{ "cells": [ { "cell_type": "markdown", "id": "ee388de5-2507-48ff-97a6-6777698d6256", "metadata": {}, "source": [ "---\n", "title: \"Diagonalisation quantique de Krylov basée sur l'échantillonnage d'un modèle de trame fermionique\"\n", "description: \"Utilisez l'algorithme de diagonalisation quantique basé sur des échantillons pour simuler le modèle d'Anderson à impureté unique à l'aide d'un matériel quantique bruité.\"\n", "---\n", "\n", "{/* cspell:ignore fontdict fontsize nocc SQKD DMRG textrm varepsilon vecs pqrs ijkl */}\n", "\n", "\n", "\n", "# Diagonalisation quantique de Krylov basée sur l'échantillonnage d'un modèle de trame fermionique\n", "\n", "*Estimation de l'utilisation : Neuf secondes sur un processeur Heron r2 (NOTE : Il s'agit uniquement d'une estimation. Votre durée d'exécution peut varier.)*\n", "\n" ] }, { "cell_type": "markdown", "id": "a1b2c3d4-e5f6-7890-abcd-ef1234567890", "metadata": {}, "source": [ "\n", "\n", "## Résultats d'apprentissage\n", "\n", "À l'issue de ce tutoriel, les utilisateurs devraient avoir compris :\n", "\n", "* Comment utiliser le [module complémentaire SQD Qiskit](https://github.com/Qiskit/qiskit-addon-sqd) pour estimer l'énergie de l'état fondamental d'un modèle de réseau à l'aide de chaînes de bits échantillonnées à partir d'une unité de traitement quantique (QPU).\n", "* Comment utiliser [ffsim](https://github.com/qiskit-community/ffsim) pour construire des circuits d'évolution temporelle destinés à la simulation fermionique.\n", "* Comment combiner des échantillons provenant de plusieurs circuits en vue d'un traitement ultérieur à l'aide de l'algorithme de diagonalisation de Krylov par échantillonnage (SKQD).\n", "\n", "\n", "\n", "## Prérequis\n", "\n", "Nous recommandons aux utilisateurs de se familiariser avec les sujets suivants avant de suivre ce tutoriel :\n", "\n", "* [Diagonalisation quantique basée sur l'échantillonnage d'un hamiltonien chimique](/docs/tutorials/sample-based-quantum-diagonalization)\n", "* [Diagonalisation quantique de Krylov des hamiltoniens sur réseau](/docs/tutorials/krylov-quantum-diagonalization)\n", "* [Qiskit primitives](/docs/guides/primitives)\n", "\n" ] }, { "cell_type": "markdown", "id": "dc5cc74e-06bf-45ac-a69b-81778138e08f", "metadata": {}, "source": [ "\n", "\n", "## Arrière-plan\n", "\n", "Ce tutoriel montre comment utiliser la diagonalisation quantique basée sur les échantillons (SQD) pour estimer l'énergie de l'état fondamental d'un modèle de réseau fermionique. Plus précisément, nous étudions le modèle d'Anderson unidimensionnel à impureté unique (SIAM), qui est utilisé pour décrire les impuretés magnétiques intégrées dans les métaux.\n", "\n", "Ce tutoriel suit un processus similaire à celui du tutoriel connexe [Diagonalisation quantique basée sur l'échantillonnage d'un hamiltonien chimique](/docs/tutorials/sample-based-quantum-diagonalization). Cependant, une différence essentielle réside dans la manière dont les circuits quantiques sont construits. L'autre tutoriel utilise un ansatz variationnel heuristique, qui est intéressant pour les hamiltoniens de la chimie avec potentiellement des millions de termes d'interaction. D'autre part, ce tutoriel utilise des circuits qui se rapprochent de l'évolution du temps par l'intermédiaire de l'hamiltonien. Ces circuits peuvent être profonds, ce qui rend cette approche plus adaptée aux applications des modèles de treillis. Les vecteurs d'état préparés par ces circuits forment la base d'un [sous-espace de Krylov](https://en.wikipedia.org/wiki/Krylov_subspace) et, par conséquent, l'algorithme converge de manière prouvée et efficace vers l'état fondamental, sous des hypothèses appropriées.\n", "\n", "L'approche utilisée dans ce tutoriel peut être considérée comme une combinaison des techniques utilisées dans la SQD et la [diagonalisation quantique de Krylov (KQD)](https://arxiv.org/abs/2407.14431). L'approche combinée est parfois appelée diagonalisation quantique de Krylov basée sur les échantillons (SQKD). Voir [Krylov quantum diagonalization of lattice Hamiltonians](/docs/tutorials/krylov-quantum-diagonalization) pour un tutoriel sur la méthode KQD.\n", "\n", "Ce tutoriel est basé sur le travail [\"Quantum-Centric Algorithm for Sample-Based Krylov Diagonalization\"](https://arxiv.org/abs/2501.09702), qui peut être consulté pour plus de détails.\n", "\n", "\n", "\n", "### Modèle d'Anderson à impureté unique (SIAM)\n", "\n", "L'hamiltonien SIAM unidimensionnel est une somme de trois termes :\n", "\n", "$$\n", "H = H_{\\textrm{imp}}+ H_\\textrm{bath} + H_\\textrm{hyb},\n", "$$\n", "\n", "où\n", "\n", "$$\n", "\\begin{align*}\n", " H_\\textrm{imp} &= \\varepsilon \\left( \\hat{n}_{d\\uparrow} + \\hat{n}_{d\\downarrow} \\right) + U \\hat{n}_{d\\uparrow}\\hat{n}_{d\\downarrow}, \\\\\n", " H_\\textrm{bath} &= -t \\sum_{\\substack{\\mathbf{j} = 0\\\\ \\sigma\\in \\{\\uparrow, \\downarrow\\}}}^{L-1} \\left(\\hat{c}^\\dagger_{\\mathbf{j}, \\sigma}\\hat{c}_{\\mathbf{j}+1, \\sigma} + \\hat{c}^\\dagger_{\\mathbf{j}+1, \\sigma}\\hat{c}_{\\mathbf{j}, \\sigma} \\right), \\\\\n", " H_\\textrm{hyb} &= V\\sum_{\\sigma \\in \\{\\uparrow, \\downarrow \\}} \\left(\\hat{d}^\\dagger_\\sigma \\hat{c}_{0, \\sigma} + \\hat{c}^\\dagger_{0, \\sigma} \\hat{d}_{\\sigma} \\right).\n", "\\end{align*}\n", "$$\n", "\n", "Ici, $c^\\dagger_{\\mathbf{j},\\sigma}/c_{\\mathbf{j},\\sigma}$ sont les opérateurs de création/annihilation fermioniques pour le site de bain $\\mathbf{j}^{\\textrm{th}}$ avec spin $\\sigma$, $\\hat{d}^\\dagger_{\\sigma}/\\hat{d}_{\\sigma}$ sont les opérateurs de création/annihilation pour le mode impureté, et $\\hat{n}_{d\\sigma} = \\hat{d}^\\dagger_{\\sigma} \\hat{d}_{\\sigma}$. $t$, $U$, et $V$ sont des nombres réels décrivant les interactions de saut, sur site, d'hybridation, et $\\varepsilon$ est un nombre réel spécifiant le potentiel chimique.\n", "\n", "Notez que le hamiltonien est une instance spécifique du hamiltonien générique interaction-électron,\n", "\n", "$$\n", "\\begin{align*}\n", " H &= \\sum_{\\substack{p, q \\\\ \\sigma}} h_{pq} \\hat{a}^\\dagger_{p\\sigma} \\hat{a}_{q\\sigma} + \\sum_{\\substack{p, q, r, s \\\\ \\sigma \\tau}} \\frac{h_{pqrs}}{2} \\hat{a}^\\dagger_{p\\sigma} \\hat{a}^\\dagger_{q\\tau} \\hat{a}_{s\\tau} \\hat{a}_{r\\sigma} \\\\\n", " &= H_1 + H_2,\n", "\\end{align*}\n", "$$\n", "\n", "où $H_1$ est constitué de termes à un corps, qui sont quadratiques dans les opérateurs de création et d'annihilation fermioniques, et $H_2$ est constitué de termes à deux corps, qui sont quartiques. Pour le SIAM,\n", "\n", "$$\n", "H_2 = U \\hat{n}_{d\\uparrow}\\hat{n}_{d\\downarrow}\n", "$$\n", "\n", "et $H_1$ contient le reste des termes de l'hamiltonien. Afin de représenter l'hamiltonien de manière programmatique, nous stockons la matrice $h_{pq}$ et le tenseur $h_{pqrs}$.\n", "\n", "\n", "\n", "### Bases de position et d'élan\n", "\n", "En raison de la symétrie translationnelle approximative dans $H_\\textrm{bath}$, nous ne nous attendons pas à ce que l'état fondamental soit peu dense dans la base de position (la base orbitale dans laquelle l'hamiltonien est spécifié ci-dessus). La performance de la SQD n'est garantie que si l'état de base est peu dense, c'est-à-dire qu'il n'a un poids significatif que sur un petit nombre d'états de base de calcul. Pour améliorer l'éparpillement de l'état fondamental, nous effectuons la simulation dans la base orbitale dans laquelle $H_\\textrm{bath}$ est diagonale. Nous appelons cette base la *base du momentum*. Comme $H_\\textrm{bath}$ est un hamiltonien fermionique quadratique, il peut être efficacement diagonalisé par une rotation orbitale.\n", "\n", "\n", "\n", "### Évolution temporelle approximative par l'hamiltonien\n", "\n", "Pour approximer l'évolution temporelle de l'hamiltonien, nous utilisons une décomposition de Trotter-Suzuki du second ordre,\n", "\n", "$$\n", " e^{-i \\Delta t H} \\approx e^{-i\\frac{\\Delta t}{2} H_2} e^{-i\\Delta t H_1} e^{-i\\frac{\\Delta t}{2} H_2}.\n", "$$\n", "\n", "Dans le cadre de la [transformation de Jordan-Wigner](https://en.wikipedia.org/wiki/Jordan%E2%80%93Wigner_transformation), l'évolution temporelle par $H_2$ équivaut à une porte [CPhase](/docs/api/qiskit/qiskit.circuit.library.CPhaseGate) unique entre les orbitales de spin ascendant et de spin descendant sur le site de l'impureté. Comme $H_1$ est un hamiltonien fermionique quadratique, l'évolution temporelle par $H_1$ équivaut à une rotation orbitale.\n", "\n", "Les états de base de Krylov $\\{ |\\psi_k\\rangle \\}_{k=0}^{D-1}$, où $D$ est la dimension du sous-espace de Krylov, sont formés par l'application répétée d'une seule étape de Trotter, de sorte que\n", "\n", "$$\n", " |\\psi_k\\rangle \\approx \\left[e^{-i\\frac{\\Delta t}{2} H_2} e^{-i\\Delta t H_1} e^{-i\\frac{\\Delta t}{2} H_2} \\right]^k\\ket{\\psi_0}.\n", "$$\n", "\n", "Dans le flux de travail suivant, basé sur la SQD, nous prélèverons des échantillons de cet ensemble de circuits et nous traiterons l'ensemble combiné de chaînes de bits à l'aide de la SQD. Cette approche contraste avec celle utilisée dans le tutoriel connexe intitulé [Sample-based quantum diagonalization of a chemistry Hamiltonian](/docs/tutorials/sample-based-quantum-diagonalization), où les échantillons ont été tirés d'un seul circuit variationnel heuristique.\n", "\n" ] }, { "cell_type": "markdown", "id": "df66d697-102c-4a2c-80f3-f67fdda05573", "metadata": {}, "source": [ "\n", "\n", "## Exigences\n", "\n", "Avant de commencer ce tutoriel, assurez-vous que les éléments suivants sont installés :\n", "\n", "* Qiskit SDK v1.0 ou plus tard, avec prise en charge de [la visualisation](/docs/api/qiskit/visualization)\n", "* Qiskit Runtime v0.22 ou plus tard (`pip install qiskit-ibm-runtime`)\n", "* Module complémentaire SQD Qiskit v0.11 ou version ultérieure (`pip install qiskit-addon-sqd`)\n", "* ffsim v0.0.72 ou version ultérieure (`pip install ffsim`)\n", "\n" ] }, { "cell_type": "markdown", "id": "c3d4e5f6-a7b8-9012-cdef-123456789012", "metadata": {}, "source": [ "\n", "\n", "## Exemple de simulateur à petite échelle\n", "\n" ] }, { "cell_type": "markdown", "id": "8540487a-8033-49c2-9f30-022336105f64", "metadata": {}, "source": [ "\n", "\n", "### Étape 1 : Mettre en correspondance le problème avec un circuit quantique\n", "\n" ] }, { "cell_type": "markdown", "id": "4e6652bd-c97d-4f2f-98a7-d44857805cf1", "metadata": {}, "source": [ "Tout d'abord, nous générons le hamiltonien SIAM dans la base de position. L'hamiltonien est représenté par la matrice $h_{pq}$ et le tenseur $h_{pqrs}$. Ensuite, nous le faisons pivoter vers la base de quantité de mouvement. Dans la base de position, nous plaçons l'impureté sur le premier site. Cependant, lorsque nous passons à la base de quantité de mouvement, nous déplaçons l'impureté vers un site central pour faciliter les interactions avec d'autres orbitales.\n", "\n" ] }, { "cell_type": "code", "execution_count": 1, "id": "b5cb9c28-4721-4141-8665-96885038e210", "metadata": {}, "outputs": [], "source": [ "import numpy as np\n", "import pyscf.fci\n", "\n", "\n", "def siam_hamiltonian(\n", " norb: int,\n", " hopping: float,\n", " onsite: float,\n", " hybridization: float,\n", " chemical_potential: float,\n", ") -> tuple[np.ndarray, np.ndarray]:\n", " \"\"\"Hamiltonian for the single-impurity Anderson model.\"\"\"\n", " # Place the impurity on the first site\n", " impurity_orb = 0\n", "\n", " # One body matrix elements in the \"position\" basis\n", " h1e = np.zeros((norb, norb))\n", " np.fill_diagonal(h1e[:, 1:], -hopping)\n", " np.fill_diagonal(h1e[1:, :], -hopping)\n", " h1e[impurity_orb, impurity_orb + 1] = -hybridization\n", " h1e[impurity_orb + 1, impurity_orb] = -hybridization\n", " h1e[impurity_orb, impurity_orb] = chemical_potential\n", "\n", " # Two body matrix elements in the \"position\" basis\n", " h2e = np.zeros((norb, norb, norb, norb))\n", " h2e[impurity_orb, impurity_orb, impurity_orb, impurity_orb] = onsite\n", "\n", " return h1e, h2e\n", "\n", "\n", "def momentum_basis(norb: int) -> np.ndarray:\n", " \"\"\"Get the orbital rotation to change from the position to the momentum basis.\"\"\"\n", " n_bath = norb - 1\n", "\n", " # Orbital rotation that diagonalizes the bath (non-interacting system)\n", " hopping_matrix = np.zeros((n_bath, n_bath))\n", " np.fill_diagonal(hopping_matrix[:, 1:], -1)\n", " np.fill_diagonal(hopping_matrix[1:, :], -1)\n", " _, vecs = np.linalg.eigh(hopping_matrix)\n", "\n", " # Expand to include impurity\n", " orbital_rotation = np.zeros((norb, norb))\n", " # Impurity is on the first site\n", " orbital_rotation[0, 0] = 1\n", " orbital_rotation[1:, 1:] = vecs\n", "\n", " # Move the impurity to the center\n", " new_index = n_bath // 2\n", " perm = np.r_[1 : (new_index + 1), 0, (new_index + 1) : norb]\n", " orbital_rotation = orbital_rotation[:, perm]\n", "\n", " return orbital_rotation\n", "\n", "\n", "def rotated(\n", " h1e: np.ndarray, h2e: np.ndarray, orbital_rotation: np.ndarray\n", ") -> tuple[np.ndarray, np.ndarray]:\n", " \"\"\"Rotate the orbital basis of a Hamiltonian.\"\"\"\n", " h1e_rotated = np.einsum(\n", " \"ab,Aa,Bb->AB\",\n", " h1e,\n", " orbital_rotation,\n", " orbital_rotation.conj(),\n", " optimize=\"greedy\",\n", " )\n", " h2e_rotated = np.einsum(\n", " \"abcd,Aa,Bb,Cc,Dd->ABCD\",\n", " h2e,\n", " orbital_rotation,\n", " orbital_rotation.conj(),\n", " orbital_rotation,\n", " orbital_rotation.conj(),\n", " optimize=\"greedy\",\n", " )\n", " return h1e_rotated, h2e_rotated\n", "\n", "\n", "# Total number of spatial orbitals, including the bath sites and the impurity\n", "# This should be an even number\n", "norb = 8\n", "\n", "# System is half-filled\n", "nelec = (norb // 2, norb // 2)\n", "# One orbital is the impurity, the rest are bath sites\n", "n_bath = norb - 1\n", "\n", "# Hamiltonian parameters\n", "hybridization = 1.0\n", "hopping = 1.0\n", "onsite = 10.0\n", "chemical_potential = -0.5 * onsite\n", "\n", "# Generate Hamiltonian in position basis\n", "h1e, h2e = siam_hamiltonian(\n", " norb=norb,\n", " hopping=hopping,\n", " onsite=onsite,\n", " hybridization=hybridization,\n", " chemical_potential=chemical_potential,\n", ")\n", "\n", "# Rotate to momentum basis\n", "orbital_rotation = momentum_basis(norb)\n", "h1e_momentum, h2e_momentum = rotated(h1e, h2e, orbital_rotation.T.conj())\n", "# In the momentum basis, the impurity is placed in the center\n", "impurity_index = n_bath // 2\n", "\n", "# Use PySCF to compute the exact ground state energy\n", "reference_energy, _ = pyscf.fci.direct_spin1.kernel(h1e, h2e, norb, nelec)" ] }, { "cell_type": "markdown", "id": "e888edf2-7865-41a8-be57-6ddb72dd0cc7", "metadata": {}, "source": [ "Ensuite, nous générons les circuits pour produire les états de la base de Krylov.\n", "Pour chaque espèce de spin, l'état initial $\\ket{\\psi_0}$ est donné par la superposition de toutes les excitations possibles des trois électrons les plus proches du niveau de Fermi dans les 4 modes vides les plus proches à partir de l'état $|00\\cdots 0011 \\cdots 11\\rangle$, et réalisé par l'application de sept [XXPlusYYGates](/docs/api/qiskit/qiskit.circuit.library.XXPlusYYGate).\n", "Les états évoluant dans le temps sont produits par des applications successives d'une étape de Trotter du second ordre.\n", "\n", "Pour une description plus détaillée de ce modèle et de la manière dont les circuits sont conçus, voir [\"Quantum-Centric Algorithm for Sample-Based Krylov Diagonalization\" (Algorithme centré sur le quantum pour la diagonalisation de Krylov à partir d'échantillons).](https://arxiv.org/abs/2501.09702)\n", "\n" ] }, { "cell_type": "code", "execution_count": 2, "id": "0f729f86-1814-4d5a-ae65-3f9614e103b3", "metadata": {}, "outputs": [], "source": [ "from typing import Sequence\n", "\n", "import ffsim\n", "import scipy\n", "from qiskit import QuantumCircuit, QuantumRegister\n", "from qiskit.circuit import CircuitInstruction, Qubit\n", "from qiskit.circuit.library import CPhaseGate, XGate, XXPlusYYGate\n", "\n", "\n", "def prepare_initial_state(qubits: Sequence[Qubit], norb: int, nocc: int):\n", " \"\"\"Prepare initial state.\"\"\"\n", " assert norb >= 8\n", " x_gate = XGate()\n", " rot = XXPlusYYGate(0.5 * np.pi, -0.5 * np.pi)\n", " for i in range(nocc):\n", " yield CircuitInstruction(x_gate, [qubits[i]])\n", " yield CircuitInstruction(x_gate, [qubits[norb + i]])\n", " for i in range(3):\n", " for j in range(nocc - i - 1, nocc + i, 2):\n", " yield CircuitInstruction(rot, [qubits[j], qubits[j + 1]])\n", " yield CircuitInstruction(\n", " rot, [qubits[norb + j], qubits[norb + j + 1]]\n", " )\n", " yield CircuitInstruction(rot, [qubits[j + 1], qubits[j + 2]])\n", " yield CircuitInstruction(\n", " rot, [qubits[norb + j + 1], qubits[norb + j + 2]]\n", " )\n", "\n", "\n", "def trotter_step(\n", " qubits: Sequence[Qubit],\n", " time_step: float,\n", " one_body_evolution: np.ndarray,\n", " h2e: np.ndarray,\n", " impurity_index: int,\n", " norb: int,\n", "):\n", " \"\"\"A Trotter step.\"\"\"\n", " # Assume the two-body interaction is just the on-site interaction of the impurity\n", " onsite = h2e[\n", " impurity_index, impurity_index, impurity_index, impurity_index\n", " ]\n", " # Two-body evolution for half the time\n", " yield CircuitInstruction(\n", " CPhaseGate(-0.5 * time_step * onsite),\n", " [qubits[impurity_index], qubits[norb + impurity_index]],\n", " )\n", " # One-body evolution for the full time\n", " yield CircuitInstruction(\n", " ffsim.qiskit.OrbitalRotationJW(norb, one_body_evolution), qubits\n", " )\n", " # Two-body evolution for half the time\n", " yield CircuitInstruction(\n", " CPhaseGate(-0.5 * time_step * onsite),\n", " [qubits[impurity_index], qubits[norb + impurity_index]],\n", " )\n", "\n", "\n", "# Time step\n", "time_step = 0.2\n", "# Number of Krylov basis states\n", "krylov_dim = 8\n", "\n", "# Initialize circuit\n", "qubits = QuantumRegister(2 * norb, name=\"q\")\n", "circuit = QuantumCircuit(qubits)\n", "\n", "# Generate initial state\n", "for instruction in prepare_initial_state(qubits, norb=norb, nocc=norb // 2):\n", " circuit.append(instruction)\n", "circuit.measure_all()\n", "\n", "# Create list of circuits, starting with the initial state circuit\n", "circuits = [circuit.copy()]\n", "\n", "# Add time evolution circuits to the list\n", "one_body_evolution = scipy.linalg.expm(-1j * time_step * h1e_momentum)\n", "for i in range(krylov_dim - 1):\n", " # Remove measurements\n", " circuit.remove_final_measurements()\n", " # Append another Trotter step\n", " for instruction in trotter_step(\n", " qubits,\n", " time_step,\n", " one_body_evolution,\n", " h2e_momentum,\n", " impurity_index,\n", " norb,\n", " ):\n", " circuit.append(instruction)\n", " # Measure qubits\n", " circuit.measure_all()\n", " # Add a copy of the circuit to the list\n", " circuits.append(circuit.copy())" ] }, { "cell_type": "code", "execution_count": 3, "id": "9f2cc4d4-ecac-457a-bcae-558319668e1f", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "\"Output" ] }, "execution_count": 3, "metadata": {}, "output_type": "execute_result" } ], "source": [ "circuits[0].draw(\"mpl\", scale=0.4, fold=-1)" ] }, { "cell_type": "code", "execution_count": 4, "id": "827976ec-4815-4707-80b1-e13fb2fef309", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "\"Output" ] }, "execution_count": 4, "metadata": {}, "output_type": "execute_result" } ], "source": [ "circuits[-1].draw(\"mpl\", scale=0.4, fold=-1)" ] }, { "cell_type": "markdown", "id": "b8ca6be4-61d9-47be-8099-8712c7ecc774", "metadata": {}, "source": [ "\n", "\n", "### Étape 2 : Optimiser le problème pour l'exécution quantique\n", "\n" ] }, { "cell_type": "markdown", "id": "a3304e1b-9c7c-4212-8744-d1c62292eced", "metadata": {}, "source": [ "Nous optimisons ensuite le circuit pour le matériel cible. Pour l'instant, nous allons créer un backend générique doté d'un nombre donné de qubits et d'un ensemble de portes dans lequel les circuits d'évolution temporelle se décomposent naturellement.\n", "\n" ] }, { "cell_type": "code", "execution_count": 5, "id": "2d2fdbff-1e22-45af-a2eb-c334e4328c59", "metadata": {}, "outputs": [], "source": [ "from qiskit.providers.fake_provider import GenericBackendV2\n", "\n", "backend = GenericBackendV2(\n", " 2 * norb, basis_gates=[\"cp\", \"xx_plus_yy\", \"p\", \"x\"]\n", ")" ] }, { "cell_type": "markdown", "id": "ce3e6a52-7b99-49b6-8294-b005efa59cfc", "metadata": {}, "source": [ "Maintenant, nous utilisons Qiskit pour transpiler les circuits vers le backend cible.\n", "\n" ] }, { "cell_type": "code", "execution_count": 6, "id": "c8643533-9fec-40bf-a307-da8839b1e444", "metadata": {}, "outputs": [], "source": [ "from qiskit.transpiler import generate_preset_pass_manager\n", "\n", "pass_manager = generate_preset_pass_manager(\n", " optimization_level=3, backend=backend\n", ")\n", "isa_circuits = pass_manager.run(circuits)" ] }, { "cell_type": "markdown", "id": "6cfd3eea-e2d9-40a5-a449-1d3d790a5f2d", "metadata": {}, "source": [ "\n", "\n", "### Étape 3 : Exécuter à l'aide de l' Qiskit primitives\n", "\n" ] }, { "cell_type": "markdown", "id": "ad48e3e6-1013-45a1-942c-63766fed5819", "metadata": {}, "source": [ "Après avoir optimisé les circuits pour l'exécution matérielle, nous sommes prêts à les exécuter sur le matériel cible et à collecter des échantillons pour l'estimation de l'énergie de l'état fondamental. Après avoir utilisé la primitive Sampler pour échantillonner les chaînes de bits de chaque circuit, nous combinons tous les résultats dans un seul dictionnaire de comptage et nous indiquons les 20 chaînes de bits les plus fréquemment échantillonnées.\n", "\n" ] }, { "cell_type": "code", "execution_count": null, "id": "80eee553-60d6-4258-88ab-d8d120418c36", "metadata": {}, "outputs": [], "source": [ "from qiskit.visualization import plot_histogram\n", "from qiskit.primitives import StatevectorSampler\n", "\n", "# Sample from the circuits\n", "sampler = StatevectorSampler()\n", "job = sampler.run(isa_circuits, shots=500)" ] }, { "cell_type": "code", "execution_count": 8, "id": "10af4663-7375-4b50-bae6-9f3d5106457b", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "\"Output" ] }, "execution_count": 8, "metadata": {}, "output_type": "execute_result" } ], "source": [ "from qiskit.primitives import BitArray\n", "\n", "# Combine the shots from the individual Trotter circuits\n", "bit_array = BitArray.concatenate_shots(\n", " [result.data.meas for result in job.result()]\n", ")\n", "\n", "plot_histogram(bit_array.get_counts(), number_to_keep=20)" ] }, { "cell_type": "markdown", "id": "2aa74455-d16b-4ac3-a354-54a79d5c5759", "metadata": {}, "source": [ "\n", "\n", "### Étape 4 : Post-traitement et restitution du résultat au format classique souhaité\n", "\n" ] }, { "cell_type": "markdown", "id": "d3f7713c-a4e6-407f-94de-0e5cfeb1134c", "metadata": {}, "source": [ "Nous exécutons maintenant l'algorithme SQD à l'aide de la fonction `diagonalize_fermionic_hamiltonian` . Voir la [documentation de l'API](https://qiskit.github.io/qiskit-addon-sqd/apidocs/qiskit_addon_sqd.fermion.html#qiskit_addon_sqd.fermion.diagonalize_fermionic_hamiltonian) pour des explications sur les arguments de cette fonction.\n", "\n" ] }, { "cell_type": "code", "execution_count": 9, "id": "7609d1e1-e8ef-48e1-a965-97927f403163", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Iteration 1\n", "\tSubsample 0\n", "\t\tEnergy: -13.4222953188441\n", "\t\tSubspace dimension: 529\n", "\tSubsample 1\n", "\t\tEnergy: -13.42237556285828\n", "\t\tSubspace dimension: 784\n", "\tSubsample 2\n", "\t\tEnergy: -13.422045397387413\n", "\t\tSubspace dimension: 529\n", "Iteration 2\n", "\tSubsample 0\n", "\t\tEnergy: -13.422379583305478\n", "\t\tSubspace dimension: 900\n", "\tSubsample 1\n", "\t\tEnergy: -13.422376197704326\n", "\t\tSubspace dimension: 841\n", "\tSubsample 2\n", "\t\tEnergy: -13.422421162849295\n", "\t\tSubspace dimension: 1089\n", "Iteration 3\n", "\tSubsample 0\n", "\t\tEnergy: -13.422421164670345\n", "\t\tSubspace dimension: 1156\n", "\tSubsample 1\n", "\t\tEnergy: -13.422421492737689\n", "\t\tSubspace dimension: 1156\n", "\tSubsample 2\n", "\t\tEnergy: -13.422421205869572\n", "\t\tSubspace dimension: 1156\n", "Iteration 4\n", "\tSubsample 0\n", "\t\tEnergy: -13.422421494558726\n", "\t\tSubspace dimension: 1225\n", "\tSubsample 1\n", "\t\tEnergy: -13.422421492737689\n", "\t\tSubspace dimension: 1156\n", "\tSubsample 2\n", "\t\tEnergy: -13.422421492737689\n", "\t\tSubspace dimension: 1156\n" ] } ], "source": [ "from qiskit_addon_sqd.fermion import (\n", " SCIResult,\n", " diagonalize_fermionic_hamiltonian,\n", ")\n", "\n", "# List to capture intermediate results\n", "result_history = []\n", "\n", "\n", "def callback(results: list[SCIResult]):\n", " result_history.append(results)\n", " iteration = len(result_history)\n", " print(f\"Iteration {iteration}\")\n", " for i, result in enumerate(results):\n", " print(f\"\\tSubsample {i}\")\n", " print(f\"\\t\\tEnergy: {result.energy}\")\n", " print(\n", " f\"\\t\\tSubspace dimension: {np.prod(result.sci_state.amplitudes.shape)}\"\n", " )\n", "\n", "\n", "rng = np.random.default_rng(24)\n", "result = diagonalize_fermionic_hamiltonian(\n", " h1e_momentum,\n", " h2e_momentum,\n", " bit_array,\n", " samples_per_batch=100,\n", " norb=norb,\n", " nelec=nelec,\n", " num_batches=3,\n", " max_iterations=5,\n", " symmetrize_spin=True,\n", " callback=callback,\n", " seed=rng,\n", ")" ] }, { "cell_type": "markdown", "id": "3dee9c61-fc42-48e9-8888-af8fc831cd5c", "metadata": {}, "source": [ "La cellule de code suivante affiche les résultats. Le premier graphique représente l'énergie calculée en fonction du nombre d'itérations de récupération de la configuration, tandis que le second graphique montre l'occupation moyenne de chaque orbite spatiale après l'itération finale. Comme il s'agit d'un problème très simple, la première itération nous rapproche déjà beaucoup de l'énergie exacte (notez l'échelle de l'axe des y).\n", "\n" ] }, { "cell_type": "code", "execution_count": 10, "id": "b6879566-8bf5-4c28-bfb6-b2686692e3d3", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Reference energy: -13.42249\n", "SQD energy: -13.42242\n", "Absolute error: 0.00007\n" ] }, { "data": { "text/plain": [ "\"Output" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "import matplotlib.pyplot as plt\n", "\n", "min_es = [\n", " min(result, key=lambda res: res.energy).energy\n", " for result in result_history\n", "]\n", "min_id, min_e = min(enumerate(min_es), key=lambda x: x[1])\n", "\n", "# Data for energies plot\n", "x1 = range(len(result_history))\n", "\n", "# Data for avg spatial orbital occupancy\n", "y2 = np.sum(result.orbital_occupancies, axis=0)\n", "x2 = range(len(y2))\n", "\n", "fig, axs = plt.subplots(1, 2, figsize=(12, 6))\n", "\n", "# Plot energies\n", "axs[0].plot(x1, min_es, label=\"energy\", marker=\"o\")\n", "axs[0].set_xticks(x1)\n", "axs[0].set_xticklabels(x1)\n", "axs[0].axhline(\n", " y=reference_energy,\n", " color=\"#BF5700\",\n", " linestyle=\"--\",\n", " label=\"reference energy\",\n", ")\n", "axs[0].set_title(\"Approximated Ground State Energy vs SQD Iterations\")\n", "axs[0].set_xlabel(\"Iteration Index\", fontdict={\"fontsize\": 12})\n", "axs[0].set_ylabel(\"Energy\", fontdict={\"fontsize\": 12})\n", "axs[0].legend()\n", "\n", "# Plot orbital occupancy\n", "axs[1].bar(x2, y2, width=0.8)\n", "axs[1].set_xticks(x2)\n", "axs[1].set_xticklabels(x2)\n", "axs[1].set_title(\"Avg Occupancy per Spatial Orbital\")\n", "axs[1].set_xlabel(\"Orbital Index\", fontdict={\"fontsize\": 12})\n", "axs[1].set_ylabel(\"Avg Occupancy\", fontdict={\"fontsize\": 12})\n", "\n", "print(f\"Reference energy: {reference_energy:.5f}\")\n", "print(f\"SQD energy: {min_e:.5f}\")\n", "print(f\"Absolute error: {abs(min_e - reference_energy):.5f}\")\n", "plt.tight_layout()\n", "plt.show()" ] }, { "cell_type": "markdown", "id": "0c9f4976-d770-426a-822e-e6756c2cfbe7", "metadata": {}, "source": [ "\n", "\n", "### Vérifier la consommation d'énergie\n", "\n", "L'énergie restituée par le SQD est garantie de constituer une borne supérieure de l'énergie réelle de l'état fondamental. La valeur de l'énergie peut être vérifiée, car SQD renvoie également les coefficients du vecteur d'état qui approxime l'état fondamental. Vous pouvez calculer l'énergie à partir du vecteur d'état en utilisant ses matrices de densité réduites à une et deux particules, comme le montre la cellule de code suivante.\n", "\n" ] }, { "cell_type": "code", "execution_count": 11, "id": "e2b9de72-61cf-49d3-a1b5-f043e4b16956", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Recomputed energy: -13.42242\n" ] } ], "source": [ "rdm1 = result.sci_state.rdm(rank=1, spin_summed=True)\n", "rdm2 = result.sci_state.rdm(rank=2, spin_summed=True)\n", "\n", "energy = np.sum(h1e_momentum * rdm1) + 0.5 * np.sum(h2e_momentum * rdm2)\n", "\n", "print(f\"Recomputed energy: {energy:.5f}\")" ] }, { "cell_type": "markdown", "id": "5221928a-79ff-4f54-90b4-1fe4d7739aae", "metadata": {}, "source": [ "\n", "\n", "## Exemple de matériel à grande échelle\n", "\n", "Nous allons maintenant exécuter un exemple plus complexe sur un véritable QPU.\n", "Pour l'énergie de référence, nous utilisons les résultats d'un calcul [DMRG](https://en.wikipedia.org/wiki/Density_matrix_renormalization_group) qui a été effectué séparément.\n", "\n" ] }, { "cell_type": "code", "execution_count": null, "id": "933037d8-847e-4986-80da-5ac8d677b2ff", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Using backend ibm_boston\n", "Iteration 1\n", "\tSubsample 0\n", "\t\tEnergy: -28.63965951544449\n", "\t\tSubspace dimension: 9801\n", "\tSubsample 1\n", "\t\tEnergy: -28.625588929202006\n", "\t\tSubspace dimension: 9409\n", "\tSubsample 2\n", "\t\tEnergy: -28.647371834135498\n", "\t\tSubspace dimension: 8281\n", "Iteration 2\n", "\tSubsample 0\n", "\t\tEnergy: -28.67213260849567\n", "\t\tSubspace dimension: 29584\n", "\tSubsample 1\n", "\t\tEnergy: -28.670340686158816\n", "\t\tSubspace dimension: 27225\n", "\tSubsample 2\n", "\t\tEnergy: -28.669976379525988\n", "\t\tSubspace dimension: 31329\n", "Iteration 3\n", "\tSubsample 0\n", "\t\tEnergy: -28.68622875601382\n", "\t\tSubspace dimension: 36100\n", "\tSubsample 1\n", "\t\tEnergy: -28.698569623143126\n", "\t\tSubspace dimension: 34225\n", "\tSubsample 2\n", "\t\tEnergy: -28.694848533971882\n", "\t\tSubspace dimension: 33856\n", "Iteration 4\n", "\tSubsample 0\n", "\t\tEnergy: -28.69883392844593\n", "\t\tSubspace dimension: 42025\n", "\tSubsample 1\n", "\t\tEnergy: -28.701289495200996\n", "\t\tSubspace dimension: 38025\n", "\tSubsample 2\n", "\t\tEnergy: -28.699319594978245\n", "\t\tSubspace dimension: 45369\n", "Iteration 5\n", "\tSubsample 0\n", "\t\tEnergy: -28.701936886834154\n", "\t\tSubspace dimension: 51076\n", "\tSubsample 1\n", "\t\tEnergy: -28.702468711812013\n", "\t\tSubspace dimension: 53824\n", "\tSubsample 2\n", "\t\tEnergy: -28.702298147575938\n", "\t\tSubspace dimension: 52900\n", "Reference energy: -28.70660\n", "SQD energy: -28.70247\n", "Absolute error: 0.00413\n" ] }, { "data": { "text/plain": [ "\"Output" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "from qiskit_ibm_runtime import SamplerV2 as Sampler\n", "from qiskit_ibm_runtime import QiskitRuntimeService\n", "\n", "# Model parameters\n", "norb = 20\n", "nelec = (norb // 2, norb // 2)\n", "n_bath = norb - 1\n", "hybridization = 1.0\n", "hopping = 1.0\n", "onsite = 10.0\n", "chemical_potential = -0.5 * onsite\n", "\n", "# Generate Hamiltonian and orbital rotation\n", "h1e, h2e = siam_hamiltonian(\n", " norb=norb,\n", " hopping=hopping,\n", " onsite=onsite,\n", " hybridization=hybridization,\n", " chemical_potential=chemical_potential,\n", ")\n", "orbital_rotation = momentum_basis(norb)\n", "h1e_momentum, h2e_momentum = rotated(h1e, h2e, orbital_rotation.T.conj())\n", "impurity_index = n_bath // 2\n", "\n", "# Set reference energy to DMRG value computed separately\n", "reference_energy = -28.70659686\n", "\n", "# Algorithm parameters\n", "time_step = 0.2\n", "krylov_dim = 8\n", "\n", "# Construct circuits\n", "qubits = QuantumRegister(2 * norb, name=\"q\")\n", "circuit = QuantumCircuit(qubits)\n", "for instruction in prepare_initial_state(qubits, norb=norb, nocc=norb // 2):\n", " circuit.append(instruction)\n", "circuit.measure_all()\n", "circuits = [circuit.copy()]\n", "one_body_evolution = scipy.linalg.expm(-1j * time_step * h1e_momentum)\n", "for i in range(krylov_dim - 1):\n", " circuit.remove_final_measurements()\n", " for instruction in trotter_step(\n", " qubits,\n", " time_step,\n", " one_body_evolution,\n", " h2e_momentum,\n", " impurity_index,\n", " norb,\n", " ):\n", " circuit.append(instruction)\n", " circuit.measure_all()\n", " circuits.append(circuit.copy())\n", "\n", "# Initialize hardware backend\n", "service = QiskitRuntimeService()\n", "backend = service.least_busy(\n", " operational=True, simulator=False, min_num_qubits=127\n", ")\n", "print(f\"Using backend {backend.name}\")\n", "\n", "# Transpile to backend\n", "pass_manager = generate_preset_pass_manager(\n", " optimization_level=3, backend=backend\n", ")\n", "isa_circuits = pass_manager.run(circuits)\n", "\n", "# Sample from the circuits\n", "sampler = Sampler(backend)\n", "sampler.options.environment.job_tags = [\"TUT_SKQD\"]\n", "job = sampler.run(isa_circuits, shots=500)\n", "\n", "# Combine the shots from the individual Trotter circuits\n", "bit_array = BitArray.concatenate_shots(\n", " [result.data.meas for result in job.result()]\n", ")\n", "\n", "# Run configuration recovery and diagonalization\n", "result_history = []\n", "\n", "\n", "def callback(results: list[SCIResult]):\n", " result_history.append(results)\n", " iteration = len(result_history)\n", " print(f\"Iteration {iteration}\")\n", " for i, result in enumerate(results):\n", " print(f\"\\tSubsample {i}\")\n", " print(f\"\\t\\tEnergy: {result.energy}\")\n", " print(\n", " f\"\\t\\tSubspace dimension: {np.prod(result.sci_state.amplitudes.shape)}\"\n", " )\n", "\n", "\n", "rng = np.random.default_rng(24)\n", "result = diagonalize_fermionic_hamiltonian(\n", " h1e_momentum,\n", " h2e_momentum,\n", " bit_array,\n", " samples_per_batch=100,\n", " norb=norb,\n", " nelec=nelec,\n", " num_batches=3,\n", " max_iterations=5,\n", " symmetrize_spin=True,\n", " callback=callback,\n", " seed=rng,\n", ")\n", "\n", "\n", "# Plot results\n", "min_es = [\n", " min(result, key=lambda res: res.energy).energy\n", " for result in result_history\n", "]\n", "min_id, min_e = min(enumerate(min_es), key=lambda x: x[1])\n", "x1 = range(len(result_history))\n", "y2 = np.sum(result.orbital_occupancies, axis=0)\n", "x2 = range(len(y2))\n", "fig, axs = plt.subplots(1, 2, figsize=(12, 6))\n", "axs[0].plot(x1, min_es, label=\"energy\", marker=\"o\")\n", "axs[0].set_xticks(x1)\n", "axs[0].set_xticklabels(x1)\n", "axs[0].axhline(\n", " y=reference_energy,\n", " color=\"#BF5700\",\n", " linestyle=\"--\",\n", " label=\"reference energy\",\n", ")\n", "axs[0].set_title(\"Approximated Ground State Energy vs SQD Iterations\")\n", "axs[0].set_xlabel(\"Iteration Index\", fontdict={\"fontsize\": 12})\n", "axs[0].set_ylabel(\"Energy\", fontdict={\"fontsize\": 12})\n", "axs[0].legend()\n", "axs[1].bar(x2, y2, width=0.8)\n", "axs[1].set_xticks(x2)\n", "axs[1].set_xticklabels(x2)\n", "axs[1].set_title(\"Avg Occupancy per Spatial Orbital\")\n", "axs[1].set_xlabel(\"Orbital Index\", fontdict={\"fontsize\": 12})\n", "axs[1].set_ylabel(\"Avg Occupancy\", fontdict={\"fontsize\": 12})\n", "print(f\"Reference energy: {reference_energy:.5f}\")\n", "print(f\"SQD energy: {min_e:.5f}\")\n", "print(f\"Absolute error: {abs(min_e - reference_energy):.5f}\")\n", "plt.tight_layout()\n", "plt.show()" ] }, { "cell_type": "markdown", "id": "482ebea3-84b8-471b-bddc-b282e23162ad", "metadata": {}, "source": [ "\n", "\n", "## Etapes suivantes\n", "\n", "\n", " Si ce travail vous a intéressé, les documents suivants pourraient vous intéresser :\n", "\n", " * [Diagonalisation quantique par échantillonnage d'un hamiltonien chimique](/docs/tutorials/sample-based-quantum-diagonalization) – tutoriel associé utilisant une approche variationnelle heuristique à la place des circuits de Trotter\n", " * [Diagonalisation quantique de Krylov des hamiltoniens sur réseau](/docs/tutorials/krylov-quantum-diagonalization) : un tutoriel sur la méthode KQD\n", " * [Documentation de l'API de l'extension SQD](https://qiskit.github.io/qiskit-addon-sqd/apidocs/qiskit_addon_sqd.fermion.html#qiskit_addon_sqd.fermion.diagonalize_fermionic_hamiltonian) - Référence pour la `diagonalize_fermionic_hamiltonian` fonction\n", " * « [*Quantum-Centric Algorithm for Sample-Based Krylov Diagonalization*](https://arxiv.org/abs/2501.09702) » – l'article sur lequel s'appuie ce tutoriel\n", "\n", "\n" ] }, { "cell_type": "markdown", "metadata": {}, "id": "a1b8767d", "source": "© IBM Corp., 2017-2026" } ], "metadata": { "kernelspec": { "display_name": "Python 3", "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" }, "hours": 1.5, "qpuSeconds": 9 }, "nbformat": 4, "nbformat_minor": 4 }