Variational Quantum Eigensolver
VQE (the Variational Quantum Eigensolver) estimates the lowest energy of a molecule or material. It runs a short quantum circuit with adjustable settings, measures the energy of the state it makes, and lets a normal computer tweak the settings to push that energy down. It is a guess-and-check loop between a quantum and a normal computer, built for today's noisy hardware. It really does run on real machines at small sizes, but it has no proven speedup, and normal chemistry methods match or beat every published VQE result.
| Task | Best classical | Quantum |
|---|---|---|
| Exact ground-state energy | Exponential in system size (full configuration interaction) | Unknown — VQE is a heuristic with no runtime or accuracy guarantee |
| Where practice stands today | CCSD(T), DFT and DMRG handle systems far larger than any quantum demo, at chemical accuracy for many of them | ~12-qubit demos with error mitigation, accuracy below classical baselines |
What problem does VQE try to solve?
VQE tries to find the ground-state energy of a quantum system. That is its lowest possible energy, the way a ball settles at the bottom of a bowl. It is the most useful single number in computer chemistry. Energies decide how fast reactions go, how strongly things stick together, and how materials behave.
The trouble is size. The number of ways to arrange electrons in a molecule grows exponentially as the molecule gets bigger. So exact normal computing only works for small molecules.
The proper quantum algorithm for this job is phase estimation. But its circuits are so deep they need error correction. VQE, proposed in 2014, is the workaround. It asks: what is the best we can do with circuits short enough to survive today's noise?
How do normal computers solve this?
Normal chemistry software is a mature and very strong field. Density functional theory (DFT) handles thousands of atoms, roughly. Coupled cluster, written CCSD(T) and called the "gold standard," reaches chemical accuracy (precise enough to predict real chemistry) for medium-sized molecules. DMRG and tensor networks handle systems where electrons are tightly linked. These methods run the world's real chemistry work today.
The truly hard cases for normal computers are strongly correlated systems. Those are molecules where many electrons strongly affect each other. The standard example is the FeMo cofactor in the enzyme nitrogenase, which contains transition metals. That is the target quantum methods aim at. Whether VQE in particular can ever beat normal methods there is an open question, not a promise.
How does VQE work? A guess-and-check loop
VQE rests on the variational principle. Any trial state's energy is always greater than or equal to the true lowest energy. So lower is always better. Think of hiking down a foggy valley. You can't see the bottom, but any step that takes you lower is progress.
The loop goes like this:
- A circuit with adjustable angles, called the ansatz (German for "starting guess"), makes a trial state. It has rotation gates plus gates that entangle qubits.
- You run and measure it many times. From the results you estimate the energy, written ⟨H⟩. It is a weighted sum of simple averages. Each group of terms needs its own way of measuring.
- A normal optimizer nudges the angles, and you repeat.
Notice what the quantum computer adds. Each run, it checks one trial state. All the searching is done by the normal computer. The hope is that an entangling circuit can describe tightly linked electrons more cheaply than normal methods can. That is possible in principle. It is unproven in practice.
Run it: one energy check inside the loop
What are its limits?
- No guarantees. Unlike phase estimation, VQE promises nothing. The ansatz may not be able to describe the lowest-energy state at all. And the optimizer may never find the best angles it could reach.
- Barren plateaus. For many ansatz types, the slope of the energy shrinks exponentially as you add qubits. The optimizer ends up wandering on a flat plain with no sense of which way is down. This is a proven problem for broad families of circuits, not a rumor.
- Measurement cost. Chemical accuracy (about 1.6 milli-Hartree, a tiny unit of energy) needs millions to billions of shots per energy check for real molecules. That is often the biggest cost, and it is often quietly ignored.
- Squeezed by normal computers. The short circuits that noise allows can often be simulated on a normal computer. New normal "stand-in" methods have copied many published VQE experiments outright. That squeezes the window for an advantage from both sides.
How does VQE do on hardware today?
VQE is one of the most-run algorithms on real hardware. Examples include H₂ on 2 qubits (2016), BeH₂ on 6 (2017), and a 12-qubit Hartree–Fock calculation (2020). Many studies of the Hubbard model and small molecules have followed, on superconducting and trapped-ion machines. See current QPUs. Every one needed error mitigation (math tricks that reduce the effect of noise). And every one found energies that a laptop gets faster and more accurately.
That is not a failure. These are honest hardware tests. But no VQE calculation has ever gone beyond what normal computers can do. The field itself has shifted toward error-corrected algorithms as the believable path to useful quantum chemistry. Run the ansatz in the lab. Notice how transpilation (rewriting into a chip's own gates) reshapes even this 5-gate circuit on real devices.