Add Tree Tensor Networks

.circleci/config.yml

@@ -40,7 +40,7 @@ jobs:
           command: |
               pip install pytest-xdist pytest-cov
               export RENO_NUM_THREADS=1
-              pytest -n 4 --durations=0 --cov=renormalizer
+              pytest -n 4 --durations=0 --cov=renormalizer renormalizer
               pip install primme==3.2.*
               pytest --durations=0 renormalizer/mps/tests/test_gs.py::test_multistate --cov=renormalizer --cov-append
       - run:
@@ -54,12 +54,6 @@ jobs:
       - run:
           name: Build docs
           command: cd doc; make html
-      - run:
-          name: Codecov report
-          command: |
-              curl -Os https://uploader.codecov.io/latest/linux/codecov
-              chmod +x codecov
-              ./codecov
       - persist_to_workspace:
           root: doc/
           paths: html

.github/workflows/ci.yml

@@ -23,7 +23,7 @@ jobs:
         run: |
           pip install pytest-xdist pytest-cov
           export RENO_NUM_THREADS=1
-          pytest -n 4 --durations=0 --cov=renormalizer
+          pytest -n 4 --durations=0 --cov=renormalizer renormalizer
           pip install primme==3.2.*
           pytest --durations=0 renormalizer/mps/tests/test_gs.py::test_multistate --cov=renormalizer --cov-append
       - name: run examples
@@ -38,4 +38,6 @@ jobs:
           cd doc
           make html
       - name: Upload coverage to Codecov
-        uses: codecov/codecov-action@v3
+        uses: codecov/codecov-action@v4
+        env:
+          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}

example/run.sh

@@ -1,6 +1,15 @@
 export PYTHONPATH=../:PYTHONPATH
 code=0
-for python_args in fmo.py sbm.py h2o_qc.py "dynamics.py std.yaml" "transport_kubo.py std.yaml"; do
+for python_args in fmo.py \
+                   sbm.py \
+                   h2o_qc.py \
+                   "dynamics.py std.yaml" \
+                   "transport_kubo.py std.yaml" \
+                   ./ttns/junction_zt.py \
+                   "./ttns/junction_ft.py 32 1 100" \
+                   "./ttns/sbm_zt.py 050 001 050"\
+                   ./ttns/sbm_ft.py
+do
     echo ============================$python_args=============================
     timeout 20s python $python_args
     exit_code=$?

example/ttns/junction_ft.py

@@ -0,0 +1,237 @@
+import sys
+
+from renormalizer import BasisHalfSpin, Op, Quantity, BasisSHO, BasisDummy
+from renormalizer.mps.mps import expand_bond_dimension_general
+from renormalizer.sbm import ColeDavidsonSDF
+from renormalizer.utils.configs import (
+    EvolveMethod,
+    EvolveConfig,
+    CompressConfig,
+    CompressCriteria,
+)
+from renormalizer.utils import constant, log
+from renormalizer.utils.log import package_logger as logger
+from renormalizer.tn import BasisTree, TreeNodeBasis, TTNS, TTNO
+
+import numpy as np
+
+
+Ms_str = sys.argv[1]  # 8 16 24 32
+initial_str = sys.argv[2]  # 1 0
+temperature_str = sys.argv[3]  # 000 100 200 300
+
+dump_dir = "./"
+job_name = f"Ms{Ms_str}_i{initial_str}_temperature{temperature_str}"  ####################
+log.register_file_output(dump_dir + job_name + ".log", mode="w")
+
+Ms = int(Ms_str)
+
+n_ph_mode = 1000
+omega_c = Quantity(500, "cm-1").as_au()
+ita = Quantity(2000, "cm-1").as_au() / 2
+beta = 0.5
+upper_limit = Quantity(1, "eV").as_au() * 10
+logger.info(("phonon parameters", omega_c, ita, beta, upper_limit))
+sdf = ColeDavidsonSDF(ita, omega_c, beta, upper_limit)
+w, c2 = sdf.Wang1(n_ph_mode)
+c = np.sqrt(c2)
+logger.info(w)
+logger.info(c)
+
+reno = sdf.reno(w[-1])
+logger.info(f"renormalization constant: {reno}")
+
+temperature = Quantity(int(temperature_str), "K").to_beta()
+logger.info(f"temperature beta: {temperature}")
+
+n_e_mode = 320
+
+beta_e = Quantity(1, "eV").as_au() * reno
+alpha_e = Quantity(0.2, "eV").as_au() * reno
+v = 0.1 * reno
+mu_l = Quantity(v / 2, "eV").as_au()
+mu_r = Quantity(-v / 2, "eV").as_au()
+
+
+e_k = np.arange(1, n_e_mode + 1) / (n_e_mode + 1) * 4 * beta_e - 2 * beta_e
+rho_e = 1 / (e_k[1] - e_k[0])
+e_k_l = e_k - mu_l
+e_k_r = e_k - mu_r
+
+mode_with_e = [(f"L{i}", e) for i, e in enumerate(e_k_l)] + [(f"R{i}", e) for i, e in enumerate(e_k_r)]
+mode_with_e.sort(key=lambda x: x[1])
+logger.info(mode_with_e)
+
+basis = []
+first_positive = True
+for mode, e in mode_with_e:
+    if e > 0 and first_positive:
+        first_positive = False
+        basis.append(BasisHalfSpin("s"))
+    basis.append(BasisHalfSpin((mode, "p")))
+    basis.append(BasisHalfSpin((mode, "q")))
+
+dofs = [b.dofs[0] for b in basis]
+logger.info(dofs)
+s_idx = dofs.index("s")
+basis_tree_l = BasisTree.binary_mctdh(basis[:s_idx], dummy_label="EL-dummy")
+basis_tree_r = BasisTree.binary_mctdh(basis[s_idx + 1 :], dummy_label="ER-dummy")
+
+
+ham_terms = []
+i_l_terms = []
+i_r_terms = []
+for mode, e in mode_with_e:
+    if mode[0] == "L":
+        mu = mu_l
+        i_terms = i_l_terms
+    else:
+        assert mode[0] == "R"
+        mu = mu_r
+        i_terms = i_r_terms
+
+    ham_terms.append(Op("+ -", (mode, "p"), e + mu))
+    ham_terms.append(Op("+ -", (mode, "q"), -(e + mu)))
+    v2 = alpha_e**2 / beta_e**2 * np.sqrt(4 * beta_e**2 - (e + mu) ** 2) / 2 / np.pi / rho_e
+    v = np.sqrt(v2)
+    theta = np.arctan(np.exp(-temperature * e / 2))
+    idx = dofs.index((mode, "p"))
+    if idx < s_idx:
+        z_idx = list(range(idx + 1, s_idx))
+    else:
+        assert s_idx < idx
+        z_idx = list(range(s_idx + 1, idx))
+    z_dofs = [dofs[i] for i in z_idx]
+    op1 = Op(
+        "+ " + "Z " * len(z_idx) + "-",
+        [(mode, "p")] + z_dofs + ["s"],
+        v * np.cos(theta),
+    )
+    op2 = Op(
+        "- " + "Z " * len(z_idx) + "+",
+        [(mode, "p")] + z_dofs + ["s"],
+        v * np.cos(theta),
+    )
+    idx = dofs.index((mode, "q"))
+    if idx < s_idx:
+        z_idx = list(range(idx + 1, s_idx))
+    else:
+        assert s_idx < idx
+        z_idx = list(range(s_idx + 1, idx))
+    z_dofs = [dofs[i] for i in z_idx]
+    op3 = Op(
+        "- " + "Z " * len(z_idx) + "-",
+        [(mode, "q")] + z_dofs + ["s"],
+        v * np.sin(theta),
+    )
+    op4 = Op(
+        "+ " + "Z " * len(z_idx) + "+",
+        [(mode, "q")] + z_dofs + ["s"],
+        v * np.sin(theta),
+    )
+    ham_terms.extend([op1, op2, op3, op4])
+    # move 1j to expectation
+    i_terms.extend(op2 - op1 + op4 - op3)
+
+initial_occupied = initial_str == "1"
+
+if initial_occupied:
+    ham_terms.append(Op("+ -", "s", qn=[0, 0], factor=-4 * (c**2 / w**2).sum()))
+
+# vibrations at last
+
+# boson energy
+for imode in range(n_ph_mode):
+    op1 = Op(r"p^2", f"v_{imode}_p", factor=0.5, qn=0)
+    op2 = Op(r"x^2", f"v_{imode}_p", factor=0.5 * w[imode] ** 2, qn=0)
+    ham_terms.extend([op1, op2])
+    op1 = Op(r"p^2", f"v_{imode}_q", factor=-0.5, qn=0)
+    op2 = Op(r"x^2", f"v_{imode}_q", factor=-0.5 * w[imode] ** 2, qn=0)
+    ham_terms.extend([op1, op2])
+
+theta_array = np.arctanh(np.exp(-w * temperature / 2))
+# system-boson coupling
+for imode in range(n_ph_mode):
+    sys_op = Op("+ -", "s", qn=[0, 0])
+    if initial_occupied:
+        sys_op = sys_op - Op.identity("s")
+
+    theta = theta_array[imode]
+    op1 = sys_op * Op(r"x", f"v_{imode}_p", factor=2 * c[imode] * np.cosh(theta), qn=[0])
+    op2 = sys_op * Op(r"x", f"v_{imode}_q", factor=2 * c[imode] * np.sinh(theta), qn=[0])
+    ham_terms.extend(op1 + op2)
+
+nbas = np.max([16 * c2 / w**3 * np.cosh(theta_array) ** 2, np.ones(n_ph_mode) * 4], axis=0)
+nbas = np.round(nbas).astype(int)
+nbas = np.min([nbas, np.ones(n_ph_mode) * 512], axis=0)
+logger.info(nbas)
+basis_list_phonon = []
+for imode in range(n_ph_mode):
+    basis_list_phonon.append(BasisSHO(f"v_{imode}_p", w[imode], int(nbas[imode])))
+    basis_list_phonon.append(BasisSHO(f"v_{imode}_q", w[imode], int(nbas[imode])))
+
+labels = np.array([[nbas > Ms], [nbas > Ms]]).T.ravel()
+basis_tree_phonon = BasisTree.binary_mctdh(
+    basis_list_phonon,
+    contract_primitive=True,
+    contract_label=labels,
+    dummy_label="phonon-dummy",
+)
+node1 = TreeNodeBasis([BasisDummy("dummy")])
+node1.add_child([basis_tree_l.root, basis_tree_r.root])
+node2 = TreeNodeBasis([basis[s_idx]])
+node2.add_child([node1, basis_tree_phonon.root])
+basis_tree = BasisTree(node2)
+basis_tree.print(logger.info)
+
+
+# model = Model(basis, ham_terms)
+ttno = TTNO(basis_tree, ham_terms)
+i_l_mpo = TTNO(basis_tree, i_l_terms)
+i_r_mpo = TTNO(basis_tree, i_r_terms)
+# n_l_mpo = TTNO(basis_tree, terms=[Op("+ -", f"L{i}") for i in range(n_e_mode)])
+# n_r_mpo = TTNO(basis_tree, terms=[Op("+ -", f"R{i}") for i in range(n_e_mode)])
+n_s_mpo = TTNO(basis_tree, terms=Op("+ -", "s"))
+ttno.print_shape(False, logger.info)
+i_l_mpo.print_shape(False, logger.info)
+i_r_mpo.print_shape(False, logger.info)
+# n_r_mpo.print_shape(False, logger.info)
+# n_s_mpo.print_shape(False, logger.info)
+# 0 - [1, 0] (spin up) means occupation, 1 - [0, 1] (spin down) means unoccupation
+# initial condition is taken care of in the Hamiltonian
+condition = {dofs[i]: 1 for i in range(len(dofs))}
+if initial_occupied:
+    condition["s"] = 0
+else:
+    condition["s"] = 1
+
+ttns = TTNS(basis_tree, condition=condition)
+ttns.compress_config = CompressConfig(CompressCriteria.fixed, max_bonddim=Ms)
+ttns = expand_bond_dimension_general(ttns, ttno, ex_mps=None)
+ttns.evolve_config = EvolveConfig(EvolveMethod.tdvp_ps)
+ttns.print_shape(print_function=logger.info, full=False)
+
+step = 0.5 * constant.fs2au
+# step = 5
+nsteps = 200
+au2muA = 6.623618237510e3
+i = 0
+current_list = []
+while True:
+    i_l = (1j * ttns.expectation(i_l_mpo)).real
+    i_r = (1j * ttns.expectation(i_r_mpo)).real
+    # n_l = mps.expectation(n_l_mpo)
+    # n_r = mps.expectation(n_r_mpo)
+    n_s = ttns.expectation(n_s_mpo)
+    logger.info((i, ttns.bond_dims))
+    current = (i_r - i_l) / 2 * au2muA
+    # logger.info((n_l, n_r, n_s, i_l*au2muA, i_r*au2muA, current))
+    logger.info((n_s, i_l * au2muA, i_r * au2muA, current))
+    current_list.append(current)
+    i += 1
+    if i == nsteps:
+        break
+    if i > 0:
+        ttns.evolve_config = EvolveConfig(EvolveMethod.tdvp_ps)
+    ttns = ttns.evolve(ttno, step)
+logger.info(current_list)