Photonic GBS Applications
Photonic samplers like Jiuzhang and TuringQ run Gaussian Boson Sampling (GBS). GBS is not just a benchmark — it underlies a family of graph and chemistry algorithms that map onto photonic sampling.
What GBS computes
Encoding a symmetric matrix into squeezed light makes photon-count patterns sample subgraphs weighted by matrix hafnians — the Gaussian-state analogue of the permanent that governs standard boson sampling — the basis of several useful algorithms.
Algorithm family
| Application | Idea |
|---|---|
| Dense subgraph | Frequent samples concentrate on dense subgraphs |
| Max clique | Post-process GBS samples to grow cliques |
| Graph similarity | Sample distributions form feature vectors for graph kernels |
| Molecular vibronic spectra | Map vibrational modes onto the interferometer |
Caveat: every entry needs classical post-processing and verification, is not guaranteed to return an optimal answer, and depends on the graph/molecule embedding and device calibration.
Finding dense subgraphs
import qalgora
# Conceptual API only. Not included in the open reference implementation.
program = qalgora.photonic.GBSProgram.from_adjacency(adjacency, n_modes=16, shots_count=20000)
samples = qalgora.photonic.run(program, target="jiuzhang")
dense = qalgora.photonic.densest_subgraph(samples, size=4)
print("densest 4-node candidate:", dense)光子 GBS 应用
Jiuzhang（九章）和 TuringQ（图灵量子）等光子采样器可运行高斯玻色采样（GBS）。GBS 不仅是一项基准测试，还支撑着一系列可映射到光子采样的图算法与化学算法。
GBS 的计算原理
把对称矩阵编码到压缩光中，光子计数模式就会按一定权重采样出子图，权重由矩阵的 hafnian（哈夫尼安）决定——它正是标准玻色采样中 permanent（积和式）在高斯态下的对应量，也是一系列实用算法的根基。
算法家族
| 应用场景 | 核心思路 |
|---|---|
| 稠密子图 | 高频采样集中于稠密子图 |
| 最大团 | 后处理 GBS 采样结果，逐步扩大团 |
| 图相似度 | 采样分布构成图核的特征向量 |
| 分子振动光谱 | 将振动模式映射至干涉仪 |
注意：表中每一项都需要经典后处理与验证，不保证返回最优解，并且依赖于图/分子的嵌入方式与设备标定。
寻找稠密子图
import qalgora
# 仅为概念性 API，未包含在开源参考实现中。
program = qalgora.photonic.GBSProgram.from_adjacency(adjacency, n_modes=16, shots_count=20000)
samples = qalgora.photonic.run(program, target="jiuzhang")
dense = qalgora.photonic.densest_subgraph(samples, size=4)
print("densest 4-node candidate:", dense)