Async subroutines.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "fa8b70aa",
   "metadata": {},
   "source": [
    "# To sync or not to sync\n",
    "\n",
    "`Estimator` and `Sampler` primitives currently only allow asynchronous calls:\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b6d26f95",
   "metadata": {},
   "outputs": [],
   "source": [
    "est = Estimator()\n",
    "job = est.run([circuit], [operator], [values])\n",
    "result = job.result()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c88fc5eb",
   "metadata": {},
   "source": [
    "In context of the algorithms, we discussed whether we want to follow the same interface for advanced subroutines like gradients or fidelities, as"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "2d32cec8",
   "metadata": {},
   "outputs": [],
   "source": [
    "grad = ParameterShiftEstimatorGradient()\n",
    "job = grad.run([circuit], [operator], [values])\n",
    "result = job.result()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "40dfd676",
   "metadata": {},
   "source": [
    " Alternatively we would only allow synchronous calls as"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "e9495b9a",
   "metadata": {},
   "outputs": [],
   "source": [
    "grad = ParameterShiftEstimatorGradient()\n",
    "result = grad.evaluate([circuit], [operator], [values])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4fef8a91",
   "metadata": {},
   "source": [
    "### Pro sync\n",
    "\n",
    "* easier syntax\n",
    "* \"where do we draw the line?\" -- if subroutines have `.run` do algorithms also need it?"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0a0a12e3",
   "metadata": {},
   "source": [
    "### Pro async\n",
    "\n",
    "* consistency with other estimator and sampler\n",
    "* subroutines and primitives are used in the same piece of code -- different signatures look weird"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9c6cfc4e",
   "metadata": {},
   "source": [
    "### Examples"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "385ec5be",
   "metadata": {},
   "source": [
    "#### VQE \"sync\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "963dab2e",
   "metadata": {},
   "outputs": [],
   "source": [
    "class VQE(MinimumEigensolver):\n",
    "    ...\n",
    "    \n",
    "    def compute_minimum_eigenvalue(self, operator):\n",
    "        \n",
    "        def energy(theta):\n",
    "            job = self.estimator.run([self.ansatz], [operator], [theta])\n",
    "            return job.result()\n",
    "            \n",
    "        def gradient(theta):\n",
    "            return self.gradient.evaluate([self.ansatz], [operator], [theta])\n",
    "        \n",
    "        result = self.optimizer.minimize(energy, initial_guess, jac=gradient)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "25f430b5",
   "metadata": {},
   "source": [
    "### VQE \"all async\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c10ca846",
   "metadata": {},
   "outputs": [],
   "source": [
    "class VQE(MinimumEigensolver):\n",
    "    ...\n",
    "    \n",
    "    def compute_minimum_eigenvalue(self, operator):\n",
    "        \n",
    "        def energy(theta):\n",
    "            job = self.estimator.run([self.ansatz], [operator], [theta])\n",
    "            return job.result()\n",
    "            \n",
    "        def gradient(theta):\n",
    "            job = self.gradient.run([self.ansatz], [operator], [theta])\n",
    "            return job.result()\n",
    "        \n",
    "        result = self.optimizer.minimize(energy, initial_guess, jac=gradient)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "80102684",
   "metadata": {},
   "source": [
    "\n",
    "\n",
    "#### VQD \"sync\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "0d2b6b00",
   "metadata": {},
   "outputs": [],
   "source": [
    "class VQD(Eigensolver):\n",
    "    ...\n",
    "    \n",
    "    def energy_evaluation(operator, theta, penalty):\n",
    "        job = self.estimator.run([self.ansatz], [operator], [theta])\n",
    "        energy = job.result()\n",
    "        \n",
    "        overlap = self.fidelity.evaluate([self.ansatz], [self.ansatz], [self.optimal_point], [theta])\n",
    "        \n",
    "        return energy + penalty * overlap"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4d592842",
   "metadata": {},
   "source": [
    "#### VQD \"all async\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d2de0598",
   "metadata": {},
   "outputs": [],
   "source": [
    "class VQD(Eigensolver):\n",
    "    ...\n",
    "    \n",
    "    def energy_evaluation(operator, theta, penalty):\n",
    "        jobs = [\n",
    "            self.estimator.run([self.ansatz], [operator], [theta]),\n",
    "            self.fidelity.run([self.ansatz], [self.ansatz], [self.optimal_point], [theta])\n",
    "        ]\n",
    "        energy, overlap = [job.result() for job in jobs]\n",
    "                \n",
    "        return energy + penalty * overlap"
   ]
  }
 ],
 "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": "3.10.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}