In 1982 Richard Feynman stood up and asked a mischievous question: if you want to simulate physics on a computer, and physics is quantum mechanical, then why are you using a classical computer? "Nature isn't classical, dammit," he said, "and if you want to make a simulation of nature, you'd better make it quantum mechanical." That single complaint is the seed the whole field grew from. Quantum computers were not first imagined to break codes or search databases — they were imagined to do one thing that ordinary machines choke on: simulate other quantum systems. Molecules, magnets, superconductors, the soup inside a neutron star. This is the application quantum computers were invented for, and still their most natural "killer app."

The exponential wall

Why can't a classical supercomputer just do it? Because a quantum state is enormous. As we saw with multi-qubit states, a system of n two-level particles (spins, qubits, electrons in orbitals) lives in a space of dimension 2^{n}, so writing the state down means storing

That number doubles with every particle you add. Ten spins is a thousand numbers; forty spins is a trillion; and by around fifty spins you have passed what the largest classical machines on Earth can hold exactly. The atoms in one methane molecule, the electrons in a stretch of copper-oxide — nature tracks all 2^{n} amplitudes effortlessly, in real time, for free. A classical simulator has to write them all out, and it simply runs out of room. That is the exponential wall.

Feynman's insight: fight fire with fire

Here is the escape. The reason the state is exponentially large is itself quantum — it comes from superposition and entanglement. So instead of fighting that with a classical machine, use a quantum system you can control to imitate a quantum system you can't. Both live in the same exponentially large space, both obey the same rules, so the controllable one can shadow the uncontrollable one move for move — without ever writing down a single amplitude. You prepare the quantum computer in the state of interest, let it evolve, and read out physical quantities. The exponential size that dooms the classical approach is exactly what the quantum simulator supplies for free.

There are two ways to build such a mimic, and it is worth knowing both.

Style 1 — digital (gate-based) simulation

A gate-based quantum computer, the kind described by the quantum circuit model, evolves a state under a Hamiltonian H for time t by applying the unitary e^{-iHt}. But e^{-iHt} is one huge operator on the whole register — you cannot build it as a single gate. The trick is Trotterization: split H into a few simple pieces you can implement, say H = A + B, chop the time into many tiny slices \Delta t, and alternate them:

Each factor is a short, buildable circuit; string enough of them together and you approximate the full evolution as closely as you like. Step through the figure to see one giant block become a repeating ladder of small ones.

Style 2 — analog simulation

The other route skips gates entirely. Instead of digitising the evolution, you build a piece of hardware whose own physical Hamiltonian can be tuned to match the target system's, and simply let it run — an analog stand-in. Cold neutral atoms held in optical lattices, arrays of Rydberg atoms, trapped ions, and superconducting circuits can each be dialled so their interactions copy those of, say, a magnetic material. There are no discrete gate steps and no Trotter error; you are letting one natural system act out another. The price is flexibility — an analog machine is purpose-built for a family of Hamiltonians, where a digital one can (in principle) simulate any local Hamiltonian by reprogramming its gates. Both are actively used today, and on near-term (NISQ) hardware even a noisy analog device can yield real physics.

Worked example 1: where the wall actually is

Let us make the wall concrete. Take a chain of n spin-½ particles. Its state is a list of 2^{n} complex numbers, and a single complex number (two double-precision floats) costs 16 bytes. So the memory to store the state exactly is

At n = 50 that is 2^{54} \approx 1.8 \times 10^{16} bytes — about 16 petabytes, rivalling the memory of the very largest supercomputers just to hold one state vector, before doing a single step of physics. Add ten more spins, n = 60, and the bill multiplies by 2^{10} = 1024 to roughly 16 exabytes — beyond any machine that exists. This is why around 50 spins is the practical ceiling for exact classical simulation, and why a modest quantum device — which stores the same state in n physical qubits, not 2^{n} numbers — is not just faster but living in a different universe of scale.

Worked example 2: a two-term Trotter step and its error

Where does the "\approx" in Trotterization come from, and how good is it? Consider one slice with H = A + B. Expand both sides to second order in the small time \Delta t. The Trotter product is

while the exact one-slice evolution is

The constant and first-order (\Delta t) terms match exactly. Subtract, and everything cancels until the \Delta t^2 term, where the product carries A^2 + 2AB + B^2 but the exact evolution carries A^2 + AB + BA + B^2. The mismatch is the commutator:

So the leading error of one step is O(\Delta t^2), set by how badly A and B fail to commute — and it vanishes exactly when [A,B] = 0 (commuting pieces can be applied independently). Cover a full time t with N = t/\Delta t such steps and the errors add up to roughly N \cdot O(\Delta t^2) = O(t\,\Delta t), which shrinks as you take more, finer slices. Accuracy is yours to buy — at the cost of a longer circuit.

What you actually simulate

The payoff is a shortlist of the hardest, most valuable problems in physical science — problems that are hard precisely because they are quantum:

• Quantum chemistry. Reaction energies, reaction rates, and the electronic structure of molecules — the ground truth behind catalyst design and drug discovery. Even modest molecules defeat exact classical methods.
• Nitrogen fixation. The enzyme nitrogenase fixes atmospheric nitrogen at room temperature via a metal cluster (FeMoco) whose electronic structure is a notorious quantum-chemistry challenge — cracking it could reinvent fertiliser production.
• High-temperature superconductivity. The Hubbard model and cuprate physics — why some materials carry current with no resistance — remain unsolved strongly-correlated systems that a quantum simulator is built to probe.
• Condensed matter & new materials. Magnetism, phase transitions, exotic states of matter — modelling lattices of interacting spins to design materials from first principles.

In every case the resource cost of the problem matches the resource the quantum computer naturally supplies. That alignment is why simulation is widely seen as the most promising near-term use of quantum hardware: unlike deep code-breaking algorithms, useful simulations can be shallow, and even noisy or analog machines can contribute.

See it: the wall, and how little more memory buys

The curve is 2^{n} — the number of amplitudes to store a state of n particles — against the particle count. It hugs the floor, then rockets up almost vertically: that near-vertical cliff is the exponential wall. The dashed line is your classical memory budget; where the curve crosses it is the largest system you can simulate exactly. Drag the slider to buy more memory and watch the crossing barely creep to the right — because doubling your memory buys only one more particle. No amount of classical hardware climbs this wall.