Phase Estimation via Compressed Time Evolution

• The paper introduces compressed controlled time evolution to overcome high resource costs in traditional quantum phase estimation by minimizing circuit depth and control overhead.
• It details sequential Hadamard-test, direct and imaginary-time phase-gradient, and tensor-network methods to enable hardware-efficient implementations on near-term quantum devices.
• The approaches leverage symmetry, physical locality, and algebraic structures to perform scalable spectral analysis in applications like quantum chemistry and many-body physics.

Phase estimation with compressed controlled time evolution addresses a central challenge in quantum simulation: the high resource cost of implementing controlled time-evolution operators as required for quantum phase estimation (QPE) protocols. By compressing or locally controlling time-evolution and circumventing or reducing the depth of controlled operations, this family of methods enables phase estimation and spectral analysis of large quantum many-body systems on near-term devices. The approaches span circuit-level compression, phase-tracking and gradient estimation, tensor-network methods, and algebraic compression, each adapting measurement strategies and circuit constructions to maximize hardware efficiency.

1. Fundamentals of Compressed Controlled Time Evolution in Phase Estimation

Traditional QPE algorithms, including textbook QFT-based QPE, require controlled application of the global unitary time-evolution operator $U(t)=e^{-iHt}$ or its powers—typically as controlled-$U^{2^k}$ gates—using deep circuits and multiple ancillas. The exponential cost in circuit depth and control fidelity is a bottleneck for practical implementation on noisy intermediate-scale quantum (NISQ) or early fault-tolerant devices.

Compressed controlled time evolution subsumes a class of protocols that:

• Replace or approximate controlled-$U(t)$ with shallower (often locally-controlled) circuits
• Extract phase or spectral information through recursive phase tracking, gradient computations, or alternative measurement primitives not reliant on global control
• Exploit symmetry, physical locality, and mathematical structure (e.g., translational invariance, tensor-network representations, Lie algebraic properties) to minimize control overhead and maximize hardware efficiency (Schiffer et al., 23 Jun 2025, Karacan, 26 Nov 2025, Kanno et al., 2024, Kökcü et al., 2023).

The core objective remains the estimation of eigenphases and spectral properties via protocols that scale polynomially (or even logarithmically) in system size $n$, time $t$, target precision $\epsilon$, and achieve sample complexity and circuit depth compatible with available quantum hardware.

2. Key Methods for Phase Estimation with Compressed Control

Three primary hardware-efficient protocol classes have been established (Schiffer et al., 23 Jun 2025):

(a) Sequential Hadamard-Test with Local Control

• Goal: Compute $g_L = \langle \psi | U_L | \psi \rangle$ and extract its phase $\phi_L = \arg g_L$ for $U_L = u_L\ldots u_2 u_1$, representing any $L$-gate time evolution or circuit.
• Method: Instead of globally controlled-$U_L$, control only the $L$-th gate at each step:
  1. Prepare ancilla in $|+\rangle$, system in $|\psi\rangle$
  2. Apply $U_{L-1}$ unconditionally to the system
  3. Apply $u_L$ controlled by the ancilla, measure $\sigma^x$ and $\sigma^y$ to obtain $\mathrm{Re}(g_L g_{L-1}^*)$ and $\mathrm{Im}(g_L g_{L-1}^*)$
  4. Recursively reconstruct $g_L$ and accumulate phase increments: $\Delta\phi_L = \arg(g_L g_{L-1}^*)$, $$\phi_L = \sum_{\ell=1}^L (-1)^{L-\ell} \Delta\phi_\ell$$.

• Resource scaling: Circuit depth is the depth of the local gate; sample complexity $$N_{\rm seq} = O(N_{\rm gates}^2 \epsilon^{-2} r_{\min}^{-2})$$ (Schiffer et al., 23 Jun 2025).

(b) Direct Phase-Gradient (Hamiltonian Phase) Method

• Goal: Track the instantaneous phase velocity $\phi'(t)$ of $g(t) = \langle \psi | e^{-iHt} | \psi \rangle$.
• Method:
  • Measure the expectation of local Pauli terms individually via circuits with locally controlled $P_j$:
  • $\phi'(t) = -\sum_j \lambda_j a_j(t)/r^2(t)$,
  • with $$a_j(t) = \frac{1}{2} \langle \psi(t)|\{P_j,|\psi\rangle\langle\psi|\}|\psi(t)\rangle$$ and $r(t) = |\langle \psi|e^{-iHt}|\psi\rangle|$.
  • Numerically integrate $\phi'(t)$ over a time grid to obtain $\phi(t)$.
• Resource scaling: Sample complexity $$N_{\rm dir} = O((n t/\epsilon)^{2+1/s} r_{\min}^{-4})$$, with $s$ the quadrature order (Schiffer et al., 23 Jun 2025).

(c) Imaginary-Time Phase-Gradient Method

• Goal: Utilize the analyticity of $g(z)$ with $z = t - i \beta$.
• Method: Finite-difference formula using survival probabilities after imaginary-time evolution:

$$\phi'(t) \approx [\ln r(t - i\tau) - \ln r(t + i\tau)]/(2\tau)$$

Implement $e^{\pm H \tau}$ via local block encodings and postselect on ancilla $|0\rangle$ outcomes.

• Resource scaling: $$N_{\mathrm{ITE}}=O(n^{2+1/s}(t/\epsilon)^{3+3/(2s)}\,r_{±,min}^{-2})$$ (Schiffer et al., 23 Jun 2025).

3. Compression Protocols and Circuit Construction

Efficient circuit-level compression transforms the implementation of controlled-$U(t)$, yielding dramatic resource reductions:

(a) Translationally-Invariant Compression (TICC)

• Decompose $H=\sum_{j} H_j$ with associated anticommuting Pauli strings $K_j$ and construct:

$$\mathcal{C}\text{-}U(t) = |0\rangle_a\!\langle0|\otimes I + |1\rangle_a\!\langle1|\otimes e^{-iHt}$$

• By leveraging TI and locality, use a brickwall circuit with two parameter sets to realize compressed versions of $e^{\pm iHt/2}$, and control only selected layers rather than every gate.
• Circuit depth: $O(t\,\mathrm{polylog}(tN/\epsilon))$ plus additive control overhead, asymptotically near-optimal (Karacan, 26 Nov 2025).

(b) Tensor-Network (MPO) Compression

• Construct a brick-wall circuit approximating $e^{-iH\Delta t}$ by optimizing over MPO representations with constrained bond dimension and SVD truncation.
• Controlled operations are embedded by promoting each block to a controlled version with respect to the ancilla.
• Circuit depth is determined by the MPO’s bond dimension and the brick-wall depth $d_{\rm evol}$; practical implementations achieve $>10\times$ gate-count compression compared to naive approaches (Kanno et al., 2024).

(c) Lie Algebraic Compression for Free Fermion Systems

• For quadratic Hamiltonians, exploit block algebra, triangle $\rightarrow$ square, and “diamond” compression, yielding fixed-depth circuits—$O(n)$ depth and $O(n^2)$ CNOTs— independent of evolution time $t$ (Kökcü et al., 2023).

4. Sample Complexity, Error Analysis, and Trade-offs

All phase-tracking and compression-based routines necessarily trade circuit depth (and often control overhead) for increased shot/sample complexity and classical postprocessing.

• Sample Complexity: For sequential and gradient methods, total measurement cost scales polynomially with system size, simulation time, and inverse error threshold, e.g., $$N_{\rm seq} = O(\log(1/\delta) (n t/\epsilon)^{2+2/p} r_{\min}^{-2})$$ (Schiffer et al., 23 Jun 2025).
• Integration Errors: For integration-based gradient methods, quadrature error is controlled by the order and fineness of the discretization grid.
• Statistical Errors: Add in quadrature across increments or time slices; error per phase step (sequential) is bounded as $$\mathrm{Var}[\Delta\phi_\ell] \leq (1/(2M_\ell))(1/r_{\ell-1}^2+1/r_\ell^2)$$ (Schiffer et al., 23 Jun 2025).
• Postprocessing: Classical reconstruction (phase addition, numerical quadrature) is computationally negligible compared to quantum runtime.

The decisive gain is in circuit depth and device suitability: compressed/locally-controlled protocols admit constant- or $O(\log n)$-depth implementations for phase increments, versus $O(n t)$ for fully controlled global unitaries.

5. Implementation Considerations and Applications

Device Architecture:

• These approaches are aligned for devices with restricted control connectivity, shallow circuit depth tolerance, and limited multi-qubit control capabilities (Schiffer et al., 23 Jun 2025, Karacan, 26 Nov 2025, Kanno et al., 2024).
• MPO compression enables state-of-the-art simulations on $>30$ qubits with exponentially suppressed noise pedestals (Kanno et al., 2024).

Algorithmic Integration:

• Compressed controlled evolution is used as a subroutine in iterative QPE/iterative phase search and for spectral analysis protocols.
• TICC has been demonstrated in IQPE on large frustrated spin systems (e.g., 6×6 triangular lattice) with gate counts (414–828 CNOTs) that match early hardware capabilities (Karacan, 26 Nov 2025).
• Tensor-network compression enabled gap calculations for extended Hubbard chains on cloud devices and molecular simulations with up to 17 qubits for hydrocarbons (Kanno et al., 2024).

Generalization Limits:

• Translational invariance and k-locality are generally required for the strongest compression (TICC); systems with disorder or long-range interactions complicate this structure (Karacan, 26 Nov 2025).
• For generic Hamiltonians, local block encodings or gradient-based approaches may still apply but with reduced compression.

6. Comparison with Alternative and Ancilla-Free Strategies

Several protocols achieve related or complementary goals without fully controlled time evolution:

Approach	Quantum Resource	Compression Mechanism	Sample Complexity
Local-control QPE/Gradient	One ancilla, local gates	Sequential/gradient phase tracking	$O((n t/\epsilon)^{2+1/p})$
TICC (TI Hamiltonians)	One ancilla	Circuit param/Pauli ref. trick	$O(t\,\mathrm{polylog}(tN/\epsilon))$
Tensor-network (MPO) compression	One ancilla	MPO/Brickwall structure	Gate count $O(N d_{\rm evol} t/\Delta t)$
Algebraic compression (free ferm.)	One ancilla	Block/diamond circuit	Gate count $O(n^2)$
Ancilla-free phase retrieval	None	Vectorial/2D PR, classical opt.	$O(NM/\epsilon^2)$

Ancilla-free phase retrieval reconstructs phase information via classical postprocessing of magnitude-only time series data from overlap measurements, leveraging interference or two-dimensional methods (Clinton et al., 2024). Trade-offs include higher classical cost, larger data sets, and stricter requirements on spectral support and noise. These methods remove all multi-qubit control and are particularly advantageous for NISQ-era devices with minimal ancilla overhead at the expense of increased measurement and classical postprocessing (Clinton et al., 2024).

7. Outlook and Open Problems

Phase estimation protocols using compressed controlled time evolution constitute a practical pathway for large-scale quantum spectral estimation on hardware-limited devices. They have achieved:

• Near-optimal circuit depth for time evolution under translationally-invariant local Hamiltonians (Karacan, 26 Nov 2025)
• Demonstrated feasibility for quantum chemistry, strongly correlated lattices, and large-scale Hubbard chains (Kanno et al., 2024)
• Eliminated all-to-all control by recasting phase estimation in locally controlled, shallow-depth, or even ancilla-free circuit primitives

Ongoing challenges include:

• Extending these compression strategies to non-TI, disordered, or highly nonlocal Hamiltonians
• Bounding the convergence and performance guarantees of optimization-based circuit compression (randomized QAOA ansatz, PEPS/Tensor-Tree alternatives)
• Enhancing phase-tracking and phase-gradient robustness at low overlap or in regimes of high spectral crowding
• Integrating compressed controlled time evolution protocols with error correction and advanced hardware-native gate sets

Collectively, these advances will enable phase estimation and quantum simulation tasks of systems at and beyond classical computational reach (Schiffer et al., 23 Jun 2025, Karacan, 26 Nov 2025, Kanno et al., 2024, Kökcü et al., 2023, Clinton et al., 2024).