Skip to main content
← back

VQE: Variational Quantum Eigensolver

Can today's quantum hardware calculate the energy of a real molecule? VQE is the algorithm we use to find out. It runs on every platform we test — IBM, Tuna-9, and emulator — connecting the Hamiltonian (the problem) to the ansatz (the circuit). This page explains how the algorithm works, then shows what happens when we run it on real hardware.

Like tuning a guitar by ear. You play a note (run a quantum circuit), listen to the pitch (measure energy), and turn the tuning peg (adjust parameters) until it sounds right. The quantum computer plays the notes; the classical computer turns the pegs.

The VQE Loop

VQE repeats four steps until the energy converges. Drag the slider to adjust θ and watch the energy update in real time. The loop highlights which step is active.

QPrepareMMeasureEComputeθOptimizeθ = -0.224E = -1.1373 Ha
Energy: -1.1373 HaFCI: -1.1373 HaError: 0.0 kcal/molOptimal: θ = -0.224

1. Prepare

Parameterized circuit (ansatz) creates a trial quantum state |ψ(θ)⟩

2. Measure

Run circuit on hardware, measure in Z, X, and Y bases

3. Compute

Classical computer calculates ⟨ψ|H|ψ⟩ from measurements

4. Optimize

Classical optimizer (COBYLA, SPSA) adjusts parameters θ

From Molecule to Energy

Before VQE can run, the molecule must be translated into a quantum circuit. This pipeline turns atomic coordinates into something a quantum computer can execute.

H₂moleculePySCFintegralsHHamiltonianJWJordan-Wigner∑gᵢPᵢPauli stringsU(θ)circuitEenergy
Assembly line. PySCF computes the raw electron integrals (the ingredients). Jordan-Wigner translates them into qubit language (Pauli strings). The ansatz is the recipe card — it tells the quantum computer how to combine those ingredients. VQE is the chef, tasting and adjusting until the dish is right.

Energy Landscape

The energy E(θ) for H at equilibrium (R = 0.735 Å). Click or drag on the chart to explore. The green band shows chemical accuracy (±1 kcal/mol around FCI). VQE's job is to find the minimum.

HFFCI-1.1373 Ha-0.75π-0.50π-0.25π0θ (rad)-1.15-1.10-1.05-1.00-0.95-0.90-0.85-0.80-0.75-0.70-0.65-0.60-0.55-0.50-0.45-0.40-0.35-0.30-0.25-0.20-0.15-0.10-0.050.000.050.100.150.200.250.300.350.400.45Energy (Ha)
E(θ) exact
FCI = -1.1373 Ha
HF = -1.1170 Ha
Chemical accuracy
Finding the lowest valley. The energy landscape is a curve with one clear minimum. VQE walks downhill from the starting guess (θ = 0, which is Hartree-Fock) toward the true ground state. For H with 1 parameter, this is trivial. For larger molecules with hundreds of parameters, the landscape becomes a high-dimensional maze.

Potential Energy Surface

We ran VQE at 14 bond distances from 0.3 to 3.0 Å (65,536 shots each) and compared against exact (FCI) and classical (Hartree-Fock) references. The green dots are real emulator measurements — not theory. Hover to see values.

TREX0.51.01.52.02.53.0Bond Distance (Å)-1.2-1.1-1.0-0.9-0.8-0.7-0.6-0.5Energy (Ha)
FCI (exact)
Hartree-Fock
VQE Emulator
IBM TREX (best hardware)
Where classical fails. At equilibrium (R = 0.735 Å), Hartree-Fock and FCI nearly agree — the correlation energy is only 12.7 kcal/mol. But stretch the bond to 3.0 Å and the gap explodes to 174 kcal/mol. This is where quantum advantage lives: strongly correlated electrons that classical mean-field theory cannot describe. The emulator VQE (green) tracks FCI across the entire curve.

Hardware Reality Check

Theory is clean; hardware is noisy. These are real measurements from our H VQE experiment at equilibrium, ranked from best to worst. The dashed line marks chemical accuracy (1 kcal/mol) — the threshold where quantum chemistry becomes useful for predicting reactions.

IBM TREX0.22 kcal/molIBM TorinoEmulator avg0.28 kcal/molQI EmulatorTuna-9 REM+PS0.92 kcal/molQI Tuna-9IBM Raw9.22 kcal/molIBM TorinoTorino Raw (worst)25.60 kcal/molIBM Torino051015202530kcal/molchem. accuracy

Best Result

0.22 kcal/mol

TREX readout mitigation on IBM Torino. Within chemical accuracy.

Key Insight

80% readout noise

Most error is readout, not gates. Mitigation recovers nearly all lost accuracy.

Emulator Baseline

0.28 kcal/mol avg

Shot noise floor with 65k shots. No hardware errors, just statistical sampling.

Paper Trail

VQE isn't just theory — we replicated three landmark VQE papers on real hardware and compared our results to the originals. Each replication tests specific claims from the paper and reports pass/fail against our measurements.

Reproducibility is the point. These papers defined VQE. We re-ran their experiments on modern hardware to see what holds up and what doesn't. The original Peruzzo 2014 paper used a photonic chip; we ran it on superconducting transmons. Kandala's 2017 circuit is exactly the ansatz in our Ansatz Explorer. Each replication page shows claim-by-claim results.

Key Terms

See the full glossary for more definitions.

Hartree (Ha)

The atomic unit of energy. 1 Ha = 627.509 kcal/mol. Molecular energies are typically between -0.5 and -150 Ha. H₂ ground state: -1.1373 Ha.

Chemical Accuracy

1 kcal/mol (1.6 mHa) — the precision needed to predict chemical reaction outcomes. Our best hardware VQE achieves 0.22 kcal/mol via TREX mitigation.

Ansatz

The parameterized quantum circuit that prepares trial states. Choice of ansatz determines expressiveness, depth, and hardware compatibility. See the Ansatz Explorer.

Hamiltonian

The operator H = ∑ gᵢ Pᵢ encoding a molecule’s energy as a sum of Pauli strings. VQE measures ⟨ψ|H|ψ⟩ to estimate the energy.

Expectation Value

⟨ψ|H|ψ⟩ — the average energy from many circuit runs. Each Pauli term needs separate measurements. More shots = lower statistical noise.

Classical Optimizer

The algorithm adjusting θ each iteration. COBYLA (gradient-free) and SPSA (stochastic gradient) are common. The optimizer never touches the quantum computer directly.

Explore More

About VQE

The Variational Quantum Eigensolver (VQE) is a hybrid quantum-classical algorithm for finding the ground-state energy of molecules. The quantum processor prepares a trial state using a parameterized circuit (ansatz), measures the energy, and a classical optimizer adjusts the parameters to minimize it. This loop repeats until convergence.

VQE is designed for near-term noisy hardware: the circuits are short enough to run before decoherence destroys the signal, and the classical optimizer handles the noise by treating energy measurements as a stochastic objective function. It was first demonstrated by Peruzzo et al. in 2014 on a photonic processor.

This page shows real VQE results from IBM Quantum and Quantum Inspire Tuna-9 for hydrogen (H₂) and lithium hydride (LiH), compared against exact classical solutions. The gap between hardware results and the exact answer reveals how noise, ansatz choice, and error mitigation affect accuracy.