Why would Carleman linearization still require such enormous gate counts despite converting to a linear form?

This is an excellent question about a fundamental bottleneck in quantum simulation of nonlinear systems.

The paradox is real: Carleman linearization mathematically transforms a nonlinear ODE into a linear one, yet the quantum circuit implementing it still requires exponentially many gates. The culprit is truncation depth and nonlocality, not the linearity itself.

Here's what happens. When you embed a nonlinear system into a linear one via Carleman's method, you're not just rewriting the equations—you're expanding the state space. The nonlinear variables get replaced by an infinite tower of polynomial moments (x, x², x³, ...). To run this on a quantum computer, you must truncate this tower at some finite level N [AG-2024.05-2264]. That truncation level directly controls the number of qubits needed and, crucially, the circuit depth.

The second killer is nonlocality. In lattice-based formulations of fluid dynamics (like Lattice Boltzmann), the Carleman approach embeds local interactions in ways that demand long-range qubit couplings on the hardware. Compiling these nonlocal dependencies into SWAP gates—which move qubits around a physical device—balloons the total gate count exponentially [AG-2024.02-2050]. Even though the underlying mathematics is linear, the *hardware constraints* force you to pay an exponential penalty.

Interestingly, different Carleman formulations differ wildly in their practical cost. Lattice Boltzmann shows "exponential depth with the number of Carleman variables" [AG-2024.02-2050], while a Navier–Stokes approach can dramatically reduce the number of variables needed—potentially yielding a tractable depth, provided locality is preserved [AG-2024.02-2050].

The takeaway: linearity in the equations is not the same as linearity in the quantum circuit. What matters is how many auxiliary degrees of freedom you introduce (truncation) and how far apart they must be coupled on real hardware.