one can set the ``ncore`` (number of core orbitals) and ``ncas`` (number of active orbitals) parameters in ``get_*_integrals``. ``ncas=None`` will include all orbitals in DMRG.\n",
         "\n",
-        "For medium to large scale DMRG calculations, it is highly recommended to use a scratch space with high IO speed rather than the ``./tmp`` used in the following example. One also needs to set a suitable ``stack_mem`` in the ``DMRGDriver`` constructor to set the memory used for storing renormalized operators (in bytes). The default is ``stack_mem=int(1024**3)`` (1 GB). For medium scale calculations 10 to 30 GB might be required.\n",
+        "For medium to large scale DMRG calculations, it is highly recommended to use a scratch space with high IO speed rather than the ``./tmp`` used in the following example. When running multiple independent DMRG jobs simultaneously, each job should have a distinct scratch folder name to avoid file I/O conflicts. One also needs to set a suitable ``stack_mem`` in the ``DMRGDriver`` constructor to set the memory used for storing renormalized operators (in bytes). The default is ``stack_mem=int(1024**3)`` (1 GB). For medium scale calculations 10 to 30 GB might be required.\n",
         "\n",
         "For the meaning of DMRG parameters, please have a look at the [Hubbard - Run DMRG](https://block2.readthedocs.io/en/latest/tutorial/hubbard.html#Run-DMRG) page."
       ]

src/dmrg/mps.hpp

@@ -536,7 +536,7 @@ template <typename S> struct MPSInfo {
                 right_probs[i]->probs[j] *= x;
             }
         }
-        // adjusted temparary fci dims
+        // adjusted temporary fci dims
         vector<shared_ptr<StateInfo<S>>> left_dims_fci_t(n_sites + 1);
         vector<shared_ptr<StateInfo<S>>> right_dims_fci_t(n_sites + 1);
         for (int i = 0; i < n_sites + 1; i++) {