The unitary brick-wall is proposed: a $k-particle fermionic architecture for nearest-neighbor hardware, combining Reconfigurable Beam Splitter gates with interleaved single-qubit phase gates and a non-Gaussian magic-state encoding.
Abstract
Designing scalable parameterized quantum circuits for machine learning faces three obstacles: barren plateaus, the absence of guarantees that the learned function class is classically hard, and prohibitive circuit evaluations per gradient step. We propose the unitary brick-wall: a $k$-particle fermionic architecture for nearest-neighbor hardware, combining Reconfigurable Beam Splitter gates with interleaved single-qubit phase gates and a non-Gaussian magic-state encoding, where $k$ is a tunable dial trading classical simulation hardness against training cost. Trainable. The brick-wall has dynamical Lie algebra $\mathfrak{u}(n)$ and is surjective onto $U(n)$ via Givens rotations, enabling Haar initialization. Two-body correlator readouts achieve gradient variance $\Theta(k^3/n^5)$, polynomial in $n$ throughout $n-2k=\Omega(n)$. Expressive. Classical hardness is controlled by $k$: best-known classical sampling algorithms run in time $2^{\Theta(k)}\mathrm{poly}(n)$, worst-case #P-hardness holds from $k=n^{\epsilon}$, and the average-case machinery of Fermion Sampling applies at $k=\Theta(n)$. At our operating point $k=60$, best-known classical simulation exceeds $10^{24}$ operations at every $n$. Efficient. A multi-layer parallel parameter-shift rule computes all $O(n^2)$ gradients from $4kn$ circuit evaluations per gradient step, a factor $n/k$ reduction over the $4n^2$ evaluations of the standard rule, growing linearly with $n$ at fixed $k$. The unitary butterfly variant targets all-to-all hardware, with depth $2\log n$ and $(3/2)n\log n$ parameters, similar hardness guarantees, and $4k\log n$ evaluations per gradient step -- the same factor-$n/k$ reduction. Its trainability holds at two levels: absence of exponential barren plateaus is unconditional, while the sharp $\Theta(k^3/n^5)$ rate holds under a two-particle approximate-2-design conjecture.
This paper invalidates the algorithm-relative exponential-cost conclusion for the supervised two-body readout, without affecting the gradient-variance, barren-plateau, parameter-shift, or sampling-hardness results.
Quantum algorithms are conventionally presented with their input state supplied for free. When the input is classical data, this convention conceals a cost that is frequently larger than the algorithm it precedes. We review what the three standard encodings, such as basis encoding, amplitude encoding, and Grover--Rudolph distribution loading, actually cost once transpiled to a hardware gate set, and argue that the resulting $\Theta(N)$ bound is a counting theorem rather than an engineering limitation that improved hardware will remove. Measured gate counts for a representative loading task are reported: an optimal library implementation requires $247$ CNOT gates at $n=8$ qubits and doubles with each additional qubit, while the classical preprocessing that produces the rotation angles requires reading the entire input vector. We show how this cost eliminates the quadratic advantage of quantum amplitude estimation for Monte Carlo integration, and argue that the same accounting constrains quantum machine learning more broadly: the strong input models that make quantum algorithms fast on classical data also enable classical dequantization, and quantum kernel methods carry a $\Theta(M^2)$ state-preparation cost for the Gram matrix that does not amortize. We explain that the efficiently preparable states, device-generated distributions, variationally learned loading, and amortized preparation are required to get advantage from quantum machine learning and close with a checklist for evaluating input-dependent advantage claims. Executable notebooks reproducing every construction and measurement discussed here are available.
Variational quantum circuits have been central to many proposed near-term applications of quantum computing, but a growing body of evidence suggests that trainability and quantum advantage are fundamentally at odds: ans\"atze expressive enough to resist efficient classical simulation tend to exhibit barren plateaus, while structures that provably rule out barren plateaus typically render them classically simulable. We propose a stacked linear combination of unitaries (S-LCU) as a variational ansatz which provides a tunable trade-off between barren plateaus and classical simulability. Using a diagrammatic analysis, we bound the loss-landscape variance of the Free Fermion S-LCU, whose elements are fermionic Gaussian unitaries. We prove a variance lower bound of $\Omega(1/(n k^{3l}))$, with a simulation cost of $O(k^{2l} n^3)$ using the best known classical algorithm, compared to a quantum gate complexity of only $O(lkn^2)$. The number of layers $l$ serves as a single dial that trades computational complexity against the rate of cost concentration. This offers practitioners a systematic method for constructing ans\"atze with a complexity-trainability trade-off that best suits their application and hardware.
This work constructs a native 2D pairwise ansatz and compares its expressibility and trainability with representative 1D ansatze at identical layer depths, despite their different circuit depths.
Qubit-efficient quantum relaxations compress classical decision variables into expectation values on substantially fewer qubits. We ask what resource tradeoffs this compression entails for quantum optimization. For the complete quadratic-Majorana encoding on $n$ qubits, pairwise correlators can represent $m=\Theta(n^2)$ binary variables. We define the universal margin as the smallest correlator magnitude that can be guaranteed with prescribed signs for every target sign assignment. We show that it is exactly $\Delta_{\rm Maj}(n)=\tan\!\left(\frac{\pi}{4n}\right)=\Theta(1/n)$, whereas uniformly random sign assignments retain $\Theta(1/\sqrt n)$ target-specific margins. The stronger $1/n$ worst-case scaling is Majorana-specific. Moreover, arbitrary density operators and fermionic Gaussian states generate the same quadratic-Majorana covariance body, so non-Gaussian state resources cannot enlarge this two-point relaxation. Beyond Majoranas, standard quantum random access code bounds provide general information-theoretic baselines. For any fixed family of $m$ designated binary observables on $n$ qubits, the universal margin is at most $\sqrt{(2\ln2\;n/m)}$, while arbitrary random access decoding from $N$ copies with constant success probability above $1/2$ requires $nN=\Omega(m)$. For a fixed Pauli correlation encoding required to work uniformly over all targets, maintaining a fixed nonzero decoded magnitude under smooth sign decoding therefore requires a rescaling parameter that grows as the available margin shrinks. Thus, while providing substantial qubit savings, compression can shift cost into restricted expectation value geometry, smaller expectation value magnitudes, or more demanding information recovery rather than eliminate it.
A Clifford+T quantum circuit construction that approximately implements any classically specified unitary to within error $\epsilon$ and achieves a worst-case $T$-count with leading exponential scaling of $2^{5n/4}$ whenever $\log(1/\epsilon)=\operatorname{poly}(n)$.