Changes on tenpy_sz_convention.py

Author: Charlespkuer

examples/dmrg_tenpy_integration.py renamed to examples/tenpy_sz_convention.py

@@ -1,8 +1,14 @@
 """
-Demonstration of TeNPy-DMRG and TensorCircuit integration
-1. Compute ground state (MPS) of 1D Transverse Field Ising model using TeNPy
-2. Convert MPS to TensorCircuit's QuOperator
-3. Initialize MPSCircuit with converted state and verify results
+Demonstrates the different internal basis conventions between
+TeNPy's TFIChain and XXZChain models and showcases a robust
+method for handling these inconsistencies when converting to TensorCircuit.
+
+1.  TFIChain: Shows a direct conversion works perfectly.
+2.  XXZChain:
+    a. Runs DMRG to obtain a non-trivial ground state.
+    b. Shows that direct conversion leads to incorrect expectation values for correlation functions.
+    c. Demonstrates that applying a layer of X-gates in TensorCircuit
+3.  Tensor Dissection: Provides definitive proof of the differing internal basis conventions between the two models.
 """
 
 import numpy as np
@@ -14,7 +20,6 @@
 import tensorcircuit as tc
 
 print("Scenario 1: Antiferromagnetic State (TFIChain)")
-
 L = 10
 afm_model_params = {"L": L, "bc_MPS": "finite", "J": 1.0, "g": 0.0, "conserve": None}
 afm_M = TFIChain(afm_model_params)
@@ -33,7 +38,6 @@
         for i in range(L)
     ]
 )
-
 print("\nAntiferromagnetic state site-by-site magnetization comparison:")
 print("TeNPy:", np.round(mag_z_afm_tenpy, 8))
 print("TC:   ", np.round(mag_z_afm_tc, 8))
@@ -42,21 +46,17 @@
     "\n[SUCCESS] TFI-based Antiferromagnetic state matches perfectly with the pure converter."
 )
 
-# --- Scenario 2: XXZChain Model ---
-print("Scenario 2: XXZChain Model")
 
+print("Scenario 2: XXZChain Model")
 xxz_model_params = {"L": L, "bc_MPS": "finite", "Jxx": 1.0, "Jz": 0.5, "hz": 0.1}
 xxz_M = XXZChain(xxz_model_params)
-rng = np.random.default_rng(42)
-random_state = rng.choice(["up", "down"], size=L).tolist()
-print(f"Testing with a random product state: {random_state}")
-
+example_state = ["up", "down", "up", "up", "down", "down", "up", "down", "down", "up"]
+print(f"Testing with a random product state: {example_state}")
 psi_rand_xxz = MPS.from_product_state(
-    xxz_M.lat.mps_sites(), random_state, bc=xxz_M.lat.bc_MPS
+    xxz_M.lat.mps_sites(), example_state, bc=xxz_M.lat.bc_MPS
 )
 tc_rand_xxz_state = tc.quantum.tenpy2qop(psi_rand_xxz)
 circuit_rand_xxz = tc.MPSCircuit(L, wavefunction=tc_rand_xxz_state)
-
 mag_z_rand_xxz_tenpy = psi_rand_xxz.expectation_value("Sz")
 mag_z_rand_xxz_tc = np.array(
     [
@@ -67,11 +67,9 @@
         for i in range(L)
     ]
 )
-
 print("\nXXZ-based random state site-by-site magnetization comparison:")
 print("TeNPy:", np.round(mag_z_rand_xxz_tenpy, 8))
 print("TC:   ", np.round(mag_z_rand_xxz_tc, 8))
-
 try:
     np.testing.assert_allclose(mag_z_rand_xxz_tenpy, mag_z_rand_xxz_tc, atol=1e-5)
 except AssertionError as e:
@@ -85,12 +83,10 @@
         "[SUCCESS] Test passes after applying the sign correction for the XXZChain model."
     )
 
-# --- Scenario 3: Tensor Dissection for Both Models ---
-print("Scenario 3: Tensor Dissection for Both Models")
 
+print("Scenario 3: Tensor Dissection for Both Models")
 simple_L = 2
 simple_labels = ["up", "down"]
-
 print("\nDissecting TFIChain-based Tensors")
 sites_tfi = afm_M.lat.mps_sites()[:simple_L]
 psi_simple_tfi = MPS.from_product_state(sites_tfi, simple_labels, bc="finite")
@@ -99,7 +95,6 @@
     print(
         f"For '{label}', TFIChain internal tensor has non-zero at physical index {np.where(B_tensor[0,:,0] != 0)[0][0]}"
     )
-
 print("\nDissecting XXZChain-based Tensors")
 sites_xxz = xxz_M.lat.mps_sites()[:simple_L]
 psi_simple_xxz = MPS.from_product_state(sites_xxz, simple_labels, bc="finite")
@@ -108,9 +103,7 @@
     print(
         f"For '{label}', XXZChain internal tensor has non-zero at physical index {np.where(B_tensor[0,:,0] != 0)[0][0]}"
     )
-
-
-print("\nFinal Conclusion")
+print("\n Conclusion")
 print("The dissection above shows the root cause of the different behaviors:")
 print(
     "  - TFIChain's 'up' maps to index 0, 'down' to index 1. This matches TC's standard."
@@ -124,4 +117,47 @@
 )
 print("or to create model-specific converters.")
 
+
+print("--- Scenario 3: Correcting XXZChain DMRG state with X-gates ---")
+
+L = 10
+xxz_model_params = {"L": L, "bc_MPS": "finite", "Jxx": 1.0, "Jz": 1.0, "conserve": None}
+xxz_M = XXZChain(xxz_model_params)
+psi0_xxz = MPS.from_product_state(
+    xxz_M.lat.mps_sites(), ["up", "down"] * (L // 2), bc=xxz_M.lat.bc_MPS
+)
+dmrg_params = {"max_sweeps": 10, "trunc_params": {"chi_max": 64}}
+eng = dmrg.TwoSiteDMRGEngine(psi0_xxz, xxz_M, dmrg_params)
+E, psi_gs_xxz = eng.run()
+print(f"XXZ DMRG finished. Ground state energy: {E:.10f}")
+
+state_raw_quvector = tc.quantum.tenpy2qop(psi_gs_xxz)
+
+i, j = L // 2 - 1, L // 2
+corr_tenpy = psi_gs_xxz.correlation_function("Sz", "Sz", sites1=[i], sites2=[j])[0, 0]
+print("\nApplying X-gate to each qubit to correct the basis convention...")
+circuit_to_be_corrected = tc.MPSCircuit(L, wavefunction=state_raw_quvector)
+
+for k in range(L):
+    circuit_to_be_corrected.x(k)
+
+corr_tc_corrected = (
+    tc.backend.real(
+        circuit_to_be_corrected.expectation((tc.gates.z(), [i]), (tc.gates.z(), [j]))
+    )
+    / 4.0
+)
+
+print(f"\nComparing <Sz_{i}Sz_{j}> correlation function for the DMRG ground state:")
+print(f"TeNPy (Ground Truth):         {corr_tenpy:.8f}")
+print(f"TC (after X-gate correction): {corr_tc_corrected:.8f}")
+np.testing.assert_allclose(corr_tenpy, corr_tc_corrected, atol=1e-5)
+print(
+    "\n[SUCCESS] The correlation functions match perfectly after applying the X-gate correction."
+)
+print(
+    "This demonstrates the recommended physical approach to handle the XXZChain's inverted basis convention."
+)
+
+
 print("\n\nWorkflow demonstration and analysis complete!")

tensorcircuit/quantum.py

@@ -1301,8 +1301,18 @@ def tenpy2qop(tenpy_obj: Any) -> QuOperator:
         out_edges = [node["p*"] for node in nodes]
         in_edges = [node["p"] for node in nodes]
         ignore_edges = [nodes[0]["wL"], nodes[-1]["wR"]]
-    else:
-        nodes = [Node(W.to_ndarray()) for W in tenpy_tensors]
+    else:  # MPS
+        nodes = []
+        for i in range(nwires):
+            B_obj = tenpy_obj.get_B(i)
+            arr = B_obj.to_ndarray()
+            labels = B_obj.get_leg_labels()
+            vL_idx = labels.index("vL")
+            p_idx = labels.index("p")
+            vR_idx = labels.index("vR")
+            arr_reordered = arr.transpose((vL_idx, p_idx, vR_idx))
+            node = Node(arr_reordered, name=f"mps_{i}", axis_names=["vL", "p", "vR"])
+            nodes.append(node)
 
         if nwires > 1:
             for i in range(nwires - 1):