"
]
@@ -687,7 +687,7 @@
],
"metadata": {
"kernelspec": {
- "display_name": "Python 3 (ipykernel)",
+ "display_name": "pq",
"language": "python",
"name": "python3"
},
@@ -701,7 +701,7 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
- "version": "3.8.13"
+ "version": "3.8.13 (default, Mar 28 2022, 06:16:26) \n[Clang 12.0.0 ]"
},
"toc": {
"base_numbering": 1,
@@ -744,6 +744,11 @@
"_Feature"
],
"window_display": false
+ },
+ "vscode": {
+ "interpreter": {
+ "hash": "08942b1340a5932ff3a93f52933a99b0e263568f3aace1d262ffa4d9a0f2da31"
+ }
}
},
"nbformat": 4,
diff --git a/tutorials/mbqc/Pattern_EN.ipynb b/tutorials/mbqc/Pattern_EN.ipynb
index 26f04c7..db301b8 100644
--- a/tutorials/mbqc/Pattern_EN.ipynb
+++ b/tutorials/mbqc/Pattern_EN.ipynb
@@ -2,17 +2,20 @@
"cells": [
{
"cell_type": "markdown",
+ "metadata": {
+ "tags": []
+ },
"source": [
"# Speeding Up Quantum Circuit Simulation by MBQC\n",
"\n",
" Copyright (c) 2021 Institute for Quantum Computing, Baidu Inc. All Rights Reserved. "
- ],
- "metadata": {
- "tags": []
- }
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {
+ "tags": []
+ },
"source": [
"## Introduction\n",
"\n",
@@ -23,13 +26,11 @@
"Another idea to solve the bottleneck of computation resources is to jump out of the framework of the quantum circuit model and simulate a circuit by its equivalents. **Measurement-Based Quantum Computation (MBQC)** [3-6] is another universal quantum computation model, that receives wide attention by its unique way of performing computation since its proposal. As is mentioned in [MBQC Quick Start Guide](MBQC_EN.ipynb), for non-adaptive measurements, not only can they be carried out simultaneously in physical implementation, but in classical simulation, they can also help to reduce the effective number of qubits in the computation, thus reducing the consumption of memory and computational resources. \n",
"\n",
"In this tutorial, we will introduce a new scheme of quantum circuit simulation by firstly translating it into its MBQC equivalent and then optimizing the order of measurements to finally improve the simulation efficiency. In the meanwhile, we will also demonstrate applications of the ``circuit`` and ``mcalculus`` modules in our MBQC package by two concrete examples. "
- ],
- "metadata": {
- "tags": []
- }
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"## Quantum Circuit Simulation\n",
"\n",
@@ -44,11 +45,11 @@
"Table 1: Quantum circuit simulation scheme by MBQC proposed in this tutorial
\n",
"\n",
"Next, we will introduce the above three steps in more details and with code implementations."
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"### Quantum circuit construction\n",
"\n",
@@ -58,12 +59,13 @@
"Figure 1: A simple example of quantum circuit
\n",
"\n",
"In the figure $Ry$ stands for a rotation-y gate, the 2-qubits gate is $CNOT$ gate, $|0\\rangle$ is the initial quantum state. The code implementation using ``Circuit`` to construct such a circuit is as follows:"
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "code",
"execution_count": null,
+ "metadata": {},
+ "outputs": [],
"source": [
"# Import common modules\n",
"from numpy import pi, random\n",
@@ -92,39 +94,38 @@
"# Add Ry gate\n",
"cir.ry(theta[2], 0)\n",
"cir.ry(theta[3], 1)"
- ],
- "outputs": [],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"Then, we add measurements to the circuit.\n",
"\n",
"**Note**: The way to add measurements in `Circuit` is different from `UAnsatz`! The former is to call `.measure` method before running the circuit, while the latter is to call `.measure` method after running the circuit."
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "code",
"execution_count": null,
+ "metadata": {},
+ "outputs": [],
"source": [
"# Input measurement information\n",
"# Measure all qubits by default\n",
"cir.measure()"
- ],
- "outputs": [],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"Then, we need to pass the constructed circuit to the translation module ``mcalculus`` for further process."
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"### Translation and optimization\n",
"\n",
@@ -137,19 +138,20 @@
"- standardization: integrate all subpatterns into a standard pattern;\n",
"\n",
"- simplification and optimization: simplify and optimize all measurement commands in the standard pattern."
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"In terms of code implementation, we provide `MCalculus` class for this task. We can call the method `set_circuit` to pass the constructed circuit ``cir`` to `MCalculus`."
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "code",
"execution_count": null,
+ "metadata": {},
+ "outputs": [],
"source": [
"# Import mcalculus module\n",
"from paddle_quantum.mbqc.mcalculus import MCalculus\n",
@@ -159,12 +161,11 @@
"\n",
"# Pass the circuit to MCalculus\n",
"mc.set_circuit(cir)"
- ],
- "outputs": [],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"#### One-by-one translation\n",
"\n",
@@ -181,21 +182,21 @@
"We arrange the above commands from left to right and get a command list $[E_{12}E_{23}E_{34}E_{45}M_1M_2M_3M_4X_5Z_5]$. The detailed parameters in the commands are omitted here for simplicity. \n",
"\n",
"Similarly, the command list of $CNOT$ gate can be given by \\[$E_{12}E_{23}E_{24}M_1 M_2 X_4 Z_3 Z_4$\\]. A measurement in the circuit model can be simply given by a command list \\[$M_1$\\] as measurements in the circuit model play the same role as $Z$ measurements on the output qubits in MBQC."
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"As for the code implementation, we provide a private method `__to_subpattern` to translate quantum gates and measurements one by one. (This method records all the one-to-one correspondence between gates and subpatterns. You can also customize this function if needed.) Once all the gates and measurements are translated, the information of all the subpatterns is stored in a list named **wild pattern** which will be used for further processes.\n",
"\n",
"\n",
"Figure 2: Translate quantum gates and measurements one by one to obtain a wild pattern
"
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"#### Standardization\n",
"\n",
@@ -205,106 +206,107 @@
"\n",
"\n",
"Figure 3: Transform a wild pattern to a standard pattern
"
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"In terms of code implementation, we can directly call the method `standardize` to complete the standarization process. "
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "code",
"execution_count": null,
+ "metadata": {},
+ "outputs": [],
"source": [
"# Standarization\n",
"mc.standardize()"
- ],
- "outputs": [],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"#### Simplification and optimization\n",
"\n",
"Once a standard pattern is obtained, one option is to directly pass it to the MBQC simulation module and run the simulation. However, a better job can be done before the simulation. Just like the quantum circuit optimization which aims to find a simpler but equivalent representation of the circuit, we can also find a refined pattern based on the standard one. Such refinement can be done in two approaches: removing the measurement dependency as much as possible and reordering the measurement commands. Both approaches aim to reduce the number of effective qubits involved in the actual simulation.\n",
"\n",
"Due to the above two considerations, we use a **signal shifting** operation to simplify the measurement dependencies and use a **row-major order optimization algorithm** to optimize the order of measurement commands."
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"**Signal shifting**\n",
"\n",
"The operation of signal shifting aims to simplify dependency by pulling out a particular type of dependency from the measurement command and compensating it with a \"signal\" command [7]. In terms of code implementation, we can directly call the method `shift_signals` to realize this."
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "code",
"execution_count": null,
+ "metadata": {},
+ "outputs": [],
"source": [
"# Signal shifting\n",
"mc.shift_signals()"
- ],
- "outputs": [],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"**Row-major order optimization algorithm**\n",
"\n",
"The row-major order optimization algorithm is a new algorithm we propose. The intention is to measure vertices with row-major order. Once all vertices in a row are measured, we can completely remove this row, thereby reducing the number of effective vertices involved in subsequent operations. Intuitively speaking, this method makes quantum gates and measurements executed by rows in some sense. Numerical experiments show that this optimization technique provides a significant improvement for the simulation of quantum shallow circuits (see below).\n",
"\n",
"In terms of code implementation, we can directly call `optimize_by_row` to implement the row-major optimization algorithm."
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "code",
"execution_count": null,
+ "metadata": {},
+ "outputs": [],
"source": [
"# Optimize the measurement order\n",
"mc.optimize_by_row()"
- ],
- "outputs": [],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"Now, we can call the method `get_pattern` to get the optimized pattern."
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "code",
"execution_count": null,
+ "metadata": {},
+ "outputs": [],
"source": [
"# Get the Pattern\n",
"pattern = mc.get_pattern()"
- ],
- "outputs": [],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"### Simulation\n",
"\n",
"Once we have the optimized pattern, we can call the method `set_pattern` in `MBQC` to pass it to the simulation module and call `run_pattern` to start the simulation process."
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "code",
"execution_count": null,
+ "metadata": {},
+ "outputs": [],
"source": [
"# Import required modules\n",
"from paddle_quantum.mbqc.simulator import MBQC\n",
@@ -329,34 +331,34 @@
"\n",
"# Obtain quantum output\n",
"quantum_output = mbqc.get_quantum_output()\n",
- "print(\"The qunatum output state is：\", quantum_output)\n",
+ "print(\"The quantum output state is：\", quantum_output)\n",
"\n",
"# Obtain classcial output\n",
"classical_output = mbqc.get_classical_output()\n",
"print(\"The classical output is：\", classical_output)"
- ],
- "outputs": [],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"At this point, we have achieved the entire process of circuit simulation."
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"### Function \"simulate_by_mbqc\"\n",
"\n",
"For convenience, we provide a function `simulate_by_mbqc` to pack the whole process of quantum circuit simulation. We can construct a circuit from ``Circuit`` first and then call `simulate_by_mbqc` to run the simulation process. This function translates the constructed circuit to its MBQC equivalent, runs the simulation, and finally returns the classical outcomes or quantum state vector equivalent to the quantum circuit model. An example is given as follows."
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "code",
"execution_count": null,
+ "metadata": {},
+ "outputs": [],
"source": [
"# Import required modules\n",
"from numpy import random, pi\n",
@@ -406,12 +408,11 @@
"# Print the returned classical and quantum outputs\n",
"print(\"Classical output is：\", classical_output)\n",
"print(\"Quantum output is：\", quantum_output)\n"
- ],
- "outputs": [],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"## Examples\n",
"\n",
@@ -436,21 +437,21 @@
"|8|inst_5x5_10_7.txt|25|**\\|**| 18|inst_5x6_10_7.txt|30|**\\|**|28|inst_6x6_10_7.txt|36|**\\|**|38|inst_6x7_10_7.txt|42|**\\|**|48|inst_7x7_10_7.txt|49|\n",
"|9|inst_5x5_10_8.txt|25|**\\|**| 19|inst_5x6_10_8.txt|30|**\\|**|29|inst_6x6_10_8.txt|36|**\\|**|39|inst_6x7_10_8.txt|42|**\\|**|49|inst_7x7_10_8.txt|49|\n",
"|10|inst_5x5_10_9.txt|25|**\\|**| 20|inst_5x6_10_9.txt|30|**\\|**|30|inst_6x6_10_9.txt|36|**\\|**|40|inst_6x7_10_9.txt|42|**\\|**|50|inst_7x7_10_9.txt|49|"
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"We compare the running time between our simulation scheme and two other simulators (`statevector` and `matrix_product_state`) in Qiskit. We choose a shorter time from `statevector` and `matrix_product_state` as the running time of the Qiskit simulator. All the numerical experiments are conducted on a standard laptop with 16G RAM and Intel Core i7 10TH GEN CPU. The time comparison is shown in Figure 4. It is clear that our new simulation scheme provides a significant improvement over the Qiskit simulators for simulating shallow circuits.\n",
"\n",
"\n",
"Figure 4: Time comparison between Qiskit and our MBQC simulation scheme for different instances
"
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"## Conclusion\n",
"\n",
@@ -459,11 +460,11 @@
"Although we adopt the state vector for underlying simulation in the current MBQC package, the idea of quantum circuit simulation by MBQC is not limited to any data structure. \n",
"\n",
"In terms of quantum circuit simulation and MBQC based algorithms, there are still lots of unknowns to explore. Welcome to join us and discover the infinite possibilities of MBQC together!"
- ],
- "metadata": {}
+ ]
},
{
"cell_type": "markdown",
+ "metadata": {},
"source": [
"---\n",
"## References\n",
@@ -485,8 +486,7 @@
"[8] Broadbent, Anne, and Elham Kashefi. \"Parallelizing quantum circuits.\" [Theoretical computer science 410.26 (2009): 2489-2510.](https://arxiv.org/abs/0704.1736)\n",
"\n",
"[9] Boixo, Sergio, et al. \"Characterizing quantum supremacy in near-term devices.\" [Nature Physics 14.6 (2018): 595-600.](https://www.nature.com/articles/s41567-018-0124-x)"
- ],
- "metadata": {}
+ ]
}
],
"metadata": {
@@ -559,4 +559,4 @@
},
"nbformat": 4,
"nbformat_minor": 4
-}
\ No newline at end of file
+}
diff --git a/tutorials/qnn_research/Noise_CN.ipynb b/tutorials/qnn_research/Noise_CN.ipynb
index 3478c27..f3eb5b4 100644
--- a/tutorials/qnn_research/Noise_CN.ipynb
+++ b/tutorials/qnn_research/Noise_CN.ipynb
@@ -465,11 +465,12 @@
]
},
{
+ "attachments": {},
"cell_type": "markdown",
"id": "comparable-athletics",
"metadata": {},
"source": [
- "**注释：** 在 [纠缠蒸馏](../locc/EntanglementDistillation_LOCCNET_CN.ipynb) 的教程中我们介绍了如何利用 Paddle Quantm 中的 LoccNet 模块来研究纠缠蒸馏，即利用多个含噪声的纠缠对来提取高保真度的纠缠对，感兴趣的读者可以前往阅读。"
+ "**注释：** 在 [纠缠蒸馏](../locc/EntanglementDistillation_LOCCNET_CN.ipynb) 的教程中我们介绍了如何利用 Paddle Quantum 中的 LoccNet 模块来研究纠缠蒸馏，即利用多个含噪声的纠缠对来提取高保真度的纠缠对，感兴趣的读者可以前往阅读。"
]
},
{
diff --git a/tutorials/quantum_simulation/BuildingMolecule_CN.ipynb b/tutorials/quantum_simulation/BuildingMolecule_CN.ipynb
index a1563b8..386bcaa 100644
--- a/tutorials/quantum_simulation/BuildingMolecule_CN.ipynb
+++ b/tutorials/quantum_simulation/BuildingMolecule_CN.ipynb
@@ -10,6 +10,7 @@
]
},
{
+ "attachments": {},
"cell_type": "markdown",
"id": "52532fe2",
"metadata": {},
@@ -34,7 +35,7 @@
"\n",
"该公式使用了[原子单位](https://en.wikipedia.org/wiki/Hartree_atomic_units)。该公式包含了 $N$ 个电子和 $M$ 个原子核，其中 $x_i$ 是第 $i$ 个电子的位置，$R_I$ 是第 $I$ 个原子核的位置。\n",
"\n",
- "本教程分为四部分，首先我们先来讨论如何使用 `qchem` 模块构造一个分子。之后，我们将简要描述如何通过调用外部量子化学来计算\n",
+ "本教程分为四部分，首先我们先来讨论如何使用 `qchem` 模块构造一个分子。之后，我们将简要描述如何通过调用外部量子化学工具来计算\n",
"[Hartree Fock](https://en.wikipedia.org/wiki/Hartree%E2%80%93Fock_method)\n",
"单粒子轨道。接下来，我们展示如何得到二次量子化的哈密顿量。最后，我们将描述如何将费米子的哈密顿量 (Fermionic Hamiltonian) 转化为适用于量子计算机的泡利字符串 (Pauli strings)。 "
]
@@ -64,8 +65,9 @@
"source": [
"# Eliminate noisy python warnings\n",
"import warnings\n",
- "\n",
- "warnings.filterwarnings(\"ignore\")"
+ "warnings.filterwarnings(\"ignore\")\n",
+ "import logging\n",
+ "logging.basicConfig(level=logging.INFO)"
]
},
{
@@ -76,25 +78,13 @@
"outputs": [],
"source": [
"# in Angstrom\n",
- "h2o_structure_direct = [[\"H\", [-0.02111417,0.8350417,1.47688078]], # H 代表着水分子中的氢元素\n",
- " [\"O\", [0.0, 0.0, 0.0]], # O 代表着水分子中的氧元素\n",
- " [\"H\", [-0.00201087,0.45191737,-0.27300254]]]"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 4,
- "id": "d354c5dd",
- "metadata": {},
- "outputs": [],
- "source": [
- "from paddle_quantum.qchem import geometry\n",
- "\n",
- "h2o_structure_xyz = geometry(file=\"h2o.xyz\")\n",
- "assert h2o_structure_xyz == h2o_structure_direct"
+ "h2o_coords = [(\"H\", [-0.02111417,0.8350417,1.47688078]), # H 代表着水分子中的氢元素\n",
+ " (\"O\", [0.0, 0.0, 0.0]), # O 代表着水分子中的氧元素\n",
+ " (\"H\", [-0.00201087,0.45191737,-0.27300254])]"
]
},
{
+ "attachments": {},
"cell_type": "markdown",
"id": "e996795a",
"metadata": {},
@@ -103,38 +93,57 @@
"\n",
"Hartree Fock 方法使用 [Slater 行列式](https://en.wikipedia.org/wiki/Slater_determinant)来表示 $N$ 电子波函数。它可以为我们提供一套单粒子轨道，这些轨道通常被作为其他量子化学方法的输入。 \n",
"\n",
- "量桨使用 psi4 [1] 作为它的量子化学引擎。我们可以使用 `qchem` 模块提供的 `get_molecular_data` 函数来执行相关计算，得到分子的所需信息。 `get_molecular_data` 函数以分子结构、分子总电荷和自旋多重性为主要输入，并返回一个 OpenFermion [2] `MolecularData` 对象。 \n",
- "\n",
- "我们继续以水分子为例。为了运行 Hartree Fock 计算，我们需要将 `method` 设置为 *scf* (Self - Consistent Field)。 我们还可以通过在 `basis` 参数中指定 [basis set](https://en.wikipedia.org/wiki/Basis_set_(chemistry)) 的类型来提高 Hartree Fock 计算的精度。"
+ "量桨当前的量子化学引擎是 PySCF [1]，未来我们也将支持 psi4 等更多的量子化学工具包（注：PySCF 目前仅支持 Mac 和 Linux 平台）。我们可以使用 `qchem` 模块提供的 `PySCFDriver` 类来进行 Hartree Fock 计算，它以分子结构、分子总电荷数和自旋多重度为主要输入，我们还可以通过在 `basis` 参数中指定不同的 [basis set](https://en.wikipedia.org/wiki/Basis_set_(chemistry)) 来控制 Hartree Fock 计算的精度。"
]
},
{
"cell_type": "code",
- "execution_count": null,
+ "execution_count": 4,
"id": "29945676",
"metadata": {},
- "outputs": [],
+ "outputs": [
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "INFO:root:\n",
+ "#######################################\n",
+ "SCF Calculation (Classical)\n",
+ "#######################################\n",
+ "INFO:root:Basis: sto-3g\n"
+ ]
+ },
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "converged SCF energy = -73.9677038774737\n"
+ ]
+ },
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "INFO:root:SCF energy: -73.96770.\n"
+ ]
+ }
+ ],
"source": [
- "from paddle_quantum.qchem import get_molecular_data\n",
- "\n",
- "h2o_moledata = get_molecular_data(\n",
- " h2o_structure_direct,\n",
- " charge=0, # 水分子是中性的，不带电\n",
- " multiplicity=1, # 水分子只有一个未配对电子\n",
- " basis=\"sto-3g\",\n",
- " method=\"scf\",\n",
- " if_save=True, # 是否将 MolecularData 中的信息存储成 hdf5 文件\n",
- " if_print=True, # 是否需要打印出水分子的基态能量\n",
- " name=\"\", # 指定 hdf5 文件的名字\n",
- " file_path=\".\" # 指定 hdf5 文件的路径 \n",
+ "from paddle_quantum.qchem import PySCFDriver\n",
+ "\n",
+ "driver = PySCFDriver()\n",
+ "driver.load_molecule(\n",
+ " atom=h2o_coords, # H2O 分子几何结构\n",
+ " basis=\"sto-3g\", # 量子化学计算基组（basis set）\n",
+ " multiplicity=1, # 分子的自旋多重度，因为水分子总自旋为S=0，所以自旋多重度2S+1=1\n",
+ " charge=0, # 分子电荷，水分子是中性分子，总电荷为0\n",
+ " unit=\"Angstrom\"\n",
")\n",
- "\n",
- "from openfermion.chem import MolecularData\n",
- "\n",
- "assert isinstance(h2o_moledata, MolecularData)"
+ "driver.run_scf() # 进行 Hartree Fock 计算"
]
},
{
+ "attachments": {},
"cell_type": "markdown",
"id": "973af7d1",
"metadata": {},
@@ -146,12 +155,12 @@
"$$\n",
"\\hat{H}=\\sum_{p,q}h_{pq}\\hat{c}^{\\dagger}_p\\hat{c}_q+\\frac{1}{2}\\sum_{p,q,r,s}v_{pqrs}\\hat{c}^{\\dagger}_p\\hat{c}^{\\dagger}_q\\hat{c}_r\\hat{c}_s,\\tag{3}$$\n",
"\n",
- "其中 $p$, $q$, $r$ 和 $s$ 是 Hartree Fock 轨道，$\\hat{c}^{\\dagger}_p$ 和 $\\hat{c}_q$ 分别是生成算子 (creation operation) 和湮灭算子 (annihilation operation)。系数 $h_{pq}$ 和 $v_{pqrs}$ 分别是单体积分 (one-body integrations) 和双体积分 (two-body integrations)，该信息可以用如下的方式从 `MolecularData` 提取出来。"
+ "其中 $p$, $q$, $r$ 和 $s$ 是 Hartree Fock 轨道，$\\hat{c}^{\\dagger}_p$ 和 $\\hat{c}_q$ 分别是生成算子 (creation operation) 和湮灭算子 (annihilation operation)。系数 $h_{pq}$ 和 $v_{pqrs}$ 分别是单体积分 (one-body integrations) 和双体积分 (two-body integrations)，该信息可以用如下的方式从 `PySCFDriver` 中计算出来。"
]
},
{
"cell_type": "code",
- "execution_count": 6,
+ "execution_count": 5,
"id": "a9a906ad",
"metadata": {},
"outputs": [
@@ -159,13 +168,13 @@
"name": "stdout",
"output_type": "stream",
"text": [
- "[[-3.2911e+01 5.5623e-01 2.8755e-01 1.4640e-15 -7.4568e-02 -9.4552e-02 2.8670e-01]\n",
- " [ 5.5623e-01 -8.0729e+00 -4.0904e-02 -1.6823e-15 1.7890e-01 3.5048e-01 -1.3460e+00]\n",
- " [ 2.8755e-01 -4.0904e-02 -7.3355e+00 -4.8424e-15 4.1911e-01 5.2109e-01 7.0928e-01]\n",
- " [ 1.4640e-15 -1.6823e-15 -4.8424e-15 -7.5108e+00 -1.4127e-14 -2.6576e-14 -1.5008e-15]\n",
- " [-7.4568e-02 1.7890e-01 4.1911e-01 -1.4127e-14 -5.7849e+00 2.0887e+00 1.2427e-01]\n",
- " [-9.4552e-02 3.5048e-01 5.2109e-01 -2.6576e-14 2.0887e+00 -5.0803e+00 1.3967e-02]\n",
- " [ 2.8670e-01 -1.3460e+00 7.0928e-01 -1.5008e-15 1.2427e-01 1.3967e-02 -5.0218e+00]]\n"
+ "[[-3.2911e+01 5.5623e-01 2.8755e-01 -5.1653e-15 -7.4568e-02 -9.4552e-02 -2.8670e-01]\n",
+ " [ 5.5623e-01 -8.0729e+00 -4.0904e-02 1.5532e-15 1.7890e-01 3.5048e-01 1.3460e+00]\n",
+ " [ 2.8755e-01 -4.0904e-02 -7.3355e+00 6.8611e-16 4.1911e-01 5.2109e-01 -7.0928e-01]\n",
+ " [-5.1650e-15 1.5617e-15 6.7472e-16 -7.5108e+00 5.4096e-16 1.7947e-15 4.5448e-16]\n",
+ " [-7.4568e-02 1.7890e-01 4.1911e-01 5.3561e-16 -5.7849e+00 2.0887e+00 -1.2427e-01]\n",
+ " [-9.4552e-02 3.5048e-01 5.2109e-01 1.8235e-15 2.0887e+00 -5.0803e+00 -1.3967e-02]\n",
+ " [-2.8670e-01 1.3460e+00 -7.0928e-01 4.8475e-16 -1.2427e-01 -1.3967e-02 -5.0218e+00]]\n"
]
}
],
@@ -173,107 +182,77 @@
"import numpy as np \n",
"np.set_printoptions(precision=4, linewidth=150)\n",
"\n",
- "hpq, vpqrs = h2o_moledata.get_integrals()\n",
- "assert np.shape(hpq)==(7, 7) # When use sto3g basis, the total number of molecular orbitals used in water calculation is 7\n",
+ "hpq = driver.get_onebody_tensor(\"int1e_kin\") + driver.get_onebody_tensor(\"int1e_nuc\")\n",
+ "vpqrs = driver.get_twobody_tensor()\n",
+ "assert np.shape(hpq)==(7, 7) # H2O 分子在STO-3G基组下的轨道数量为7个\n",
"assert np.shape(vpqrs)==(7, 7, 7, 7)\n",
"\n",
"print(hpq)\n",
- "# print(vpqrs)"
+ "# print(vpqrs) # vpqrs张量规模较大，为使输出更简洁，我们默认将这一行注释，感兴趣的用户可以将这个张量打印出来"
]
},
{
+ "attachments": {},
"cell_type": "markdown",
"id": "de4f1bf2",
"metadata": {},
"source": [
- "大多数情况下，我们并不需要把这些积分信息手动的提取出来，*qchem* 模块已经将该步骤融入函数 `fermionic_hamiltonian` 中，我们可以直接进行下一步的运算，获得哈密顿量。"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 7,
- "id": "069cfcd5",
- "metadata": {},
- "outputs": [],
- "source": [
- "from paddle_quantum.qchem import fermionic_hamiltonian\n",
- "\n",
- "H_of_water = fermionic_hamiltonian(\n",
- " h2o_moledata,\n",
- " multiplicity=1,\n",
- " active_electrons=4,\n",
- " active_orbitals=4\n",
- ")\n",
- "\n",
- "from openfermion.ops import FermionOperator\n",
- "\n",
- "assert isinstance(H_of_water, FermionOperator)"
- ]
- },
- {
- "cell_type": "markdown",
- "id": "47f24b4b",
- "metadata": {},
- "source": [
- "通过指定 `active_electrons` 和 `active_orbitals`，我们可以限制哈密顿的自由度，从而可以减少哈密顿量的的项数，加速运算。我们还可以使用 *qchem* 中 `active_space` 函数来查看哪些轨道是**核心 (core)** 轨道，哪些轨道是**活跃 (active)** 轨道。"
+ "大多数情况下，我们并不需要把这些积分信息手动的提取出来，*qchem* 模块已经将该过程封装在 `Molecule` 类中，我们可以直接从它里面得到水分子的哈密顿量。"
]
},
{
"cell_type": "code",
"execution_count": 8,
- "id": "c136344d",
+ "id": "069cfcd5",
"metadata": {},
"outputs": [
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "INFO:root:\n",
+ "#######################################\n",
+ "SCF Calculation (Classical)\n",
+ "#######################################\n",
+ "INFO:root:Basis: sto-3g\n"
+ ]
+ },
{
"name": "stdout",
"output_type": "stream",
"text": [
- "List of core orbitals: [0, 1, 2]\n",
- "List of active orbitals: [3, 4, 5, 6]\n"
+ "converged SCF energy = -73.9677038774737\n"
+ ]
+ },
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "INFO:root:SCF energy: -73.96770.\n"
]
}
],
"source": [
- "from paddle_quantum.qchem import active_space\n",
+ "# 构建 H2O 的 `Molecule` 实例\n",
+ "from paddle_quantum.qchem import Molecule\n",
"\n",
- "core_orbits_list, act_orbits_list = active_space(\n",
- " 10, # number of electrons in water molecule\n",
- " 7, # number of molecular orbitals in water molecule\n",
- " active_electrons=4,\n",
- " active_orbitals=4\n",
+ "mol = Molecule(\n",
+ " geometry=h2o_coords,\n",
+ " basis=\"sto-3g\",\n",
+ " driver=PySCFDriver()\n",
")\n",
"\n",
- "print(\"List of core orbitals: {:}\".format(core_orbits_list))\n",
- "print(\"List of active orbitals: {:}\".format(act_orbits_list))"
+ "\n",
+ "H_of_water = mol.get_molecular_hamiltonian()"
]
},
{
+ "attachments": {},
"cell_type": "markdown",
- "id": "05b6aefc",
+ "id": "e71ea17b",
"metadata": {},
"source": [
- "## 从费米子哈密顿量到自旋哈密顿量\n",
- "\n",
- "在量子计算中，我们只能使用由泡利矩阵构成的算符(operator)\n",
- "\n",
- "$$\n",
- "\\boldsymbol{\\sigma}_x=\\begin{pmatrix}\n",
- "0 & 1\\\\\n",
- "1 & 0\n",
- "\\end{pmatrix},\\quad \\boldsymbol{\\sigma}_y=\\begin{pmatrix}\n",
- "0 & -i\\\\\n",
- "i & 0\n",
- "\\end{pmatrix},\\quad \\boldsymbol{\\sigma}_z=\\begin{pmatrix}\n",
- "1 & 0\\\\\n",
- "0 & -1\n",
- "\\end{pmatrix}.\\tag{4}\n",
- "$$\n",
- "\n",
- "因此，我们需要将现有的哈密顿量（二次量子化形式）变换成量子比特算符 (qubit operator)，这种变换被称为 [Jordan-Wigner 变换](https://en.wikipedia.org/wiki/Jordan%E2%80%93Wigner_transformation)。\n",
- "\n",
- "> 此外，Bravyi-Kitaev 变换也可以达到相同的效果，只需要将参数 mapping_method 改成 'bravyi_kitaev' 即可。\n",
- "\n",
- "在量桨中，我们提供 `spin_hamiltonian` 函数来实现从费米子哈密顿量到自旋哈密顿量的变换，代码如下："
+ "由于量子计算中允许的操作是基于泡利算符$\\hat{\\sigma}_x$、$\\hat{\\sigma}_y$、$\\hat{\\sigma}_z$的，因此，我们需要将上式中的哈密顿量（二次量子化形式）变换成量子比特算符 (qubit operator)，这可以通过[Jordan-Wigner 变换](https://en.wikipedia.org/wiki/Jordan%E2%80%93Wigner_transformation)实现。在*qchem*中，这一功能已经被集成到 `get_molecular_hamiltonian` 方法中了。"
]
},
{
@@ -286,34 +265,24 @@
"name": "stdout",
"output_type": "stream",
"text": [
- "There are 193 terms in H2O Hamiltonian in total.\n",
+ "There are 2110 terms in H2O Hamiltonian in total.\n",
"The first 10 terms are \n",
- " -72.10615980544185 I\n",
- "-0.007310917992546774 X0, X1, Y2, Y3\n",
- "0.0052460870730834325 X0, X1, Y2, Z3, Z4, Y5\n",
- "0.0016283548447087654 X0, X1, Y2, Z3, Z4, Z5, Z6, Y7\n",
- "0.0052460870730834325 X0, X1, X3, X4\n",
- "0.0016283548447087654 X0, X1, X3, Z4, Z5, X6\n",
- "-0.005994544380559027 X0, X1, Y4, Y5\n",
- "0.001387644178102622 X0, X1, Y4, Z5, Z6, Y7\n",
- "0.001387644178102622 X0, X1, X5, X6\n",
- "-0.009538223793221182 X0, X1, Y6, Y7\n"
+ " -44.808115297466266 I\n",
+ "12.537584025014317 Z0\n",
+ "12.537584025014313 Z1\n",
+ "1.8115178684479893 Z2\n",
+ "1.8115178684479893 Z3\n",
+ "1.4546840922727915 Z4\n",
+ "1.4546840922727915 Z5\n",
+ "1.4299903414012243 Z6\n",
+ "1.429990341401224 Z7\n",
+ "1.0849294605602529 Z8\n"
]
}
],
"source": [
- "from paddle_quantum.qchem import spin_hamiltonian\n",
- "\n",
- "pauli_H_of_water_ = spin_hamiltonian(\n",
- " h2o_moledata,\n",
- " multiplicity=1,\n",
- " active_electrons=4,\n",
- " active_orbitals=4,\n",
- " mapping_method='jordan_wigner'\n",
- ")\n",
- "\n",
- "print('There are', pauli_H_of_water_.n_terms, 'terms in H2O Hamiltonian in total.')\n",
- "print('The first 10 terms are \\n', pauli_H_of_water_[:10])"
+ "print('There are', H_of_water.n_terms, 'terms in H2O Hamiltonian in total.')\n",
+ "print('The first 10 terms are \\n', H_of_water[:10])"
]
},
{
@@ -325,6 +294,7 @@
]
},
{
+ "attachments": {},
"cell_type": "markdown",
"id": "d7417802",
"metadata": {},
@@ -332,19 +302,13 @@
"---\n",
"## 参考文献\n",
"\n",
- "[1] [Psi4: an open-source ab initio electronic structure program](https://wires.onlinelibrary.wiley.com/doi/abs/10.1002/wcms.93)\n",
- "\n",
- "[2] [OpenFermion: the electronic structure package for quantum computers\n",
- "](https://iopscience.iop.org/article/10.1088/2058-9565/ab8ebc)"
+ "[1] [PySCF: Python-based Simulations of Chemistry Framework](https://pyscf.org/)"
]
}
],
"metadata": {
- "interpreter": {
- "hash": "f7cfecff1ef1940b21a48efa1b88278bb096bd916f13c2df11af4810c38b47e1"
- },
"kernelspec": {
- "display_name": "Python 3.8.0 ('pq')",
+ "display_name": "modellib",
"language": "python",
"name": "python3"
},
@@ -358,7 +322,12 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
- "version": "3.8.0"
+ "version": "3.7.15"
+ },
+ "vscode": {
+ "interpreter": {
+ "hash": "8f24120f890011f53feb4ed62c47961d8565ec1de8b7cb23548c15bd6da8f2d2"
+ }
}
},
"nbformat": 4,
diff --git a/tutorials/quantum_simulation/BuildingMolecule_EN.ipynb b/tutorials/quantum_simulation/BuildingMolecule_EN.ipynb
index 5079640..eff7ae0 100644
--- a/tutorials/quantum_simulation/BuildingMolecule_EN.ipynb
+++ b/tutorials/quantum_simulation/BuildingMolecule_EN.ipynb
@@ -9,12 +9,13 @@
]
},
{
+ "attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"## Overview\n",
"\n",
- "In this tutorial, we will demonstrate how to use Paddle Quantum's `qchem` module to build valid Hamiltonian for simulating chemical molecules on a quantum computer. We will go step by step how to build the second quantized Hamiltonian from a molecular structure and how to transform it to a set of Pauli matrices. \n",
+ "In this tutorial, we will demonstrate how to use Paddle Quantum's *qchem* module to build valid Hamiltonian for simulating chemical molecules on a quantum computer. We will go step by step how to build the second quantized Hamiltonian from a molecular structure and how to transform it to a set of Pauli matrices. \n",
"\n",
"Hamiltonian is a physical quantity related to the total energy of a physical system. In general, it can be represented as \n",
"\n",
@@ -32,7 +33,7 @@
"\n",
"when we use [atomic units](https://en.wikipedia.org/wiki/Hartree_atomic_units). Our electronic problem contains $N$ electrons and $M$ nucleus. We use $x_i$ to denote position of the $i$-th electron, and use $R_I$ to denote position of the $I$-th nuclei. \n",
"\n",
- "This tutorial will have the following parts. Let's first talk about how to construct a molecule in `qchem`. After that, we will briefly describe how to calculate [Hartree Fock](https://en.wikipedia.org/wiki/Hartree%E2%80%93Fock_method) single particle orbitals by calling external quantum chemistry within Paddle Quantum. Next, we show how we can obtain the Hamiltonian in second quantization representation. Finally, we describe how to transform the Fermionic Hamiltonian to Pauli strings recognized by quantum computer."
+ "This tutorial will have the following parts. Let's first talk about how to construct a molecule in *qchem*. After that, we will briefly describe how to calculate [Hartree Fock](https://en.wikipedia.org/wiki/Hartree%E2%80%93Fock_method) single particle orbitals by calling external quantum chemistry within Paddle Quantum. Next, we show how we can obtain molecular Hamiltonian using *qchem*'s `Molecule` class."
]
},
{
@@ -57,8 +58,9 @@
"source": [
"# Eliminate noisy python warnings\n",
"import warnings\n",
- "\n",
- "warnings.filterwarnings(\"ignore\")"
+ "warnings.filterwarnings(\"ignore\")\n",
+ "import logging\n",
+ "logging.basicConfig(level=logging.INFO)"
]
},
{
@@ -68,68 +70,69 @@
"outputs": [],
"source": [
"# in Angstrom\n",
- "h2o_structure_direct = [[\"H\", [-0.02111417,0.8350417,1.47688078]], # H stands for hydrogen element in water\n",
- " [\"O\", [0.0, 0.0, 0.0]], # O stands for oxygen element in water\n",
- " [\"H\", [-0.00201087,0.45191737,-0.27300254]]]"
- ]
- },
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "Instead of specifying molecular structure directly, we can also pass the \\*.xyz file to the `geometry` function to get the same structure."
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 4,
- "metadata": {},
- "outputs": [],
- "source": [
- "from paddle_quantum.qchem import geometry\n",
- "\n",
- "h2o_structure_xyz = geometry(file=\"h2o.xyz\")\n",
- "assert h2o_structure_xyz == h2o_structure_direct"
+ "h2o_coords = [(\"H\", [-0.02111417,0.8350417,1.47688078]), # H stands for hydrogen element in water\n",
+ " (\"O\", [0.0, 0.0, 0.0]), # O stands for oxygen element in water\n",
+ " (\"H\", [-0.00201087,0.45191737,-0.27300254])]"
]
},
{
+ "attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"## Calculate Hartree Fock orbitals\n",
"Hartree Fock method uses the [Slater determinant](https://en.wikipedia.org/wiki/Slater_determinant) to represent the $N$-electron wavefunction. It could provide us with a set of single particle orbitals which are often taken as input to more advanced quantum chemistry methods. \n",
"\n",
- "Paddle Quantum uses psi4 [1] as its quantum chemistry engine. We could use the `get_molecular_data` function provided in `qchem` module to manage the quantum chemistry calculation and get the necessary information about the molecule. `get_molecular_data` function takes molecular structure, total molecular charge, and spin multiplicity as its major inputs, it will return an OpenFermion [2] `MolecularData` object. \n",
- "\n",
- "Let's continue with our water molecule example. To run the Hartree Fock calculation, we need to set the `method` keyword argument to *scf* (Self Consistent Field). We can also improve the quality of Hartree Fock calculation by specifying the type of [basis set](https://en.wikipedia.org/wiki/Basis_set_(chemistry)) in the `basis` argument. "
+ "Currently, Paddle Quantum uses PySCF [1] as its quantum chemistry engine, we will support more quantum chemistry toolkits, such as psi4, in the future (NOTE: PySCF currently only support Mac and Linux platform). We could use the `PySCFDriver` provided in `qchem` module to manage the quantum chemistry calculation and get the necessary information about the molecule. It takes molecular structure, total molecular charge, and spin multiplicity as its major inputs. We can also control the precision of Hartree Fock calculation by setting different [basis set](https://en.wikipedia.org/wiki/Basis_set_(chemistry)) for `basis` keyword. "
]
},
{
"cell_type": "code",
- "execution_count": null,
+ "execution_count": 4,
"metadata": {},
- "outputs": [],
+ "outputs": [
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "INFO:root:\n",
+ "#######################################\n",
+ "SCF Calculation (Classical)\n",
+ "#######################################\n",
+ "INFO:root:Basis: sto-3g\n"
+ ]
+ },
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "converged SCF energy = -73.9677038774737\n"
+ ]
+ },
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "INFO:root:SCF energy: -73.96770.\n"
+ ]
+ }
+ ],
"source": [
- "from paddle_quantum.qchem import get_molecular_data\n",
- "\n",
- "h2o_moledata = get_molecular_data(\n",
- " h2o_structure_direct,\n",
- " charge=0, # Water molecule is charge neutral\n",
- " multiplicity=1, # In the ground state, the lowest 5 molecular orbitals of water molecular will be occupied by a pair of electrons with opposite spin\n",
- " basis=\"sto-3g\",\n",
- " method=\"scf\",\n",
- " if_save=True, # Whether to save information contained in MolecularData object to a hdf5 file\n",
- " if_print=True, # Wheter to print the ground state energy of water molecule\n",
- " name=\"\", # Specifies the name of the hdf5 file\n",
- " file_path=\".\" # Specifies where to store the hdf5 file \n",
+ "from paddle_quantum.qchem import PySCFDriver\n",
+ "\n",
+ "driver = PySCFDriver()\n",
+ "driver.load_molecule(\n",
+ " atom=h2o_coords, # Geometry of H2O molecule\n",
+ " basis=\"sto-3g\", # Basis set for quantum chemistry calculation\n",
+ " multiplicity=1, # Spin multiplicity for molecule, since the total spin of H2O is S=0，its spin multiplicity is 2S+1=1\n",
+ " charge=0, # Total charge of molecule, since H2O is charge neutral, its charge=0\n",
+ " unit=\"Angstrom\"\n",
")\n",
- "\n",
- "from openfermion.chem import MolecularData\n",
- "\n",
- "assert isinstance(h2o_moledata, MolecularData)"
+ "driver.run_scf() # Perform Hartree Fock calculation"
]
},
{
+ "attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
@@ -139,25 +142,25 @@
"$$\n",
"\\hat{H}=\\sum_{p,q}h_{pq}\\hat{c}^{\\dagger}_p\\hat{c}_q+\\frac{1}{2}\\sum_{p,q,r,s}v_{pqrs}\\hat{c}^{\\dagger}_p\\hat{c}^{\\dagger}_q\\hat{c}_r\\hat{c}_s,\\tag{3}$$\n",
"\n",
- "where $p$, $q$, $r$ and $s$ are Hartree Fock orbitals computed in the previous section. $\\hat{c}^{\\dagger}_p$ and $\\hat{c}_q$ are creation and annihilation operations, respectively. The two coefficients $h_{pq}$ and $v_{pqrs}$ are called molecular integrals, and can be obtained from `MolecularData` object in the following way."
+ "where $p$, $q$, $r$ and $s$ are Hartree Fock orbitals computed in the previous section. $\\hat{c}^{\\dagger}_p$ and $\\hat{c}_q$ are creation and annihilation operations, respectively. The two coefficients $h_{pq}$ and $v_{pqrs}$ are called molecular integrals, and can be obtained from `PySCFDriver` in the following way."
]
},
{
"cell_type": "code",
- "execution_count": 6,
+ "execution_count": 5,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "[[-3.2911e+01 5.5623e-01 2.8755e-01 1.4640e-15 -7.4568e-02 -9.4552e-02 2.8670e-01]\n",
- " [ 5.5623e-01 -8.0729e+00 -4.0904e-02 -1.6823e-15 1.7890e-01 3.5048e-01 -1.3460e+00]\n",
- " [ 2.8755e-01 -4.0904e-02 -7.3355e+00 -4.8424e-15 4.1911e-01 5.2109e-01 7.0928e-01]\n",
- " [ 1.4640e-15 -1.6823e-15 -4.8424e-15 -7.5108e+00 -1.4127e-14 -2.6576e-14 -1.5008e-15]\n",
- " [-7.4568e-02 1.7890e-01 4.1911e-01 -1.4127e-14 -5.7849e+00 2.0887e+00 1.2427e-01]\n",
- " [-9.4552e-02 3.5048e-01 5.2109e-01 -2.6576e-14 2.0887e+00 -5.0803e+00 1.3967e-02]\n",
- " [ 2.8670e-01 -1.3460e+00 7.0928e-01 -1.5008e-15 1.2427e-01 1.3967e-02 -5.0218e+00]]\n"
+ "[[-3.2911e+01 5.5623e-01 2.8755e-01 -5.1653e-15 -7.4568e-02 -9.4552e-02 -2.8670e-01]\n",
+ " [ 5.5623e-01 -8.0729e+00 -4.0904e-02 1.5532e-15 1.7890e-01 3.5048e-01 1.3460e+00]\n",
+ " [ 2.8755e-01 -4.0904e-02 -7.3355e+00 6.8611e-16 4.1911e-01 5.2109e-01 -7.0928e-01]\n",
+ " [-5.1650e-15 1.5617e-15 6.7472e-16 -7.5108e+00 5.4096e-16 1.7947e-15 4.5448e-16]\n",
+ " [-7.4568e-02 1.7890e-01 4.1911e-01 5.3561e-16 -5.7849e+00 2.0887e+00 -1.2427e-01]\n",
+ " [-9.4552e-02 3.5048e-01 5.2109e-01 1.8235e-15 2.0887e+00 -5.0803e+00 -1.3967e-02]\n",
+ " [-2.8670e-01 1.3460e+00 -7.0928e-01 4.8475e-16 -1.2427e-01 -1.3967e-02 -5.0218e+00]]\n"
]
}
],
@@ -165,139 +168,103 @@
"import numpy as np \n",
"np.set_printoptions(precision=4, linewidth=150)\n",
"\n",
- "hpq, vpqrs = h2o_moledata.get_integrals()\n",
- "assert np.shape(hpq)==(7, 7) # When use sto3g basis, the total number of molecular orbitals used in water calculation is 7\n",
+ "hpq = driver.get_onebody_tensor(\"int1e_kin\") + driver.get_onebody_tensor(\"int1e_nuc\")\n",
+ "vpqrs = driver.get_twobody_tensor()\n",
+ "assert np.shape(hpq)==(7, 7) # H2O has 7 orbitals when using STO-3G basis.\n",
"assert np.shape(vpqrs)==(7, 7, 7, 7)\n",
"\n",
"print(hpq)\n",
- "# print(vpqrs)"
+ "# print(vpqrs) # vpqrs is a large tensor，for brevity, we comment this line by default, interested users can uncomment this line to see it."
]
},
{
+ "attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
- "Most of the time, we don't need to extract those integrals and assemble the Hamiltonian manually, *qchem* module has already helped us take care of this by providing the `fermionic_hamiltonian` function."
+ "Most of the time, we don't need to extract those integrals and assemble the Hamiltonian manually, *qchem* has already ecapsulated this procedure in `Molecule` class, we can get the Hamiltonian of water molecule from it."
]
},
{
"cell_type": "code",
- "execution_count": 7,
- "metadata": {},
- "outputs": [],
- "source": [
- "from paddle_quantum.qchem import fermionic_hamiltonian\n",
- "\n",
- "H_of_water = fermionic_hamiltonian(\n",
- " h2o_moledata,\n",
- " multiplicity=1,\n",
- " active_electrons=4,\n",
- " active_orbitals=4\n",
- ")\n",
- "\n",
- "from openfermion.ops import FermionOperator\n",
- "\n",
- "assert isinstance(H_of_water, FermionOperator)"
- ]
- },
- {
- "cell_type": "markdown",
- "metadata": {},
- "source": [
- "By specifying `active_electrons` and `active_orbitals` keyword arguments, we can reduce the number of freedom of our Hamiltonian and thus reduce the number of terms in the spin Hamiltonian described in the next section. We can also use `active_space` function in *qchem* to return a list of *core* orbitals and *active* orbitals. "
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 8,
+ "execution_count": 6,
"metadata": {},
"outputs": [
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "INFO:root:\n",
+ "#######################################\n",
+ "SCF Calculation (Classical)\n",
+ "#######################################\n",
+ "INFO:root:Basis: sto-3g\n"
+ ]
+ },
{
"name": "stdout",
"output_type": "stream",
"text": [
- "List of core orbitals: [0, 1, 2]\n",
- "List of active orbitals: [3, 4, 5, 6]\n"
+ "converged SCF energy = -73.9677038774737\n"
+ ]
+ },
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "INFO:root:SCF energy: -73.96770.\n"
]
}
],
"source": [
- "from paddle_quantum.qchem import active_space\n",
+ "# Build `Molecule` for water molecule.\n",
+ "from paddle_quantum.qchem import Molecule\n",
"\n",
- "core_orbits_list, act_orbits_list = active_space(\n",
- " 10, # number of electrons in water molecule\n",
- " 7, # number of molecular orbitals in water molecule\n",
- " active_electrons=4,\n",
- " active_orbitals=4\n",
+ "mol = Molecule(\n",
+ " geometry=h2o_coords,\n",
+ " basis=\"sto-3g\",\n",
+ " driver=PySCFDriver()\n",
")\n",
"\n",
- "print(\"List of core orbitals: {:}\".format(core_orbits_list))\n",
- "print(\"List of active orbitals: {:}\".format(act_orbits_list))"
+ "\n",
+ "H_of_water = mol.get_molecular_hamiltonian()"
]
},
{
+ "attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
- "## From Fermionic Hamiltonian to spin Hamiltonian\n",
- "In quantum computing, we only have qubit operators composed of Pauli matrices\n",
- "\n",
- "$$\n",
- "\\boldsymbol{\\sigma}_x=\\begin{pmatrix}\n",
- "0 & 1\\\\\n",
- "1 & 0\n",
- "\\end{pmatrix},\\quad \\boldsymbol{\\sigma}_y=\\begin{pmatrix}\n",
- "0 & -i\\\\\n",
- "i & 0\n",
- "\\end{pmatrix},\\quad \\boldsymbol{\\sigma}_z=\\begin{pmatrix}\n",
- "1 & 0\\\\\n",
- "0 & -1\n",
- "\\end{pmatrix}.\\tag{4}\n",
- "$$\n",
- "\n",
- "Therefore, we need to transform our Hamiltonian in the previous section to qubit operators, [Jordan-Wigner transform](https://en.wikipedia.org/wiki/Jordan%E2%80%93Wigner_transformation) is one of the well-known methods to realize the transformation.\n",
- "> Alternatively, we also provide Bravyi-Kitaev transformation, by changing the argument, mapping_method, to 'bravyi_kitaev'.\n",
- "\n",
- "In *paddle quantum*, Hamiltonian is encoded in *pauli_str*. To avoid tedious manipulation of *string* object, we have provided `spin_hamiltonian` function which can generate the needed *pauli_str* from molecular structure on the fly."
+ "Since quantum computing only allows operation based on Pauli operators $\\hat{\\sigma}_x$, $\\hat{\\sigma}_y$, $\\hat{\\sigma}_z$, we have to transform Hamiltonian in above expression to a form represented by Pauli operators. This can be achieved via [Jordan-Wigner 变换](https://en.wikipedia.org/wiki/Jordan%E2%80%93Wigner_transformation) and we have already integrated it in `get_molecular_hamiltonian` method."
]
},
{
"cell_type": "code",
- "execution_count": 9,
+ "execution_count": 7,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "There are 193 terms in H2O Hamiltonian in total.\n",
+ "There are 2110 terms in H2O Hamiltonian in total.\n",
"The first 10 terms are \n",
- " -72.10615980544185 I\n",
- "-0.007310917992546774 X0, X1, Y2, Y3\n",
- "0.0052460870730834325 X0, X1, Y2, Z3, Z4, Y5\n",
- "0.0016283548447087654 X0, X1, Y2, Z3, Z4, Z5, Z6, Y7\n",
- "0.0052460870730834325 X0, X1, X3, X4\n",
- "0.0016283548447087654 X0, X1, X3, Z4, Z5, X6\n",
- "-0.005994544380559027 X0, X1, Y4, Y5\n",
- "0.001387644178102622 X0, X1, Y4, Z5, Z6, Y7\n",
- "0.001387644178102622 X0, X1, X5, X6\n",
- "-0.009538223793221182 X0, X1, Y6, Y7\n"
+ " -44.808115297466266 I\n",
+ "12.537584025014317 Z0\n",
+ "12.537584025014313 Z1\n",
+ "1.8115178684479893 Z2\n",
+ "1.8115178684479893 Z3\n",
+ "1.4546840922727915 Z4\n",
+ "1.4546840922727915 Z5\n",
+ "1.4299903414012243 Z6\n",
+ "1.429990341401224 Z7\n",
+ "1.0849294605602529 Z8\n"
]
}
],
"source": [
- "from paddle_quantum.qchem import spin_hamiltonian\n",
- "\n",
- "pauli_H_of_water_ = spin_hamiltonian(\n",
- " h2o_moledata,\n",
- " multiplicity=1,\n",
- " active_electrons=4,\n",
- " active_orbitals=4,\n",
- " mapping_method='jordan_wigner'\n",
- ")\n",
- "\n",
- "print('There are ', pauli_H_of_water_.n_terms, 'terms in H2O Hamiltonian in total.')\n",
- "print('The first 10 terms are \\n', pauli_H_of_water_[:10])"
+ "print('There are', H_of_water.n_terms, 'terms in H2O Hamiltonian in total.')\n",
+ "print('The first 10 terms are \\n', H_of_water[:10])"
]
},
{
@@ -308,25 +275,20 @@
]
},
{
+ "attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"---\n",
"## References\n",
"\n",
- "[1] [Psi4: an open-source ab initio electronic structure program](https://wires.onlinelibrary.wiley.com/doi/abs/10.1002/wcms.93)\n",
- "\n",
- "[2] [OpenFermion: the electronic structure package for quantum computers\n",
- "](https://iopscience.iop.org/article/10.1088/2058-9565/ab8ebc)"
+ "[1] [PySCF: Python-based Simulations of Chemistry Framework](https://pyscf.org/)"
]
}
],
"metadata": {
- "interpreter": {
- "hash": "f7cfecff1ef1940b21a48efa1b88278bb096bd916f13c2df11af4810c38b47e1"
- },
"kernelspec": {
- "display_name": "Python 3.8.0 ('pq')",
+ "display_name": "modellib",
"language": "python",
"name": "python3"
},
@@ -340,7 +302,12 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
- "version": "3.8.0"
+ "version": "3.7.15"
+ },
+ "vscode": {
+ "interpreter": {
+ "hash": "8f24120f890011f53feb4ed62c47961d8565ec1de8b7cb23548c15bd6da8f2d2"
+ }
}
},
"nbformat": 4,
diff --git a/tutorials/quantum_simulation/VQE_CN.ipynb b/tutorials/quantum_simulation/VQE_CN.ipynb
index 349e137..13bdf69 100644
--- a/tutorials/quantum_simulation/VQE_CN.ipynb
+++ b/tutorials/quantum_simulation/VQE_CN.ipynb
@@ -103,6 +103,7 @@
]
},
{
+ "attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
@@ -110,7 +111,7 @@
"\n",
"### 构造电子哈密顿量\n",
"\n",
- "首先，让我们通过下面几行代码引入必要的 library 和 package。量桨的量子化学工具包是基于 `psi4` 和 `openfermion` 进行开发的，所以需要读者先行安装这两个语言包。在进入下面的教程之前，我们强烈建议您先阅读[哈密顿量的构造](./BuildingMolecule_CN.ipynb)教程，该教程介绍了如何使用量桨的量子化学工具包。\n",
+ "首先，让我们通过下面几行代码引入必要的 library 和 package。量桨的量子化学工具包是基于 *openfermion* 进行开发的，目前使用的量子化学引擎是 *PySCF*，所以需要读者先行安装这两个语言包（注：*PySCF* 当前仅支持 Mac 和 Linux 平台，我们正在努力开发支持更多量子化学工具的引擎，将在 Paddle Quantum 下一次更新中完善对 Windows 平台的支持）。在进入下面的教程之前，我们强烈建议您先阅读[哈密顿量的构造](./BuildingMolecule_CN.ipynb)教程，该教程介绍了如何使用量桨的量子化学工具包。\n",
"\n",
"**注意：关于环境设置，请参考 [README_CN.md](https://github.com/PaddlePaddle/Quantum/blob/master/README_CN.md).**"
]
@@ -148,10 +149,11 @@
]
},
{
+ "attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
- "对于具体需要分析的分子，我们需要其**几何构型** (geometry)、**基组**（basis set，例如 STO-3G 基于高斯函数）、**多重度**（multiplicity）以及**分子的净电荷数** (charge) 等多项信息来建模计算出该分子单体积分 (one-body integrations)，双体积分(two-body integrations) 以及哈密顿量等信息。接下来，通过量桨的量子化学工具包将分子的哈密顿量提取出来并储存为 paddle quantum 的 `Hamiltonian` 类，方便我们下一步的操作。"
+ "对于具体需要分析的分子，我们需要其**几何构型** (geometry)、**基组**（basis set，例如 STO-3G 基于高斯函数）、**多重度**（multiplicity）以及**分子的净电荷数** (charge) 等多项信息来建模计算出该分子单体积分 (one-body integrations)，双体积分(two-body integrations) 以及哈密顿量等信息。利用量桨量子化学模块的 `Molecule` 类可以方便地获得储存为paddle quantum 的 `Hamiltonian` 类的分子哈密顿量，方便我们下一步的操作。"
]
},
{
@@ -168,57 +170,50 @@
"name": "stdout",
"output_type": "stream",
"text": [
- "FCI energy for H2_sto-3g_singlet (2 electrons) is -1.137283834485513.\n",
+ "converged SCF energy = -1.11675930739643\n",
"\n",
"The generated h2 Hamiltonian is \n",
- " -0.09706626861762556 I\n",
- "-0.04530261550868938 X0, X1, Y2, Y3\n",
- "0.04530261550868938 X0, Y1, Y2, X3\n",
- "0.04530261550868938 Y0, X1, X2, Y3\n",
- "-0.04530261550868938 Y0, Y1, X2, X3\n",
- "0.1714128263940239 Z0\n",
- "0.16868898168693286 Z0, Z1\n",
- "0.12062523481381837 Z0, Z2\n",
- "0.16592785032250773 Z0, Z3\n",
- "0.17141282639402394 Z1\n",
- "0.16592785032250773 Z1, Z2\n",
- "0.12062523481381837 Z1, Z3\n",
- "-0.2234315367466399 Z2\n",
- "0.17441287610651626 Z2, Z3\n",
- "-0.2234315367466399 Z3\n"
+ " -0.09706626816763092 I\n",
+ "0.17141282644776917 Z0\n",
+ "0.17141282644776917 Z1\n",
+ "-0.2234315369081346 Z2\n",
+ "-0.2234315369081346 Z3\n",
+ "0.16868898170361205 Z0, Z1\n",
+ "0.12062523483390414 Z0, Z2\n",
+ "0.1659278503377034 Z0, Z3\n",
+ "0.1659278503377034 Z1, Z2\n",
+ "0.12062523483390414 Z1, Z3\n",
+ "0.1744128761226159 Z2, Z3\n",
+ "-0.04530261550379926 X0, X1, Y2, Y3\n",
+ "0.04530261550379926 X0, Y1, Y2, X3\n",
+ "0.04530261550379926 Y0, X1, X2, Y3\n",
+ "-0.04530261550379926 Y0, Y1, X2, X3\n"
]
}
],
"source": [
- "geo = qchem.geometry(structure=[['H', [-0., 0., 0.0]], ['H', [-0., 0., 0.74]]])\n",
- "# geo = qchem.geometry(file='h2.xyz')\n",
- "\n",
- "# 将分子信息存储在 molecule 里，包括单体积分（one-body integrations），双体积分（two-body integrations），分子的哈密顿量等\n",
- "molecule = qchem.get_molecular_data(\n",
- " geometry=geo,\n",
- " basis='sto-3g',\n",
- " charge=0,\n",
- " multiplicity=1,\n",
- " method=\"fci\",\n",
- " if_save=True,\n",
- " if_print=True\n",
+ "mol = qchem.Molecule(\n",
+ " geometry=[('H', [-0., 0., 0.0]), ('H', [-0., 0., 0.74])], # 分子几何结构\n",
+ " basis=\"sto-3g\", # 基组\n",
+ " multiplicity=1, # 多重度\n",
+ " charge=0, # 净电荷数\n",
+ " driver=qchem.PySCFDriver() # 量子化学积分计算引擎\n",
")\n",
"# 提取哈密顿量\n",
- "molecular_hamiltonian = qchem.spin_hamiltonian(molecule=molecule,\n",
- " filename=None, \n",
- " multiplicity=1, \n",
- " mapping_method='jordan_wigner',)\n",
+ "molecular_hamiltonian = mol.get_molecular_hamiltonian()\n",
+ "\n",
"# 打印结果\n",
"print(\"\\nThe generated h2 Hamiltonian is \\n\", molecular_hamiltonian)"
]
},
{
+ "attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
"**注释4：** 生成这个哈密顿量的几何构型中，两个氢原子间的原子间隔（interatomic distance）为 $d = 74$ pm。\n",
"\n",
- "除了输入分子的几何结构外，我们还支持读取分子的几何构型文件 (`.xyz` 文件)，关于量子化学工具包更多的用法请参考[哈密顿量的构造](./BuildingMolecule_CN.ipynb)教程。如果你需要测试更多分子的几何构型，请移步至这个[数据库](http://smart.sns.it/molecules/index.html)。"
+ "关于量子化学工具包更多的用法请参考[哈密顿量的构造](./BuildingMolecule_CN.ipynb)教程。如果你需要测试更多分子的几何构型，请移步至这个[数据库](http://smart.sns.it/molecules/index.html)。"
]
},
{
@@ -349,23 +344,23 @@
"name": "stdout",
"output_type": "stream",
"text": [
- "iter: 20 loss: -1.0700\n",
- "iter: 20 Ground state energy: -1.0700 Ha\n",
- "iter: 40 loss: -1.1309\n",
- "iter: 40 Ground state energy: -1.1309 Ha\n",
- "iter: 60 loss: -1.1362\n",
- "iter: 60 Ground state energy: -1.1362 Ha\n",
+ "iter: 20 loss: -1.0784\n",
+ "iter: 20 Ground state energy: -1.0784 Ha\n",
+ "iter: 40 loss: -1.1300\n",
+ "iter: 40 Ground state energy: -1.1300 Ha\n",
+ "iter: 60 loss: -1.1366\n",
+ "iter: 60 Ground state energy: -1.1366 Ha\n",
"iter: 80 loss: -1.1372\n",
"iter: 80 Ground state energy: -1.1372 Ha\n",
"\n",
"训练后的电路：\n",
- "--Ry(7.856)----*--------------x----Ry(4.698)----*--------------x----Ry(6.277)--\n",
+ "--Ry(1.589)----*--------------x----Ry(4.701)----*--------------x----Ry(3.124)--\n",
" | | | | \n",
- "--Ry(1.548)----x----*---------|----Ry(-1.56)----x----*---------|----Ry(5.041)--\n",
+ "--Ry(4.703)----x----*---------|----Ry(1.573)----x----*---------|----Ry(1.524)--\n",
" | | | | \n",
- "--Ry(3.441)---------x----*----|----Ry(4.474)---------x----*----|----Ry(1.745)--\n",
+ "--Ry(3.096)---------x----*----|----Ry(4.493)---------x----*----|----Ry(4.904)--\n",
" | | | | \n",
- "--Ry(-0.17)--------------x----*----Ry(1.646)--------------x----*----Ry(3.152)--\n",
+ "--Ry(3.340)--------------x----*----Ry(1.555)--------------x----*----Ry(3.157)--\n",
" \n"
]
}
@@ -437,14 +432,12 @@
"outputs": [
{
"data": {
- "image/png": "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",
+ "image/png": "iVBORw0KGgoAAAANSUhEUgAAAksAAAGwCAYAAAC5ACFFAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/NK7nSAAAACXBIWXMAAA9hAAAPYQGoP6dpAABegElEQVR4nO3deVhU1eMG8HfYBmRVQRZFEFfcFZNwSRNNxT1zKco1bdHSMkuz1OpraqmVVpotYj81tVKzzC1Fc8sFRUURFVFMQVIEBGSbOb8/jjMwssgywzDD+3meeWa4c+695zIw8845556rEEIIEBEREVGRLIxdASIiIqKqjGGJiIiIqAQMS0REREQlYFgiIiIiKgHDEhEREVEJGJaIiIiISsCwRERERFQCK2NXwByo1WrcvHkTjo6OUCgUxq4OERERlYIQAvfu3YOXlxcsLIpvP2JY0oObN2/C29vb2NUgIiKicrh+/Trq1atX7PMMS3rg6OgIQP6ynZycjFwbIiIiKo20tDR4e3trP8eLw7CkB5quNycnJ4YlIiIiE/OoITQc4E1ERERUAoYlIiIiohIwLBERERGVgGOWiIjMnFqtRk5OjrGrQVTprK2tYWlpWeHtMCwREZmxnJwcxMXFQa1WG7sqREbh4uICDw+PCs2DyLBERGSmhBBISEiApaUlvL29S5x0j8jcCCGQmZmJpKQkAICnp2e5t8WwRERkpvLy8pCZmQkvLy/UqFHD2NUhqnR2dnYAgKSkJNSpU6fcXXL8mkFEZKZUKhUAwMbGxsg1ITIezReF3Nzccm+DYYmIyMzxmpVUnenj759hiYiIiKgEDEtEREREJTCrsJScnIzQ0FA4OTnBxcUF48ePR3p6eonlX3vtNTRt2hR2dnaoX78+Xn/9daSmplZirYmIiKgqM6uwFBoainPnzmH37t34448/8Pfff2PixInFlr958yZu3ryJRYsWISoqCmFhYdixYwfGjx9fibUuQV4eEBVl7FoQERFVa2YTlqKjo7Fjxw589913CAwMRJcuXbBs2TKsX78eN2/eLHKdli1b4tdff8WAAQPQsGFD9OjRA/PmzcPvv/+OvLy8YveVnZ2NtLQ0nZve5eQAo0cDM2cCCQn63z4RkQlYtmxZse/hxblz5w7q1KmDq1evapcJIbBkyRI0aNAANWrUwODBg3V6EUaOHInFixfrbGffvn3w9fUtV70rsm5R9QdKPoai6g8AP/zwAy5cuFCuelA+swlLR44cgYuLCzp06KBd1rNnT1hYWODo0aOl3k5qaiqcnJxgZVX8FFTz58+Hs7Oz9ubt7V2huhfJxgbw85OPDx3S//aJiKq4y5cv45133kHNmjXLtN68efMwaNAgnbAyffp0LF++HKtXr8aBAwcQERGBuXPnap9/7733MG/evFINw+jWrRvGjRtXaPnXX38NBweHCs+WXlT9H3UMxdX/6NGj+OabbypUHzKjsJSYmIg6deroLLOyskKtWrWQmJhYqm3cvn0bH330UYlddwAwc+ZMpKamam/Xr18vd71L1LmzvD940DDbJ6LqRQggK8s4NyHKXN3ffvsNvXr10k4sWBqZmZn4/vvvdYZTHD16FEuWLMGGDRvwxBNPICAgABMmTMCff/6pLdOyZUs0bNgQa9asecSvUODUqVMICAgo9NyJEyfQtm3bCs2UXlT9S3MMxdV/0KBB2Lp1a7nrQ1KVn8F7xowZWLhwYYlloqOjK7yftLQ09OvXD82bN9f5tlEUpVIJpVJZ4X0+UlAQsHw5EBsru+IqMFU7ERGys4Fhw4yz759/Bmxty7TKb7/9htGjR2t/3rJlC8aOHYu7d+8iNjYWjRo1QkJCAlxdXeHo6IjNmzcjPT0dSqUSjz/+uHa9RYsWITg4GO3bt9cuc3d3x+3bt3X2N2DAAKxfvx6TJk0qtk6XLl3CvXv3ig1LTz75ZLHrlrf+pT2GouofHByMW7duISoqCi1btiy2blSyKt+yNG3aNERHR5d48/Pzg4eHh/b6Lxp5eXlITk6Gh4dHifu4d+8e+vTpo/1jtba2NuQhlZ6zM9C6tXzMrjgiqkZu376Nf/75B/3799cui4yMRJs2bQAAp0+fhru7Ozw8PHDhwgVkZWWhbdu2OHDggE6Qyc7OxrZt2zBkyBCd7WdlZcHZ2VlnWceOHXHs2DFkZ2cXW6+IiAhYWlpq66Fx//59nD9/XifMPKw89S/LMRRVf6VSiaeeeoqtSxVU5VuW3Nzc4Obm9shyQUFBSElJQUREhPYPbe/evVCr1QgMDCx2vbS0NPTu3RtKpRJbt26FbRm/+Rhc585AZKTsinvmGWPXhohMmVIpW3iMte8y+OOPP9ChQwe4u7trl50+fVonbBQVPK5duwYvLy/tOidPnsT9+/cxbdo0vP3229rlubm5hVqBvLy8kJOTg8TERPj4+BRZr5MnT0KlUhV7rb2SwlJ56l+WYyiu/oMGDcLy5cvx7rvvFls3KlmVD0ul5e/vjz59+mDChAlYsWIFcnNzMXnyZIwcOVL7h3fjxg0EBwfjxx9/RMeOHZGWloannnoKmZmZWLNmjc6ZbW5ubuW+4J5esSuOiPRFoShzV5ix/PnnnwgJCdFZFhkZiQEDBgDQDRuRkZFo27YtANnCU/BL78WLF2Fvb4/IyEidbfXr1w+dNeNCH9CMjcrMzCy2XidPnsSQIUMwe/ZsneXr16/H0qVL0bx582LXLU/9y3IMxdU/JCQEY8eOxe3bt+Hq6lps/ah4Vb4brizWrl2LZs2aITg4GCEhIejSpQtWrlypfT43NxcxMTHaP6STJ0/i6NGjOHv2LBo1agRPT0/tzWCDtsuKXXFEVA35+voiLi5O+3NaWhquXr2qHXdTMGycPHkS7dq1AwC4urri7t27Ouu5urqiUaNG2pu1tTUuXbqEoUOH6uwzOTkZAErszTh58iS6d++Otm3b6tySk5PRunXrYr9kl7f+ZTmG4uofFxcHFxcXuLi4FHtcVDKzCku1atXCunXrcO/ePaSmpuKHH36Ag4OD9nlfX18IIdC9e3cAQPfu3SGEKPJW3vkxDKJLF3nPs+KIqJoYNGgQtm3bpj0NP+HBfHOOjo5ITU3F1atX0aZNGyQlJeHgwYPo2bMnAKBdu3Y4f/68djuurq5ITU2FKHA23rx58xASElKoFSgqKgr16tUrtvXlypUrSElJKbKr7eTJk0UO+tYob/3LcgzF1X/r1q0ICQkpcUocKplZhSWzFRQEWFjkd8UREZm5oKAgCCG08+TVrVsXdnZ2WLJkCfbt2wdra2vcv38fQ4YMQWBgIHr06AEA6N27N86dO6dtnenRoweysrKwYMECxMXF4X//+x9+//13LF++vNA+Dxw4gKeeeqrYOkVERMDCwkLbZaaRm5uLqKioEscrlbf+ZTmG4uq/detWDBo0qNi60aMxLJkCJyd2xRFRtWJhYYH+/fvjt99+AwA4ODhg48aN2Lt3LwYPHozc3Fz07dsXnTp1wrZt26BQKAAArVq1Qvv27bFx40YA8vT6sLAwLF++HC1atMA///yDgwcPFppMOCsrC1u2bMGECROKrdPJkyfRuHFjnR4LADh//jyys7NLDEvlrX9pj6G4+sfFxSEmJgZ9+vQptm5UCoIqLDU1VQAQqamphtvJjh1C9O8vxOuvG24fRGRW7t+/L86fPy/u379v7KqUy2+//Sb8/f0LLX/22WfFs88+K9RqdZHr/fHHH8Lf31+oVKpS7+vrr78WvXr10lkWHh4ufHx8ylTn0qxbWfUXQojPP/9cPPXUU6Xejjkq6f+gtJ/fbFkyFZquuCtX2BVHRNVCr169cO3aNVy+fFlneUxMDAIDA7WtMQ/r168fJk6ciBs3bpR6X9bW1li2bFmF6ltalVn/rVu3YuDAgeWuK0kc7WUqNF1xkZGyK45zLhGRmbOzs0NGRobOsry8PJw7d67QuKGHTZ06tUz7evHFF8tYu/Kp7Prv2bOnTNuhojEsmZIuXWRYOnCAYYmIqiUrKytkZWVVyr58fX3LHFoetW5l1p/0h91wpoRdcURElcYQYYlME8OSKSl4VhznXCIiIqoUDEumpmtXec8pBIiIiCoFw5KpeTBVPspwlgQRERGVH8OSqdFc6TorCygw9T0REREZBsOSqXlwVWkAQHa28epBRERUTTAsmRobG0AzkVlmpnHrQkREVA0wLJkahSK/dYlzdRARVSlz58595ISTZHoYlkyRra28Z1giIjOVmJiIKVOmoFGjRrC1tYW7uzs6d+6M5cuXI9OMW9XHjBmDwYMHl3k9hjTD4gzepkjTsmTGbxhEVH1duXIFnTt3houLCz7++GO0atUKSqUSZ8+excqVK1G3bt1ir3eWm5sLa2vrSq4xVZRKpYJCoYCFRdVsw6mataKSsRuOiMzYq6++CisrK5w4cQLDhw+Hv78//Pz8MGjQIGzbtg0DBgzQllUoFFi+fDkGDhwIe3t7zJs3DwCwfPlyNGzYEDY2NmjatCn+7//+T7vO1atXoVAoEBkZqV2WkpIChUKBffv2AQD27dsHhUKBPXv2oEOHDqhRowY6deqEmJgYnbouWLAA7u7ucHR0xPjx40t1KZNffvkFrVq1gp2dHWrXro2ePXsiIyMDc+fOxerVq/Hbb79BoVDo1Oedd95BkyZNUKNGDfj5+eH9999Hbm4uACAsLAwffPABTp8+rV0vLCxMe1wvvvgi3Nzc4OTkhB49euD06dMl1u/69esYPnw4XFxcUKtWLQwaNAhXr17VPq9p/Vq0aBE8PT1Ru3ZtTJo0SVsfAMjOzsZbb72FunXrwt7eHoGBgdpj0dTZxcUFW7duRfPmzaFUKhEfH4+EhAT069cPdnZ2aNCgAdatWwdfX198/vnnAIBx48ahf//+OvXNzc1FnTp18P333z/yd19eDEumSNMNd/++cetBRCYpK6vw7CN5eXJZgc+7R5bNySld2bK4c+cOdu3ahUmTJsHe3r7IMgrNSS4PzJ07F0OGDMHZs2cxbtw4bN68GVOmTMG0adMQFRWFl156CWPHjkV4eHjZKgNg1qxZWLx4MU6cOAErKyuMGzdO+9zGjRsxd+5cfPzxxzhx4gQ8PT3x9ddfl7i9hIQEPPvssxg3bhyio6Oxb98+PP300xBC4K233sLw4cPRp08fJCQkICEhAZ06dQIAODo6IiwsDOfPn8cXX3yBb7/9Fp999hkAYMSIEZg2bRpatGihXW/EiBEAgGHDhiEpKQnbt29HREQE2rdvj+DgYCQnJxdZv9zcXPTu3RuOjo44cOAADh06BAcHB/Tp0wc5BV7w8PBwxMbGIjw8HKtXr0ZYWJg2oAHA5MmTceTIEaxfvx5nzpzBsGHD0KdPH1y6dElbJjMzEwsXLsR3332Hc+fOoU6dOhg1ahRu3ryJffv24ddff8XKlSuRlJSkXefFF1/Ejh07kFDgkl9//PEHMjMztcdsEIIqLDU1VQAQqamplbPDDz4Qon9/IXbtqpz9EZFJun//vjh//ry4f/++zvL+/eUtJSV/2YYNctnSpbrbGDpULr91K3/Zli1y2aef6pZ97jm5/Nq1/GU7dpStzv/8848AIDZt2qSzvHbt2sLe3l7Y29uLt99+W7scgJg6dapO2U6dOokJEyboLBs2bJgICQkRQggRFxcnAIhTp05pn797964AIMLDw4UQQoSHhwsA4q+//tKW2bZtmwCg/X0GBQWJV199VWc/gYGBok2bNsUeX0REhAAgrl69WuTzo0ePFoMGDSp2fY1PP/1UBAQEaH+eM2dOof0eOHBAODk5iaysLJ3lDRs2FN98802R2/2///s/0bRpU6FWq7XLsrOzhZ2dndi5c6e2jj4+PiIvL09bZtiwYWLEiBFCCCGuXbsmLC0txY0bN3S2HRwcLGbOnCmEEGLVqlUCgIiMjNQ+Hx0dLQCI48ePa5ddunRJABCfffaZdlnz5s3FwoULtT8PGDBAjBkzpsjjEaL4/wMhSv/5zZYlU8QxS0RUzRw7dgyRkZFo0aIFsh+aY65Dhw46P0dHR6Nz5846yzp37ozo6Ogy77e15nqcADw9PQFA29IRHR2NwMBAnfJBQUHaxwcOHICDg4P2tnbtWrRp0wbBwcFo1aoVhg0bhm+//RZ37959ZD02bNiAzp07w8PDAw4ODnjvvfcQHx9f4jqnT59Geno6ateurVOPuLg4xMbGFrvO5cuX4ejoqC1fq1YtZGVl6azTokULWFpa6vxuNL+Xs2fPQqVSoUmTJjr73b9/v842bGxsdH6/MTExsLKyQvv27bXLGjVqhJo1a+rU8cUXX8SqVasAALdu3cL27dt1WvwMgQO8TRHHLBFRBfz8s7xXKvOXPf00MHAgUODzDwCwZk3hsv36Ab17Aw+PxdUMGSlYNji4bHVr1KgRFApFobFBfn5+AAC7ghPzPlBcd11xNIOIRYH+wtyH+x8fKDhYXNP9p1arS7WfDh066IyLcnd3h6WlJXbv3o3Dhw9j165dWLZsGWbNmoWjR4+iQYMGRW7nyJEjCA0NxQcffIDevXvD2dkZ69evx+LFi0vcf3p6Ojw9PXXGCmm4uLgUu05AQADWrl1b6Dk3Nzft44cH0SsUCu3vJT09HZaWloiIiNAJVADg4OCgfWxnZ1eoS7U0Ro0ahRkzZuDIkSM4fPgwGjRogK6a66YaCMOSKeLUAURUAZq3kIKsrOTNEGXLonbt2ujVqxe+/PJLvPbaa2UOQgDg7++PQ4cOYfTo0dplhw4dQvPmzQHkf+gnJCSgXbt2AKATasqyn6NHj2LUqFHaZf/884/2sZ2dHRo1alRoPYVCgc6dO6Nz586YPXs2fHx8sHnzZrz55puwsbGBSqXSKX/48GH4+Phg1qxZ2mXXrl3TKVPUeu3bt0diYiKsrKzg6+tbqmNq3749NmzYgDp16sDJyalU6zysXbt2UKlUSEpKKlOIadq0KfLy8nDq1CkEBAQAAC5fvlyo5a127doYPHgwVq1ahSNHjmDs2LHlqmdZsBvOFLEbjojM2Ndff428vDx06NABGzZsQHR0NGJiYrBmzRpcuHChUGvFw6ZPn46wsDAsX74cly5dwpIlS7Bp0ya89dZbAGSIefzxx7FgwQJER0dj//79eO+998pczylTpuCHH37AqlWrcPHiRcyZMwfnzp0rcZ2jR49qB4THx8dj06ZN+O+//+Dv7w8A8PX1xZkzZxATE4Pbt28jNzcXjRs3Rnx8PNavX4/Y2FgsXboUmzdv1tmur68v4uLiEBkZidu3byM7Oxs9e/ZEUFAQBg8ejF27duHq1as4fPgwZs2ahRMnThRZv9DQULi6umLQoEE4cOAA4uLisG/fPrz++uv4999/S/V7adKkCUJDQzFq1Chs2rQJcXFxOHbsGObPn49t27YVu16zZs3Qs2dPTJw4EceOHcOpU6cwceLEIlugXnzxRaxevRrR0dE6odhgShzRRKVS6QO8f/1VjqJcsqRy9kdEJqmkga1V3c2bN8XkyZNFgwYNhLW1tXBwcBAdO3YUn376qcjIyNCWAyA2b95caP2vv/5a+Pn5CWtra9GkSRPx448/6jx//vx5ERQUJOzs7ETbtm3Frl27ihzgfffuXe06p06dEgBEXFycdtm8efOEq6urcHBwEKNHjxZvv/12iQO8z58/L3r37i3c3NyEUqkUTZo0EcuWLdM+n5SUJHr16iUcHBx06jN9+nRRu3Zt4eDgIEaMGCE+++wz4ezsrF0vKytLDB06VLi4uAgAYtWqVUIIIdLS0sRrr70mvLy8hLW1tfD29hahoaEiPj6+2DomJCSIUaNGCVdXV6FUKoWfn5+YMGGC9jOuqEHoU6ZMEd26ddP+nJOTI2bPni18fX2FtbW18PT0FEOGDBFnzpwRQsgB3gXrr3Hz5k3Rt29foVQqhY+Pj1i3bp2oU6eOWLFihU45tVotfHx8tIP2S6KPAd4KIXjp+opKS0uDs7MzUlNTy91sWSZ//gksXw4EBQHvvmv4/RGRScrKykJcXBwaNGgA26L6yIiquH///Rfe3t7466+/EFxgAFx6ejrq1q2LVatW4emnny5xGyX9H5T285tjlkyRphuO8ywREZEZ2bt3L9LT09GqVSskJCTg7bffhq+vL5544gkAcnD97du3sXjxYri4uBQ7k7u+MSyZIoYlIiIyQ7m5uXj33Xdx5coVODo6olOnTli7dq327Lv4+Hg0aNAA9erVQ1hYGKzKegZBOTEsmSJOHUBERGaod+/e6N27d7HP+/r6whijh3g2nCni5U6IiIgqDcOSKWI3HBGVAc/joepMH3//DEumiN1wRFQKmvmIch6+4i1RNZL5YE7Ch2cdLwuOWTJFmm643Fx5Se9KGuBGRKbFysoKNWrUwH///Qdra2vtZT6IqgMhBDIzM5GUlAQXF5dHTmZaEn7KmqKC10bKygIKXGuHiEhDoVDA09MTcXFxhS6PQVRduLi4wMPDo0LbYFgyRZoLM+XlyXFLDEtEVAwbGxs0btyYXXFULVlbW1eoRUmDYclU2dkB9+5x3BIRPZKFhQVn8CaqAHZgmyqeEUdERFQpGJZMFedaIiIiqhQMS6aK0wcQERFVCoYlU6UJSw/mjyAiIiLDYFgyVWxZIiIiqhQMS6aKY5aIiIgqBcOSqeLZcERERJWCYclUMSwRERFVCoYlU6XphuOYJSIiIoNiWDJVbFkiIiKqFAxLpophiYiIqFIwLJkqTh1ARERUKRiWTBWnDiAiIqoUDEumit1wRERElYJhyVQxLBEREVUKhiVTxakDiIiIKgXDkqkq2LIkhHHrQkREZMYYlkyVJiwJAeTkGLcuREREZoxhyVRpuuEAdsUREREZEMOSqVIo8gNTZqZx60JERGTGGJZMGSemJCIiMjiGJVPGiSmJiIgMjmHJlHGuJSIiIoNjWDJl7IYjIiIyOIYlU8ZuOCIiIoMzq7CUnJyM0NBQODk5wcXFBePHj0d6enqp1hVCoG/fvlAoFNiyZYthK6ov7IYjIiIyOLMKS6GhoTh37hx2796NP/74A3///TcmTpxYqnU///xzKBQKA9dQzxiWiIiIDM7K2BXQl+joaOzYsQPHjx9Hhw4dAADLli1DSEgIFi1aBC8vr2LXjYyMxOLFi3HixAl4enpWVpUrjmOWiIiIDM5sWpaOHDkCFxcXbVACgJ49e8LCwgJHjx4tdr3MzEw899xz+Oqrr+Dh4VGqfWVnZyMtLU3nZhQcs0RERGRwZhOWEhMTUadOHZ1lVlZWqFWrFhITE4td74033kCnTp0waNCgUu9r/vz5cHZ21t68vb3LXe8KYTccERGRwVX5sDRjxgwoFIoSbxcuXCjXtrdu3Yq9e/fi888/L9N6M2fORGpqqvZ2/fr1cu2/wtgNR0REZHBVfszStGnTMGbMmBLL+Pn5wcPDA0lJSTrL8/LykJycXGz32t69exEbGwsXFxed5UOHDkXXrl2xb9++ItdTKpVQKpWlPQTDYTccERGRwVX5sOTm5gY3N7dHlgsKCkJKSgoiIiIQEBAAQIYhtVqNwMDAIteZMWMGXnzxRZ1lrVq1wmeffYYBAwZUvPKGxm44IiIig6vyYam0/P390adPH0yYMAErVqxAbm4uJk+ejJEjR2rPhLtx4waCg4Px448/omPHjvDw8Ciy1al+/fpo0KBBZR9C2TEsERERGVyVH7NUFmvXrkWzZs0QHByMkJAQdOnSBStXrtQ+n5ubi5iYGGRmZhqxlnrEMUtEREQGZzYtSwBQq1YtrFu3rtjnfX19IYQocRuPer5K4ZglIiIigzOrlqVqh91wREREBsewZMo0YSknB1CpjFsXIiIiM8WwZMo03XAAxy0REREZCMOSKbO2Biwt5WOGJSIiIoNgWDJlCgUHeRMRERkYw5Kp4yBvIiIig2JYMnUMS0RERAbFsGTqNN1wHLNERERkEAxLpo4tS0RERAbFsGTqGJaIiIgMimHJ1PH6cERERAbFsGTqOHUAERGRQTEsmTp2wxERERkUw5KpY1giIiIyKIYlU8epA4iIiAyKYcnUsWWJiIjIoBiWTB3DEhERkUExLJk6Th1ARERkUAxLpo5TBxARERkUw5KpYzccERGRQTEsmTqGJSIiIoNiWDJ1BacOEMK4dSEiIjJDDEumTtOypFIBubnGrQsREZEZYlgydZqWJYBnxBERERkAw5Kps7QEbGzkY4YlIiIivWNYMgearrjMTOPWg4iIyAwxLJkDTkxJRERkMAxL5oATUxIRERkMw5I54FxLREREBsOwZA4YloiIiAyGYckcFJyYkoiIiPSKYckcsGWJiIjIYBiWzAHDEhERkcEwLJkDTh1ARERkMAxL5oBTBxARERkMw5I5YDccERGRwTAsmQOGJSIiIoNhWDIHnDqAiIjIYBiWzAFbloiIiAyGYckcMCwREREZDMOSOeDUAURERAbDsGQOOHUAERGRwTAsmYMaNeR9VhYghHHrQkREZGYYlsyBpmUJYFccERGRnjEsmQMbG0ChkI/ZFUdERKRXDEvmQKHgGXFEREQGwrBkLjgxJRERkUEwLJkLtiwREREZBMOSuWBYIiIiMgiGJXPBiSmJiIgMwkrfG1Sr1di/fz8OHDiAa9euITMzE25ubmjXrh169uwJb29vfe+SAE5MSUREZCB6a1m6f/8+/ve//8Hb2xshISHYvn07UlJSYGlpicuXL2POnDlo0KABQkJC8M8//+hrt6TBbjgiIiKD0FvLUpMmTRAUFIRvv/0WvXr1grW1daEy165dw7p16zBy5EjMmjULEyZM0NfuiWGJiIjIIPQWlnbt2gV/f/8Sy/j4+GDmzJl46623EB8fr69dE8CpA4iIiAxEb91wjwpKBVlbW6Nhw4b62jUBbFkiIiIyEL0P8C4oMzMT8fHxyMnJ0VneunVrQ+62emJYIiIiMgiDhKX//vsPY8eOxfbt24t8XqVSGWK31VuNGvI+M9O49SAiIjIzBplnaerUqUhJScHRo0dhZ2eHHTt2YPXq1WjcuDG2bt1qiF2SJiyxZYmIiEivDNKytHfvXvz222/o0KEDLCws4OPjg169esHJyQnz589Hv379DLHb6o0tS0RERAZhkJaljIwM1KlTBwBQs2ZN/PfffwCAVq1a4eTJk4bYJTEsERERGYRBwlLTpk0RExMDAGjTpg2++eYb3LhxAytWrICnp6chdkmasJSRYdx6EBERmRmDhKUpU6YgISEBADBnzhxs374d9evXx9KlS/Hxxx8bYpcAgOTkZISGhsLJyQkuLi4YP3480tPTH7nekSNH0KNHD9jb28PJyQlPPPEE7pva2B+OWSIiIjIIg4xZev7557WPAwICcO3aNVy4cAH169eHq6urIXYJAAgNDUVCQgJ2796N3NxcjB07FhMnTsS6deuKXefIkSPo06cPZs6ciWXLlsHKygqnT5+GhYWJXWNYM3VATg6QlwdYGXRWCCIiompDIYQQxq6EPkRHR6N58+Y4fvw4OnToAADYsWMHQkJC8O+//8LLy6vI9R5//HH06tULH330Ubn3nZaWBmdnZ6SmpsLJyanc26mQvDxgyBD5eN06wNHROPUgIiIyEaX9/NZr88Obb75ZqnJLlizR524ByBYiFxcXbVACgJ49e8LCwgJHjx7FEE2QKCApKQlHjx5FaGgoOnXqhNjYWDRr1gzz5s1Dly5dit1XdnY2srOztT+npaXp92DKw8oKUCqB7Gw5yJthiYiISC/0GpZOnTql8/PBgwcREBAAO00XEQCFQqHPXWolJiZqz8DTsLKyQq1atZCYmFjkOleuXAEAzJ07F4sWLULbtm3x448/Ijg4GFFRUWjcuHGR682fPx8ffPCBfg9AH2rUyA9LREREpBd6DUvh4eE6Pzs6OmLdunXw8/Mr9zZnzJiBhQsXllgmOjq6XNtWq9UAgJdeegljx44FALRr1w579uzBDz/8gPnz5xe53syZM3Va0dLS0uDt7V2uOuhVjRrA3bs8I46IiEiPqvwo4GnTpmHMmDEllvHz84OHhweSkpJ0lufl5SE5ORkeHh5FrqeZxqB58+Y6y/39/REfH1/s/pRKJZRKZSlqX8l4RhwREZHeVfmw5ObmBjc3t0eWCwoKQkpKCiIiIhAQEABAziSuVqsRGBhY5Dq+vr7w8vLSzgmlcfHiRfTt27fila9smu5OdsMRERHpjYmdH188f39/9OnTBxMmTMCxY8dw6NAhTJ48GSNHjtSeCXfjxg00a9YMx44dAyDHT02fPh1Lly7FL7/8gsuXL+P999/HhQsXMH78eGMeTvlwFm8iIiK902vL0pkzZ3R+FkLgwoULhSaGbN26tT53q7V27VpMnjwZwcHBsLCwwNChQ7F06VLt87m5uYiJiUFmgTAxdepUZGVl4Y033kBycjLatGmD3bt3o2HDhgapo0HZ28t7hiUiIiK90es8SxYWFlAoFChqk5rlCoUCKpVKX7usEqrEPEsAsHIl8PvvwPDhwAsvGK8eREREJsAo8yzFxcXpc3NUVhyzREREpHd6DUs+Pj763ByVFccsERER6Z3eBniXdKp9UW7cuKGvXZMGwxIREZHe6S0sPfbYY3jppZdw/PjxYsukpqbi22+/RcuWLfHrr7/qa9ekwbBERESkd3rrhjt//jzmzZuHXr16wdbWFgEBAfDy8oKtrS3u3r2L8+fP49y5c2jfvj0++eQThISE6GvXpMFJKYmIiPROby1LtWvXxpIlS5CQkIAvv/wSjRs3xu3bt3Hp0iUAQGhoKCIiInDkyBEGJUNhyxIREZHe6X0Gbzs7OzzzzDN45pln9L1pehRNWOK14YiIiPTGbGbwJrBliYiIyAAYlsyJJizl5AB5ecatCxERkZlgWDInmkkpAQ7yJiIi0hOGJXNiZQXY2MjHDEtERER6YZCwlMEBxsbDcUtERER6ZZCw5O7ujnHjxuHgwYOG2DyVhGGJiIhIrwwSltasWYPk5GT06NEDTZo0wYIFC3Dz5k1D7IoexukDiIiI9MogYWnw4MHYsmULbty4gZdffhnr1q2Dj48P+vfvj02bNiGPZ2oZDmfxJiIi0iuDDvB2c3PDm2++iTNnzmDJkiX466+/8Mwzz8DLywuzZ89GJruK9I/dcERERHql9xm8C7p16xZWr16NsLAwXLt2Dc888wzGjx+Pf//9FwsXLsQ///yDXbt2GbIK1Q/DEhERkV4ZJCxt2rQJq1atws6dO9G8eXO8+uqreP755+Hi4qIt06lTJ/j7+xti99UbwxIREZFeGSQsjR07FiNHjsShQ4fw2GOPFVnGy8sLs2bNMsTuqzeGJSIiIr0ySFhKSEhADc2HdjHs7OwwZ84cQ+y+etPM4s2wREREpBcGCUt5eXlIS0srtFyhUECpVMJGM8s06R9bloiIiPTKIGHJxcUFCoWi2Ofr1auHMWPGYM6cObCw4BVX9MreXt4zLBEREemFQcJSWFgYZs2ahTFjxqBjx44AgGPHjmH16tV477338N9//2HRokVQKpV49913DVGF6ostS0RERHplkLC0evVqLF68GMOHD9cuGzBgAFq1aoVvvvkGe/bsQf369TFv3jyGJX3jmCUiIiK9Mkgf2OHDh9GuXbtCy9u1a4cjR44AALp06YL4+HhD7L564wzeREREemWQsOTt7Y3vv/++0PLvv/8e3t7eAIA7d+6gZs2ahth99cZrwxEREemVQbrhFi1ahGHDhmH79u3aeZZOnDiBCxcu4JdffgEAHD9+HCNGjDDE7qs3TVjKzgZUKsDS0rj1ISIiMnEGCUsDBw5ETEwMvvnmG8TExAAA+vbtiy1btsDX1xcA8Morrxhi11Rwfqv79wEHB+PVhYiIyAzoPSzl5uaiT58+WLFiBebPn6/vzdOjWFkBNjZATo4c5M2wREREVCF6H7NkbW2NM2fO6HuzVBY8I46IiEhvDDLA+/nnny9ygDdVEp4RR0REpDcGu9zJDz/8gL/++gsBAQGw18wq/cCSJUsMsVvS4BlxREREemOQsBQVFYX27dsDAC5evKjzXEmXQSE94SzeREREemOQsBQeHm6IzVJpMSwRERHpjUGvYnv58mXs3LkT9x+MnRFCGHJ3pMExS0RERHpjkLB0584dBAcHo0mTJggJCUFCQgIAYPz48Zg2bZohdkkFsWWJiIhIbwwSlt544w1YW1sjPj4eNQpMkjhixAjs2LHDELukghiWiIiI9MYgY5Z27dqFnTt3ol69ejrLGzdujGvXrhlil1QQz4YjIiLSG4O0LGVkZOi0KGkkJydDqVQaYpdUEMcsERER6Y1BwlLXrl3x448/an9WKBRQq9X45JNP8OSTTxpil1QQu+GIiIj0xiDdcJ988gmCg4Nx4sQJ5OTk4O2338a5c+eQnJyMQ4cOGWKXVBDDEhERkd4YpGWpZcuWuHjxIrp06YJBgwYhIyMDTz/9NE6dOoWGDRsaYpdUEMMSERGR3hikZQkAnJ2dMWvWLENtnkrCsERERKQ3BgtLKSkpOHbsGJKSkqBWq3WeGzVqlKF2SwBgZyfvGZaIiIgqzCBh6ffff0doaCjS09Ph5OSkcz04hULBsGRompalrCxArQYsDDpROxERkVkzyKfotGnTMG7cOKSnpyMlJQV3797V3pKTkw2xSyqo4LQNnD6AiIioQgwSlm7cuIHXX3+9yLmWqBJYW8sbwK44IiKiCjJIWOrduzdOnDhhiE1TaXGQNxERkV4YZMxSv379MH36dJw/fx6tWrWCtaaV44GBAwcaYrdUUI0aQGoqwxIREVEFGSQsTZgwAQDw4YcfFnpOoVBApVIZYrdUEM+IIyIi0guDhKWHpwogI2A3HBERkV7wnHJzZW8v7xmWiIiIKkSvYSkkJASpqananxcsWICUlBTtz3fu3EHz5s31uUsqDluWiIiI9EKvYWnnzp3Izs7W/vzxxx/rzKuUl5eHmJgYfe6SisMxS0RERHqh17AkhCjxZ6pEmpYlTkpJRERUIRyzZK7YDUdERKQXeg1LCoVC5zpwmmVkBJqwlJFh3HoQERGZOL1OHSCEwJgxY6BUKgEAWVlZePnll2H/4MysguOZyMDYskRERKQXeg1Lo0eP1vn5+eefL1Rm1KhR+twlFYdjloiIiPRCr2Fp1apV+twcVQRbloiIiPSCA7zNFcMSERGRXphVWEpOTkZoaCicnJzg4uKC8ePHIz09vcR1EhMT8cILL8DDwwP29vZo3749fv3110qqsQExLBEREemFWYWl0NBQnDt3Drt378Yff/yBv//+GxMnTixxnVGjRiEmJgZbt27F2bNn8fTTT2P48OE4depUJdXaQAqOWeK1+oiIiMrNbMJSdHQ0duzYge+++w6BgYHo0qULli1bhvXr1+PmzZvFrnf48GG89tpr6NixI/z8/PDee+/BxcUFERERxa6TnZ2NtLQ0nVuVowlLAAd5ExERVYDZhKUjR47AxcUFHTp00C7r2bMnLCwscPTo0WLX69SpEzZs2IDk5GSo1WqsX78eWVlZ6N69e7HrzJ8/H87Oztqbt7e3Pg9FP6ytAasH4/cZloiIiMrNbMJSYmIi6tSpo7PMysoKtWrVQmJiYrHrbdy4Ebm5uahduzaUSiVeeuklbN68GY0aNSp2nZkzZyI1NVV7u379ut6OQ684bomIiKjCqnxYmjFjhnZm8OJuFy5cKPf233//faSkpOCvv/7CiRMn8Oabb2L48OE4e/ZssesolUo4OTnp3KokhiUiIqIK0+s8S4Ywbdo0jBkzpsQyfn5+8PDwQFJSks7yvLw8JCcnw8PDo8j1YmNj8eWXXyIqKgotWrQAALRp0wYHDhzAV199hRUrVujlGIyGYYmIiKjCqnxYcnNzg5ub2yPLBQUFISUlBREREQgICAAA7N27F2q1GoGBgUWuk/kgRFhY6DawWVpaQm0OZ5AxLBEREVVYle+GKy1/f3/06dMHEyZMwLFjx3Do0CFMnjwZI0eOhJeXFwDgxo0baNasGY4dOwYAaNasGRo1aoSXXnoJx44dQ2xsLBYvXozdu3dj8ODBRjwaPWFYIiIiqjCzCUsAsHbtWjRr1gzBwcEICQlBly5dsHLlSu3zubm5iImJ0bYoWVtb488//4SbmxsGDBiA1q1b48cff8Tq1asREhJirMPQH4YlIiKiCqvy3XBlUatWLaxbt67Y5319fSGE0FnWuHFj85ixuygMS0RERBVmVi1L9BCGJSIiogpjWDJndnbynmGJiIio3BiWzJm9vbznDN5ERETlxrBkzjTdcBkZxq0HERGRCWNYMmccs0RERFRhDEvmjGOWiIiIKoxhyZxpWpY4ZomIiKjcGJbMGbvhiIiIKoxhyZwVPBvuock4iYiIqHQYlsyZZsySEOyKIyIiKieGJXNmbQ1YPbiiDbviiIiIyoVhyZwpFPmtS2xZIiIiKheGJXPHQd5EREQVwrBk7hiWiIiIKoRhydxpzoi7dcu49SAiIjJRDEvmLiBA3m/ZAqhURq0KERGRKWJYMnf9+gGOjsCNG8Dffxu7NkRERCaHYcnc2dkBTz8tH//0E1uXiIiIyohhqTro3x9wcgISEoB9+4xdGyIiIpPCsFQd2Nrmty5t2MDWJSIiojJgWKou+vUDnJ1l61J4uLFrQ0REZDIYlqoLW1tg6FD5eP16IC/PuPUhIiIyEQxL1UnfvrJ16dYtti4RERGVEsNSdWJrCzzzjHzM1iUiIqJSYViqbvr2BVxcgKQkYO9eY9eGiIioymNYqm6UyvzWpQ0bALXauPUhIiKq4hiWqqO+feU145KSgEuXjF0bIiKiKo1hqTqysQHatpWPT540alWIiIiqOoal6kpzgd2ICOPWg4iIqIpjWKqu2rWT9xcvAvfuGbcuREREVRjDUnXl6gr4+ABCAJGRxq4NERFRlcWwVJ21by/v2RVHRERULIal6qzguCUhjFsXIiKiKophqTpr3lzOu5SSAsTFGbs2REREVRLDUnVmbQ20bi0fcwoBIiKiIjEsVXcdOsh7jlsiIiIqEsNSdacZ5B0dDWRmGrcuREREVRDDUnXn4QF4eQEqFXDmjLFrQ0REVOUwLBFn8yYiIioBwxLld8WdPMkpBIiIiB7CsERAy5byzLikJODGDWPXhoiIqEphWCLA1hZo0UI+ZlccERGRDoYlkjTjlsoy31JsLPD330BenmHqREREVAUwLJGkCUtRUUBOzqPLp6YCM2cCn34KvPIKcOAAxzsREZFZYlgiqV49wM1NBqWzZx9dfsMG4P59+TgxEfjkE2DaNE4/QEREZodhiSSFIr916cSJkssmJgLbt8vH770HhIbKcU+XLgGzZgEffMCB4kREZDYYlijfY4/J+127gGvXii+3Zo0cp9SuHRAYCIwcCXz3HdCvH2BpKcPW9Ony7DoiIiITx7BE+R57TLYu5eQACxcC2dmFy8TGAvv3y8djxuQvd3YGXn4Z+PprwM8PuHcPmD+/dOOfiIiIqjCGJcqnUABvvAHUrAlcvy5bix4WFibvu3WToehhXl6yK87REbh8GVi+nAO/iYjIpDEskS5nZzlQW6EAduwADh7Mfy4yUt6srIAXXih+G3XqAG+/Lbfx119yO0RERCaKYYkKa9MGeOYZ+fjLL+XYIyGAVavkspAQwN295G20bQuMHi0fr1wJXLhgsOoSEREZEsMSFe2554BmzYCMDDktQHg4cOUKYGcHDB9eum08/TTQubMcDD5/PnD3rmHrTEREZABWxq4AVVFWVvKMttdfB2Ji5PgjABg6VHbVlYZCAUyZAsTHyzFQCxYAEybIgeNZWfI+OxtwcAA6dJDliYiIqhiFEBx9W1FpaWlwdnZGamoqnJycjF0d/Tp0SIYcQA78XrlSzqlUFjduAG++CWRmFl+mfXtZprRBjIiIqIJK+/nNbjgqWefOwIAB8vGYMWUPSgBQt64c8O3qKm9168oz6fz95dgmGxt5TbrXX5eXWyEiIqpC2LKkB2bdsgTIwd1paYZr9bl2Tc7rdP267Ip77jk5LsqCWZ6IiAyHLUukPwqFYbvHfHyAJUuAnj1lMFu7Fnj/fQ4IJyKiKoFhiaoGW1s5GPzNN+XjM2dk111enrFrRkRE1RzDElUtTz4JfP454OIiL9hbcFJMIiIiI2BYoqqnbt38QeW//27cuhARUbVnVmFp3rx56NSpE2rUqAEXF5dSrSOEwOzZs+Hp6Qk7Ozv07NkTly5dMmxF6dF69wasrYGLF+U8T0REREZiVmEpJycHw4YNwyuvvFLqdT755BMsXboUK1aswNGjR2Fvb4/evXsjKyvLgDWlR3J2Bp54Qj7eutW4dSEiomrNrMLSBx98gDfeeAOtWrUqVXkhBD7//HO89957GDRoEFq3bo0ff/wRN2/exJYtWwxbWXq0gQPl/aFDQHKycetCRETVllmFpbKKi4tDYmIievbsqV3m7OyMwMBAHDlypNj1srOzkZaWpnMjA/DzA1q0AFQqYPt2Y9eGiIiqqWodlhITEwEA7u7uOsvd3d21zxVl/vz5cHZ21t68vb0NWs9qTTPQe/t2IDfXuHUhIqJqqcqHpRkzZkChUJR4u3DhQqXWaebMmUhNTdXerl+/Xqn7r1Yef1xeIiU1FThwwNi1ISKiasjK2BV4lGnTpmHMmDEllvHz8yvXtj08PAAAt27dgqenp3b5rVu30LZt22LXUyqVUCqV5donlZGlJdCvH7B6tRzo/eSTckZxIiKiSlLlw5Kbmxvc3NwMsu0GDRrAw8MDe/bs0YajtLQ0HD16tExn1JGB9e4N/PQTEBsLXLggL8BLRERUSap8N1xZxMfHIzIyEvHx8VCpVIiMjERkZCTS09O1ZZo1a4bNmzcDABQKBaZOnYr//e9/2Lp1K86ePYtRo0bBy8sLgwcPNtJRUCGOjkD37vIxpxEgIqJKVuVblspi9uzZWL16tfbndu3aAQDCw8PR/cGHbUxMDFJTU7Vl3n77bWRkZGDixIlISUlBly5dsGPHDtja2lZq3ekRBgwAdu0CDh8Gbt+W45iIiIgqgUIIIYxdCVOXlpYGZ2dnpKamwsnJydjVMV/vvgucPQs88wwwerSxa0NERCautJ/fZtUNR2ZOM0nl9u1AZqZx60JERNUGwxKZjsBAoF49ICODk1QSEVGlYVgi06FQyC44ANiyBcjJMWp1iIioemBYItPSrRvg5gakpAB//WXs2hARUTXAsESmxcoKGDJEPt60SV43joiIyIAYlsj0PPUU4OwM3LrFS6AQEZHBMSyR6VEq88+M+/lngLNfEBGRATEskWnq1w+wswPi44Fjx4xdGyIiMmMMS2Sa7O1lYALYukRERAbFsESma9AgwMYGiImRM3sTEREZAMMSmS4XF6BnT/n455+NWhUiIjJfDEtk2oYOBSwsgMhI4NAhY9eGiIjMEMMSmbY6dfJblxYsAJYu5XXjiIhIrxiWyPS99JKcqFKhAHbvBl57DYiKMnatiIjITCiE4GlEFZWWlgZnZ2ekpqbCycnJ2NWpvqKigM8+A5KSZHAaPBh49lngzh3gxg3g+nV5f+MGcP++nP07L0/eq1RynSZNgHbtgPbtZasVERGZrdJ+fjMs6QHDUhVy/z7w3XfArl0V31bdujI4BQQAbdvKS60QEZHZYFiqRAxLVdDx48CyZcDdu4CtrQw+desC3t7y3tERsLSUAcjCQt5nZckpCE6dAi5cANTq/O05OwPduwM9egB+fkY7LCIi0h+GpUrEsFRF5eUBaWlAzZqyi60sMjJkcDp5Ejh8GEhNzX/O1xcIDga6dZPbJiIik8SwVIkYlsxcXp5sbdqzBzh6VP4MyADWtq1scQoKkpdfISIik8GwVIkYlqqRe/eAAweAvXvlzOEaNjZAYKAMTu3bc3wTEZEJYFiqRAxL1dTNm8DffwP79skz7DQcHGRLU9euQOvWcmwUERFVOQxLlYhhqZoTArh8Gdi/X4anu3fzn3NyAjp1kuObWrQo+9gpIiIyGIalSsSwRFpqNXDunAxNhw/LAeYavr7A00/LFid20xERGR3DUiViWKIiqVTAmTNyjNOBA3JqAgBwdZUTZj71FAeFExEZEcNSJWJYoke6dw/Yvh3YujV/GgJ7e2DAAGD4cMDa2rj1IyKqhhiWKhHDEpVaTg4QHg5s3pw/KNzXF5g2Td4TEVGlYViqRAxLVGZCAAcPAitWyHFNVlbAqFGye46DwImIKkVpP78tKrFORKShUMiB3l99BTz2mJzo8ocfgFmzgP/+M8w+c3KAK1fkPRERlRpblvSALUtUIUIAO3fKCwBnZwM1agAvvQQ8+WTFW5lSU4ETJ+TM46dOyUHmNWsCQ4YAffvK6+YREVVT7IarRAxLpBcJCcDixfkzgwcGApMmlf36c9nZMnwdPCgvCFzwX9zKKv9yLY6OwMCBQP/+ciJNIqJqhmGpEjEskd6oVMCvvwI//SRDjaMj8MorssuuNOvu3QusXQvcuZO/3M9PBq/AQMDHR844/vPPcgZyQLYu9e8PjBwJKJUGOSwioqqIYakSMSyR3l29Cnz2mRxjBABdugAvvww4OxcuKwRw7BiwejVw/bpcVqeOHCweFCTndXqYWg0cOgRs3Cj3BQANGwLvvVd0eSIiM8SwVIkYlsgg8vJkmNm4UbYaOToCDRrI+ZkcHPLvIyKA6Gi5jqMjMGKEHI9kY/PofQghxzMtWybPynNxAd59F/D3N+ihERFVBQxLlYhhiQwqNhZYsgSIjy++jI0NMGgQMHSoDFFllZQEfPSRbGWysgJefRXo1avcVSYiMgUMS5WIYYkMLjdXXnMuLQ1IT9e9aQZq165dsX1kZcmuv8OH5c8DBgDjxvE6dkRkthiWKhHDEpkNIYANG+QgcQBo0iR/7FNZQ1NyMnDpkjzLLyNDN+BlZMiB5a6uMuS5uubfvL0BC04BR0SGV2lhadMmOQlxRIR8bzx1CmjbtuR1zp0DZs+W61y7Jr/MTp2qW2b5cnnTjD1t0UKu07ev/PnqVTl8oygbNwLDhgGnTwMLFsgzqG/flleTePllYMoU3fJr1wKffCLf152d5T4+/bT0X9QZlsjs/POPnMZAc/FfZ2fZLde7N+DhoVtWrZb//AkJ8p8oJga4eFH+05WHgwPQpo18I2nfXg5Wr6i8PFmfot7unJzK13VJRCavtJ/fFW5fz8iQJ+oMHw5MmFC6dTIz5dnMw4YBb7xRdJl69WTQadxYvr+tXi2HZJw6JYOTt7d8by5o5UoZcjSBKiJCvs+uWSPLHz4MTJwIWFoCkyfLMocOyatMfPaZ7HW4cUMGqgkTZBAkqpYef1x+W9m5E9i1S4ahX36Rt7ZtZaC5c0fONp6cLAPTwxQKoH59OV2Bg4Puzd4euH9fbuP2bXm7c0f+U6eny3/MQ4fkdry8ZHjy9weaNgU8PR89Wee9e3KOqQsX5OD3ixfl/FNFUSjkN6mWLYFWreS9o2NFfntEZGb01g2naekpTctSQb6+slXp4ZalotSqJcPQ+PFFP9+unfwi+v33xW9j0iT53rl3r/x50SL5mRAbm19m2TJg4ULg339LdwxsWSKzlpcHHD8O7Ngh/8GLesuwsJBdaI0aya67pk3lVAR2dmXbl0oFXL4s93PqlAw7DwcxR0e5/WbN5OOUFDlTueb+zh0gMbHwtm1s5DclQPcYNK1nBfn6As2by300ayZb03jNPiKzU2ktS5VBpZJz6GVkyKETRYmIACIj5aW2SpKaKkOXRlCQPFP6zz9li1RSkvzyHBKit+oTmTYrK/mPEhQE3Lol+7WtrPLHGLm5ySkH9DHOyNJSBqGmTeUkmZmZwNmzQFSU7N67fFm2Gp04IW8lqVdPtkY1aybv69UrOvDcvSu3HxUl93X9uvz2d/WqfGMAZDdks2YyCLq6yq67gjc7u/xtFwxiubkyjD18y86WLWvZ2fLn+/dlKFUq5VguW9v8xw4Ochb3mjVLNx1EaahU8hqBKpWsr1otb5rHVla6N0tLhkWq1qp0WDp7Vr4/Z2XJ94vNm+WXvaJ8/718P+zUqfjtHT4sx65u25a/rHNnOWZpxAi5n7w82R33qNBVWTRfepXK/PeqvDx5s7QErK1LV9bCQvd9tixls7Ple6iNTf7noUolPwcqUjYnR74vW1vnf+FXq+XyspRVKHQnntaU1bzPl7WsEPk9NgUvnZabK49FH2WL+r2XpWxZXvuK/J0Uej3d3ZE3aGiF/040v/dHlq1RAzntAqFuEyhfe3UuEBcHdXQMcs5fhiI3B0pXRxlmnJ2RY18TakdnWDfygaWL7ErTvvY5xbz2jjVh1bUr0LWr/L0npQLnz8P2ynnZsnX5MnLuZkB9+CSsjhyHlYU6/zVSy1+srWVu/muvtoRKWMBKoapwWaVFbv5rpLZAnp0jLGs6wdrVWf7BWVoiC7aAlRWUtgooLC2A7GzkZWQj734uLLMzYZ2TIV+c7Gxk3RdATg6U6vu62xWWsFAI2Fjk5f+dqB6qg40N8pT2yLNzhIW9HWwclbI71c4O2QpbCKUtbGpYwcJOCdjYQJWrRm6WChaqXNggR764ubnIvq+GyM6BjToLFqpcICcHqjyBXAslLKwsYGNroQ1oORa2UCvtYF3DGpZ2NoBSCbW1EjnCOr+spaUsm2cBtUrA2lINS4UMf2qVkK+9WgWlZZ42FOZkC6iFAlaWQvc9QmUJhZUllLaK/O2qraCGhfz/VMrQKKBAdp4lIARsbR60fAqR/39fYLtCANk58pdtqxTyD1yhQK5KAZXaIv//XqHQKVvo/1P1oL7WcqGAAtm58p9HaauAwkIuz1MpkKe2kP/3NvkBNytH7ldnu7kCeSoFLC0ErK3yQ35W9oM62AhZVqGQ21Up5P+n8sEGFApk58jfh40NYGGpkGWLez/JEvnvEQq5v0e+RygV+d/DPDyMdnZumfa6dq28vqfG9u2luwpDeTVtKluLUlNla8/o0cD+/YUD0/37wLp1wPvvF7+tqCg55mnOHOCpp/KXnz8vB3zPni3HriYkANOny3FLJXXnVZZhw+T9mjX5kzdv2gT83//J43jttfyyzz8v3xO//z5/TOy2bfL6rN26AW+9lV92/Hh5FvpXX8lhJQCwZw/w5ZfyqhjvvZdf9tVXZYvbkiVyDBkAHDggx/+2bSun59F44w35xfzjj+XwD0D24MybJ8PsJ5/kl50xQ44Hnj0beOwxuezMGfk6NmgALF2aX3bOHPkavvOOHCMHyIaGt9+WQ1hWrswvO3++bHSYOhUIDpbLrl6Vr3OtWnL8m8aSJXJozMsvA/36yWUJCfLv3N4eWL8+v+xXX8nf0dixwNNPy2XJycCYMfJ9dcuW/LLffScbJZ59FnjuObksM1M2lgAy+Gv+53/8Uf48ZIg8Ux+QbyCa1379+vzxxxs3yiuhhITIq6BojBwp1wkLyz8xYetWYNUq+Tso2M09Zoxspf3mGzkcCJBDk1askF8eZszILztxojzGL76Q4wwB+T/4+edAhw7yddF47TX5u/vkk/w5LY8ckV3aLVvK10XjrbeAuDj5t6Pptj95EvjwQ/k3tmRJftn33pNd57NmAY8/bg00aYJz2U3w7ndyLOLXs/PLfvS+fM+YNg3o3l0ui40F3nxT/k8U/J/+5BM5H+fkyfJ/H5B/u5MmOcPJKQhr1z5oxs7NxdJZd7E/XI0XWx3FoHoRQFoa/rslMH7fC1AiG790WKDd7oprfbHrv3Z4oV44hnsdBACkKZzx/Ok3AUsL/N5/pbb1KCwqCFsv+2N4wBW8EHQZyM5G9r1cDFs7CFCp8HPv72GblQLcvYuf4rpg480uGOh+FBN8dmn3N+yYfONb024xnK0zAQCbbnbB//3bC0+5ncJrDf7Qln3+xAxkq63xfZulqKNMBQBsS3oM38U/hW6uUXir0W/yk1SlwvjTryMtrwa+arUC9e3+A3JysOffFvjyan8E1ozBe403arf76unXkJTtgiXNv0NjBzmY9MDtllh8ZQjaOl/BR01/1ZZ94+wruH7fFR83+xGtnK4BAI7fbYp5lwbD3+E6Pmkepi0749x4XMrwwuwm6/GYyyUAwJnUBng/5nk0qHELS1vm/+PPiR6FqHs+eKfRr+hS6zwAIOZePbwdPRaetslY2Tr/G/D8iyNxIqUxpvr9hmDXMwCAqxnumHJuImrZ3MPqtp9ryy65/AwOJfvjZZ/t6OcuWzMTsmrhpTOTYG+VhfXtP9WW/erKQOy53QZjvffgac8jAIDkHEeMiZwKS4UaWx6bpy373dW++DOpA56t+zeeq7sfAJCZp8TIk28DADZ3mKcN0D/G98TmxCAM8TiCcfX/AgCo1BYYdmIWAGB9+09gbyW/XW280Q0/3XgCIXVO4BXf7dr9jTw+CyphgbC2n6O2zT0AwNaEIKy63hPBrqcx1W+rtuyYk9ORkWeLb1p/BS/bZADAzlsdsOJaX3SuFY0ZjX7Rlp0YORXJOY74osVK+NnfAgDsv90an18ZhA4ulzCnSf6b6GtnJiEhqxY+8V8Ff0c5zuVIcnMsvDwULR2vYb7/j9qyb0VNRFymOz5qugZtneMe/NK+A9zdYQxlCksDB8oPUo26dfVdHV02NnIIBAAEBMgP3S++kG/yBf3yi/wgGjWq6O2cPy8/MCZO1A0BgHwD79xZBiQAaN1afjB17Qr873/yg5iIqjFra5m0PCC/cQ0aJJcnARgPwFoF/PB/+V/Xv7EB9loBI1sCw1+TTQTplsDzD7b39df52/4WwFYATzUCXnjwLS4LwPEHzy9eDNhCNk98nwNsUAOd6wLBbeVX79xc4JaPfP7ZZwG7B81nx3yAcG8gqA4w6nEZzmxsgMkeQJ4lsHwFUNdaJvbfLYDvAXTrCLz1IK0LATwngDQBLFoGeOXJb2Lb1cAqe6CZFzDSW77xZmUByxoCd22AHj0Al/9k2Tgf4J4bUN9STj9hbS1vYU2BZHvZnN80Ry6LdgZW1wW86wBj7fObOr/zB27YA088AXg1fLBdZyDJBXAWcqCqSiWbIP6rDcBRpvr6VrKp4rYbkOAMOFrKDy9LS7k8ywewrCW/0TVylsd7xxlIqAXY2QIdO8rtqlRAugeQ6yiTuV+6LJvmAMTWAKyt8r9BKhRAeh0gywGo6wU0biqX368BXHYAFGrZjavpps1wA9JryL8tzandOdbAhQfj/Hx9gQdhCZmuQJqd/CZUv/6D7lIL4JwtICA/qB6EJaQ7AbdtAEcH2UWu2Z+NDSAsZJe53YMmnHRHudzeXvcyRzZKwMJGfrusYSG3kekI2FjLLmcXl/ztWlkDaiu5Dc23unQ7wMoSsH6wbc3/htWDvzl7+/yTKLJqyGU2NronVthYAzkPlTViV7BJDfDu0UP+nYSF6S7v3l2+zr/8Unidc+fkeqNH67ZqaAwdKl+nDRvylx05IrvzbtzI/+ZdEkMO8GY33KPLshuuErvhyli2wt1wKN9rX96/k+Jez6r0d1JZrz3fI4ovy/eI/LKV+R5hCJU2wDs5WV6FQXMB85gYee/hkT8dy6hRshVK0wyfkyNbezSPb9yQTecODvktSTNnygHX9evL8Zzr1smLpe/cqbv/y5eBv//OH4dZUFSUDEq9e8tmeM0JMpaWMnADcnzShAnyjDhNN9zUqfKLRWmCkqEV/OfSKPgPUxlli7oQ/YPu/AqVLWqsqoVF0XWrzLIKRdFlNV+M9V22qN97WcoClfva6+PvpKjfe1X4Oynu915V/06AqlGW7xES3yPKXrYsr70xVTgsbd0qx3BoaMZkzJkDzJ0rH8fH6ybDmzdl66nGokXy1q2bDESAHCMzapQML87Osnts587Cl6v64Qd5kkvBcUgav/wip4FZs0beNHx88ie7HDNGhrEvv5TjHFxcZMBauLDMvwoiIiIyQ7zciR5wniUiIiLTU9rPb16AiYiIiKgEDEtEREREJWBYIiIiIioBwxIRERFRCRiWiIiIiErAsERERERUAoYlIiIiohIwLBERERGVgGGJiIiIqAQMS0REREQlYFgiIiIiKkGFL6RLgObyemlpaUauCREREZWW5nP7UZfJZVjSg3v37gEAvL29jVwTIiIiKqt79+7B2dm52OcV4lFxih5JrVbj5s2bcHR0hEKhKPP6aWlp8Pb2xvXr10u86rGpqw7HyWM0DzxG88BjNA+GPEYhBO7duwcvLy9YWBQ/MoktS3pgYWGBevXqVXg7Tk5OZvvHXlB1OE4eo3ngMZoHHqN5MNQxltSipMEB3kREREQlYFgiIiIiKgHDUhWgVCoxZ84cKJVKY1fFoKrDcfIYzQOP0TzwGM1DVThGDvAmIiIiKgFbloiIiIhKwLBEREREVAKGJSIiIqISMCwRERERlYBhqQr46quv4OvrC1tbWwQGBuLYsWPGrlK5/f333xgwYAC8vLygUCiwZcsWneeFEJg9ezY8PT1hZ2eHnj174tKlS8apbDnNnz8fjz32GBwdHVGnTh0MHjwYMTExOmWysrIwadIk1K5dGw4ODhg6dChu3bplpBqX3fLly9G6dWvtJHBBQUHYvn279nlTP76iLFiwAAqFAlOnTtUuM/XjnDt3LhQKhc6tWbNm2udN/fg0bty4geeffx61a9eGnZ0dWrVqhRMnTmifN/X3HV9f30Kvo0KhwKRJkwCYx+uoUqnw/vvvo0GDBrCzs0PDhg3x0Ucf6VyzzaivoyCjWr9+vbCxsRE//PCDOHfunJgwYYJwcXERt27dMnbVyuXPP/8Us2bNEps2bRIAxObNm3WeX7BggXB2dhZbtmwRp0+fFgMHDhQNGjQQ9+/fN06Fy6F3795i1apVIioqSkRGRoqQkBBRv359kZ6eri3z8ssvC29vb7Fnzx5x4sQJ8fjjj4tOnToZsdZls3XrVrFt2zZx8eJFERMTI959911hbW0toqKihBCmf3wPO3bsmPD19RWtW7cWU6ZM0S439eOcM2eOaNGihUhISNDe/vvvP+3zpn58QgiRnJwsfHx8xJgxY8TRo0fFlStXxM6dO8Xly5e1ZUz9fScpKUnnNdy9e7cAIMLDw4UQ5vE6zps3T9SuXVv88ccfIi4uTvz888/CwcFBfPHFF9oyxnwdGZaMrGPHjmLSpEnan1UqlfDy8hLz5883Yq304+GwpFarhYeHh/j000+1y1JSUoRSqRQ//fSTEWqoH0lJSQKA2L9/vxBCHpO1tbX4+eeftWWio6MFAHHkyBFjVbPCatasKb777juzO7579+6Jxo0bi927d4tu3bppw5I5HOecOXNEmzZtinzOHI5PCCHeeecd0aVLl2KfN8f3nSlTpoiGDRsKtVptNq9jv379xLhx43SWPf300yI0NFQIYfzXkd1wRpSTk4OIiAj07NlTu8zCwgI9e/bEkSNHjFgzw4iLi0NiYqLO8To7OyMwMNCkjzc1NRUAUKtWLQBAREQEcnNzdY6zWbNmqF+/vkkep0qlwvr165GRkYGgoCCzO75JkyahX79+OscDmM/reOnSJXh5ecHPzw+hoaGIj48HYD7Ht3XrVnTo0AHDhg1DnTp10K5dO3z77bfa583tfScnJwdr1qzBuHHjoFAozOZ17NSpE/bs2YOLFy8CAE6fPo2DBw+ib9++AIz/OvJCukZ0+/ZtqFQquLu76yx3d3fHhQsXjFQrw0lMTASAIo9X85ypUavVmDp1Kjp37oyWLVsCkMdpY2MDFxcXnbKmdpxnz55FUFAQsrKy4ODggM2bN6N58+aIjIw0i+MDgPXr1+PkyZM4fvx4oefM4XUMDAxEWFgYmjZtioSEBHzwwQfo2rUroqKizOL4AODKlStYvnw53nzzTbz77rs4fvw4Xn/9ddjY2GD06NFm976zZcsWpKSkYMyYMQDM4+8UAGbMmIG0tDQ0a9YMlpaWUKlUmDdvHkJDQwEY//ODYYmoAiZNmoSoqCgcPHjQ2FXRu6ZNmyIyMhKpqan45ZdfMHr0aOzfv9/Y1dKb69evY8qUKdi9ezdsbW2NXR2D0HwrB4DWrVsjMDAQPj4+2LhxI+zs7IxYM/1Rq9Xo0KEDPv74YwBAu3btEBUVhRUrVmD06NFGrp3+ff/99+jbty+8vLyMXRW92rhxI9auXYt169ahRYsWiIyMxNSpU+Hl5VUlXkd2wxmRq6srLC0tC521cOvWLXh4eBipVoajOSZzOd7Jkyfjjz/+QHh4OOrVq6dd7uHhgZycHKSkpOiUN7XjtLGxQaNGjRAQEID58+ejTZs2+OKLL8zm+CIiIpCUlIT27dvDysoKVlZW2L9/P5YuXQorKyu4u7ubxXEW5OLigiZNmuDy5ctm8zp6enqiefPmOsv8/f213Y3m9L5z7do1/PXXX3jxxRe1y8zldZw+fTpmzJiBkSNHolWrVnjhhRfwxhtvYP78+QCM/zoyLBmRjY0NAgICsGfPHu0ytVqNPXv2ICgoyIg1M4wGDRrAw8ND53jT0tJw9OhRkzpeIQQmT56MzZs3Y+/evWjQoIHO8wEBAbC2ttY5zpiYGMTHx5vUcT5MrVYjOzvbbI4vODgYZ8+eRWRkpPbWoUMHhIaGah+bw3EWlJ6ejtjYWHh6eprN69i5c+dCU3dcvHgRPj4+AMznfQcAVq1ahTp16qBfv37aZebyOmZmZsLCQjeSWFpaQq1WA6gCr6PBh5BTidavXy+USqUICwsT58+fFxMnThQuLi4iMTHR2FUrl3v37olTp06JU6dOCQBiyZIl4tSpU+LatWtCCHnqp4uLi/jtt9/EmTNnxKBBg0zqFF4hhHjllVeEs7Oz2Ldvn87pvJmZmdoyL7/8sqhfv77Yu3evOHHihAgKChJBQUFGrHXZzJgxQ+zfv1/ExcWJM2fOiBkzZgiFQiF27dolhDD94ytOwbPhhDD945w2bZrYt2+fiIuLE4cOHRI9e/YUrq6uIikpSQhh+scnhJz2wcrKSsybN09cunRJrF27VtSoUUOsWbNGW8Yc3ndUKpWoX7++eOeddwo9Zw6v4+jRo0XdunW1Uwds2rRJuLq6irfffltbxpivI8NSFbBs2TJRv359YWNjIzp27Cj++ecfY1ep3MLDwwWAQrfRo0cLIeTpn++//75wd3cXSqVSBAcHi5iYGONWuoyKOj4AYtWqVdoy9+/fF6+++qqoWbOmqFGjhhgyZIhISEgwXqXLaNy4ccLHx0fY2NgINzc3ERwcrA1KQpj+8RXn4bBk6sc5YsQI4enpKWxsbETdunXFiBEjdOYfMvXj0/j9999Fy5YthVKpFM2aNRMrV67Ued4c3nd27twpABRZb3N4HdPS0sSUKVNE/fr1ha2trfDz8xOzZs0S2dnZ2jLGfB0VQhSYHpOIiIiIdHDMEhEREVEJGJaIiIiISsCwRERERFQChiUiIiKiEjAsEREREZWAYYmIiIioBAxLRERERCVgWCIiIiIqAcMSEVU5V69ehUKhQGRkpLGronXhwgU8/vjjsLW1Rdu2bYss0717d0ydOrVS61UaCoUCW7ZsMXY1iEwWwxIRFTJmzBgoFAosWLBAZ/mWLVugUCiMVCvjmjNnDuzt7RETE6NzMc+CNm3ahI8++kj7s6+vLz7//PNKqiEwd+7cIoNcQkIC+vbtW2n1IDI3DEtEVCRbW1ssXLgQd+/eNXZV9CYnJ6fc68bGxqJLly7w8fFB7dq1iyxTq1YtODo6lnsfxalIvQHAw8MDSqVST7Uhqn4YloioSD179oSHhwfmz59fbJmiWjI+//xz+Pr6an8eM2YMBg8ejI8//hju7u5wcXHBhx9+iLy8PEyfPh21atVCvXr1sGrVqkLbv3DhAjp16gRbW1u0bNkS+/fv13k+KioKffv2hYODA9zd3fHCCy/g9u3b2ue7d++OyZMnY+rUqXB1dUXv3r2LPA61Wo0PP/wQ9erVg1KpRNu2bbFjxw7t8wqFAhEREfjwww+hUCgwd+7cIrdTsBuue/fuuHbtGt544w0oFAqdFrmDBw+ia9eusLOzg7e3N15//XVkZGRon/f19cVHH32EUaNGwcnJCRMnTgQAvPPOO2jSpAlq1KgBPz8/vP/++8jNzQUAhIWF4YMPPsDp06e1+wsLC9PWv2A33NmzZ9GjRw/Y2dmhdu3amDhxItLT0wu9ZosWLYKnpydq166NSZMmafdFVN0wLBFRkSwtLfHxxx9j2bJl+Pfffyu0rb179+LmzZv4+++/sWTJEsyZMwf9+/dHzZo1cfToUbz88st46aWXCu1n+vTpmDZtGk6dOoWgoCAMGDAAd+7cAQCkpKSgR48eaNeuHU6cOIEdO3bg1q1bGD58uM42Vq9eDRsbGxw6dAgrVqwosn5ffPEFFi9ejEWLFuHMmTPo3bs3Bg4ciEuXLgGQ3VgtWrTAtGnTkJCQgLfeeuuRx7xp0ybUq1cPH374IRISEpCQkABAtlD16dMHQ4cOxZkzZ7BhwwYcPHgQkydP1ll/0aJFaNOmDU6dOoX3338fAODo6IiwsDCcP38eX3zxBb799lt89tlnAIARI0Zg2rRpaNGihXZ/I0aMKFSvjIwM9O7dGzVr1sTx48fx888/46+//iq0//DwcMTGxiI8PByrV69GWFiYNnwRVTuCiOgho0ePFoMGDRJCCPH444+LcePGCSGE2Lx5syj4tjFnzhzRpk0bnXU/++wz4ePjo7MtHx8foVKptMuaNm0qunbtqv05Ly9P2Nvbi59++kkIIURcXJwAIBYsWKAtk5ubK+rVqycWLlwohBDio48+Ek899ZTOvq9fvy4AiJiYGCGEEN26dRPt2rV75PF6eXmJefPm6Sx77LHHxKuvvqr9uU2bNmLOnDklbqdbt25iypQp2p99fHzEZ599plNm/PjxYuLEiTrLDhw4ICwsLMT9+/e16w0ePPiR9f70009FQECA9ueiXg8hhAAgNm/eLIQQYuXKlaJmzZoiPT1d+/y2bduEhYWFSExMFELkv2Z5eXnaMsOGDRMjRox4ZJ2IzJGVcaMaEVV1CxcuRI8ePUrVmlKcFi1awMIivyHb3d0dLVu21P5saWmJ2rVrIykpSWe9oKAg7WMrKyt06NAB0dHRAIDTp08jPDwcDg4OhfYXGxuLJk2aAAACAgJKrFtaWhpu3ryJzp076yzv3LkzTp8+XcojLL3Tp0/jzJkzWLt2rXaZEAJqtRpxcXHw9/cHAHTo0KHQuhs2bMDSpUsRGxuL9PR05OXlwcnJqUz7j46ORps2bWBvb69d1rlzZ6jVasTExMDd3R2AfM0sLS21ZTw9PXH27Nky7YvIXDAsEVGJnnjiCfTu3RszZ87EmDFjdJ6zsLCAEEJnWVHjWqytrXV+VigURS5Tq9Wlrld6ejoGDBiAhQsXFnrO09NT+7hgKKgK0tPT8dJLL+H1118v9Fz9+vW1jx+u95EjRxAaGooPPvgAvXv3hrOzM9avX4/FixcbpJ4VfX2IzAnDEhE90oIFC9C2bVs0bdpUZ7mbmxsSExMhhNAOYNbn3Ej//PMPnnjiCQBAXl4eIiIitGNr2rdvj19//RW+vr6wsir/W5mTkxO8vLxw6NAhdOvWTbv80KFD6NixY4Xqb2NjA5VKpbOsffv2OH/+PBo1alSmbR0+fBg+Pj6YNWuWdtm1a9ceub+H+fv7IywsDBkZGdpAdujQIVhYWBR6fYlI4gBvInqkVq1aITQ0FEuXLtVZ3r17d/z333/45JNPEBsbi6+++grbt2/X236/+uorbN68GRcuXMCkSZNw9+5djBs3DgAwadIkJCcn49lnn8Xx48cRGxuLnTt3YuzYsY8MDA+bPn06Fi5ciA0bNiAmJgYzZsxAZGQkpkyZUqH6+/r64u+//8aNGze0Z+m98847OHz4MCZPnozIyEhcunQJv/32W6EB1g9r3Lgx4uPjsX79esTGxmLp0qXYvHlzof3FxcUhMjISt2/fRnZ2dqHthIaGwtbWFqNHj0ZUVBTCw8Px2muv4YUXXtB2wRGRLoYlIiqVDz/8sFA3jL+/P77++mt89dVXaNOmDY4dO1ahsU0PW7BgARYsWIA2bdrg4MGD2Lp1K1xdXQFA2xqkUqnw1FNPoVWrVpg6dSpcXFx0xkeVxuuvv44333wT06ZNQ6tWrbBjxw5s3boVjRs3rlD9P/zwQ1y9ehUNGzaEm5sbAKB169bYv38/Ll68iK5du6Jdu3aYPXs2vLy8StzWwIED8cYbb2Dy5Mlo27YtDh8+rD1LTmPo0KHo06cPnnzySbi5ueGnn34qtJ0aNWpg586dSE5OxmOPPYZnnnkGwcHB+PLLLyt0rETmTCEeHnBARERERFpsWSIiIiIqAcMSERERUQkYloiIiIhKwLBEREREVAKGJSIiIqISMCwRERERlYBhiYiIiKgEDEtEREREJWBYIiIiIioBwxIRERFRCRiWiIiIiErw/4iPtVTGt+Y/AAAAAElFTkSuQmCC",
"text/plain": [
- ""
+ ""
]
},
- "metadata": {
- "needs_background": "light"
- },
+ "metadata": {},
"output_type": "display_data"
}
],
@@ -520,7 +513,7 @@
],
"metadata": {
"kernelspec": {
- "display_name": "Python 3 (ipykernel)",
+ "display_name": "modellib",
"language": "python",
"name": "python3"
},
@@ -534,7 +527,7 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
- "version": "3.8.13"
+ "version": "3.7.15"
},
"toc": {
"base_numbering": 1,
@@ -548,6 +541,11 @@
"toc_position": {},
"toc_section_display": true,
"toc_window_display": false
+ },
+ "vscode": {
+ "interpreter": {
+ "hash": "8f24120f890011f53feb4ed62c47961d8565ec1de8b7cb23548c15bd6da8f2d2"
+ }
}
},
"nbformat": 4,
diff --git a/tutorials/quantum_simulation/VQE_EN.ipynb b/tutorials/quantum_simulation/VQE_EN.ipynb
index 9d20b85..36b263d 100644
--- a/tutorials/quantum_simulation/VQE_EN.ipynb
+++ b/tutorials/quantum_simulation/VQE_EN.ipynb
@@ -95,6 +95,7 @@
]
},
{
+ "attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
@@ -102,7 +103,7 @@
"\n",
"### Building electronic Hamiltonian\n",
"\n",
- "First of all, let us import the necessary libraries and packages.`qchem` module in Paddle Quantum is developed basing on `psi4` and `openfermion`, so you need to install these two packages before executing the following codes. We strongly encourage you to read the tutorial [Building molecular Hamiltonian](./BuildingMolecule_EN.ipynb) first, which introduces how to utilize our quantum chemistry toolkit `qchem`.\n",
+ "First of all, let us import the necessary libraries and packages. *qchem* module in Paddle Quantum is developed on top of *openfermion*, and currently, its quantum chemistry backend is *PySCF* so you need to install these two packages before executing the following codes (NOTE: PySCF only support Mac and Linux platform, we are adding support for more quantum chemistry backends, and will improve our support for Windows in our next release). We strongly encourage you to read the tutorial [Building molecular Hamiltonian](./BuildingMolecule_EN.ipynb) first, which introduces how to utilize our quantum chemistry toolkit `qchem`.\n",
"\n",
"**Note: As to the environment setting, please refer to [README.md](https://github.com/PaddlePaddle/Quantum/blob/master/README.md).**"
]
@@ -140,10 +141,11 @@
]
},
{
+ "attachments": {},
"cell_type": "markdown",
"metadata": {},
"source": [
- "To analyze specific molecules, we need several key information such as **geometry**, **basis set** (such as STO-3G), **multiplicity**, and **charge** to model the molecule and achieve more information about the molecule, such as one-body integrations, two-body integrations, and others. Next, through our built-in quantum chemistry toolkit, we could set a `Hamiltonian` class to carry molecule Hamiltonian information, which may simplify the following calculations."
+ "To analyze specific molecules, we need several key information such as **geometry**, **basis set** (such as STO-3G), **multiplicity**, and **charge** to model the molecule and to calculate its one-body integral, two-body integral and the Hamiltonian. We can use `Molecule` class provided by *qchem* in Paddle Quantum to easily obtain molecular Hamiltonian stored as Paddle Quantum's `Hamiltonian` object. This will facilitates our further operations."
]
},
{
@@ -160,47 +162,39 @@
"name": "stdout",
"output_type": "stream",
"text": [
- "FCI energy for H2_sto-3g_singlet (2 electrons) is -1.137283834485513.\n",
+ "converged SCF energy = -1.11675930739643\n",
"\n",
"The generated h2 Hamiltonian is \n",
- " -0.09706626861762556 I\n",
- "-0.04530261550868938 X0, X1, Y2, Y3\n",
- "0.04530261550868938 X0, Y1, Y2, X3\n",
- "0.04530261550868938 Y0, X1, X2, Y3\n",
- "-0.04530261550868938 Y0, Y1, X2, X3\n",
- "0.1714128263940239 Z0\n",
- "0.16868898168693286 Z0, Z1\n",
- "0.12062523481381837 Z0, Z2\n",
- "0.16592785032250773 Z0, Z3\n",
- "0.17141282639402394 Z1\n",
- "0.16592785032250773 Z1, Z2\n",
- "0.12062523481381837 Z1, Z3\n",
- "-0.2234315367466399 Z2\n",
- "0.17441287610651626 Z2, Z3\n",
- "-0.2234315367466399 Z3\n"
+ " -0.09706626816763092 I\n",
+ "0.17141282644776917 Z0\n",
+ "0.17141282644776917 Z1\n",
+ "-0.2234315369081346 Z2\n",
+ "-0.2234315369081346 Z3\n",
+ "0.16868898170361205 Z0, Z1\n",
+ "0.12062523483390414 Z0, Z2\n",
+ "0.1659278503377034 Z0, Z3\n",
+ "0.1659278503377034 Z1, Z2\n",
+ "0.12062523483390414 Z1, Z3\n",
+ "0.1744128761226159 Z2, Z3\n",
+ "-0.04530261550379926 X0, X1, Y2, Y3\n",
+ "0.04530261550379926 X0, Y1, Y2, X3\n",
+ "0.04530261550379926 Y0, X1, X2, Y3\n",
+ "-0.04530261550379926 Y0, Y1, X2, X3\n"
]
}
],
"source": [
- "geo = qchem.geometry(structure=[['H', [-0., 0., 0.0]], ['H', [-0., 0., 0.74]]])\n",
- "# geo = qchem.geometry(file='h2.xyz')\n",
- "\n",
- "# Save molecule information in to variable molecule, including one-body integrations, one-body integrations, molecular and Hamiltonian\n",
- "molecule = qchem.get_molecular_data(\n",
- " geometry=geo,\n",
- " basis='sto-3g',\n",
- " charge=0,\n",
- " multiplicity=1,\n",
- " method=\"fci\",\n",
- " if_save=True,\n",
- " if_print=True\n",
+ "mol = qchem.Molecule(\n",
+ " geometry=[('H', [-0., 0., 0.0]), ('H', [-0., 0., 0.74])], # Molecular geometry\n",
+ " basis=\"sto-3g\", # Basis set\n",
+ " multiplicity=1, # Spin multiplicity\n",
+ " charge=0, # Total charge\n",
+ " driver=qchem.PySCFDriver() # Quantum chemistry engine\n",
")\n",
- "# Recall Hamiltonian\n",
- "molecular_hamiltonian = qchem.spin_hamiltonian(molecule=molecule,\n",
- " filename=None, \n",
- " multiplicity=1, \n",
- " mapping_method='jordan_wigner',)\n",
- "# Print results\n",
+ "# extract Hamiltonian\n",
+ "molecular_hamiltonian = mol.get_molecular_hamiltonian()\n",
+ "\n",
+ "# print result\n",
"print(\"\\nThe generated h2 Hamiltonian is \\n\", molecular_hamiltonian)"
]
},
@@ -356,23 +350,23 @@
"name": "stdout",
"output_type": "stream",
"text": [
- "iter: 20 loss: -1.0628\n",
- "iter: 20 Ground state energy: -1.0628 Ha\n",
- "iter: 40 loss: -1.1323\n",
- "iter: 40 Ground state energy: -1.1323 Ha\n",
- "iter: 60 loss: -1.1361\n",
- "iter: 60 Ground state energy: -1.1361 Ha\n",
+ "iter: 20 loss: -1.0769\n",
+ "iter: 20 Ground state energy: -1.0769 Ha\n",
+ "iter: 40 loss: -1.1309\n",
+ "iter: 40 Ground state energy: -1.1309 Ha\n",
+ "iter: 60 loss: -1.1365\n",
+ "iter: 60 Ground state energy: -1.1365 Ha\n",
"iter: 80 loss: -1.1372\n",
"iter: 80 Ground state energy: -1.1372 Ha\n",
"\n",
"Circuit after training:\n",
- "--Ry(1.603)----*--------------x----Ry(4.714)----*--------------x----Ry(3.110)--\n",
+ "--Ry(4.710)----*--------------x----Ry(1.564)----*--------------x----Ry(6.274)--\n",
" | | | | \n",
- "--Ry(1.548)----x----*---------|----Ry(1.566)----x----*---------|----Ry(-1.65)--\n",
+ "--Ry(4.674)----x----*---------|----Ry(4.707)----x----*---------|----Ry(3.918)--\n",
" | | | | \n",
- "--Ry(-0.07)---------x----*----|----Ry(4.486)---------x----*----|----Ry(1.732)--\n",
+ "--Ry(2.376)---------x----*----|----Ry(5.028)---------x----*----|----Ry(4.713)--\n",
" | | | | \n",
- "--Ry(0.148)--------------x----*----Ry(7.876)--------------x----*----Ry(0.031)--\n",
+ "--Ry(-0.00)--------------x----*----Ry(4.937)--------------x----*----Ry(0.044)--\n",
" \n"
]
}
@@ -434,7 +428,7 @@
},
{
"cell_type": "code",
- "execution_count": 41,
+ "execution_count": 8,
"metadata": {
"ExecuteTime": {
"end_time": "2021-04-30T09:15:18.096944Z",
@@ -444,14 +438,12 @@
"outputs": [
{
"data": {
- "image/png": "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",
+ "image/png": "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",
"text/plain": [
- ""
+ ""
]
},
- "metadata": {
- "needs_background": "light"
- },
+ "metadata": {},
"output_type": "display_data"
}
],
@@ -525,7 +517,7 @@
],
"metadata": {
"kernelspec": {
- "display_name": "Python 3",
+ "display_name": "modellib",
"language": "python",
"name": "python3"
},
@@ -539,7 +531,7 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
- "version": "3.8.13"
+ "version": "3.7.15"
},
"toc": {
"base_numbering": 1,
@@ -558,6 +550,11 @@
},
"toc_section_display": true,
"toc_window_display": false
+ },
+ "vscode": {
+ "interpreter": {
+ "hash": "8f24120f890011f53feb4ed62c47961d8565ec1de8b7cb23548c15bd6da8f2d2"
+ }
}
},
"nbformat": 4,
diff --git a/tutorials/quantum_simulation/output/summary_data.npz b/tutorials/quantum_simulation/output/summary_data.npz
index 1de85fd..88a36f9 100644
Binary files a/tutorials/quantum_simulation/output/summary_data.npz and b/tutorials/quantum_simulation/output/summary_data.npz differ
+ 
![](\"./med_image_example.png\")
"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## 使用 QNNMIC 模型进行医学图像分类\n",
+ "\n",
+ "### QNNMIC 模型简介\n",
+ "QNNMIC 模型是一个可以用于医学图像分类的量子机器学习模型（Quantum Machine Learning，QML）。我们具体称其为一种量子神经网络 (Quantum Neural Network, QNN)，它结合了参数化量子电路（Parameterized Quantum Circuit, PQC）和经典神经网络。对于医学图像数据，QNNMIC 可以达到 85% 以上的分类准确率。模型主要分为量子和经典两部分，结构图如下：\n",
+ "\n",
+ "![](\"./qnnmic_model_cn.png\")
\n",
+ "\n",
+ "\n",
+ "注：\n",
+ "- 通常我们使用主成分分析将图片数据进行降维处理，使其更容易通过编码电路将经典数据编码为量子态。\n",
+ "- 参数化电路的作用是特征提取，其电路参数可以在训练中调整。\n",
+ "- 量子测量由一组测量算子表示，是将量子态转化为经典数据的过程，我们可以对得到的经典数据做进一步处理。\n"
+ ]
+ },
+ {
+ "attachments": {},
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## 如何使用\n",
+ "\n",
+ "### 使用模型进行预测\n",
+ "\n",
+ "这里，我们已经给出了一个训练好的模型，可以直接用于医学图片的预测。只需要在 `test.toml` 这个配置文件中进行对应的配置，然后输入命令 `python qnn_medical_image.py --config test.toml` 即可使用训练好的医学图片分类模型对输入的图片进行测试。\n",
+ "\n",
+ "### 在线演示\n",
+ "\n",
+ "这里，我们给出一个在线演示的版本，可以在线进行测试。首先定义配置文件的内容对测试集中图片进行预测：\n"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 1,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "import toml\n",
+ "\n",
+ "test_toml = r\"\"\"\n",
+ "# 模型的整体配置文件。\n",
+ "# 图片的文件路径。\n",
+ "image_path = 'pneumoniamnist'\n",
+ "\n",
+ "# 训练集中的数据个数，默认值为 -1 即使用全部数据。\n",
+ "num_samples = 20\n",
+ "\n",
+ "# 训练好的模型参数文件的文件路径。\n",
+ "model_path = 'qnnmic.pdparams'\n",
+ "\n",
+ "# 量子电路所包含的量子比特的数量。\n",
+ "num_qubits = [8, 8]\n",
+ "\n",
+ "# 每一层量子电路中的电路深度。\n",
+ "num_depths = [2, 2]\n",
+ "\n",
+ "# 量子电路中可观测量的设置。\n",
+ "observables = [['Z0', 'Z1', 'Z2', 'Z3'], ['X0', 'X1', 'X2', 'X3']]\n",
+ "\"\"\"\n",
+ "\n",
+ "config = toml.loads(test_toml)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 3,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "图片的预测结果分别为 1, 1, 1, 1, 0, 0, 1, 1, 1, 1, 0, 0, 1, 0, 1, 1, 1, 1, 0, 0\n",
+ "图片的实际标签分别为 0, 1, 1, 0, 0, 0, 1, 1, 1, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 0\n"
+ ]
+ }
+ ],
+ "source": [
+ "from paddle_quantum.qml.qnnmic import inference\n",
+ "\n",
+ "prediction, prob, label = inference(**config)\n",
+ "print(f\"图片的预测结果分别为 {str(prediction)[1:-1]}\")\n",
+ "print(f\"图片的实际标签分别为 {str(label)[1:-1]}\")\n"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "其中标签 0 代表肺部异常，标签 1 代表正常。"
+ ]
+ },
+ {
+ "attachments": {},
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "在 `test_toml` 配置文件中：\n",
+ "- `model_path`: 为训练好的模型，这里固定为 `qnnmic.pdparams`；\n",
+ "- `num_qubits`、`num_depths`、`observables` 三个参数应与训练好的模型 ``qnnmic.pdparams`` 相匹配。`num_qubits = [8,8]` 表示量子部分一共两层电路；每层电路为 8 的量子比特；`num_depths = [2,2]` 表示每层参数化电路深度为 2；`observables` 表示每层测量算子的具体形式。"
+ ]
+ },
+ {
+ "attachments": {},
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "对于数据集中的某张肺部异常的图片：\n",
+ "\n",
+ "![](\"./medical_image/image_label_0.png\")
"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 4,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "对于上述输入的图片，模型有 98.30% 的置信度检测出肺部异常。\n"
+ ]
+ }
+ ],
+ "source": [
+ "# 使用模型进行预测并得到对应概率值\n",
+ "msg = f'对于上述输入的图片，模型有 {prob[10][1]:3.2%} 的置信度检测出肺部异常。'\n",
+ "print(msg)\n"
+ ]
+ },
+ {
+ "attachments": {},
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## 注意事项\n",
+ "\n",
+ "我们通常考虑调整 `num_qubits`，`num_depths`，`observables` 三个主要超参数，对模型的影响较大。"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## 引用信息\n",
+ "\n",
+ "```\n",
+ "@article{medmnistv2,\n",
+ " title={MedMNIST v2: A Large-Scale Lightweight Benchmark for 2D and 3D Biomedical Image Classification},\n",
+ " author={Yang, Jiancheng and Shi, Rui and Wei, Donglai and Liu, Zequan and Zhao, Lin and Ke, Bilian and Pfister, Hanspeter and Ni, Bingbing},\n",
+ " journal={arXiv preprint arXiv:2110.14795},\n",
+ " year={2021}\n",
+ "}\n",
+ "```\n",
+ "\n",
+ "## 参考文献\n",
+ "\\[1\\] Yang, Jiancheng, et al. \"Medmnist v2: A large-scale lightweight benchmark for 2d and 3d biomedical image classification.\" arXiv preprint arXiv:2110.14795 (2021)."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": []
+ }
+ ],
+ "metadata": {
+ "kernelspec": {
+ "display_name": "modellib",
+ "language": "python",
+ "name": "python3"
+ },
+ "language_info": {
+ "codemirror_mode": {
+ "name": "ipython",
+ "version": 3
+ },
+ "file_extension": ".py",
+ "mimetype": "text/x-python",
+ "name": "python",
+ "nbconvert_exporter": "python",
+ "pygments_lexer": "ipython3",
+ "version": "3.7.15 (default, Nov 24 2022, 18:44:54) [MSC v.1916 64 bit (AMD64)]"
+ },
+ "orig_nbformat": 4,
+ "vscode": {
+ "interpreter": {
+ "hash": "dfa0523b1e359b8fd3ea126fa0459d0c86d49956d91b464930b80cba21582eac"
+ }
+ }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}
diff --git a/applications/medical_image_classification/introduction_en.ipynb b/applications/medical_image_classification/introduction_en.ipynb
new file mode 100644
index 0000000..63617b7
--- /dev/null
+++ b/applications/medical_image_classification/introduction_en.ipynb
@@ -0,0 +1,210 @@
+{
+ "cells": [
+ {
+ "attachments": {},
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Medical image classification\n",
+ "\n",
+ "*Copyright (c) 2022 Institute for Quantum Computing, Baidu Inc. All Rights Reserved.*\n",
+ "\n",
+ "Medical image classification is the key technology of computer-aided diagnosis systems. The problem of medical image classification is how to extract features from images and classify them, so as to identify and understand which parts of the human body are affected by specific diseases. Here, we mainly use a quantum neural network to classify the chest data in the open data set MedMNIST, which has the following form\n",
+ "\n",
+ "![](\"./qnnmic_model_en.png\")
\n",
+ "\n",
+ "\n",
+ "Remarks：\n",
+ "- In general, we use principal component analysis (PCA) to reduce the dimension of the image data, making it easier to encode classical data into quantum states through coding circuits.\n",
+ "- The parameterized circuit is used for feature extraction, and its circuit parameters can be adjusted during training.\n",
+ "- Quantum measurement, represented by a set of measurement operators, is the process of converting quantum states into classical data, which can be further processed.\n"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "## Quick start\n",
+ "\n",
+ "### Use the model to make predictions\n",
+ "\n",
+ "Here, we have given a trained model saved in the format `qnnmic.pdparams` which can be directly used to distinguish medical images. One only needs to do the corresponding configuration in this file `test.toml`, and enter the command `python qnn_medical_image.py --config test.toml` to predict the input images.\n",
+ "\n",
+ "### Online Test\n",
+ "\n",
+ "The following shows how to configure the test file `test_toml` to make medical image prediction."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 5,
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "import toml\n",
+ "\n",
+ "test_toml = r\"\"\"\n",
+ "# The config for testing the QNNMIC model.\n",
+ "\n",
+ "# The path of the input image.\n",
+ "image_path = 'pneumoniamnist'\n",
+ "\n",
+ "# The number of data in the test dataset.\n",
+ "# The value defaults to -1 which means using all data.\n",
+ "num_samples = 20\n",
+ "\n",
+ "# The path of the trained model, which will be loaded.\n",
+ "model_path = 'qnnmic.pdparams'\n",
+ "\n",
+ "# The number of qubits of the quantum circuit in each layer.\n",
+ "num_qubits = [8, 8]\n",
+ "\n",
+ "# The depth of the quantum circuit in each layer.\n",
+ "num_depths = [2, 2]\n",
+ "\n",
+ "# The observables of the quantum circuit in each layer.\n",
+ "observables = [['Z0','Z1','Z2','Z3'], ['X0','X1','X2','X3']]\n",
+ "\"\"\"\n",
+ "\n",
+ "config = toml.loads(test_toml)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 9,
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "The prediction results of the input images are 1, 1, 1, 1, 0, 0, 1, 1, 1, 1, 0, 0, 1, 0, 1, 1, 1, 1, 0, 0 respectively.\n",
+ "The labels of the input images are 0, 1, 1, 0, 0, 0, 1, 1, 1, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 0 respectively.\n"
+ ]
+ }
+ ],
+ "source": [
+ "from paddle_quantum.qml.qnnmic import inference\n",
+ "\n",
+ "prediction, prob, label = inference(**config)\n",
+ "print(f\"The prediction results of the input images are {str(prediction)[1:-1]} respectively.\")\n",
+ "print(f\"The labels of the input images are {str(label)[1:-1]} respectively.\")"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "Here label 0 means abnormal lungs, and label 1 means normal lungs."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "In `test_toml` :\n",
+ "- `model_path`: this is the trained model, here we set it as `qnnmic.pdparams`.\n",
+ "- `num_qubits`, `num_depths`, `observables` these parameters correspond to the model ``qnnmic.pdparams``, `num_qubits = [8,8]` represents the quantum part of a total of two layers of circuit, each layer of the circuit has 8 qubits; `num_depths = [2,2]` represents the depth of parameterized circuit of each layer is 2；`observables` is the specific form of the measurement operator at each layer.\n"
+ ]
+ },
+ {
+ "attachments": {},
+ "cell_type": "markdown",
+ "metadata": {},
+ "source": [
+ "For an abnormal image from the dataset\n",
+ "\n",
+ "![](\"./surface_crack_example.png\")
\n",
+ "\n",
+ "我们具体使用的训练集包含 1000 张如上图所示的图片，为了测试模型的检测能力，我们选择了 200 张图片作为测试集。"
+ ]
+ },
+ {
+ "attachments": {},
+ "cell_type": "markdown",
+ "id": "15abdd24",
+ "metadata": {},
+ "source": [
+ "## 使用 QNNQD 模型进行裂缝图片检测"
+ ]
+ },
+ {
+ "attachments": {},
+ "cell_type": "markdown",
+ "id": "977c0662",
+ "metadata": {},
+ "source": [
+ "### QNNQD 模型简介\n",
+ "\n",
+ "QNNQD 模型是一个可以用于图片分类的量子机器学习模型（Quantum Machine Learning，QML）。我们具体称其为一种量子神经网络 (Quantum Neural Network, QNN)，它结合了参数化量子电路（Parameterized Quantum Circuit, PQC）和经典神经网络。对于表面裂缝图片数据，QNNQD 可以达到 90% 以上的分类准确率。模型主要分为量子和经典两部分，结构图如下：\n",
+ "\n",
+ "![](\"./qnnqd_model_cn.png\")
\n",
+ "\n",
+ "模型处理数据的主要过程：\n",
+ "1. 通常我们使用主成分分析将图片数据进行降维处理，使得编码电路更容易将经典数据编码为量子态。\n",
+ "2. 参数化电路的作用是特征提取，其电路参数可以在训练中被调整，可类比与经典神经网络中的可训练参数。\n",
+ "3. 量子测量由一组测量算子表示，是将量子态转化为经典数据的过程，我们可以对得到的经典数据做进一步处理。"
+ ]
+ },
+ {
+ "attachments": {},
+ "cell_type": "markdown",
+ "id": "17661d06",
+ "metadata": {},
+ "source": [
+ "## 模型快速使用"
+ ]
+ },
+ {
+ "attachments": {},
+ "cell_type": "markdown",
+ "id": "f8a3ebf3",
+ "metadata": {},
+ "source": [
+ "### 使用模型进行预测\n",
+ "\n",
+ "我们可以通过在命令行输入 `python qnn_quality_detection.py --config train.toml` 即可开始训练模型； 这里，我们已经给出了一个训练好的模型，保存格式为 `qnnqd.pdparams`，可以直接用于区分表面裂缝数据。只需要在 `test.toml` 这个配置文件中进行对应的配置，然后在终端输入命令 `python qnn_quality_detection.py --config test.toml` 即可对图片进行预测。为了避免配置过多参数，我们提供了 `example.toml` 文件，可以简单的配置之后执行 `python qnn_quality_detection.py --config example.toml` 完成图片预测。"
+ ]
+ },
+ {
+ "attachments": {},
+ "cell_type": "markdown",
+ "id": "4f739274",
+ "metadata": {},
+ "source": [
+ "### 在线演示\n",
+ "\n",
+ "下面展示如果通过配置测试文件 `test_toml` 进行裂缝图片预测。"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 1,
+ "id": "3e3772fe",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "import toml\n",
+ "\n",
+ "test_toml = r\"\"\"\n",
+ "# 模型的整体配置文件。\n",
+ "# 图片的文件路径。\n",
+ "image_path = 'SurfaceCrack/test_data/'\n",
+ "\n",
+ "# 训练集中的数据个数，默认值为 -1 即使用全部数据。\n",
+ "num_samples = 20\n",
+ "\n",
+ "# 训练好的模型参数文件的文件路径。\n",
+ "model_path = 'qnnqd.pdparams'\n",
+ "\n",
+ "# 量子电路所包含的量子比特的数量。\n",
+ "num_qubits = [4, 4]\n",
+ "\n",
+ "# 每一层量子电路中的电路深度。\n",
+ "num_depths = [2, 2]\n",
+ "\n",
+ "# 量子电路中可观测量的设置。\n",
+ "observables = [['Z0', 'Z1', 'Z2', 'Z3'], ['Z0', 'Z1', 'Z2', 'Z3']]\n",
+ "\"\"\"\n",
+ "\n",
+ "config = toml.loads(test_toml)"
+ ]
+ },
+ {
+ "attachments": {},
+ "cell_type": "markdown",
+ "id": "e5abd75c",
+ "metadata": {},
+ "source": [
+ "在 `test_toml` 配置文件中：\n",
+ "- `model_path`: 为训练好的模型，这里固定为 `qnnqd.pdparams`；\n",
+ "- `num_qubits`、`num_depths`、`observables` 三个参数应与训练好的模型 ``qnnqd.pdparams`` 相匹配。`num_qubits = [4,4]` 表示量子部分一共两层电路；每层电路为 4 的量子比特；`num_depths = [2,2]` 表示每层参数化电路深度为 2；`observables` 表示每层测量算子的具体形式。"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 2,
+ "id": "c6c9f96b",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "d:\\anaconda3\\envs\\modellib\\lib\\site-packages\\paddle\\tensor\\creation.py:125: DeprecationWarning: `np.object` is a deprecated alias for the builtin `object`. To silence this warning, use `object` by itself. Doing this will not modify any behavior and is safe. \n",
+ "Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations\n",
+ " if data.dtype == np.object:\n"
+ ]
+ },
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "图片的预测结果分别为 0, 0, 0, 1, 0, 0, 1, 0, 1, 1, 1, 0, 0, 0, 0, 1, 1, 0, 1, 1\n",
+ "图片的实际标签分别为 1, 0, 0, 1, 0, 0, 1, 0, 1, 0, 1, 0, 0, 0, 1, 1, 1, 0, 1, 1\n"
+ ]
+ }
+ ],
+ "source": [
+ "from paddle_quantum.qml.qnnqd import inference\n",
+ "\n",
+ "prediction, prob, label = inference(**config)\n",
+ "print(f\"图片的预测结果分别为 {str(prediction)[1:-1]}\")\n",
+ "print(f\"图片的实际标签分别为 {str(label)[1:-1]}\")"
+ ]
+ },
+ {
+ "attachments": {},
+ "cell_type": "markdown",
+ "id": "5b1bc8f9",
+ "metadata": {},
+ "source": [
+ "以上是模型对于测试集给出的预测结果。下面具体给出两张图片的预测细节"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 3,
+ "id": "230c5283",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "# 导入所需要的包\n",
+ "import os\n",
+ "import warnings\n",
+ "\n",
+ "warnings.filterwarnings('ignore')\n",
+ "os.environ['PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION'] = 'python'\n",
+ "\n",
+ "import numpy as np\n",
+ "import paddle\n",
+ "import matplotlib.pyplot as plt\n",
+ "from paddle_quantum.qml.qnnqd import QNNQD, ImageDataset\n",
+ "\n",
+ "# 设置模型参数\n",
+ "num_qubits = [4,4]\n",
+ "num_depths = [2,2]\n",
+ "observables = [['Z0','Z1','Z2','Z3'], ['Z0','Z1','Z2','Z3']]\n",
+ "\n",
+ "# 加载已训练的模型\n",
+ "model = QNNQD(\n",
+ " num_qubits=num_qubits,\n",
+ " num_depths=num_depths,\n",
+ " observables=observables,\n",
+ ")\n",
+ "state_dict = paddle.load('./qnnqd.pdparams')\n",
+ "model.set_state_dict(state_dict)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 4,
+ "id": "db11f93d",
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/plain": [
+ "![](\"./qnnqd_model_en.png\")
\n",
+ "\n",
+ "\n",
+ "Main process：\n",
+ "- In general, we use principal component analysis (PCA) to reduce the dimension of image data, making it easier to encode classical data into quantum states through encoding circuits.\n",
+ "- The parameterized circuit is used for feature extraction, and its circuit parameters can be adjusted during training. Similar to the trainable parameters in traditional neural networks.\n",
+ "- Quantum measurement, represented by a set of measurement operators, is the process of converting quantum states into classical data, which can be further processed."
+ ]
+ },
+ {
+ "attachments": {},
+ "cell_type": "markdown",
+ "id": "17661d06",
+ "metadata": {},
+ "source": [
+ "## Quick start"
+ ]
+ },
+ {
+ "attachments": {},
+ "cell_type": "markdown",
+ "id": "f8a3ebf3",
+ "metadata": {},
+ "source": [
+ "### Use the model to make predictions\n",
+ "\n",
+ "Entering the command `python qnn_quality_detection.py --config train.toml` will initiate the model training. Here, we have given a trained model saved in the format `qnnqd.pdparams`, which can be directly used to distinguish surface crack data. One only needs to do the corresponding configuration in this file `test.toml`, and enter the command `python qnn_quality_detection.py --config test.toml` to predict the image. To avoid configuring too many parameters, we provide the `example.toml` file which can be more easily configured. One can execute `python qnn_quality_detection.py --config example.toml` to complete image prediction."
+ ]
+ },
+ {
+ "attachments": {},
+ "cell_type": "markdown",
+ "id": "4f739274",
+ "metadata": {},
+ "source": [
+ "### Online Test\n",
+ "\n",
+ "The following shows how to configure the test file `test_toml` to make crack image prediction."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 1,
+ "id": "3e3772fe",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "import toml\n",
+ "\n",
+ "test_toml = r\"\"\"\n",
+ "# The config for test the QNNQD model.\n",
+ "# The path of the input image.\n",
+ "image_path = 'SurfaceCrack/test_data/'\n",
+ "\n",
+ "# The number of the data in the test dataset.\n",
+ "# The value defaults to -1 which means using all data.\n",
+ "num_samples = 20\n",
+ "\n",
+ "# The path of the trained model, which will be loaded.\n",
+ "model_path = 'qnnqd.pdparams'\n",
+ "\n",
+ "# The number of qubits of quantum circuit in each layer.\n",
+ "num_qubits = [4, 4]\n",
+ "\n",
+ "# The depth of quantum circuit in each layer.\n",
+ "num_depths = [2, 2]\n",
+ "\n",
+ "# The observables of quantum circuit in each layer.\n",
+ "observables = [['Z0', 'Z1', 'Z2', 'Z3'], ['Z0', 'Z1', 'Z2', 'Z3']]\n",
+ "\"\"\"\n",
+ "\n",
+ "config = toml.loads(test_toml)"
+ ]
+ },
+ {
+ "attachments": {},
+ "cell_type": "markdown",
+ "id": "e5abd75c",
+ "metadata": {},
+ "source": [
+ "In `test_toml` :\n",
+ "- `model_path`: this is the trained model, here we set it as `qnnqd.pdparams`；\n",
+ "- `num_qubits`、`num_depths`、`observables` these parameters correspond to the model ``qnnqd.pdparams`` , `num_qubits = [4,4]` represents the quantum part of a total of two layers of circuit, each layer of the circuit has 4 qubits; `num_depths = [2,2]` represents the depth of parameterized circuit of each layer is 2；`observables` is the specific form of the measurement operator at each layer.\n",
+ "\n",
+ "Test:"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 3,
+ "id": "c6c9f96b",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "The prediction results of the input images are 0, 0, 0, 1, 0, 0, 1, 0, 1, 1, 1, 0, 0, 0, 0, 1, 1, 0, 1, 1 respectively.\n",
+ "The labels of the input images are 1, 0, 0, 1, 0, 0, 1, 0, 1, 0, 1, 0, 0, 0, 1, 1, 1, 0, 1, 1 respectively.\n"
+ ]
+ }
+ ],
+ "source": [
+ "from paddle_quantum.qml.qnnqd import inference\n",
+ "\n",
+ "prediction, prob, label = inference(**config)\n",
+ "print(f\"The prediction results of the input images are {str(prediction)[1:-1]} respectively.\")\n",
+ "print(f\"The labels of the input images are {str(label)[1:-1]} respectively.\")"
+ ]
+ },
+ {
+ "attachments": {},
+ "cell_type": "markdown",
+ "id": "5b1bc8f9",
+ "metadata": {},
+ "source": [
+ "The above is the prediction result of the model for the test set. The prediction details of the two images are given below."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 4,
+ "id": "8d4191f4",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "# import\n",
+ "import os\n",
+ "import warnings\n",
+ "\n",
+ "warnings.filterwarnings('ignore')\n",
+ "os.environ['PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION'] = 'python'\n",
+ "\n",
+ "import numpy as np\n",
+ "import paddle\n",
+ "import matplotlib.pyplot as plt\n",
+ "from paddle_quantum.qml.qnnqd import QNNQD, ImageDataset\n",
+ "\n",
+ "# Set model parameters\n",
+ "num_qubits = [4,4]\n",
+ "num_depths = [2,2]\n",
+ "observables = [['Z0','Z1','Z2','Z3'], ['Z0','Z1','Z2','Z3']]\n",
+ "\n",
+ "# Load the trained model\n",
+ "model = QNNQD(\n",
+ " num_qubits=num_qubits,\n",
+ " num_depths=num_depths,\n",
+ " observables=observables,\n",
+ ")\n",
+ "state_dict = paddle.load('./qnnqd.pdparams')\n",
+ "model.set_state_dict(state_dict)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 5,
+ "id": "d74ec0d5",
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/plain": [
+ "