Commit 73c70b4b authored by Dmitry I. Lyakh's avatar Dmitry I. Lyakh
Browse files

Bug fix in exatn-gen visitor: It was calling getExpectationValueZ prematurily. 



Signed-off-by: default avatarDmitry I. Lyakh <quant4me@gmail.com>
parent 0554a791

tnqvm/visitors/exatn-gen/ExatnGenVisitor.cpp

+27 −46
Original line number Diff line number Diff line
@@ -29,7 +29,6 @@ 
 *

 **********************************************************************************/

#ifdef TNQVM_HAS_EXATN

#define _DEBUG_DIL



#include "ExatnGenVisitor.hpp"

#include "exatn.hpp"
@@ -268,8 +267,8 @@ void ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::initialize( 
    }

  }



  m_reconstructTol = 1e-3;

  m_maxBondDim = 512;

  m_reconstructTol = 1e-4;

  m_maxBondDim = 16;

  m_reconstructionFidelity = 1.0;

  m_initReconstructionRandom = false;

  if (options.keyExists<bool>("init-random")) {
@@ -300,9 +299,8 @@ void ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::initialize( 
        i + 1, exatn::getTensor(generateQubitTensorName(i)),

        std::vector<std::pair<unsigned int, unsigned int>>{});

  }

  exatn::TensorExpansion tensorEx("QuantumCircuit");

  tensorEx.appendComponent(m_qubitNetwork, TNQVM_COMPLEX_TYPE{1.0});

  m_tensorExpansion = tensorEx;

  m_tensorExpansion = exatn::TensorExpansion("QuantumCircuit");

  m_tensorExpansion.appendComponent(m_qubitNetwork, TNQVM_COMPLEX_TYPE{1.0});

  m_gateTensorBodies.clear();

  m_measuredBits.clear();

  m_obsTensorOperator.reset();
@@ -386,11 +384,8 @@ void ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::initialize( 


template <typename TNQVM_COMPLEX_TYPE>

void ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::finalize() {

  if (m_layerCounter > 0) {

    reconstructCircuitTensor(true);

  }

  if (m_layerCounter > 0) reconstructCircuitTensor(true);

  m_buffer->addExtraInfo("reconstruction-fidelity", m_reconstructionFidelity);

  assert(m_reconstructionFidelity > 0.8);

  if (m_shots > 0 && !m_measuredBits.empty()) {

    for (int i = 0; i < m_shots; ++i) {

      const auto convertToBitString =
@@ -624,6 +619,7 @@ void ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::visit(Measure &in_MeasureGate) { 
  m_measuredBits.emplace_back(in_MeasureGate.bits()[0]);

}

// === END: Gate Visitor Impls ===



template <typename TNQVM_COMPLEX_TYPE>

template <tnqvm::CommonGates GateType, typename... GateParams>

void ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::appendGateTensor(
@@ -654,8 +650,8 @@ void ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::appendGateTensor( 
                                      ? exatn::TensorShape{2, 2}

                                      : exatn::TensorShape{2, 2, 2, 2});

    // Create the tensor

    const bool created = exatn::createTensor(

        uniqueGateName, getExatnElementType(), gateTensorShape);

    const bool created = exatn::createTensor(uniqueGateName,

                                             getExatnElementType(), gateTensorShape);

    assert(created);

    // Init tensor body data

    exatn::initTensorData(uniqueGateName, flattenGateMatrix(gateMatrix));
@@ -663,12 +659,10 @@ void ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::appendGateTensor( 
    // For rank-2 gate isometric leg groups are: {0}, {1}.

    // For rank-4 gate isometric leg groups are: {0,1}, {2,3}.

    if (in_gateInstruction.nRequiredBits() == 1) {

      const bool registered =

          exatn::registerTensorIsometry(uniqueGateName, {0}, {1});

      const bool registered = exatn::registerTensorIsometry(uniqueGateName, {0}, {1});

      assert(registered);

    } else if (in_gateInstruction.nRequiredBits() == 2) {

      const bool registered =

          exatn::registerTensorIsometry(uniqueGateName, {0, 1}, {2, 3});

      const bool registered = exatn::registerTensorIsometry(uniqueGateName, {0, 1}, {2, 3});

      assert(registered);

    }

  }
@@ -676,8 +670,7 @@ void ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::appendGateTensor( 
  // Because the qubit location and gate pairing are of different integer types,

  // we need to reconstruct the qubit vector.

  std::vector<unsigned int> gatePairing;

  for (const auto &qbitLoc :

       const_cast<xacc::Instruction &>(in_gateInstruction).bits()) {

  for (const auto &qbitLoc : const_cast<xacc::Instruction &>(in_gateInstruction).bits()) {

    gatePairing.emplace_back(qbitLoc);

  }
@@ -712,9 +705,7 @@ void ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::appendGateTensor( 


template <typename TNQVM_COMPLEX_TYPE>

void ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::reconstructCircuitTensor(bool forced) {

  if (m_layersReconstruct <= 0) {

    return;

  }

  if (m_layersReconstruct <= 0) return;

  if (m_layerCounter > m_layersReconstruct || forced) {

    xacc::info("Reconstruct Tensor Expansion");

    // Flush the count every reconstruct:
@@ -729,8 +720,7 @@ void ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::reconstructCircuitTensor(bool forced) 
    const std::vector<int> qubitTensorDim(m_buffer->size(), 2);

    auto rootTensor = std::make_shared<exatn::Tensor>("ROOT", qubitTensorDim);

    auto &networkBuildFactory = *(exatn::numerics::NetworkBuildFactory::get());

    auto builder =

        networkBuildFactory.createNetworkBuilderShared(m_reconstructBuilder);

    auto builder = networkBuildFactory.createNetworkBuilderShared(m_reconstructBuilder);

    builder->setParameter("max_bond_dim", m_maxBondDim);

    auto approximant = [&]() {

      if (m_initReconstructionRandom || !m_previousOptExpansion) {
@@ -752,8 +742,7 @@ void ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::reconstructCircuitTensor(bool forced) 
          }

        }

        approximantTensorNetwork->markOptimizableAllTensors();

        auto approximant_expansion =

            std::make_shared<exatn::TensorExpansion>("Approx");

        auto approximant_expansion = std::make_shared<exatn::TensorExpansion>("Approx");

        approximant_expansion->appendComponent(approximantTensorNetwork,

                                               TNQVM_COMPLEX_TYPE{1.0, 0.0});

        approximant_expansion->conjugate();
@@ -777,20 +766,16 @@ void ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::reconstructCircuitTensor(bool forced) 
    // Run the reconstructor:

    bool reconstructSuccess = exatn::sync();

    assert(reconstructSuccess);

    //exatn::TensorNetworkReconstructor::resetDebugLevel(1,0); //debug

    exatn::TensorNetworkReconstructor::resetDebugLevel(1,0); //debug

    reconstructor.resetLearningRate(1.0);

    double residual_norm, fidelity;

    const auto startOpt = std::chrono::system_clock::now();

    bool reconstructed =

        reconstructor.reconstruct(&residual_norm, &fidelity, true);

    reconstructSuccess = exatn::sync();

    assert(reconstructSuccess);

    bool reconstructed = reconstructor.reconstruct(&residual_norm, &fidelity, true);

    reconstructSuccess = exatn::sync(); assert(reconstructSuccess);

    if (reconstructed) {

      const auto endOpt = std::chrono::system_clock::now();

      const int elapsedMs =

          std::chrono::duration_cast<std::chrono::milliseconds>(endOpt -

                                                                startOpt)

              .count();

        std::chrono::duration_cast<std::chrono::milliseconds>(endOpt - startOpt).count();

      std::stringstream ss;

      ss << "Reconstruction succeeded: Residual norm = " << residual_norm

         << "; Fidelity = " << fidelity << "; elapsed time = " << elapsedMs
@@ -917,6 +902,7 @@ ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::constructObsTensorOperator( 
template <typename TNQVM_COMPLEX_TYPE>

const double ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::getExpectationValueZ(

             std::shared_ptr<CompositeInstruction> in_function) {

  if (m_layerCounter > 0) reconstructCircuitTensor(true);

  if (!m_evaluatedExpansion) {

    exatn::TensorExpansion ketvector(m_tensorExpansion);

    // std::cout << "Before renormalize:\n";
@@ -943,11 +929,9 @@ const double ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::getExpectationValueZ( 
    assert(accumInitialized);

    m_gateTensorBodies["ExpVal"] = std::vector<TNQVM_COMPLEX_TYPE>{{0.0, 0.0}};

    auto accumulator = exatn::getTensor("ExpVal");

    const bool evaluated =

        exatn::evaluateSync(bratimesopertimesket, accumulator);

    const bool evaluated = exatn::evaluateSync(bratimesopertimesket, accumulator);

    assert(evaluated);

    m_evaluatedExpansion =

        std::make_shared<exatn::TensorExpansion>(bratimesopertimesket);

    m_evaluatedExpansion = std::make_shared<exatn::TensorExpansion>(bratimesopertimesket);

  }

  // std::cout << "There are: " << m_evaluatedExpansion->getNumComponents() << "

  // components in the expansion.\n";
@@ -1091,9 +1075,7 @@ void ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::updateLayerCounter( 


    // Update layer counter = max qubit counter across all qubits.

    for (const auto &[qId, counter] : m_qubitToGateCount) {

      if (counter > m_layerCounter) {

        m_layerCounter = counter;

      }

      if (counter > m_layerCounter) m_layerCounter = counter;

    }

    {

      std::stringstream ss;
@@ -1101,7 +1083,6 @@ void ExatnGenVisitor<TNQVM_COMPLEX_TYPE>::updateLayerCounter( 
      xacc::info(ss.str());

    }





    // for (const auto& [bit1, bit2]: m_layerTracker) {

    //   if ((q1 == bit1 || q1 == bit2) || (q2 == bit1 || q2 == bit2)) {

    //     canCombine = false;

tnqvm/visitors/exatn-gen/ExatnGenVisitor.hpp

+1 −1
Original line number Diff line number Diff line
@@ -36,7 +36,7 @@ 
// +-----------------------------+------------------------------------------------------------------------+-------------+--------------------------+

// | reconstruct-tolerance       | Reconstruction convergence tolerance                                   |    double   | 1e-4                     |

// +-----------------------------+------------------------------------------------------------------------+-------------+--------------------------+

// | max-bond-dim                | Reconstruction max bond dimension (inside approximating tensor network)|    int      | 512                      |

// | max-bond-dim                | Reconstruction max bond dimension (inside approximating tensor network)|    int      | 16                       |

// +-----------------------------+------------------------------------------------------------------------+-------------+--------------------------+

// | reconstruct-builder         | Reconstruction network builder (builds the tensor network ansatz)      |    string   | "MPS"                    |

// +-----------------------------+------------------------------------------------------------------------+-------------+--------------------------+