I want to reproduce the results of the paper: Neural-network-based multistate solver for a static Schrödinger equation,Hong Li, Qilong Zhai, and Jeff Z. Y. Chen, Phys. Rev. A 103, 032405 – Published 8 March 2021. This is the deep NN used in the paperenter image description here. I tried to simple harmonic oscillator, the input is the x coordinate value, the output is N+1 wave function phi_n(x), N is the main quantum number. The loss function enter image description here contains 3 parts: total energy, orthogonal normalizationenter image description here, L2 regularization enter image description here. This is my python code (considering total energy only):
import tensorflow as tfimport numpy as np# Constantshbar = tf.constant(1.0, dtype=tf.float32)m = tf.constant(1.0, dtype=tf.float32)omega = tf.constant(1.0, dtype=tf.float32)# Potential energy functiondef potential_energy(x): return 0.5 * m * omega ** 2 * x ** 2# Kinetic energy termdef kinetic_energy_term(psi, x): with tf.GradientTape(persistent=True) as tape: tape.watch(x) dpsi_dx = tape.gradient(psi, x) d2psi_dx2 = tape.gradient(dpsi_dx, x) del tape # Delete the tape after using it return -0.5 * hbar ** 2 / m * d2psi_dx2# Expected energy functiondef expected_energy(x, psi_n): ke = kinetic_energy_term(psi_n, x) pe = potential_energy(x) hamiltonian_psi = ke + pe * psi_n numerator = tf.reduce_sum(psi_n * hamiltonian_psi) denominator = tf.reduce_sum(psi_n * psi_n) return numerator / denominator# Custom loss functiondef unsupervised_loss(x, y_pred): energies = [expected_energy(x, y_pred[:, i:i + 1]) for i in range(y_pred.shape[-1])] energies=tf.convert_to_tensor(energies,dtype=tf.float32) energy_sum = tf.reduce_sum(energies) return energy_sum# Neural network modelN = 10 # Number of quantum statesmodel = tf.keras.Sequential([ tf.keras.layers.Dense(50, activation='sigmoid', input_shape=(1,)), tf.keras.layers.Dense(50, activation='sigmoid'), tf.keras.layers.Dense(N, activation='linear')])model.compile(optimizer='adam', loss=lambda y_true, y_pred: unsupervised_loss(y_true, y_pred))# Training datax_train = np.linspace(-5, 5, 1000).reshape(-1, 1)x_train_tensor = tf.convert_to_tensor(x_train, dtype=tf.float32)# %%# Train the modelhistory = model.fit(x_train_tensor, x_train_tensor, epochs=100, batch_size=32, validation_split=0.2)error:
Traceback (most recent call last): File "D:\software\work\python\anaconda\lib\site-packages\IPython\core\interactiveshell.py", line 3437, in run_code exec(code_obj, self.user_global_ns, self.user_ns) File "<ipython-input-62-c0c5a991e6f1>", line 2, in <module> history = model.fit(x_train_tensor, x_train_tensor, epochs=100, batch_size=32, validation_split=0.2) File "D:\software\work\python\anaconda\lib\site-packages\keras\utils\traceback_utils.py", line 70, in error_handler raise e.with_traceback(filtered_tb) from None File "C:\Users\ADMINI~1\AppData\Local\Temp\__autograph_generated_filendq0th2a.py", line 15, in tf__train_function retval_ = ag__.converted_call(ag__.ld(step_function), (ag__.ld(self), ag__.ld(iterator)), None, fscope) File "C:\Users\ADMINI~1\AppData\Local\Temp\__autograph_generated_filea5lnci17.py", line 5, in <lambda> tf__lam = (lambda y_true, y_pred: ag__.with_function_scope((lambda lscope: ag__.converted_call(unsupervised_loss, (y_true, y_pred), None, lscope)), 'lscope', ag__.STD)) File "C:\Users\ADMINI~1\AppData\Local\Temp\__autograph_generated_filea5lnci17.py", line 5, in <lambda> tf__lam = (lambda y_true, y_pred: ag__.with_function_scope((lambda lscope: ag__.converted_call(unsupervised_loss, (y_true, y_pred), None, lscope)), 'lscope', ag__.STD)) File "C:\Users\ADMINI~1\AppData\Local\Temp\__autograph_generated_filedrz1d0_q.py", line 10, in tf__unsupervised_loss energies = [ag__.converted_call(ag__.ld(expected_energy), (ag__.ld(x), ag__.ld(y_pred)[:, ag__.ld(i):(ag__.ld(i) + 1)]), None, fscope) for i in ag__.converted_call(ag__.ld(range), (ag__.ld(y_pred).shape[(- 1)],), None, fscope)] File "C:\Users\ADMINI~1\AppData\Local\Temp\__autograph_generated_filedrz1d0_q.py", line 10, in <listcomp> energies = [ag__.converted_call(ag__.ld(expected_energy), (ag__.ld(x), ag__.ld(y_pred)[:, ag__.ld(i):(ag__.ld(i) + 1)]), None, fscope) for i in ag__.converted_call(ag__.ld(range), (ag__.ld(y_pred).shape[(- 1)],), None, fscope)] File "C:\Users\ADMINI~1\AppData\Local\Temp\__autograph_generated_file40wmfh26.py", line 10, in tf__expected_energy ke = ag__.converted_call(ag__.ld(kinetic_energy_term), (ag__.ld(psi_n), ag__.ld(x)), None, fscope) File "C:\Users\ADMINI~1\AppData\Local\Temp\__autograph_generated_filemre1bw7b.py", line 13, in tf__kinetic_energy_term d2psi_dx2 = ag__.converted_call(ag__.ld(tape).gradient, (ag__.ld(dpsi_dx), ag__.ld(x)), None, fscope)TypeError: in user code: File "D:\software\work\python\anaconda\lib\site-packages\keras\engine\training.py", line 1160, in train_function * return step_function(self, iterator) File "<ipython-input-61-b273cbda69ea>", line 46, in None * lambda y_true, y_pred: unsupervised_loss(y_true, y_pred) File "<ipython-input-61-b273cbda69ea>", line 33, in unsupervised_loss * energies = [expected_energy(x, y_pred[:, i:i + 1]) for i in range(y_pred.shape[-1])] File "<ipython-input-50-7a5ddb0aa6ce>", line 24, in expected_energy * ke = kinetic_energy_term(psi_n, x) File "<ipython-input-50-7a5ddb0aa6ce>", line 18, in kinetic_energy_term * d2psi_dx2 = tape.gradient(dpsi_dx, x) TypeError: Argument `target` should be a list or nested structure of Tensors, Variables or CompositeTensors to be differentiated, but received None.How to fix the error?
