A hardware-efficient variational ansatz with an exact diagonal metric for real- and imaginary-time evolution and Haar sampling
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.
Kai: I'm Kai.
Mira: And I'm Mira, here with Lev.
Kai: We're looking at the paper: "A hardware-efficient variational ansatz with an exact diagonal metric for real- and imaginary-time evolution and Haar sampling".
Mira: Here's the gist of the paper. Variational quantum algorithms, particularly those requiring metric awareness for time evolution or optimization, traditionally rely on costly auxiliary measurements and ill-conditioned matrix inversions when using metrics like the Fubini–Study metric.
Kai: This work introduces a hardware-efficient n-qubit ansatz that parametrises states via a binary tree structure.
Mira: A key structural result is that the pullback Fubini–Study metric for this tree structure is exactly diagonal in closed form (Theorem 1).
Kai: This diagonal property enables quantum natural gradient and variational imaginary-time evolution to utilize elementwise division, thereby eliminating the need for auxiliary metric circuits or matrix inversion.
Paper summary: Kai: To summarize this paper, it’s about taking those hard parts of variational quantum algorithms—the metric awareness needed for time evolution or optimization—and making them way more efficient. The core thesis is that by using a binary tree structure to parameterize the states, you can get an exact diagonal form for the Fubini–Study pullback metric without needing auxiliary measurements or matrix inversions.
Mira: That means you can run quantum natural gradient and imaginary-time evolution using simple elementwise division, which cuts out the usual massive overhead that scales with the number of parameters P. This is a huge simplification for anyone trying to use these methods on actual hardware.
Lev: From my perspective, that reduction in overhead is what actually matters because those auxiliary tests scale with P squared and are regularized by a pseudoinverse, which limits how many problems you can reach in practice two <ref:2607.07942#pg1>. If you eliminate that scaling issue, the overhead lands much lower than we currently expect.
Kai: They also show this structure lets you perform exact unitary-invariant Haar sampling on a symmetry sector with zero auxiliary metric circuits, achieving it at an O of k cost per sample. That’s a direct speedup for benchmarking states in those specific areas.
Mira: And when the target state is sparse, supported on only k computational-basis states, they leverage the gauge freedom of the redundant tree parameters to create a pruning compiler that builds circuits with two-qubit counts that grow linearly with k. This leads to near optimal O of n times k divided by log of n scaling for those specific circuits one <ref:2607.07942#pg1>.
Lev: That linear growth in CNOTs is critical because it shows a path toward making these algorithms runnable on real hardware, even when you’re dealing with complex structures like those in electronic structure problems. It shifts the complexity from being potentially prohibitive to being manageable.
Kai: So, they’re showing a way to build primitives that are inherently better behaved geometrically than what we currently have, which should make preparing and sampling those sector states much faster.
Conclusion: Mira: Thinking about the title, "A hardware-efficient variational ansatz with an exact diagonal metric," it really captures the essence: they’ve built a state representation that is efficient for hardware, and they solved the metric problem by making it diagonal. That’s what makes the real-time evolution and Haar sampling so much cleaner.
Kai: I think what stands out is how they combine all these elements into one structure—the binary tree, the diagonal metric, and then the pruning compiler for sparsity. It feels like they’ve created a really solid foundation that simplifies everything from gradient calculation to circuit compilation.
Lev: What this implies for me is that if we can implement this kind of structural simplification on real hardware, it drastically reduces the required gate depth and the necessary classical control overhead for running simulations, which is exactly what error correction systems need to manage one <ref:2607.07942#pg1>.
Mira: It moves us away from needing dense metric estimations that scale quadratically with parameters P, toward something that scales much better when you have sparse states or when using this specific tree ansatz. It’s about making the geometry work *with* the hardware constraints instead of fighting them.
Kai: So, in simple terms, they’ve shown a way to build a state preparation tool that is fundamentally well-behaved geometrically, which lets us run simulations for things like molecular ground states and dynamics with fewer gates and less classical calculation.
Lev: It’s not a magic fix for every problem, but it provides a much better starting point—a bare ansatz—that is exactly controllable and doesn't suffer from barren plateaus when you dress it up correctly. That’s where the real practical work lies right now zero <ref:2607.07942#pg1>.
University College London · London Centre for Nanotechnology
quant-ph
Submitted: 2026-07-08
Updated: 2026-11-05
Comments: 58 pages. Revised metric terminology and phase-gauge conventions; clarified the projective-volume derivation; updated references. Numerical results unchanged
Code: https://github.com/dariopicozzi/quantumsymmetry
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 92/100
The gist: Variational quantum algorithms, particularly those requiring metric awareness for time evolution or optimization…
Key concepts
- Pullback Fubini–Study Metric
- This is a specific mathematical metric used to measure distance or change in quantum states within the variational algorithm. For this tree structure, the key result is that this metric simplifies into a diagonal form, which makes calculations much faster and avoids needing complex matrix inversions or extra circuits.
- Quantum Natural Gradient
- This is an optimization technique used to find the best parameters for a quantum state. Because the metric is diagonal, calculating this gradient becomes very efficient. Instead of requiring many costly steps involving auxiliary measurements, it can be computed using only elementwise division, significantly speeding up the optimization process.
- Pruning Compiler
- When the target state is sparse (supported on few basis states), some tree parameters become redundant. The pruning compiler exploits this redundancy by converting these extra parameters into circuits whose complexity grows linearly with the number of active states ($k$), achieving near-optimal scaling for circuit size.
Terminology
Summary
Variational quantum algorithms, particularly those requiring metric awareness for time evolution or optimization, traditionally rely on costly auxiliary measurements and ill-conditioned matrix inversions when using metrics like the Fubini–Study metric. p. 0 A hardware-efficient variational ansatz with an exact diagonal metric for real- and
imaginary-time evolution and Haar sampling
Dario Picozzi1, 2, ∗
1Department of Physics and Astronomy, University College London (UCL),
Gower Street, London, WC1E 6BT, United Kingdom
2London Centre for Nanotechnology, 19 Gordon St, London, WC1H 0AH, United Kingdom
Variational quantum algorithms depend on the geometry of their parametrised circuits: metricaware optimisation and time evolution require the Fubini–Study metric, which has hitherto demanded costly auxiliary measurements and ill-conditioned inversions. This work introduces a
hardware-efficient n-qubit ansatz, which parametrises states by a binary tree and whose Fubini–
Study pullback metric is diagonal in closed form. A hardware-efficient variational ansatz with an exact diagonal metric for real- and
imaginary-time evolution and Haar sampling
Dario Picozzi1, 2, ∗
1Department of Physics and Astronomy, University College London (UCL),
Gower Street, London, WC1E 6BT, United Kingdom
2London Centre for Nanotechnology, 19 Gordon St, London, WC1H 0AH, United Kingdom
Variational quantum algorithms depend on the geometry of their parametrised circuits: metricaware optimisation and time evolution require the Fubini–Study metric, which has hitherto demanded costly auxiliary measurements and ill-conditioned inversions. This work introduces a
hardware-efficient n-qubit ansatz, which parametrises states by a binary tree and whose Fubini–
Study pullback metric is diagonal in closed form. A hardware-efficient variational ansatz with an exact diagonal metric for real- and
imaginary-time evolution and Haar sampling
Dario Picozzi1, 2, ∗
1Department of Physics and Astronomy, University College London (UCL),
Gower Street, London, WC1E 6BT, United Kingdom
2London Centre for Nanotechnology, 19 Gordon St, London, WC1H 0AH, United Kingdom
Variational quantum algorithms depend on the geometry of their parametrised circuits: metricaware optimisation and time evolution require the Fubini–Study metric, which has hitherto demanded costly auxiliary measurements and ill-conditioned inversions. This work introduces a
hardware-efficient n-qubit ansatz, which parametrises states by a binary tree and whose Fubini–
Study pullback metric is diagonal in closed form. This work introduces a
hardware-efficient n-qubit ansatz, which parametrises states by a binary tree and whose Fubini–
Study pullback metric is diagonal in closed form. Quantum natural gradient on the tree parameters,
variational imaginary- and real-time evolution, and exact unitary-invariant (Haar) sampling on a
symmetry sector run with no auxiliary metric circuits or matrix inversion.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
The method also allows for exact unitary-invariant (Haar) sampling on a symmetry sector without auxiliary metric circuits, achieving this with O(k) cost per sample. p. 1 We show that the metric admits the diagonal closed form via a direct calculation that
uses no Hadamard tests; that an exact four-term parameter-shift rule yields analytic Euclidean gradients on the
same circuit at four evaluations per parameter; and that, when the target state is sparsely supported on k ≪ 2
n
computational-basis states, the redundant parameters of the full tree carry a gauge freedom made explicit through an
active/fixed/inactive node classification and exploit to compile pruned circuits whose CNOT count provably grows
linearly with k (at most 2nAs = O(n
2k) for any ordering, with the tight constant conjectured). The diagonal metric
has one further consequence, exploited throughout: drawing the parameters from the Fubini–Study volume measure
dµFS ∝
p
det g dλ (1)
reproduces the unitary-invariant (Haar) ensemble on a chosen symmetry sector exactly and rejection-free, the diagonal
metric reducing the draw to one independent sample per tree node. Quantum natural gradient on the tree parameters,
variational imaginary- and real-time evolution, and exact unitary-invariant (Haar) sampling on a
symmetry sector run with no auxiliary metric circuits or matrix inversion. When the target state
is supported on a subspace of k computational-basis states, the redundant tree parameters carry
a gauge freedom a pruning compiler converts into circuits whose two-qubit count provably grows
linearly in k; a variant reaches near-optimal O(nk/ log n) scaling with the closed-form metric intact. Quantum natural gradient on the tree parameters,
variational imaginary- and real-time evolution, and exact unitary-invariant (Haar) sampling on a
symmetry sector run with no auxiliary metric circuits or matrix inversion. When the target state
is supported on a subspace of k computational-basis states, the redundant tree parameters carry
a gauge freedom a pruning compiler converts into circuits whose two-qubit count provably grows
linearly in k; a variant reaches near-optimal O(nk/ log n) scaling with the closed-form metric intact.Where the paper says it
Where the paper says it
Where the paper says it
For applications such as electronic-structure calculations on small molecules and half-filled Hubbard quench dynamics, the method achieves reference-level accuracy using one to three orders of magnitude fewer two-qubit gates compared to leading alternatives. p. 0 When the target state
is supported on a subspace of k computational-basis states, the redundant tree parameters carry
a gauge freedom a pruning compiler converts into circuits whose two-qubit count provably grows
linearly in k; a variant reaches near-optimal O(nk/ log n) scaling with the closed-form metric intact. On electronic-structure calculations for small molecules and half-filled Hubbard quench dynamics,
the method reaches reference-level accuracy with one to three orders of magnitude fewer two-qubit
gates than leading alternatives. Interchangeable constructions (a Schur-transform dressing or internal reparameterisations) make the ansatz exactly spin-adapted, with fixed total spin at every
parameter and no penalty terms. The bare ansatz is an exactly controllable, well-conditioned and
barren-plateau-free primitive for preparing and sampling sector states: on its own, it is classically
simulable in k (a boundary proved for a general class of sector-sparse ans¨atze); composed with
a classically hard dressing, it yields molecular ground states, sector-Haar benchmarking, thermal
correlators, and exact effective Hamiltonians trained from energy measurements alone, with the
composed circuit carrying the potential for quantum advantage.Where the paper says it
Where the paper says it
The central structural result hinges on the binary tree chart pullback metric yielding an exactly diagonal closed form. p. 0 A hardware-efficient variational ansatz with an exact diagonal metric for real- and
imaginary-time evolution and Haar sampling
Dario Picozzi1, 2, ∗
1Department of Physics and Astronomy, University College London (UCL),
Gower Street, London, WC1E 6BT, United Kingdom
2London Centre for Nanotechnology, 19 Gordon St, London, WC1H 0AH, United Kingdom
Variational quantum algorithms depend on the geometry of their parametrised circuits: metricaware optimisation and time evolution require the Fubini–Study metric, which has hitherto demanded costly auxiliary measurements and ill-conditioned inversions. This work introduces a
hardware-efficient n-qubit ansatz, which parametrises states by a binary tree and whose Fubini–
Study pullback metric is diagonal in closed form. The diagonal metric first raises this covector by the Hadamard reciprocal g
−1
, giving
the metric gradient ∇gE = g
−1dE, which is already the entire imaginary-time / natural-gradient update,
λ˙
imag = − g
−1
dE = − ∇gE. (6)
Real-time evolution uses the same raised gradient, but rotated by the K¨ahler complex structure: the binary tree
applies J through one bottom-up and one top-down pass over the active split tree, supplying the 90◦
turn that selects
Hamiltonian rather than gradient motion. The diagonal metric first raises this covector by the Hadamard reciprocal g
−1
, giving
the metric gradient ∇gE = g
−1dE, which is already the entire imaginary-time / natural-gradient update,
λ˙
imag = − g
−1
dE = − ∇gE. (6)
Real-time evolution uses the same raised gradient, but rotated by the K¨ahler complex structure: the binary tree
applies J through one bottom-up and one top-down pass over the active split tree, supplying the 90◦
turn that selects
Hamiltonian rather than gradient motion. We show that the metric admits the diagonal closed form via a direct calculation that
uses no Hadamard tests; that an exact four-term parameter-shift rule yields analytic Euclidean gradients on the
same circuit at four evaluations per parameter; and that, when the target state is sparsely supported on k ≪ 2
n
computational-basis states, the redundant parameters of the full tree carry a gauge freedom made explicit through an
active/fixed/inactive node classification and exploit to compile pruned circuits whose CNOT count provably grows
linearly with k (at most 2nAs = O(n
2k) for any ordering, with the tight constant conjectured). The diagonal metric
has one further consequence, exploited throughout: drawing the parameters from the Fubini–Study volume measure
dµFS ∝
p
det g dλ (1)
reproduces the unitary-invariant (Haar) ensemble on a chosen symmetry sector exactly and rejection-free, the diagonal
metric reducing the draw to one independent sample per tree node. Quantum natural gradient on the tree parameters,
variational imaginary- and real-time evolution, and exact unitary-invariant (Haar) sampling on a
symmetry sector run with no auxiliary metric circuits or matrix inversion. When the target state
is supported on a subspace of k computational-basis states, the redundant tree parameters carry
a gauge freedom a pruning compiler converts into circuits whose two-qubit count provably grows
linearly in k; a variant reaches near-optimal O(nk/ log n) scaling with the closed-form metric intact. As proved in Appendix S, these bare computations are classically simulable in the active-leaf count, so their
accuracy can be verified exactly; they establish the ansatz as an efficient, exactly controllable primitive whose composition with a hard dressing is where a quantum–classical separation can arise. Composing that primitive with
a non-classically-simulable dressing U(ϕ) then yields several further applications: dressed molecular ground states
(Sec.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
The derivation of the gradient evaluation shows that the real-target tree gradient costs 2 theta energy evaluations, and one quantum natural-gradient or imaginary-time step is computable from 2P+1 evaluations instead of 4P+1. p. 6 One full Euclidean energy gradient therefore costs 4|θ| + 2|ω| circuit evaluations
exactly, so one full quantum-natural-gradient step is computable from at most 4P + 1 energy evaluations, where P is
the number of variational parameters. When the prepared state and the observable are real, the amplitude-only chart ω = 0 that underlies ground-state
search and imaginary-time evolution, each tree derivative collapses further to a two-evaluation rule [30, 31],
∂θiE = E˜(θi +
π
4
) − E˜(θi −
π
4
), (11)
where E˜(θi ±
π
4
) is the expectation of H in the state the circuit prepares at the shifted angle, followed by a fixed
gate U
(±)
0 = exp(±i
π
4
Pi), with Pi = |bi⟩⟨bi
|
ctrl ⊗ I the control projector of Eq. This U
(±)
0
is diagonal in the
computational basis and acts only on the ancestor qubits, and is realised by the same Rz phase block the complex
ansatz already compiles, truncated above the target qubit qi
; it therefore adds at most the linear-in-k CNOT overhead
of the pruned ansatz, with one such block per derivative. The full real-target gradient then costs 2|θ| evaluations and
one natural-gradient step 2P + 1, half the generic figure; the reduction is specific to the real chart and does not apply
to the complex-amplitude real-time runs, which retain the four-term rule. For all but a handful of fine-tuned ans¨atze, however, the metric is
dense, must be estimated by O(P
2
) auxiliary Hadamard tests, and is then regularised by a pseudoinverse on the
classical side, an overhead that scales with the parameter count P and ultimately limits which problems are reachable
in practice [6–9]. Here we cut this overhead away at its source by designing an ansatz whose Fubini–Study pullback metric on the tree
coordinates is diagonal in closed form, so that the linear system reduces to elementwise division and the auxiliary metric
circuits disappear entirely. One full Euclidean energy gradient therefore costs 4|θ| + 2|ω| circuit evaluations
exactly, so one full quantum-natural-gradient step is computable from at most 4P + 1 energy evaluations, where P is
the number of variational parameters. When the prepared state and the observable are real, the amplitude-only chart ω = 0 that underlies ground-state
search and imaginary-time evolution, each tree derivative collapses further to a two-evaluation rule [30, 31],
∂θiE = E˜(θi +
π
4
) − E˜(θi −
π
4
), (11)
where E˜(θi ±
π
4
) is the expectation of H in the state the circuit prepares at the shifted angle, followed by a fixed
gate U
(±)
0 = exp(±i
π
4
Pi), with Pi = |bi⟩⟨bi
|
ctrl ⊗ I the control projector of Eq.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Metricaware optimization can be performed without costly auxiliary measurements or ill-conditioned inversions because an exact diagonal metric for real- and imaginary-time evolution is in closed form, reducing the linear system to elementwise division.
-
Variational quantum algorithms can utilize a natural gradient update that requires only one explicit Euler step of the real-time flow, which is exactly the
real-time twin
of one quantum natural-gradient or imaginary-time step, avoiding dense matrix assembly or classical inversion. -
A pruning compiler can convert redundant tree parameters into circuits whose two-qubit count provably grows linearly in k, achieving near optimal O(nk/ log n) scaling when the target state is supported on a subspace of k computational-basis states.
-
The exact Fubini–Study sampling on a symmetry sector can be performed exactly and without rejection using the diagonal metric, reducing the draw to one independent sample per tree node at O(k) cost per sample even when the sector is exponentially large.
-
A two-block construction allows for decoupled state preparation and training, where the core parameters maintain an exactly diagonal metric while a dressing block handles complex dynamics or correlation outside the active space, enabling training in a way that isolates trainability from hardness.
Abstract
Variational quantum algorithms depend on the geometry of their parametrised circuits: metric-aware optimisation and time evolution require the Fubini--Study metric, which has hitherto demanded costly auxiliary measurements and ill-conditioned inversions. This work introduces a hardware-efficient n-qubit ansatz, which parametrises states by a binary tree and whose Hilbert-space pullback metric is diagonal in closed form. Quantum natural gradient on the tree parameters, variational imaginary- and real-time evolution, and exact unitary-invariant (Haar) sampling on a symmetry sector run with no auxiliary metric circuits or matrix inversion. When the target state is supported on a subspace of k computational-basis states, the redundant tree parameters carry a gauge freedom a pruning compiler converts into circuits whose two-qubit count provably grows linearly in k; a variant reaches near-optimal O(nk/ n) scaling with the closed-form metric intact. On electronic-structure calculations for small molecules and half-filled Hubbard quench dynamics, the method reaches reference-level accuracy with one to three orders of magnitude fewer two-qubit gates than leading alternatives. Interchangeable constructions (a Schur-transform dressing or internal reparameterisations) make the ansatz exactly spin-adapted, with fixed total spin at every parameter and no penalty terms. The bare ansatz is an exactly controllable, well-conditioned and barren-plateau-free primitive for preparing and sampling sector states: on its own, it is classically simulable in k (a boundary proved for a general class of sector-sparse ansätze); composed with a classically hard dressing, it yields molecular ground states, sector-Haar benchmarking, thermal correlators, and exact effective Hamiltonians trained from energy measurements alone, with the composed circuit carrying the potential for quantum advantage.
Sources
- On the practical usefulness of the Hardware Efficient Ansatz
- Transformation of quantum states using uniformly controlled rotations
- Geometrical Formulation of Quantum Mechanics
- Equating quantum imaginary time evolution, Riemannian gradient flows, and stochastic implementations
- A Quantum Approximate Optimization Algorithm
- Nearly Optimal Circuit Size for Sparse Quantum State Preparation
- Optimal Circuit Size for Fixed-Hamming-Weight Quantum States Preparation
- Periodic Symmetry-Adapted Encoding: Qubit Reduction in Crystalline Electronic Structure
- Tapering off qubits to simulate fermionic Hamiltonians
- Symmetry-adapted qubit encoding with complete active space and Bravyi--Kitaev mapping for quantum chemistry on a quantum computer
- The Quantum Schur Transform: I. Efficient Qudit Circuits
- The Quantum Paldus Transform: Efficient Circuits with Applications
- Trainability and Expressivity of Hamming-Weight Preserving Quantum Circuits for Machine Learning
- Simulating quantum computers with probabilistic methods
- Lie-algebraic classical simulations for quantum computing
- Does provable absence of barren plateaus imply classical simulability?
- Fast and converged classical simulations of evidence for the utility of quantum computing before fault tolerance
- Classically estimating observables of noiseless quantum circuits
- The Church of the Symmetric Subspace
- Simulating Quantum Circuits with Sparse Output Distributions
Related papers
- Computational regimes in matrix-product-state-based quantum trajectory simulations
- Toward chemically accurate quantum simulations on quantum hardware: A quantum-HPC hybrid approach
- Transversal surface-code game with reconfigurable qubits
- Fulqrum: a generalized framework for quantum subspace diagonalization
- Flux-Activated Resonant Control of a Bosonic Quantum Memory
- Machine learning the effects of many quantum measurements