ruqu_algorithms/vqe.rs
1//! Variational Quantum Eigensolver (VQE)
2//!
3//! Finds the ground-state energy of a Hamiltonian using a classical-quantum
4//! hybrid optimization loop:
5//!
6//! 1. A parameterized **ansatz** circuit prepares a trial state on the quantum
7//! processor (or simulator).
8//! 2. The **expectation value** of the Hamiltonian is measured for that state.
9//! 3. A **classical optimizer** (gradient descent with parameter-shift rule)
10//! updates the circuit parameters to minimize the energy.
11//! 4. Steps 1-3 repeat until convergence or the iteration budget is exhausted.
12//!
13//! The ansatz used here is "hardware-efficient": each layer applies Ry and Rz
14//! rotations on every qubit, followed by a linear CNOT entangling chain.
15
16use ruqu_core::circuit::QuantumCircuit;
17use ruqu_core::simulator::{SimConfig, Simulator};
18use ruqu_core::types::{Hamiltonian, PauliOp, PauliString};
19
20// ---------------------------------------------------------------------------
21// Configuration and result types
22// ---------------------------------------------------------------------------
23
24/// Configuration for a VQE run.
25pub struct VqeConfig {
26 /// The Hamiltonian whose ground-state energy we seek.
27 pub hamiltonian: Hamiltonian,
28 /// Number of qubits in the ansatz circuit.
29 pub num_qubits: u32,
30 /// Number of ansatz layers (depth). Each layer contributes
31 /// `2 * num_qubits` parameters (Ry + Rz per qubit).
32 pub ansatz_depth: u32,
33 /// Maximum number of classical optimizer iterations.
34 pub max_iterations: u32,
35 /// Stop early when the absolute energy change between successive
36 /// iterations falls below this threshold.
37 pub convergence_threshold: f64,
38 /// Step size for gradient descent.
39 pub learning_rate: f64,
40 /// Optional RNG seed for reproducible simulation.
41 pub seed: Option<u64>,
42}
43
44/// Result returned by [`run_vqe`].
45pub struct VqeResult {
46 /// Lowest energy found during the optimization.
47 pub optimal_energy: f64,
48 /// Parameter vector that produced `optimal_energy`.
49 pub optimal_parameters: Vec<f64>,
50 /// Energy at each iteration (length = `num_iterations`).
51 pub energy_history: Vec<f64>,
52 /// Total number of iterations executed.
53 pub num_iterations: u32,
54 /// Whether the optimizer converged before exhausting `max_iterations`.
55 pub converged: bool,
56}
57
58// ---------------------------------------------------------------------------
59// Ansatz construction
60// ---------------------------------------------------------------------------
61
62/// Return the total number of variational parameters for the given ansatz
63/// dimensions. Each layer uses `2 * num_qubits` parameters (one Ry and one
64/// Rz rotation per qubit).
65pub fn num_parameters(num_qubits: u32, depth: u32) -> usize {
66 (2 * num_qubits as usize) * (depth as usize)
67}
68
69/// Build a hardware-efficient ansatz circuit.
70///
71/// Each layer consists of:
72/// 1. **Rotation sub-layer**: Ry(theta) on every qubit.
73/// 2. **Rotation sub-layer**: Rz(theta) on every qubit.
74/// 3. **Entangling sub-layer**: Linear CNOT chain (0->1, 1->2, ..., n-2->n-1).
75///
76/// `params` must have exactly [`num_parameters`]`(num_qubits, depth)` entries.
77///
78/// # Panics
79///
80/// Panics if `params.len()` does not equal the expected parameter count.
81pub fn build_ansatz(num_qubits: u32, depth: u32, params: &[f64]) -> QuantumCircuit {
82 let expected = num_parameters(num_qubits, depth);
83 assert_eq!(
84 params.len(),
85 expected,
86 "build_ansatz: expected {} parameters, got {}",
87 expected,
88 params.len()
89 );
90
91 let mut circuit = QuantumCircuit::new(num_qubits);
92 let mut idx = 0;
93
94 for _layer in 0..depth {
95 // Ry rotations
96 for q in 0..num_qubits {
97 circuit.ry(q, params[idx]);
98 idx += 1;
99 }
100 // Rz rotations
101 for q in 0..num_qubits {
102 circuit.rz(q, params[idx]);
103 idx += 1;
104 }
105 // Linear CNOT entangling chain
106 for q in 0..num_qubits.saturating_sub(1) {
107 circuit.cnot(q, q + 1);
108 }
109 }
110
111 circuit
112}
113
114// ---------------------------------------------------------------------------
115// Energy evaluation
116// ---------------------------------------------------------------------------
117
118/// Evaluate the expectation value of the Hamiltonian for a given set of
119/// ansatz parameters.
120///
121/// Builds the ansatz, simulates it, and returns `<psi|H|psi>`.
122pub fn evaluate_energy(
123 config: &VqeConfig,
124 params: &[f64],
125) -> ruqu_core::error::Result<f64> {
126 let circuit = build_ansatz(config.num_qubits, config.ansatz_depth, params);
127 let sim_config = SimConfig {
128 seed: config.seed,
129 noise: None,
130 shots: None,
131 };
132 let result = Simulator::run_with_config(&circuit, &sim_config)?;
133 Ok(result.state.expectation_hamiltonian(&config.hamiltonian))
134}
135
136// ---------------------------------------------------------------------------
137// VQE optimizer
138// ---------------------------------------------------------------------------
139
140/// Run the VQE optimization loop.
141///
142/// Uses gradient descent with the **parameter-shift rule** to compute
143/// analytical gradients. For each parameter theta_i the gradient is:
144///
145/// ```text
146/// dE/d(theta_i) = [ E(theta_i + pi/2) - E(theta_i - pi/2) ] / 2
147/// ```
148///
149/// This requires 2 circuit evaluations per parameter per iteration, so the
150/// total cost is `O(max_iterations * 2 * num_parameters)` circuit runs.
151pub fn run_vqe(config: &VqeConfig) -> ruqu_core::error::Result<VqeResult> {
152 let n_params = num_parameters(config.num_qubits, config.ansatz_depth);
153
154 // Initialize parameters with small values to break symmetry.
155 let mut params = vec![0.1_f64; n_params];
156
157 let mut energy_history: Vec<f64> = Vec::with_capacity(config.max_iterations as usize);
158 let mut converged = false;
159
160 let mut best_energy = f64::MAX;
161 let mut best_params = params.clone();
162
163 for iteration in 0..config.max_iterations {
164 // ------------------------------------------------------------------
165 // Forward pass: compute current energy
166 // ------------------------------------------------------------------
167 let energy = evaluate_energy(config, ¶ms)?;
168 energy_history.push(energy);
169
170 if energy < best_energy {
171 best_energy = energy;
172 best_params = params.clone();
173 }
174
175 // ------------------------------------------------------------------
176 // Convergence check (skip first iteration since we need a delta)
177 // ------------------------------------------------------------------
178 if iteration > 0 {
179 let prev = energy_history[iteration as usize - 1];
180 if (prev - energy).abs() < config.convergence_threshold {
181 converged = true;
182 break;
183 }
184 }
185
186 // ------------------------------------------------------------------
187 // Backward pass: compute gradient via parameter-shift rule
188 // ------------------------------------------------------------------
189 let shift = std::f64::consts::FRAC_PI_2;
190 let mut gradient = vec![0.0_f64; n_params];
191
192 for i in 0..n_params {
193 let mut params_plus = params.clone();
194 let mut params_minus = params.clone();
195 params_plus[i] += shift;
196 params_minus[i] -= shift;
197
198 let e_plus = evaluate_energy(config, ¶ms_plus)?;
199 let e_minus = evaluate_energy(config, ¶ms_minus)?;
200 gradient[i] = (e_plus - e_minus) / 2.0;
201 }
202
203 // ------------------------------------------------------------------
204 // Parameter update (gradient descent -- minimize energy)
205 // ------------------------------------------------------------------
206 for i in 0..n_params {
207 params[i] -= config.learning_rate * gradient[i];
208 }
209 }
210
211 let num_iterations = energy_history.len() as u32;
212 Ok(VqeResult {
213 optimal_energy: best_energy,
214 optimal_parameters: best_params,
215 energy_history,
216 num_iterations,
217 converged,
218 })
219}
220
221// ---------------------------------------------------------------------------
222// Hamiltonian helpers
223// ---------------------------------------------------------------------------
224
225/// Create an approximate H2 (molecular hydrogen) Hamiltonian in the STO-3G
226/// basis mapped to 2 qubits via the Bravyi-Kitaev transformation.
227///
228/// ```text
229/// H = -1.0523 II + 0.3979 IZ + -0.3979 ZI + -0.0112 ZZ + 0.1809 XX
230/// ```
231///
232/// The exact ground-state energy of this Hamiltonian is approximately -1.137
233/// Hartree (at equilibrium bond length ~0.735 angstrom).
234pub fn h2_hamiltonian() -> Hamiltonian {
235 Hamiltonian {
236 terms: vec![
237 // Identity term (constant offset)
238 (-1.0523, PauliString { ops: vec![] }),
239 // IZ: Pauli-Z on qubit 1
240 (
241 0.3979,
242 PauliString {
243 ops: vec![(1, PauliOp::Z)],
244 },
245 ),
246 // ZI: Pauli-Z on qubit 0
247 (
248 -0.3979,
249 PauliString {
250 ops: vec![(0, PauliOp::Z)],
251 },
252 ),
253 // ZZ: Pauli-Z on both qubits
254 (
255 -0.0112,
256 PauliString {
257 ops: vec![(0, PauliOp::Z), (1, PauliOp::Z)],
258 },
259 ),
260 // XX: Pauli-X on both qubits
261 (
262 0.1809,
263 PauliString {
264 ops: vec![(0, PauliOp::X), (1, PauliOp::X)],
265 },
266 ),
267 ],
268 num_qubits: 2,
269 }
270}
271
272/// Create a simple single-qubit Z Hamiltonian: `H = -1.0 Z`.
273///
274/// The ground state is |0> with energy -1.0. Useful for smoke-testing VQE.
275pub fn single_z_hamiltonian() -> Hamiltonian {
276 Hamiltonian {
277 terms: vec![(
278 -1.0,
279 PauliString {
280 ops: vec![(0, PauliOp::Z)],
281 },
282 )],
283 num_qubits: 1,
284 }
285}
286
287// ---------------------------------------------------------------------------
288// Tests
289// ---------------------------------------------------------------------------
290
291#[cfg(test)]
292mod tests {
293 use super::*;
294
295 #[test]
296 fn test_num_parameters() {
297 assert_eq!(num_parameters(2, 1), 4);
298 assert_eq!(num_parameters(4, 3), 24);
299 assert_eq!(num_parameters(1, 5), 10);
300 }
301
302 #[test]
303 fn test_build_ansatz_gate_count() {
304 let n = 3;
305 let depth = 2;
306 let params = vec![0.0; num_parameters(n, depth)];
307 let circuit = build_ansatz(n, depth, ¶ms);
308 assert_eq!(circuit.num_qubits(), n);
309 // Each layer: 3 Ry + 3 Rz + 2 CNOT = 8 gates, times 2 layers = 16
310 assert_eq!(circuit.gates().len(), 16);
311 }
312
313 #[test]
314 #[should_panic(expected = "expected 4 parameters")]
315 fn test_build_ansatz_wrong_param_count() {
316 build_ansatz(2, 1, &[0.0; 3]);
317 }
318
319 #[test]
320 fn test_h2_hamiltonian_structure() {
321 let h = h2_hamiltonian();
322 assert_eq!(h.num_qubits, 2);
323 assert_eq!(h.terms.len(), 5);
324 }
325
326 #[test]
327 fn test_single_z_hamiltonian() {
328 let h = single_z_hamiltonian();
329 assert_eq!(h.num_qubits, 1);
330 assert_eq!(h.terms.len(), 1);
331 }
332}