S2E Simulation Data (oscillator, pulse shaper, amplifier, noncolinear SFG) for NeurIPS Submission

This data is created using the Start-to-End Software model for chirped pulse amplifier and nonlinear optics systems that we developed (see https://arxiv.org/abs/2211.09640 for more information on the model). The simulation involved an oscillator, pulse shaper, amplifier, and noncolinear sum-frequency generation (SFG) upconversion process. The stationary parameters for the sub-sections are included in the parameter files. Each example in the dataset uses a different set of pulse shaper parameters (effectively modulating the input to the amplifier and then to the SFG. The pulse shaper parameter space included scanning second- and third-order dispersion (SOD and TOD) values from -1e4 to 1e4 (fs)^2 and -1e5 to 1e5 (fs)^3, respectively. For spectral amplitude shaping, hole positions from 1022 nm to 1036 nm in steps of 0.35 nm, hole depths from 0 to 0.95 in steps of 0.0475, and hole widths from 0.1 nm to 4 nm in steps of 0.195 nm were all scanned. Out of all these combinations 10,000 parameter combinations were selected (enforcing at least 400 had no hole ie hole depth was zero). The simulation for each parameter combination was run and the three output fields in frequency domain from the final SFG process were stored as complex values in a concatenated vector. For data storage purposes, these vectors were all downsampled and cut compared to the original simulation output. The accompanying Neurips paper submission describes this data reduction process. In total, there are 100 files each containing 100 simulation runs. Because each simulation for the SFG process involves 100 outputs, each simulation run itself as 100 of these output vectors. In total, this means there are 100x100x100 = 1000000 field vectors for 10000 unique pulse shaper parameter set combinations.

本数据集基于我们开发的啁啾脉冲放大器与非线性光学系统端到端软件模型构建（模型详情参见https://arxiv.org/abs/2211.09640）。
本次仿真涵盖振荡器、脉冲整形器、放大器以及非共线和频产生（noncolinear sum-frequency generation, SFG）上转换过程。各子模块的固定参数已收录于参数文件中。
数据集中的每个样本均采用一套独特的脉冲整形器参数，可有效调制输入至放大器、再至SFG过程的光脉冲。脉冲整形器参数空间的扫描范围如下：二阶色散（second-order dispersion, SOD）与三阶色散（third-order dispersion, TOD）分别取值为-1×10⁴至1×10⁴ (fs)²和-1×10⁵至1×10⁵ (fs)³；针对光谱振幅整形，我们扫描了1022 nm至1036 nm范围内、步长0.35 nm的光谱凹谷位置，0至0.95、步长0.0475的凹谷深度，以及0.1 nm至4 nm、步长0.195 nm的凹谷宽度。从所有参数组合中筛选出10000组有效组合（要求至少400组的凹谷深度为0，即无光谱凹谷）。
针对每组参数组合运行仿真后，将最终SFG过程输出的三个频域光场以复数值形式存储为拼接向量。为优化数据存储效率，所有向量均相对于原始仿真输出进行了下采样与截断，相关的数据降维流程已在伴随的NeurIPS投稿论文中详细说明。
本数据集共包含100个文件，每个文件内含100次仿真结果。由于单次SFG过程的仿真会输出100组结果，因此单次仿真运行本身对应100个上述输出向量。综上，针对10000组独特的脉冲整形器参数组合，本数据集总计包含100×100×100 = 1000000个光场向量。