Conserved and relaxing modes enable polynomial-success coherent Carleman lattice Boltzmann evolution

AI summary

Overview

Research area: Quantum algorithms for computational fluid dynamics — specifically quantum lattice Boltzmann methods built on Carleman linearization and quantum block encodings.

Technical level: Advanced. The paper assumes familiarity with lattice Boltzmann theory, moment-space (multi-relaxation-time) collision operators, Carleman linearization, block encodings, linear combinations of unitaries, and postselection-based quantum linear algebra.

Scope (one sentence): The paper shows that performing the lattice Boltzmann collision in a weighted Hermite moment basis splits the Carleman-lifted, pair-scaled collision stage into a diagonal part plus three small 2×2 blocks (six in three dimensions), whose operator norm approaches one as the pair-sector scale λ grows, yielding a near-unity per-step success probability and polynomial rather than exponential overhead for coherent multi-step runs.

Note: the supplied text ends partway through Section IV, so the detailed resource estimates and observable-accuracy results announced for Sections V and VI are not present in the available content and are flagged as "not reported" where relevant below.

What This Paper Is About

Quantum computers cannot directly simulate the nonlinear, dissipative equations of fluid dynamics, so researchers linearize them — for lattice Boltzmann, by "lifting" the population vector into tensor powers (Carleman linearization) and truncating at second order. Each step of that lifted evolution must be embedded as a block of a larger unitary, and the run only succeeds when a postselection measurement passes; with existing population-space encodings that pass probability is tiny (6×10⁻⁵ or about 10⁻² per step), so a T-step coherent run succeeds with probability that decays exponentially in T. The paper's goal is to find a representation of the collision in which the block encoding is nearly lossless, and then to show that the resulting complete-run cost — not just the single-step cost — grows polynomially with the number of steps.

Key Contributions

1. A structure theorem for the scaled collision stage. The authors prove that if the conserved moments are orthogonal to the relaxing moments in a positive metric, the quadratic equilibrium feeds only relaxing moments from products of conserved moments, the coupling rows are mutually orthogonal, and streaming is an isometry, then the scaled level-2 collision stage is a direct sum of a diagonal contraction and one 2×2 block per coupled relaxing moment. The resulting operator norm is 1 + c_s/λ² + O(λ⁻⁴), with c_s = σ_s²/[2ω_s(2−ω_s)], and the theorem is verified to machine precision (residual 5×10⁻¹⁶ after rotation; norm agreement over 85 rate-and-scale combinations to 7×10⁻¹⁶).

2. A norm-attaining block encoding with three 2×2 dilations. Using Halmos dilation of each contracted 2×2 block, plus uniformly controlled rotations for the diagonal part, the authors build a unitary whose block equals 𝒦_λ/‖𝒦_λ‖₂ with subnormalization equal to the operator norm, requiring 4n+15 qubits for L = 2ⁿ (one more than the linear-combination-of-unitaries circuit) and no linear combination of terms.

3. A telescoped complete-run success probability with polynomial scaling laws. Section IV derives the success probability of a T-step run followed by output recovery, showing that optimizing the pair-sector scale for a prescribed horizon gives 1/T² and 1/T success laws for the two structured encodings, with the coupled-block encoding gaining a further factor linear in T. It also supplies a rest-shifted encoding, an explicit preparation circuit for few-mode initial data with exact amplification, and a validated protocol with deferred measurement, amplification and overlap readout.

4. Validation of nonlinear observables and instance-specific resource accounting. Section V establishes two nonlinearly generated observables of a nondegenerate two-mode flow and an accuracy window; Section VI gives a complete gate-level resource estimate for extracting one of them, from state preparation to readout, compared against matched quantum constructions and efficient classical evolutions. (The numerical details of these two sections are not in the supplied text.)

Main Findings

• Norm is fixed by conserved-to-relaxing coupling, not by lattice size or dissipation. In the weighted Hermite basis the only off-diagonal part of the lifted collision is a 3×4 per-site block from same-site products of the two conserved momenta (m₁m₁, m₂m₂, m₁m₂, m₂m₁) into the trace, deviator and shear moments, with singular values ω_e/(√2 λ) and ω_ν/(√2 λ). For D2Q9, σ_s = ω_s/√2, giving c_s = ω_s/[4(2−ω_s)] and c_{4,s} = −ω_s(3ω_s²−6ω_s+4)/[32(2−ω_s)³].

• Norm values collapse rapidly toward unity with λ. At ω = 1.5, ‖𝒦_λ‖₂ is 1.504861, 1.076738, 1.007436, 1.000833 and 1.000075 at λ = 1, 3, 10, 30 and 100, identically on D2Q9, D3Q19 and D3Q27.

• The block encoding beats a two-term linear combination of unitaries. At ω = 1.5 and λ = 10 the two-term LCU has α_L − 1 = 0.106 while the block structure gives ‖𝒦_λ‖₂ − 1 = 0.0074; the ratio grows with λ because one is first order and the other second order in 1/λ.

• Three distinct factors separate the single-step success probability. On a 32×32 Taylor-Green state at Ma = 0.087 and ω = 1.5: population space with sparse-access oracles gives 1.5×10⁻⁵ (seven-ancilla subnormalization of Sanavio et al.) and 3.4×10⁻⁵ with α_s = √(s_r s_c) max|a_ij|; the two-term SVD construction of Bastida Zamora et al., modeled as the optimal subnormalization σ_max = 9.4, gives 1.1×10⁻²; the weighted encoding f/√w moves singular values into [0.25, 1.5] and raises sparse access to 1.2×10⁻³; the Euclidean moment basis with plain encoding gives 1.4×10⁻³; the weighted Hermite representation (sparsity three, entries at most one, α_s = 3) gives 0.11 with the same generic construction; the structured constructions on the coupling give 0.82 and 0.98 for the two-term LCU at λ = 10 and 100, and 0.985 and 0.9998 for the coupled block at the same scales. The metric, basis and coupling encoding thus contribute factors of order 10², 10² and 10.

• Statevector verification on 23 qubits. On a 4×4 lattice the postselected state matches 𝒦_λψ/α in both sectors to 2×10⁻¹⁴ and the measured success probability matches r_λ/α² to six digits: p = 0.984965, 0.965744 and 0.980375 over three consecutive steps at λ = 10, where the LCU gives 0.8171, 0.8012 and 0.8133; 0.999518 at λ = 100; 0.9788 and 0.862257 at λ = 3 against 0.5456; and 0.980263 for MRT rates ω_ν/ω_e/ω_q/ω_ε = 1.3/1.6/1.1/1.8 at λ = 10 against 0.8068.

• Residual loss is physical, not an encoding artifact. The remaining 1 − r_λ lies between 3×10⁻⁴ and 2×10⁻² over the tested steps and equals the non-equilibrium norm that relaxation removes from the scaled state; on the 32×32 two-mode run at Ma = 0.173 it is at most 2×10⁻³ per step. The norm change of the level-1 sector is given explicitly, including an indefinite cross term that is at most 10⁻³ of the norm on the trajectories considered, stays below the relaxed norm for unperturbed Taylor-Green runs at rates up to 1.7, and exceeds it in a few steps at rates 1.9 and 1.95 where both are below 10⁻⁵ of the norm.

• Hundred-step runs succeed with orders-of-magnitude higher probability than population-space encodings, and the combined encoding-plus-sector-selection overhead grows polynomially in the number of steps rather than exponentially.

• Truncation accuracy is adequate over the horizon where the cost is favorable. Over horizons of several tens of steps the truncated dynamics reproduces the nonlinearly generated modes of the benchmark flow to within a few percent.

• The structure extends to three dimensions. The same diagonal-plus-2×2-block decomposition holds on D3Q19 and D3Q27, with six symmetric conserved-momentum products coupling into six second-order moments with σ_s = ω_s/√2, so the norm is the same function of ω and λ as on D2Q9.

• Expansion limits are stated honestly. The 1/λ² series overestimates near ω → 2⁻, giving 1.098 against the exact 1.051 at ω = 1.95 and λ = 10; the exact singular value of the 2×2 block is used for all finite-λ costs.

• Not reported in the supplied text: the explicit numerical accuracy windows for the two observables, the gate counts and qubit counts of the full pipeline, the comparison against classical CFD cost, and the precise prefactors of the 1/T² and 1/T success laws.

Methodology in Plain English

The authors keep the standard lattice Boltzmann split between a local collision and a streaming step, but change three things about how the collision is represented on a quantum computer.

First, instead of storing the nine raw populations per site, they store them in a weighted Hermite moment basis (essentially mass, momentum, energy, stress and higher ghost moments), scaled by the square roots of the lattice weights. This makes the transformation from populations to moments an orthogonal rotation, so quantum state norms behave like physical moment magnitudes.

Second, they use a multi-relaxation-time collision, where each moment decays at its own rate, rather than a single-rate BGK collision. In the moment basis the linear part is simply diagonal, and the quadratic equilibrium contribution touches only the three second-order moments. That means the Carleman-lifted collision is almost diagonal: a fully contractive diagonal piece plus a small coupling from products of the conserved momenta into the relaxing second-order moments.

Third, they rotate the four pair slots (products of the two momentum components) into symmetric combinations, which decouples the antisymmetric slot and breaks the leftover coupling into independent 2×2 blocks. They then choose the scale λ of the pair sector so the operator norm of the whole stage sits just above one, and use a standard trick (Halmos dilation) to embed each contracted 2×2 block exactly inside a 2×2 unitary acting on one ancilla. Because the block encoding then has subnormalization equal to the true operator norm, the postselection probability is as close to one as the norm allows. They check the whole construction by simulating the circuit's statevector on a 4×4 lattice and comparing against a classical Carleman lift.

For the multi-step problem they scale the pair sector for the specific run length, so the per-step normalization loss and the final sector-selection loss trade against each other; telescoping the product of per-step factors turns the exponential decay into a polynomial overhead. The benchmark used to test everything is a two-mode Taylor-Green-like flow on a 32×32 lattice at Ma = 0.087, whose quadratic nonlinearity generates an acoustic mode at k = κ(1,0) peaking near 0.4 t_adv (horizon T = 40 steps) and a vortical mode at k = κ(1,2) peaking near 0.76 t_adv (horizon T = 77 steps), with mode amplitudes A₂ = 0.6 and phase φ = 0.3.

Why This Matters

Impact on research. Previous assessments of the Carleman lattice Boltzmann route treated the single-step postselection probability as the cost metric. This paper shows that the physically meaningful quantity is the cost of a specified observable at a specified accuracy over a specified horizon, and that the collision's own conservation-and-relaxation structure — not the number of velocities — sets that cost. The block-structure theorem also gives a constructive recipe for norm-attaining encodings that the authors argue applies to any lattice Boltzmann model with a quadratic equilibrium, including three-dimensional lattices.

Real-world applications (as motivated by the cited literature):

• Aerodynamic flow simulation, including the nozzle-flow problem previously targeted by quantum differential-equation algorithms.
• Canonical wall-bounded flows such as Poiseuille and Couette flows, relevant to microfluidics and channel design.
• Turbulent and vortical flow modelling where the nonlinear mode generation demonstrated here (acoustic and vortical modes generated by mode interaction) is the physics of interest.
• General kinetic-theory transport problems with quadratic local equilibria, since the structure theorem is stated at the level of moment couplings rather than a specific lattice.

Industry relevance. The paper supplies the kind of gate-level, instance-specific resource accounting that quantum-computing roadmaps and fault-tolerant architecture planning require, and it provides a fair comparison against efficient classical evolution of the same problem. That matters to organizations deciding whether fluid-dynamics workloads belong on a quantum roadmap, and to hardware teams sizing qubit counts (4n+15 qubits for an L = 2ⁿ lattice) and non-Clifford gate budgets.

Future Directions

• Full fault-tolerant costing. The paper works in logical-qubit and gate terms; incorporating error correction overhead, non-Clifford gate distillation and circuit depth under noise would show what the polynomial success law costs in physical qubits.
• Extending beyond level-2 truncation. The truncation error is characterized only for second-order Carleman truncation; whether higher orders preserve the same block structure, or whether the polynomial overhead survives them, is open.
• Three-dimensional end-to-end validation. The block structure is proven and numerically checked on D3Q19 and D3Q27, but the full preparation-evolution-readout pipeline and its resource estimate are developed for the two-dimensional case.
• Which properties of the collision fix the cost, for other kinetic models. The authors frame the theorem as a statement about conserved-versus-relaxing coupling; mapping which further collision models satisfy assumption (iii) (mutually orthogonal coupling rows) — and what happens when they do not, since the output space then rotates and the stage no longer splits — is a natural next question.
• Hybrid pipelines. Combining this amplitude-encoded, coherent collision encoding with deterministic completely-positive-trace-preserving treatments of relaxation (as in Khan et al.) is an obvious complementary route.

Target Audience

Quantum-algorithm researchers working on linear-systems and differential-equation solvers; lattice Boltzmann and kinetic-theory specialists interested in quantum implementations; quantum resource-estimation and architecture teams; and computational fluid dynamics practitioners evaluating whether quantum hardware could eventually address their workloads. The paper is written for readers comfortable with block encodings and moment-space collision operators, and is not an introductory treatment.

Authors’ abstract

Carleman linearization represents nonlinear lattice Boltzmann collision dynamics in a form amenable to quantum block encoding, but low postselection probabilities in existing population-space encodings make coherent multi-step evolution exponentially unlikely to succeed. We present a quantum Carleman lattice Boltzmann scheme based on the multi-relaxation-time collision operator in a weighted Hermite moment basis. In this basis, the lifted collision operator separates into small coupled blocks whose norm exceeds unity because of the coupling between conserved and relaxing modes. This block structure enables a norm-attaining block encoding with a single-step success probability close to unity. Crucially, scaling the pair sector for a prescribed evolution horizon makes the combined encoding and sector-selection overhead grow polynomially rather than exponentially with the number of steps. For a hundred-step evolution, the resulting complete-run success probability exceeds that of population-space encodings by orders of magnitude. Over horizons of several tens of steps, the truncated dynamics reproduces the nonlinearly generated modes of a benchmark flow to within a few percent. We verify the combined preparation, evolution and readout protocol as a single unitary circuit and give a complete gate-level resource estimate for extracting a nonlinear observable of the benchmark flow from a coherent Carleman evolution. The block structure of conserved and relaxing modes holds equally on three-dimensional lattices and provides a constructive route to norm-attaining encodings for lattice Boltzmann models with quadratic equilibria.