burn_quantum 0.4.0

Differentiable qubit and qudit simulation with Burn's native tensor and module APIs
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211

//! Independent analytic and finite-shot references for physical photonic learning.
mod common;
#[path = "../support/device.rs"]
mod runtime;
use burn::tensor::{DType, Tensor, TensorData};
use burn_quantum::{ComplexTensor, ResourceLimits, photonic::optics::*};
use common::{close, values};
fn device() -> burn::tensor::Device {
    runtime::execution().unwrap().device
}
fn preparation(batch: usize) -> FockState {
    let bs = Interferometer::beam_splitter(2, ModeId(0), ModeId(1), std::f64::consts::FRAC_PI_2)
        .unwrap();
    FockState::number(
        FockBasis::new(2, PhotonSector::UpTo(1), 3).unwrap(),
        &[1, 0],
        batch,
        &device(),
        DType::F32,
        Default::default(),
    )
    .unwrap()
    .apply(&bs, Default::default())
    .unwrap()
}
fn experiment(batch: usize, efficiency: f64) -> SinglePhotonPhaseExperiment {
    let bs = Interferometer::beam_splitter(2, ModeId(0), ModeId(1), std::f64::consts::FRAC_PI_2)
        .unwrap();
    SinglePhotonPhaseExperiment::new(
        preparation(batch),
        ModeId(0),
        bs.adjoint(),
        vec![Detector::threshold(efficiency, 0.).unwrap(); 2],
        4,
        Default::default(),
    )
    .unwrap()
}
#[test]
fn finite_shot_phase_probes_have_correct_derivative_error_and_physical_cost() {
    let phase_experiment = experiment(3, 0.8);
    let x = Tensor::from_data([0.3f32, 0.8, 1.4], &device());
    let exact = phase_experiment.evaluate(x.clone()).unwrap();
    let readout: Vec<_> = exact
        .records()
        .iter()
        .map(|r| f64::from(r == &[1, 0]))
        .collect();
    let gradients = phase_experiment
        .gradient(x.clone(), &readout, 20_000, 151, 120_000)
        .unwrap();
    let again = phase_experiment
        .gradient(x.clone(), &readout, 20_000, 151, 120_000)
        .unwrap();
    for ((g, r), angle) in gradients.iter().zip(again).zip([0.3f64, 0.8, 1.4]) {
        assert_eq!(g.gradient, r.gradient);
        assert_eq!(g.attempts, 40_000);
        assert!((g.gradient + 0.4 * angle.sin()).abs() < 7. * g.standard_error + 1e-6);
    }
    assert!(
        phase_experiment
            .gradient(x, &readout, 20_000, 151, 119_999)
            .is_err()
    );
    let histograms = exact.sample_counts(20_000, 17, 60_000).unwrap();
    let no_click = exact.records().iter().position(|r| r == &[0, 0]).unwrap();
    for h in histograms {
        assert_eq!(h.counts.iter().sum::<usize>(), h.attempts);
        assert!((h.counts[no_click] as f64 / 20_000. - 0.2).abs() < 0.02);
    }
    let zero = experiment(1, 0.)
        .evaluate(Tensor::from_data([0.7f32], &device()))
        .unwrap();
    let estimate = zero.sample_readout(&[0., 1., 1., 2.], 16, 1, 16).unwrap()[0];
    assert_eq!(
        (estimate.mean, estimate.variance, estimate.attempts),
        (0., 0., 16)
    );
    let two = FockState::number(
        FockBasis::new(2, PhotonSector::Fixed(2), 3).unwrap(),
        &[1, 1],
        1,
        &device(),
        DType::F32,
        Default::default(),
    )
    .unwrap();
    assert!(
        SinglePhotonPhaseExperiment::new(
            two,
            ModeId(0),
            Interferometer::phase(2, ModeId(0), 0.).unwrap(),
            vec![Detector::ideal(2).unwrap(); 2],
            9,
            Default::default()
        )
        .is_err()
    );
}
#[test]
fn finite_shot_classical_fisher_matches_lossy_mz_reference() {
    // P=(1-eta, eta sin²(theta/2), eta cos²(theta/2), 0) gives I=eta.
    let experiment = experiment(1, 0.8);
    let theta = Tensor::from_data([0.7f32], &device());
    let metric = experiment
        .classical_fisher(theta.clone(), 100_000, 1e-4, 97, 300_000)
        .unwrap()[0];
    assert!((metric.value - 0.8).abs() < 0.03, "{}", metric.value);
    assert_eq!(metric.attempts, 300_000);
    assert_eq!(metric.regularized_outcomes, 1); // structurally absent double click
    assert!(
        experiment
            .classical_fisher(theta, 100_000, 1e-4, 97, 299_999)
            .is_err()
    );
    assert!(
        experiment
            .classical_fisher(Tensor::from_data([0.7f32], &device()), 2, 1e-320, 97, 6)
            .is_err()
    );
}
fn feature(phases: &[f32], eta: f64) -> FockDensity {
    let phase = Tensor::from_data(TensorData::new(phases.to_vec(), [phases.len()]), &device());
    preparation(phases.len())
        .phase_shift(ModeId(0), phase, Default::default())
        .unwrap()
        .density(Default::default())
        .unwrap()
        .lose(ModeId(0), eta, Default::default())
        .unwrap()
        .lose(ModeId(1), eta, Default::default())
        .unwrap()
}
#[test]
fn native_mixed_kernel_matches_scalar_feature_map_and_retains_loss_mass() {
    let a = [-0.9f32, 0.2, 1.1];
    let b = [-0.2f32, 0.8];
    let eta = 0.7f64;
    let kernel = feature(&a, eta)
        .hilbert_schmidt_kernel(&feature(&b, eta), ResourceLimits::default())
        .unwrap();
    let oracle: Vec<_> = a
        .iter()
        .flat_map(|&x| {
            b.iter().map(move |&y| {
                (1. - eta).powi(2) + eta.powi(2) * ((x as f64 - y as f64) / 2.).cos().powi(2)
            })
        })
        .collect();
    close(&values(kernel), &oracle, 3e-6);
    let self_kernel = values(
        feature(&a, eta)
            .hilbert_schmidt_kernel(&feature(&a, eta), ResourceLimits::default())
            .unwrap(),
    );
    // Purity is below one: the vacuum and surviving-photon mixture was not normalized away.
    close(
        &[self_kernel[0], self_kernel[4], self_kernel[8]],
        &[0.58; 3],
        3e-6,
    );
}
fn bloch(radius: f64, angle: f64) -> FockDensity {
    let re = vec![
        (1. + radius * angle.cos()) / 2.,
        radius * angle.sin() / 2.,
        radius * angle.sin() / 2.,
        (1. - radius * angle.cos()) / 2.,
    ];
    let tensor = Tensor::from_data(TensorData::new(re, [1, 2, 2]), (&device(), DType::F32));
    FockDensity::from_entries(
        FockBasis::new(2, PhotonSector::Fixed(1), 2).unwrap(),
        ComplexTensor::new(tensor.clone(), tensor.zeros_like()).unwrap(),
        Default::default(),
    )
    .unwrap()
}
#[test]
fn squared_uhlmann_and_bures_match_independent_mixed_qubit_formula() {
    // A rotating Bloch vector of fixed radius r has F=1-r² sin²(delta/2), g_B=r²/4.
    for radius in [0., 0.4, 0.9, 1.] {
        let delta = 0.1f64;
        let center = bloch(radius, 0.3);
        let plus = center
            .uhlmann_fidelity(&bloch(radius, 0.3 + delta), 2, 2e-6)
            .unwrap()[0];
        let minus = center
            .uhlmann_fidelity(&bloch(radius, 0.3 - delta), 2, 2e-6)
            .unwrap()[0];
        let expected = 1. - radius.powi(2) * (delta / 2.).sin().powi(2);
        close(&[plus, minus], &[expected; 2], 2e-6);
        let metric = bures_metric_from_fidelities(plus, minus, delta).unwrap();
        assert!(
            (metric - radius.powi(2) / 4.).abs() < 5e-4,
            "radius={radius} metric={metric}"
        );
    }
    let a = bloch(0.4, 0.);
    let b = bloch(0.8, 0.);
    let commuting = (0.7f64.sqrt() * 0.9f64.sqrt() + 0.3f64.sqrt() * 0.1f64.sqrt()).powi(2);
    close(
        &a.uhlmann_fidelity(&b, 2, 2e-6).unwrap(),
        &[commuting],
        2e-6,
    );
    assert!(a.uhlmann_fidelity(&b, 1, 2e-6).is_err());
    assert!(a.uhlmann_fidelity(&bloch(1.2, 0.), 2, 2e-6).is_err());
    assert!(bures_metric_from_fidelities(1.1, 0.9, 0.1).is_err());
    assert!(bures_metric_from_fidelities(0.9, 0.9, 1e200).is_err());
    assert!(bures_metric_from_fidelities(0.9, 0.9, 1e-200).is_err());
}