serie 03

Kuengjoe_S03/Kuengjoe_S03_Aufg3.py

@@ -0,0 +1,92 @@
+import numpy as np
+
+
+def function_values(x1_value, x2_value, x3_value):
+    first_function_value = x1_value + x2_value**2 - x3_value**2 - 13
+    second_function_value = np.log(x2_value**4 / 4) + np.exp(0.5 * x3_value - 1) - 1
+    third_function_value = (x2_value - 3)**2 - x3_value**3 + 7
+
+    return np.array([
+        first_function_value,
+        second_function_value,
+        third_function_value
+    ], dtype=float)
+
+
+def jacobian_matrix(x1_value, x2_value, x3_value):
+    derivative_f1_x1 = 1
+    derivative_f1_x2 = 2 * x2_value
+    derivative_f1_x3 = -2 * x3_value
+
+    derivative_f2_x1 = 0
+    derivative_f2_x2 = 4 / x2_value
+    derivative_f2_x3 = 0.5 * np.exp(0.5 * x3_value - 1)
+
+    derivative_f3_x1 = 0
+    derivative_f3_x2 = 2 * (x2_value - 3)
+    derivative_f3_x3 = -3 * x3_value**2
+
+    return np.array([
+        [derivative_f1_x1, derivative_f1_x2, derivative_f1_x3],
+        [derivative_f2_x1, derivative_f2_x2, derivative_f2_x3],
+        [derivative_f3_x1, derivative_f3_x2, derivative_f3_x3]
+    ], dtype=float)
+
+
+def damped_newton_method(start_vector, tolerance=1e-5, maximum_iterations=100):
+    current_vector = np.array(start_vector, dtype=float)
+
+    for iteration_index in range(maximum_iterations):
+        current_function_values = function_values(
+            current_vector[0],
+            current_vector[1],
+            current_vector[2]
+        )
+
+        current_function_norm = np.linalg.norm(current_function_values, 2)
+
+        if current_function_norm < tolerance:
+            return current_vector, iteration_index, current_function_norm
+
+        current_jacobian_matrix = jacobian_matrix(
+            current_vector[0],
+            current_vector[1],
+            current_vector[2]
+        )
+
+        newton_step = np.linalg.solve(current_jacobian_matrix, -current_function_values)
+
+        damping_factor = 1.0
+
+        while damping_factor > 1e-8:
+            candidate_vector = current_vector + damping_factor * newton_step
+            candidate_function_values = function_values(
+                candidate_vector[0],
+                candidate_vector[1],
+                candidate_vector[2]
+            )
+
+            if np.linalg.norm(candidate_function_values, 2) < current_function_norm:
+                current_vector = candidate_vector
+                break
+
+            damping_factor = damping_factor / 2
+
+    final_function_values = function_values(
+        current_vector[0],
+        current_vector[1],
+        current_vector[2]
+    )
+    final_function_norm = np.linalg.norm(final_function_values, 2)
+
+    return current_vector, maximum_iterations, final_function_norm
+
+
+start_vector = [1.5, 3.0, 2.5]
+
+solution_vector, number_of_iterations, final_norm = damped_newton_method(start_vector)
+
+print("Start vector =", start_vector)
+print("Approximate solution =", solution_vector)
+print("Iterations =", number_of_iterations)
+print("||f(x^(k))||_2 =", final_norm)