betaFeedback

WorksPhysics and Astronomy

Success Probability of Block-Encoding Methods

Abstract

Block-encoding methods apply nonunitary functions f(H~)f(\widetilde H) of a normalized Hermitian operator on a quantum computer, but only upon a successful postselection, which introduces a shot overhead rarely included in resource estimates. We analyze the success probability psuccess(f)=∑nwn∣f(E~n)∣2p^{(f)}_{\mathrm{success}}=\sum_n w_n|f(\tilde E_n)|^2 of three constructions, namely linear combination of unitaries (LCU), qubitization and quantum singular value transformation (QSVT), where wnw_n are the weights of the input state on the eigenstates of H~\widetilde H. Using the one-dimensional Hubbard model with up to 14 sites (35 qubits), we show that for LCU, the success probability is bounded by the squared ratio between the spectral norm and the 11-norm of the Pauli decomposition, and that for a Hartree–Fock input it plateaus with system size while the ground-state overlap keeps decreasing. For qubitization, it oscillates with the Chebyshev degree, and favorable degrees can be identified from a few dominant spectral components or predicted from shallow-degree data. For QSVT, the polynomial normalization constant directly sets the success probability, and a tight certified bound balances it against phase-finding robustness. An application to quantum Krylov methods shows a nonuniform shot overhead across Krylov moments. The success probability should therefore be part of the resource estimation of block-encoding-based algorithms.

Introduction

Block-encoding methods provide a general route for applying nonunitary transformations of an operator HH on quantum computers. Such transformations are needed in many applications, such as solving linear systems of equations [1, 2, 3], solving partial differential equations [4], or solving the Schrödinger equation for strongly correlated many-electron systems, e.g., using quantum Krylov subspace methods [5, 6, 7, 8, 9, 10, 11, 12]. In the latter case, the task is to determine the low-energy spectrum of the Hamiltonian HH of the electronic system, in particular its ground-state energy, which can also be targeted by block-encoding-based ground-state preparation and energy estimation [13, 14, 15]. Quantum Krylov subspace methods address this problem by constructing a subspace from repeated applications of HH to an input state. Because the Hamiltonian HH is not unitary, these applications cannot be realized directly on a quantum computer.

Let us first consider a block-encoding of a general Hermitian operator HH. One constructs a unitary UU acting on the system together with an ancilla register aa, such that a designated block of UU (the top-left block, selected by projecting the ancilla onto ∣0⟩a\lvert0\rangle_a) satisfies [16]

(⟨0∣a⊗I)U(∣0⟩a⊗I)=H~,(1)(\langle0|_a \otimes I)U(|0\rangle_a \otimes I)=\widetilde{H}, \tag*{(1)}

where

H~=H/λ(2)\widetilde{H}=H/\lambda \tag*{(2)}

is a renormalized version of HH such that its spectral norm is less than or equal to one. H~\widetilde{H} can also include, when necessary, a shift. The application of H~\widetilde{H} to an input state ∣ψ⟩|\psi\rangle is recovered when the ancilla register is measured in the ∣0⟩a|0\rangle_a state. The probability of obtaining this outcome, which we refer to as the success probability, is a central feature investigated in this work.

The Linear Combination of Unitaries (LCU) framework [16, 17, 18] provides a constructive way to build such a block-encoding. Writing the Hermitian operator HH as a weighted sum of Pauli strings PjP_j,

H=∑jαjPj,(3)H=\sum_j \alpha_j P_j, \tag*{(3)}

the LCU block-encoding is assembled from two operations acting on an ancilla register of nan_a qubits. A “prepare” operator creates a coefficient-weighted superposition on the ancilla,

PREP∣0⟩a=∑j∣αj∣λ∣j⟩a≡∣PREP⟩,(4)\mathrm{PREP}|0\rangle_a=\sum_j\sqrt{\frac{|\alpha_j|}{\lambda}}|j\rangle_a\equiv|\mathrm{PREP}\rangle, \tag*{(4)}

and a “select” operator applies each Pauli string unitary conditioned on the ancilla index,

SELECT=∑j∣j⟩⟨j∣⊗sgn⁡(αj)Pj,(5)\mathrm{SELECT}=\sum_j|j\rangle\langle j|\otimes\operatorname{sgn}(\alpha_j)P_j, \tag*{(5)}

where we write a=0a=0 for the measurement outcome in which all ancilla qubits are found in ∣0⟩|0\rangle. Together, these operations directly give a block-encoding satisfying Eq. (1) through

U=PREP†SELECT PREP.(6)U=\mathrm{PREP}^{\dagger}\mathrm{SELECT}\,\mathrm{PREP}. \tag*{(6)}

An important point is that the necessary unitarity of the preparation operator in Eq. (4) imposes that the renormalization factor equals the so-called “LCU 1-norm”

λ=∑j∣αj∣(LCU 1-norm).(7)\lambda=\sum_j|\alpha_j|\quad\text{(LCU 1-norm)}. \tag*{(7)}

Further refinements of the block-encoding construction give access to increasingly general functions of the Hermitian operator HH. For instance, qubitization builds on this same prepare and select pair: combining SELECT with a reflection about the ∣PREP⟩|\mathrm{PREP}\rangle state, RPREP=2∣PREP⟩⟨PREP∣−IR_{\mathrm{PREP}} = 2|\mathrm{PREP}\rangle\langle\mathrm{PREP}| - I, gives the quantum walk operator W=RPREP SELECTW = R_{\mathrm{PREP}}\ \mathrm{SELECT} used here [16]. Powers of WW generate Chebyshev polynomials of H~\widetilde{H},

(⟨PREP∣⊗I)Wk(∣PREP⟩⊗I)=Tk(H~),(8)(\langle\mathrm{PREP}| \otimes I)W^k(|\mathrm{PREP}\rangle\otimes I) = T_k(\widetilde{H}), \tag*{(8)}

so that a single block-encoding form gives access to an entire family of transformations indexed by a degree kk. The LCU case is recovered at k=1k = 1, since T1(x)=xT_1(x) = x. Finally, Quantum Singular Value Transformation (QSVT) [3, 19, 20] generalizes this construction to a broad class of polynomial transformations P(H~)P(\widetilde{H}) of definite parity, renormalized by a factor C≥sup⁡x∈[−1,1]∣P(x)∣C \ge\sup_{x\in[-1,1]} |P(x)| so that ∣P(x)/C∣≤1|P(x)/C| \le1 on [−1,1][-1,1]. For a polynomial of degree dd and a phase sequence Φ=(φ0,φ1,…,φd)\Phi= (\varphi_0, \varphi_1, \ldots, \varphi_d), we define the phase-modulated walk-operator product

UΦ=Z(φ0) W Z(φ1) W ⋯ W Z(φd),Z(φ)=eiφRPREP,(9)\begin{aligned} U_\Phi&= Z(\varphi_0)\,W\,Z(\varphi_1)\,W\,\cdots\,W\,Z(\varphi_d),\\ Z(\varphi) &= e^{i\varphi R_{\mathrm{PREP}}}, \tag*{(9)} \end{aligned}

where the walk operator WW is applied dd times and RPREPR_{\mathrm{PREP}} is the reflection introduced above. Considering real polynomials, so that P(H~)P(\widetilde{H}) is Hermitian, QSVT guarantees that there exists a phase sequence Φ\Phi such that [3]

(⟨PREP∣⊗I)UΦ(∣PREP⟩⊗I)=P(H~)C.(10)(\langle\mathrm{PREP}| \otimes I)U_\Phi(|\mathrm{PREP}\rangle\otimes I) = \frac{P(\widetilde{H})}{C}. \tag*{(10)}

The phases Φ\Phi are determined classically beforehand from the coefficients of P/CP/C [21, 22, 23]. Qubitization and QSVT block-encodings thus give access to different encoded functions f(H~)f(\widetilde{H}).

The three constructions discussed above, i.e., LCU, qubitization, and QSVT block-encodings, form the scope of this work. They correspond to f=If = I (the identity, f(x)=xf(x) = x) for LCU, f(x)=Tk(x)f(x) = T_k(x) for qubitization, and f(x)=P(x)/Cf(x) = P(x)/C for QSVT. An important shared property of these three constructions is that the desired application of f(H~)f(\widetilde{H}) to an input state ∣ψ⟩|\psi\rangle is obtained only upon a successful postselection of the ancilla register. The corresponding success probabilities are denoted by

psuccess(I)(LCU),psuccess(Tk)(qubitization),psuccess(P/C)(QSVT),(11)\begin{aligned} p_{\mathrm{success}}^{(I)} &\quad\text{(LCU)},\\ p_{\mathrm{success}}^{(T_k)} &\quad\text{(qubitization)},\\ p_{\mathrm{success}}^{(P/C)} &\quad\text{(QSVT)}, \tag*{(11)} \end{aligned}

and psuccess(f)p_{\mathrm{success}}^{(f)} refers to a generic encoded function. Throughout, lowercase pp denotes a success probability, while PP denotes a polynomial and PjP_j a Pauli string. For a direct postselection protocol, a success probability psuccess(f)p_{\mathrm{success}}^{(f)} implies on average an overhead of 1/psuccess(f)1/p_{\mathrm{success}}^{(f)} in the number of circuit repetitions, or shots.

Understanding the success probability in detail is not straightforward, since it is determined by more than the operator HH itself. Indeed, it depends on the spectral properties of the renormalized operator, the considered function f(H~)f(\widetilde{H}), and also the renormalization used to construct the block-encoding and the input state to which the f(H)f(H) is applied. We write the spectral decomposition of the Hamiltonian, of its normalized form and of the input state as

H∣En⟩=En∣En⟩,H~∣En⟩=E~n∣En⟩,∣ψ⟩=∑nβn∣En⟩,wn=∣βn∣2,(12)\begin{aligned} H|E_n\rangle&= E_n|E_n\rangle, \qquad\widetilde{H}|E_n\rangle= \widetilde{E}_n|E_n\rangle,\\ |\psi\rangle&= \sum_n \beta_n|E_n\rangle, \qquad w_n = |\beta_n|^2, \tag*{(12)} \end{aligned}

where EnE_n are the eigenvalues of HH, ordered so that E0E_0 is the ground-state energy in the half-filled sector to which all input states considered here belong. We denote by Emin⁡E_{\min} and Emax⁡E_{\max} the extremal eigenvalues of HH over the full Fock space of the 2N2N system qubits, on which the block-encoding acts. E~n∈[−1,1]\widetilde{E}_n \in[-1,1] corresponds to normalized eigenvalues of H~\widetilde{H} (E~n=En/λ\widetilde{E}_n = E_n/\lambda, or E~n=(En−Emax⁡)/λs\widetilde{E}_n = (E_n - E_{\max})/\lambda_s if a shift is introduced, with λs\lambda_s the 1-norm of the shifted decomposition). Interestingly, the block-encoding implementing an encoded function f(H~)f(\widetilde{H}) succeeds with probability

psuccess(f)=∥f(H~)∣ψ⟩∥2=∑nwn∣f(E~n)∣2.(13)p_{\mathrm{success}}^{(f)} = \|f(\widetilde{H})|\psi\rangle\|^2 = \sum_n w_n |f(\widetilde{E}_n)|^2. \tag*{(13)}

Equation (13) provides the common spectral framework used throughout this work: for a given encoded function ff, the success probability is determined jointly by the spectral weights of the input state, the spectrum of the normalized Hamiltonian, and the normalization required to implement ff.

Resource estimates for block-encodings of electronic Hamiltonians have focused on reducing the coefficient 1-norm λ\lambda, through factorized decompositions [24, 25, 26, 27, 28], orbital transformations [29] or Hamiltonian partitionings [30], in order to reduce the circuit depth; to our knowledge, the shot overhead due to the success probability has usually not been reported. In this article, we analyze this success probability for the three constructions above, using the one-dimensional Hubbard model as a controlled testbed, with the Hartree–Fock (HF) state as main input state, compared with the exact ground state and random states. To reach system sizes beyond those accessible with a dense Hamiltonian, we evaluate the action of HH on state vectors directly from its Pauli decomposition, which reduces the memory scaling from O(4nsys)O(4^{n_{\mathrm{sys}}}) to O(2nsys)O(2^{n_{\mathrm{sys}}}) and allows us to study up to 14 sites (28 system qubits, 35 qubits including the LCU ancilla register). For each construction we address one question: for LCU, how much the 1-norm limits the success probability; for qubitization, whose success probability oscillates with the degree kk, whether favorable degrees can be identified without computing the full spectrum, with an application to quantum Krylov methods; and for QSVT, how the polynomial scaling factor CC trades success probability against the robustness of phase finding. For the latter, we use a certified upper bound CMarkovC_{\mathrm{Markov}} on the supremum of the target polynomial, obtained from grid samples through the Markov inequality; to our knowledge, such a certificate has not previously been used to normalize QSVT targets.

The article is structured as follows. Section II summarizes how the LCU 1-norm, the features of HH and the input state determine the success probability. Section III presents the numerical results on the Hubbard model: the computational setup (Sec. III A), LCU (Sec. III B), qubitization, including degree selection and the application to quantum Krylov (Sec. III C), and QSVT (Sec. III D). We conclude in the last section.

Summary of main findings

This section collects the analytical results that organize our findings; the numerical evidence is given in Sec. III. The three constructions differ in encoded function, implementation primitive and normalization (Table I). Within Eq. (13), the success probability then depends on the normalization (Sec. II A), on the features of HH and the system size (Sec. II B), and on the input state (Sec. II C).

Dependence of the success probability on the LCU 1-norm

We consider a Pauli decomposition as in Eq. (3), and use the notations above, together with the spectral norm of HH (shifted if necessary),

Ω=∥H∥,λ≥Ω.(14)\Omega= \lVert H\rVert,\qquad\lambda\ge\Omega. \tag*{(14)}

For a fixed input state ∣ψ⟩\lvert\psi\rangle as in Eq. (12), the success probability of a general block-encoding [Eq. (1)] increases as the renormalization factor λ\lambda (which in this general setting is not necessarily the LCU 1-norm) approaches the spectral norm Ω\Omega of HH [Eq. (14)]. Replacing the LCU 1-norm [Eq. (7)] by the tighter Ω\Omega therefore provides an upper benchmark for the achievable success probability [24]. The corresponding success probabilities under the two renormalization choices are

psuccess, 1-norm(I)=1λ2∑nwnEn2,p^{(I)}_{\mathrm{success,\,1\text{-}norm}} = \frac{1}{\lambda^2}\sum_n w_n E_n^2,
psuccess, spectral(I)=1Ω2∑nwnEn2.(15)p^{(I)}_{\mathrm{success,\,spectral}} = \frac{1}{\Omega^2}\sum_n w_n E_n^2. \tag*{(15)}

where EnE_n are the eigenvalues of HH before renormalization. This leads to:

psuccess, spectral(I)psuccess, 1-norm(I)=λ2Ω2≥1.(16)\frac{p^{(I)}_{\mathrm{success,\,spectral}}}{p^{(I)}_{\mathrm{success,\,1\text{-}norm}}} = \frac{\lambda^2}{\Omega^2} \ge1. \tag*{(16)}

A standard technique is to pin the spectral maximum to zero [16]:

Hs≡H−Emax⁡I,Emax⁡=max⁡nEn.(17)H_s \equiv H-E_{\max}I,\qquad E_{\max}=\max_n E_n. \tag*{(17)}

After this shift all eigenvalues lie in [Emin⁡−Emax⁡,0][E_{\min}-E_{\max},0] and Ωs=Emax⁡−Emin⁡\Omega_s=E_{\max}-E_{\min}, the spectral norm of HsH_s over the full Fock space. For the Hubbard model, Emin⁡=E0E_{\min}=E_0 at weak coupling, but beyond a crossover (U=3tU=3t for N=2N=2, between tt and 2t2t for N=4N=4) the global minimum lies in a sector with fewer electrons, which avoids the double-occupancy cost, so that Emin⁡<E0E_{\min}<E_0. Since all input states belong to a single particle-number sector, the normalization could in principle be tightened by symmetry-adapted shifts that leave this sector unchanged, such as the block-invariant symmetry shift of Ref. [31]. The spectral norm Ωs\Omega_s is the smallest normalization compatible with ∥H~∥2≤1\lVert\widetilde{H}\rVert_2\le1, leading to ∥H~∥2=1\lVert\widetilde{H}\rVert_2=1. We thus have psuccess, spectral(I)≥psuccess, 1-norm(I)p^{(I)}_{\mathrm{success,\,spectral}}\ge p^{(I)}_{\mathrm{success,\,1\text{-}norm}} and psuccess, spectral(I)=1p^{(I)}_{\mathrm{success,\,spectral}}=1 iff ∣ψ⟩=∣Emin⁡⟩\lvert\psi\rangle=\lvert E_{\min}\rangle (up to a phase).

The PREP-SELECT construction [Eq. (6)] related to the LCU block-encoding specific form, however, constrains the usage of the LCU 1-norm, i.e., λs=∑j∣αj∣\lambda_s=\sum_j\lvert\alpha_j\rvert of the shifted decomposition, because the PREP operator must be unitary. The physically realized success probability is therefore always

psuccess(I)=∥Hs∣ψ⟩∥2λs2=(Ωsλs)2∥Hs∣ψ⟩∥2Ωs2≤(Ωsλs)2≡pbound,(18)\begin{aligned} p^{(I)}_{\mathrm{success}}=\frac{\lVert H_s\lvert\psi\rangle\rVert^2}{\lambda_s^2}=\left(\frac{\Omega_s}{\lambda_s}\right)^2\frac{\lVert H_s\lvert\psi\rangle\rVert^2}{\Omega_s^2} \\ &\le\left(\frac{\Omega_s}{\lambda_s}\right)^2 \equiv p_{\mathrm{bound}}, \tag*{(18)} \end{aligned}

where the first factor, (Ωs/λs)2≤1(\Omega_s/\lambda_s)^2\le1, is the loss due to the LCU 1-norm, and the second, ∥Hs∣ψ⟩∥2/Ωs2\lVert H_s\lvert\psi\rangle\rVert^2/\Omega_s^2, is the success probability that a block-encoding normalized by the spectral norm would reach; it depends only on the input state. Improving the success probability can only be achieved by reducing λs\lambda_s or by the choice of ∣ψ⟩\lvert\psi\rangle.

For qubitization [Eq. (8)], the realized value uses the LCU 1-norm, E~n=(En−Emax⁡)/λs\widetilde{E}_n=(E_n-E_{\max})/\lambda_s,

psuccess(Tk)=∑nwnTk(En−Emax⁡λs)2.(19)p^{(T_k)}_{\mathrm{success}}=\sum_n w_n T_k\left(\frac{E_n-E_{\max}}{\lambda_s}\right)^2. \tag*{(19)}

Unlike the LCU case, the LCU 1-norm does not affect the qubitization success probability in a clearly predictable way because TkT_k is nonlinear. The same holds for the QSVT success probability. For QSVT, with target polynomial P(x)P(x) and scaling factor CC, the implemented polynomial is P(x)/CP(x)/C, giving

psuccess(P/C)=∥P(H~)C∣ψ⟩∥2=1C2∑nwn∣P(E~n)∣2.(20)p^{(P/C)}_{\mathrm{success}}=\left\lVert\frac{P(\widetilde{H})}{C}\lvert\psi\rangle\right\rVert^2=\frac{1}{C^2}\sum_n w_n\lvert P(\widetilde{E}_n)\rvert^2. \tag*{(20)}

As with qubitization, PP is nonlinear, so the HH renormalization does not affect the QSVT success probability in a clearly predictable way. In addition, QSVT introduces its own scaling factor

PropertyLCUQubitizationQSVT
Encoded function f(x)f(x)xxTk(x)T_k(x)P(x)/CP(x)/C (PP real, definite parity)
PrimitivePREP-SELECTWalk operator WWQSVT phase sequence
Main control parameterLCU 1-norm λ\lambdaDegree kkPolynomial PP and scaling CC
Success probabilitypsuccess(I)=∑nwnE~n2p_{\mathrm{success}}^{(I)}=\sum_n w_n\widetilde{E}_n^2psuccess(Tk)=∑nwnTk(E~n)2p_{\mathrm{success}}^{(T_k)}=\sum_n w_n T_k(\widetilde{E}_n)^2psuccess(P/C)=∑nwn∣P(E~n)∣2/C2p_{\mathrm{success}}^{(P/C)}=\sum_n w_n\lvert P(\widetilde{E}_n)\rvert^2/C^2

Table 1. Comparison of the block-encoding constructions considered in this work.

C≥Copt=sup⁡x∈[−1,1]∣P(x)∣,(21)C \ge C_{\mathrm{opt}} = \sup_{x\in[-1,1]} \lvert P(x)\rvert, \tag*{(21)}

required for ∣P(x)/C∣≤1\lvert P(x)/C\rvert\le1 on [−1,1][-1,1]; any bound looser than the supremum reduces the success probability as 1/C21/C^2 (Sec. III D).

In most of the remaining analysis, we work with the shifted and normalized Hamiltonian, whose eigenvalues lie in [−1,0][-1,0]. This one-sided spectrum simplifies the analysis of the Chebyshev transformations and provides a convenient setting in which to derive bounds of the success probability.

Dependence of the success probability on system features

Beyond the choice of normalization, the features of HH and the system size affect both the spectral distribution and the normalization itself.

In the Hubbard model, the important feature of HH is the interaction strength. When it grows, the coefficient 1-norm λs\lambda_s accumulates contributions from every term that the interaction activates, whereas the spectral norm Ωs\Omega_s tracks only the largest spectral gap. The bound (Ωs/λs)2(\Omega_s/\lambda_s)^2 of Eq. (18) therefore decreases with U/tU/t, from its free-fermion value (1 for N=2N=2, ≈0.63\approx0.63 for N=14N=14, tending to (4/(π+2))2≈0.61(4/(\pi+2))^2 \approx0.61 at large NN) to 4/94/9 for every NN: at large UU, Ωs=Emax⁡−Emin⁡→NU\Omega_s=E_{\max}-E_{\min}\to NU while λs→3NU/2\lambda_s\to3NU/2 (Sec. III B). A central and initially counterintuitive finding concerns the dependence on system size: for the Hubbard systems and structured input states considered, the success probability shows no systematic decay with system size, both for LCU (Sec. III B) and QSVT (Sec. III D 4) block-encodings. Although λs\lambda_s increases with system size, the spectral distribution of the input state changes with it, and these two effects compensate: for the HF input state, psuccess(I)p_{\mathrm{success}}^{(I)} settles onto a plateau rather than decaying toward zero. In the Hubbard model this plateau is reached already at N=4N=4: the success probability separates most sharply between N=2N=2 and N≥4N\ge4 in the intermediate-correlation crossover, while the curves for N≥4N\ge4 are nearly indistinguishable and converge to the same large-U/tU/t value (Figure 1(b)).

Figure 1. LCU success probability psuccess(I)p_{\mathrm{success}}^{(I)} versus U/tU/t (axis scaled as U/(U+4t)U/(U+4t)). (a) N=2N=2, for the half-filled ground state ∣E0⟩|E_0\rangle (green) and the HF state (red): exact value (Ωs/λs)2∥Hs∣ψ⟩∥2/Ωs2(\Omega_s/\lambda_s)^2\|\mathcal{H}_s|\psi\rangle\|^2/\Omega_s^2 (lines) and compiled PREP-SELECT circuit sampled with Nshots=104N_{\mathrm{shots}}=10^4 (open markers). The dashed line is pbound=(Ωs/λs)2p_{\mathrm{bound}}=(\Omega_s/\lambda_s)^2, reached by ∣E0⟩|E_0\rangle only for U≲3tU\lesssim3t, where the global minimum of HH lies at half filling. Blue: random input states (mean ±\pm standard deviation over 200 states, individual states shown as dots). In the circuit simulations the input state is injected directly as the initial statevector of the simulator. (b) HF input state for N=2N=2–1414 (color): psuccess(I)(∣ψHF⟩)p_{\mathrm{success}}^{(I)}(|\psi_{\mathrm{HF}}\rangle) (solid, left axis) and HF–ground-state (GS) overlap ∣⟨ψHF∣E0⟩∣2|\langle\psi_{\mathrm{HF}|E_0\rangle}|^2 (dashed, right axis, log scale). The dash-dotted line marks the large-UU limit pbound→4/9p_{\mathrm{bound}}\to4/9.

Dependence of the success probability on the input state and ground-state overlap

The input state itself plays a role that differs between the LCU and qubitization block-encodings. For the LCU block-encoding, the success probability is maximized by the eigenvector ∣Emin⁡⟩\lvert E_{\min}\rangle, which after the spectral shift carries the eigenvalue of largest magnitude: psuccess(I)(∣ψ⟩)≤(Ωs/λs)2=psuccess(I)(∣Emin⁡⟩)p_{\mathrm{success}}^{(I)}(\lvert\psi\rangle)\le(\Omega_s/\lambda_s)^2=p_{\mathrm{success}}^{(I)}(\lvert E_{\min}\rangle) for every normalized ∣ψ⟩\lvert\psi\rangle. Within the half-filled sector the maximizer is the ground state ∣E0⟩\lvert E_0\rangle, which reaches the bound when E0=Emin⁡E_0=E_{\min} and lies slightly below it otherwise (Sec. III B). For qubitization, the state dependence is qualitatively richer. Writing E~n=cos⁡θn\widetilde{E}_n=\cos\theta_n, so that Tk(E~n)=cos⁡(kθn)T_k(\widetilde{E}_n)=\cos(k\theta_n), and using cos⁡2(kθn)=[1+cos⁡(2kθn)]/2\cos^2(k\theta_n)=[1+\cos(2k\theta_n)]/2, the qubitization success probability becomes

psuccess(Tk)=12+12∑nwncos⁡(2kθn).(22)p_{\mathrm{success}}^{(T_k)}=\frac{1}{2}+\frac{1}{2}\sum_n w_n\cos(2k\theta_n). \tag*{(22)}

Its average over degrees is therefore

psuccess(Tk)‾=lim⁡K→∞1K∑k=1Kpsuccess(Tk)=12+12∑n:θn=πwn,(23)\overline{p_{\mathrm{success}}^{(T_k)}}=\lim_{K\to\infty}\frac{1}{K}\sum_{k=1}^{K}p_{\mathrm{success}}^{(T_k)}=\frac{1}{2}+\frac{1}{2}\sum_{n:\theta_n=\pi}w_n, \tag*{(23)}

since (1/K)∑kcos⁡(2kθn)→0(1/K)\sum_k\cos(2k\theta_n)\to0 unless θn\theta_n is a multiple of π\pi ([32, 33]). The input state controls the fluctuations around this average. Writing psuccess(Tk)=[1+∑nwncos⁡(2kθn)]/2p_{\mathrm{success}}^{(T_k)}=[1+\sum_n w_n\cos(2k\theta_n)]/2, the variance over kk is (1/8)∑nwn2(1/8)\sum_n w_n^2 for distinct θn\theta_n, so the inverse participation ratio ∑nwn2\sum_n w_n^2 directly sets the amplitude of the oscillations. A state with broad spectral support, e.g., a random state, for which ∑nwn2≈2/(4N+1)\sum_n w_n^2\approx2/(4^N+1), remains close to 1/21/2, whereas a state concentrated on a few eigenstates (large ∑nwn2\sum_n w_n^2) explores a much broader range of values even at large kk. This oscillation makes the degree an additional lever: for concentrated input states, favorable degrees can be identified from a few dominant spectral components, or predicted from shallow-degree data without classical diagonalization (Sec. III C 1). In the Hubbard model, the HF input state exhibits such a concentrated spectral structure across the system sizes studied, whereas a random state has broad spectral support.

The success probability should also be distinguished from the fidelity of the input state, i.e., its overlap ∣⟨ψHF∣E0⟩∣2\lvert\langle\psi_{\mathrm{HF}}\lvert E_0\rangle\rvert^2 with the ground state. The fidelity measures alignment with a single eigenvector, whereas psuccessp_{\mathrm{success}} is a weighted average over the whole spectrum. As a result, psuccess(I)p_{\mathrm{success}}^{(I)} is quasi-independent of the fidelity of the HF input: it plateaus with system size while the overlap keeps falling (Sec. III B). This differs from filtering-based ground-state preparation, where the success probability is set by the ground-state overlap [13].

QuantitySetting
Model1D Hubbard chain at half filling
Hoppingt=1t = 1
Interactionvaried through U/tU/t
System sizeN=2N = 2–14 sites
System qubitsnsys=2Nn_{\mathrm{sys}} = 2N
LCU terms7N−37N - 3
Spectral shiftHs=H−Emax⁡IH_s = H - E_{\max}I
LCU normalization coefficient1-norm λs\lambda_s
Input statesHF, ground state, random
QubitizationTk(H~)T_k(\tilde{H})
QSVTdefinite-parity Chebyshev targets
Largest system28 system qubits + 7 ancilla = 35 qubits
ValidationDense matrix for N≤6N \le6; matrix-free beyond

Table 2. Summary of the numerical setup used throughout the study.

Success probability in resource estimation

In a direct postselection protocol, the shot overhead per accepted outcome is 1/psuccess(f)1/p_{\mathrm{success}}^{(f)}, and when the block-encoding is applied repeatedly, e.g., to the states spanning a Krylov subspace, this overhead accumulates over the course of the algorithm [34, 35, 36]. Amplitude amplification and its oblivious and fixed-point variants reduce the number of repetitions to O(1/psuccess(f))O(1/\sqrt{p_{\mathrm{success}}^{(f)}}) [37, 38, 39], at the cost of deeper circuits. A method with shallower circuits but a lower success probability may therefore be more costly overall than a deeper one with a higher acceptance probability, so the success probability should be reported alongside circuit depth, gate count and qubit count, which are the quantities usually estimated [27, 40, 41].

Numerical results

We now turn to the numerical characterization of the success probability for the three methods considered here: LCU, qubitization and QSVT. As a controlled testbed, we use the one-dimensional Hubbard model. This model is well suited for this analysis: it is amenable to exact diagonalization for validation, its correlation strength is tunable through a single ratio, and it admits known analytic limits against which the success probability can be checked.

The Hubbard model and computational setup

We study the one-dimensional Hubbard model on a chain of NN sites at half filling [42, 43, 44, 45],

H=−t∑⟨i,j⟩,σ(ci,σ†cj,σ+h.c.)+U∑ini,↑ni,↓(24)H = -t \sum_{\langle i,j\rangle,\sigma} \left(c_{i,\sigma}^{\dagger}c_{j,\sigma} + \mathrm{h.c.}\right) + U \sum_i n_{i,\uparrow}n_{i,\downarrow} \tag*{(24)}

where tt is the nearest-neighbor hopping amplitude, UU is the on-site interaction strength, ci,σ†c_{i,\sigma}^{\dagger} and ci,σc_{i,\sigma} are fermionic creation and annihilation operators, and

ni,σ=ci,σ†ci,σ(25)n_{i,\sigma} = c_{i,\sigma}^{\dagger}c_{i,\sigma} \tag*{(25)}

is the occupation operator. The ratio U/tU/t controls the crossover from the weakly interacting regime (t≫Ut \gg U) to the strongly correlated Mott insulating regime with localized electrons (U≫tU \gg t). Throughout this work, we set t=1t = 1 and vary U/tU/t.

We map the fermionic operators to nsys=2Nn_{\mathrm{sys}} = 2N qubits using the Jordan–Wigner transformation [46], with the spin-up orbitals ordered first. For the nearest-neighbor hopping terms, the Jordan–Wigner strings cancel between adjacent sites, yielding

H=−t2∑σ∑i=0N−2(Xi+δ,σXi+1+δ,σ+Yi+δ,σYi+1+δ,σ)+NU4I−U4∑i=0N−1(Zi+ZN+i)+U4∑i=0N−1ZiZN+i.(26)H = -\frac{t}{2}\sum_{\sigma}\sum_{i=0}^{N-2}\left(X_{i+\delta,\sigma}X_{i+1+\delta,\sigma}+Y_{i+\delta,\sigma}Y_{i+1+\delta,\sigma}\right) + \frac{NU}{4}I - \frac{U}{4}\sum_{i=0}^{N-1}\left(Z_i+Z_{N+i}\right) + \frac{U}{4}\sum_{i=0}^{N-1}Z_iZ_{N+i}. \tag*{(26)}

This decomposition contains 7N−37N - 3 Pauli strings, from which the LCU 1-norm is obtained directly. We use the spectral shift of Eq. (17), so that the normalized eigenvalues lie in [−1,0][-1,0].

The PREP operator loads the amplitudes ∣αj∣/λ\sqrt{|\alpha_j|/\lambda} onto an ancilla register of width na=⌈log⁡2(7N−3)⌉n_a = \lceil\log_2(7N - 3)\rceil using a Grover–Rudolph binary tree of rotation angles constructed from partial sums of the coefficient distribution [47]. The sign factors sgn⁡(αj)\operatorname{sgn}(\alpha_j) are incorporated into SELECT rather than PREP. SELECT applies each signed Pauli string sgn⁡(αj)Pj\operatorname{sgn}(\alpha_j)P_j controlled on the ancilla index ∣j⟩a|j\rangle_a. Each term is implemented as a multicontrolled Pauli controlled by the ancilla index, with the control complexity determined by the ancilla width nan_a. The compiled PREP-SELECT-PREP†^\dagger circuit is validated at two levels. First, for N≤6N \le6, where the dense Hamiltonian remains computationally accessible, the projected circuit block-encodes with the explicitly constructed H~\tilde{H} to machine precision (∼10−15\sim10^{-15}). Second, beyond this range, the exact HF success probability, evaluated matrix-free by applying HsH_s term by term to ∣ψHF⟩|\psi_{\mathrm{HF}}\rangle, psuccess(I)(∣ψHF⟩)=∥Hs∣ψHF⟩∥2/λs2p_{\mathrm{success}}^{(I)}(|\psi_{\mathrm{HF}}\rangle) = \|H_s|\psi_{\mathrm{HF}}\rangle\|^2/\lambda_s^2 is compared with the shot-based estimate p^success(I)(∣ψHF⟩)\hat{p}_{\mathrm{success}}^{(I)}(|\psi_{\mathrm{HF}}\rangle), i.e., the fraction of shots in which the ancilla register is measured in ∣0⟩a\lvert0\rangle_a when the compiled circuit is applied to ∣ψHF⟩\lvert\psi_{\mathrm{HF}}\rangle; the two agree within the binomial standard error psuccess(I)(1−psuccess(I))/Nshots\sqrt{p_{\mathrm{success}}^{(I)}(1-p_{\mathrm{success}}^{(I)})/N_{\mathrm{shots}}}, with Nshots=20000N_{\mathrm{shots}}=20000, which validates the matrix-free implementation used for N>6N>6. The numerical setup is summarized in Table II.

LCU success probability

Ground state as input state

The LCU block-encoding success probability obeys the state-independent upper bound in Eq. (18), which is controlled entirely by the ratio Ωs/λs\Omega_s/\lambda_s with pbound=(Ωs/λs)2p_{\mathrm{bound}}=(\Omega_s/\lambda_s)^2, and two limits fix its range. At U=0U=0, the spectrum is symmetric Emin⁡=−Emax⁡E_{\min}=-E_{\max} and the hopping strings contribute 2t(N−1)2t(N-1) to the 1-norm, while the constant introduced by the shift contributes Emax⁡E_{\max}:

Ωs∣U=0=2Emax⁡,λs∣U=0=2t(N−1)+Emax⁡,\Omega_s\big|_{U=0}=2E_{\max},\qquad\lambda_s\big|_{U=0}=2t(N-1)+E_{\max},
pbound∣U=0=(2Emax⁡2t(N−1)+Emax⁡)2.(27)p_{\mathrm{bound}}\big|_{U=0}=\left(\frac{2E_{\max}}{2t(N-1)+E_{\max}}\right)^2. \tag*{(27)}

For N=2N=2 the two scales coincide (Ωs=λs=4t\Omega_s=\lambda_s=4t) and pbound=1p_{\mathrm{bound}}=1, but for larger chains the hopping strings do not all saturate the operator norm simultaneously and the bound falls below unity: pbound≈0.73, 0.68, 0.66p_{\mathrm{bound}\approx0.73,\ 0.68,\ 0.66} and 0.630.63 for N=4, 6, 8N=4,\ 6,\ 8 and 1414. These values converge to a constant at large NN: at U=0U=0, Emax⁡=2∑εq>0εqE_{\max}=2\sum_{\varepsilon_q>0}\varepsilon_q with εq=−2tcos⁡(qπ/(N+1))\varepsilon_q=-2t\cos(q\pi/(N+1)), so that Emax⁡→4tN/πE_{\max}\to4tN/\pi, Ωs=2Emax⁡→8tN/π\Omega_s=2E_{\max}\to8tN/\pi and λs→2tN+4tN/π\lambda_s\to2tN+4tN/\pi, giving

pbound∣U=0→(4π+2)2≈0.61.p_{\mathrm{bound}}\big|_{U=0}\to\left(\frac{4}{\pi+2}\right)^2\approx0.61.

In the strong-coupling limit the two scales separate. For U≫tU\gg t, the highest-energy configuration is the fully doubly occupied state, which gives Emax⁡=NUE_{\max}=NU, while ∣Emin⁡∣=O(t)\lvert E_{\min}\rvert=O(t) is negligible on the scale of UU. For the shifted Pauli decomposition of Eq. (26), the coefficients are:

Ωs→U≫tNU,\Omega_s\xrightarrow{U\gg t}NU,
λs=2t(N−1)+NU2⏟Z+NU4⏟ZZ+3NU4⏟const→U≫t3NU2.(28)\lambda_s=2t(N-1)+\underbrace{\frac{NU}{2}}_{Z}+\underbrace{\frac{NU}{4}}_{ZZ}+\underbrace{\frac{3NU}{4}}_{\mathrm{const}}\xrightarrow{U\gg t}\frac{3NU}{2}. \tag*{(28)}

and therefore

Ωsλs→U≫t23,pbound→49≈0.44.(29)\frac{\Omega_s}{\lambda_s}\xrightarrow{U\gg t}\frac{2}{3},\qquad p_{\mathrm{bound}}\to\frac{4}{9}\approx0.44. \tag*{(29)}

Here the ZZ and ZZZZ terms contribute NU/2NU/2 and NU/4NU/4 respectively, while the constant term, NU/4NU/4 before shifting, contributes 3NU/43NU/4 in magnitude after the spectral shift, since Emax⁡→NUE_{\max}\to NU sends it to NU/4−Emax⁡→−3NU/4NU/4-E_{\max}\to-3NU/4. The limit 4/94/9 is therefore independent of both NN and U/tU/t. The decrease of the bound with U/tU/t, from its free-fermion value to 4/94/9, is the decomposition overhead of the 1-norm: the Pauli strings per site partially cancel in the operator norm but count additively in λs\lambda_s. Unlike the spectral norm, the 1-norm depends on the orbital representation used for the Pauli decomposition, so that pboundp_{\mathrm{bound}} can be increased a priori by optimizing this representation [29]; the site and molecular-orbital bases are compared below. After the spectral shift, the state of lowest energy Emin⁡E_{\min} carries the eigenvalue of largest magnitude of HsH_s, ∣Emin⁡−Emax⁡∣=Ωs\lvert E_{\min}-E_{\max}\rvert=\Omega_s, and therefore saturates the bound. The half-filled ground state saturates it only when E0=Emin⁡E_0=E_{\min}, i.e., for U≲tU\lesssim t (N=2N=2) and U≲tU\lesssim t (N=4N=4). Beyond, psuccess(∣E0⟩)=(Ωs/λs)2(Emax⁡−E0)2/(Emax⁡−Emin⁡)2p_{\mathrm{success}}(\lvert E_0\rangle)=(\Omega_s/\lambda_s)^2(E_{\max}-E_0)^2/(E_{\max}-E_{\min})^2 falls slightly below pboundp_{\mathrm{bound}}, by at most ≈8%\approx8\% for N≤4N\leq4 (minimum 0.92 pbound0.92\,p_{\mathrm{bound}} at U≈6tU\approx6t for N=4N=4), and returns to it at large UU, where E0E_0 and Emin⁡E_{\min} are both O(t)O(t) against Emax⁡=NUE_{\max}=NU. As a consistency check, Figure 1(a) compares, for N=2N=2, the success probability computed from (Ωs/λs)2∥Hs∣ψ⟩∥2/Ωs2(\Omega_s/\lambda_s)^2\lVert H_s\lvert\psi\rangle\rVert^2/\Omega_s^2 with the one measured on the compiled PREP-SELECT circuit, for the half-filled ground state and for the HF state; the two agree, confirming that the circuit realizes the 1-norm normalization.

HF as input state

We use the HF input state ∣ψHF⟩\lvert\psi_{\mathrm{HF}}\rangle, taken here as the single Slater determinant minimizing the energy for the noninteracting problem [48]. In the site basis it is a superposition of computational basis states. At U=0U=0 the HF approximation is exact (∣ψHF⟩=∣E0⟩\lvert\psi_{\mathrm{HF}}\rangle=\lvert E_0\rangle), so psuccess(I)(∣ψHF⟩)=pboundp_{\mathrm{success}}^{(I)}(\lvert\psi_{\mathrm{HF}}\rangle)=p_{\mathrm{bound}}, which equals 1 only for N=2N=2 (Eq. (27)). For U>0U>0, the exact ground state acquires correlations beyond a single Slater determinant [43]. Expanding ∣ψHF⟩=∑nβn∣En⟩\lvert\psi_{\mathrm{HF}}\rangle=\sum_n\beta_n\lvert E_n\rangle in the exact eigenbasis of Eq. (12) (with ∣β0∣2<1\lvert\beta_0\rvert^2<1 for U>0U>0 once correlations make the HF input state different from the exact ground state):

psuccess(I)(∣ψHF⟩)=∑nwnE~n2≤E~02=pbound.(30)p_{\mathrm{success}}^{(I)}(\lvert\psi_{\mathrm{HF}}\rangle)=\sum_n w_n\widetilde{E}_n^2\leq\widetilde{E}_0^2=p_{\mathrm{bound}}. \tag*{(30)}

The difference pbound−psuccess(I)(∣ψHF⟩)p_{\mathrm{bound}}-p_{\mathrm{success}}^{(I)}(\lvert\psi_{\mathrm{HF}}\rangle) therefore reflects electron correlation, but it is not identical to the loss of ground-state overlap: it is a spectrally weighted measure of how the HF spectral weight is distributed away from the extremal eigenvalue.

In order to study the effect of the basis on the success probability, we also computed the same metric after a change of basis from the site basis to the molecular-orbital (MO) basis. In the MO basis, the HF state is represented as a single computational-basis occupation string (e.g., ∣ψHF⟩=∣1010⟩\lvert\psi_{\mathrm{HF}}\rangle=\lvert1010\rangle for N=2N=2), requiring only single-qubit XX gates for state preparation. However, this simplification does not improve LCU success probability, because the relevant cost is controlled by the Hamiltonian decomposition rather than by state-preparation complexity alone. In the site basis the interaction is local, with fewer Pauli strings contributing to the 1-norm. The orbital rotation to the MO basis introduces additional nonlocal terms with no site-basis counterpart, so the MO interaction has more terms and we have λMO>λsite\lambda_{\mathrm{MO}}>\lambda_{\mathrm{site}}, consistent with the observation that localized orbitals reduce the 1-norm of molecular Hamiltonians [29]. Since Ωs\Omega_s is basis independent:

pboundsite=(Ωs/λs,site)2>(Ωs/λs,MO)2=pboundMO∀U/t>0.(31)\begin{aligned} p_{\mathrm{bound}}^{\mathrm{site}}=(\Omega_s/\lambda_{s,\mathrm{site}})^2>(\Omega_s/\lambda_{s,\mathrm{MO}})^2 \\ &=p_{\mathrm{bound}}^{\mathrm{MO}}\quad\forall U/t>0. \tag*{(31)} \end{aligned}

Thus, for the success probability considered here, the optimal basis is not necessarily the one that minimizes state-preparation complexity, but rather one that minimizes the ratio λs/Ωs\lambda_s/\Omega_s.

Overlap between the ground state and the HF state

Unlike the overlap ∣β0∣2|\beta_0|^2, which depends only on the ground-state component of ∣ψHF⟩|\psi_{\mathrm{HF}}\rangle, psuccess(I)(∣ψHF⟩)p_{\mathrm{success}}^{(I)}(|\psi_{\mathrm{HF}}\rangle) of Eq. (30) is a second moment over the whole HF spectral distribution. As UU increases, the interaction inflates λs\lambda_s relative to Ωs\Omega_s: pbound=(Ωs/λs)2p_{\mathrm{bound}=(\Omega_s/\lambda_s)^2} decreases, compressing all normalized eigenvalues E~n\tilde{E}_n uniformly. The success probability is therefore affected not only by the loss of ground-state weight but also by the rescaling of the entire spectrum induced by the normalization. The two quantities can therefore differ substantially: in Figure 1(b) we can see that already at N=2N=2, U=4tU=4t the HF state retains a substantial overlap ∣β0∣2≈0.85|\beta_0|^2\approx0.85 with the exact ground state, yet psuccess(I)(∣ψHF⟩)p_{\mathrm{success}}^{(I)}(|\psi_{\mathrm{HF}}\rangle) has dropped to ≈0.35\approx0.35.

Since the shot overhead scales as 1/psuccess(I)(∣ψHF⟩)1/p_{\mathrm{success}}^{(I)}(|\psi_{\mathrm{HF}}\rangle), this is a more severe requirement than the overlap alone would suggest. More generally, the success probability is quasi-independent of the fidelity of the input state: at large UU, psuccess(I)(∣ψHF⟩)p_{\mathrm{success}}^{(I)}(|\psi_{\mathrm{HF}}\rangle) plateaus at ≈0.25\approx0.25–0.260.26 for N≥4N\geq4, while the HF–GS overlap keeps falling with NN (Figure 1(b)). For our purposes, this means that the shot overhead of the LCU block-encoding with a HF input remains bounded, over all system sizes studied (N≤14N\leq14), even in the strongly correlated regime where the HF state becomes a poor approximation of the ground state, while the bound itself stays bounded away from zero in both limits, tending to (4/(π+2))2≈0.61(4/(\pi+2))^2\approx0.61 at U=0U=0 for large NN and to 4/94/9 for every NN as U→∞U\to\infty. The success probability therefore does not penalize the use of a simple HF input state in larger systems; conversely, it cannot serve as a proxy for the fidelity of the input state, which must be assessed separately.

Qubitization success probability

Whereas LCU block-encoding applies H~\tilde{H} directly, qubitization realizes the Chebyshev transformation Tk(H~)T_k(\tilde{H}). The corresponding success probability is therefore degree dependent:

psuccess(Tk)=∥Tk(H~)∣ψ⟩∥2=∑nwnTk(E~n)2(32)p_{\mathrm{success}}^{(T_k)} = \lVert T_k(\widetilde{H}) \lvert\psi\rangle\rVert^2 = \sum_n w_n T_k(\widetilde{E}_n)^2 \tag*{(32)}

Writing each eigenvalue as E~n=cos⁡θn\widetilde{E}_n = \cos\theta_n thus

θn=arccos⁡(E~n),(33)\theta_n = \arccos(\widetilde{E}_n), \tag*{(33)}

the Chebyshev identity Tk(cos⁡θn)=cos⁡(kθn)T_k(\cos\theta_n) = \cos(k\theta_n) turns Eq. (32) into a weighted sum of oscillating terms,

psuccess(Tk)=∑nwncos⁡2(kθn).(34)p_{\mathrm{success}}^{(T_k)} = \sum_n w_n \cos^2(k\theta_n). \tag*{(34)}

Two features of Eq. (34) are particularly relevant to the analysis that follows. First, as UU increases, the LCU 1-norm grows faster than the spectral norm, compressing the normalized spectrum into [−Ωs/λs,0][-\Omega_s/\lambda_s,0], which tends to [−2/3,0][-2/3,0] at large UU (Sec. III B). Second, at fixed UU the probability oscillates with kk: a single term of Eq. (34) vanishes when

kθn=(m+12)π,m∈Z,(35)k\theta_n = \left(m+\frac{1}{2}\right)\pi,\qquad m\in\mathbb{Z}, \tag*{(35)}

so psuccess(Tk)p_{\mathrm{success}}^{(T_k)} can become small when the dominant spectral components lie near such nodes. It is this second feature that might make the choice of degree an operational lever. At k=1k=1 the Chebyshev transformation is the identity, T1(H~)=H~T_1(\widetilde{H})=\widetilde{H}, so

psuccess(T1)=∥H~∣ψ⟩∥2=psuccess(I),(36)p_{\mathrm{success}}^{(T_1)} = \lVert\widetilde{H}\lvert\psi\rangle\rVert^2 = p_{\mathrm{success}}^{(I)}, \tag*{(36)}

for every input state and every U/tU/t: the LCU success probability is the k=1k=1 slice of the qubitization framework. The features specific to qubitization (oscillation with kk, nodes, and the resulting freedom in choosing the degree) emerge for k>1k>1, and make kk an additional feature that affects resource estimation: different degrees of the same walk can have substantially different shot overheads.

Selecting the Chebyshev polynomial degree

Because psuccess(Tk)p_{\mathrm{success}}^{(T_k)} oscillates with kk, a blindly chosen degree can fall near a node, as expressed in Eq. (35), where the success probability can become very small. For the HF input state, whose spectral weight is concentrated on a small number of eigenstates, this raises a practical question: can a favorable degree with respect to the success probability be selected in advance, without having to scan many degrees and retain the best one? This question is relevant whenever the degree is not fixed by the algorithm. The most direct application arises when the target accuracy only imposes a minimum degree k≥kmin⁡k \ge k_{\min}, as is generic for polynomial methods whose degree is set by the spectral gap and the required precision [13], while the Chebyshev degree itself remains free above this threshold:

since psuccess(Tk)p_{\mathrm{success}}^{(T_k)} varies strongly between neighboring degrees, a slightly deeper circuit can reduce the shot overhead substantially. For instance, for N=8N=8 at U/t=16U/t=16 (Figure 2), increasing the degree from k=2k=2 to k=3k=3 raises psuccess(Tk)p_{\mathrm{success}}^{(T_k)} from ≈0.23\approx0.23 to ≈0.94\approx0.94, i.e., a fourfold reduction in shot overhead at the cost of a single additional walk step. A second, more specific application is the construction of a Krylov space from a noncontiguous set of degrees, where the degrees themselves are free parameters; we illustrate it in detail in Sec. III C.2. In both cases, identifying favorable degrees without scanning them all, which would itself cost shots at every degree, is operationally useful. To this end, we propose below two inexpensive strategies to identify, a priori, Chebyshev degrees with a high success probability.

Figure 2. Qubitization success probability psuccess(Tk)p_{\mathrm{success}}^{(T_k)} as a function of the Chebyshev degree kk at U/t=16U/t=16 for the HF input state and N=2N=2–1212 sites (color as in Figure 1(b) and Figure 10). Lines and dots show exact values, computed without sampling from the matrix-free action of HH (Sec. III A). Open triangles show sampled-circuit estimates with Nshots=104N_{\mathrm{shots}}=10^4, available for k≤10k\leq10 (N≤6N\leq6) and k≤6k\leq6 (N=8N=8); their binomial standard deviations (≤0.005\leq0.005) are smaller than the markers. The sampled estimates agree with the exact values within two binomial standard deviations (rms deviation ≈0.005\approx0.005). The dotted black line marks the long-degree average 1/21/2. The oscillatory dependence on kk produces degrees with both high and near-vanishing success probabilities.

We quantify the effective number of eigenstates contributing to the input state through the participation ratio

PR=1∑nwn2.(37)\mathrm{PR} = \frac{1}{\sum_n w_n^2}. \tag*{(37)}

Recall from Eq. (12) that ∑nwn=1\sum_n w_n=1 while ∑nwn2≤1\sum_n w_n^2\le1, so that PR≥1\mathrm{PR}\ge1. The participation ratio is close to 1 when the weight is concentrated on a few eigenstates, giving a few dominant frequencies and strong coherent oscillations in psuccess(Tk)p_{\mathrm{success}}^{(T_k)}, and large when the weight is spread over the spectrum, in which case the oscillations average out over kk. Once the dominant pairs {wn,θn}\{w_n,\theta_n\} are known, Eq. (34) determines psuccess(Tk)p_{\mathrm{success}}^{(T_k)} for every degree, allowing the optimal degree to be identified without executing the corresponding deep circuits. Writing k∗k^* for the selected degree and pmedp_{\mathrm{med}} for the median of psuccess(Tk)p_{\mathrm{success}}^{(T_k)} over the considered degree window, we use psuccess(Tk∗)/pmedp_{\mathrm{success}}^{(T_{k^*})}/p_{\mathrm{med}} as the figure of merit. The median represents the success probability expected from a degree chosen without spectral information, so the ratio measures the gain brought by the selection strategy. The largest achievable value of this ratio is pmax⁡/pmedp_{\max}/p_{\mathrm{med}}, with pmax⁡=max⁡kpsuccess(Tk)p_{\max}=\max_k p_{\mathrm{success}}^{(T_k)} over the same window; we call pmax⁡p_{\max} the oracle optimum, since reaching it requires knowing psuccess(Tk)p_{\mathrm{success}}^{(T_k)} at every degree. All results below use the window 1≤k≤2001\le k\le200 and the HF input state for N=2,4,6,8N=2,4,6,8 and U/t∈{4,8,16}U/t\in\{4,8,16\}, i.e., 12 cases.

The spectrum-informed strategy keeps only the mm eigenstates with the largest weights wnw_n, whose index set we denote Mm\mathcal{M}_m, and optimizes the resulting partial sum,

psuccess,m(Tk)=∑n∈Mmwncos⁡2(kθn),(38)p_{\mathrm{success},m}^{(T_k)} = \sum_{n\in\mathcal{M}_m} w_n\cos^2(k\theta_n), \tag*{(38)}
kA∗=arg⁡max⁡k psuccess,m(Tk).k_A^* = \underset{k}{\arg\max}\ p_{\mathrm{success},m}^{(T_k)}.

It requires the full eigendecomposition and therefore serves as a spectrum-informed benchmark rather than a practical spectrum-free protocol. With m=5m=5, the selected degree lies within 5% of the oracle optimum in 11 of the 12 cases and reaches on average 98% of pmax⁡p_{\max}, showing that a small number of spectral components captures most of the relevant kk-dependence.

The spectrum-free strategy uses no spectral input. It requires psuccess(Tk)p_{\mathrm{success}}^{(T_k)} only at shallow degrees k≤kfitk\leq k_{\mathrm{fit}}, fits mm weights and angles {wnfit,θnfit}\{w_n^{\mathrm{fit}},\theta_n^{\mathrm{fit}}\} to those points by nonlinear least squares against the model in Eq. (34), and extrapolates the fitted sum to select

kB⋆=arg⁡max⁡k ∑n≤mwnfitcos⁡2(kθnfit).(39)k_B^\star=\underset{k}{\arg\max}\ \sum_{n\leq m}w_n^{\mathrm{fit}}\cos^2\left(k\theta_n^{\mathrm{fit}}\right). \tag*{(39)}

Because psuccess(Tk)p_{\mathrm{success}}^{(T_k)} is a weighted sum of a small number of squared oscillatory components, a window of shallow circuits can constrain their frequencies and amplitudes, and the fitted model can predict a favorable deeper degree without classical diagonalization. We use kfit=25k_{\mathrm{fit}}=25 and m=5m=5, so that the 2m=102m=10 fitted parameters are fewer than the kfitk_{\mathrm{fit}} data points. In the results below, the shallow values of psuccess(Tk)p_{\mathrm{success}}^{(T_k)} are computed exactly; the effect of shot noise on the fit is not assessed here.

For the HF input state, the median-degree baseline is pmed≈1/2p_{\mathrm{med}}\approx1/2 (between 0.49 and 0.51) in all cases. The spectrum-informed strategy raises the per-circuit success probability to 98% of the oracle optimum on average, corresponding to a gain of 1.7–2.0×\times, nearly independently of the participation ratio over the range studied (PR ≈1.4\approx1.4–14). The spectrum-free strategy reaches 91% of the oracle optimum on average, with gains of 1.4–2.0×\times, and never selects a degree below the median (Figure 3). Its performance depends on the length of the shallow window and on the number of fitted components. With kfit=10k_{\mathrm{fit}}=10, which allows at most m=4m=4 components, the gain ranges from 0.84×\times to 2.0×\times: when nearby frequencies are not resolved, the extrapolation identifies the wrong revival and can select a degree below the median. The required number of components also grows with the fitting window, since a longer window resolves more of the beating between frequencies: at kfit=25k_{\mathrm{fit}}=25, m=3m=3 gives a minimum gain of only 0.46×\times, whereas m=5m=5 gives 1.4×\times.

Figure 3. Degree-selection performance for HF input states at N≤8N \le8 and U/t∈{4,8,16}U/t \in\{4, 8, 16\}, over the window 1≤k≤2001 \le k \le200. (a) Success probability at the degree selected by the spectrum-informed strategy (blue, Eq. (38), m=5m = 5) and the spectrum-free strategy (red, Eq. (39), kfit=25k_{\mathrm{fit}} = 25, m=5m = 5), compared with the median-degree baseline pmedp_{\mathrm{med}} (grey) and the oracle optimum pmax⁡p_{\max} (black horizontal lines). Vertical lines separate the three values of U/tU/t. (b) Gain psuccess(Tk⋆)/pmedp_{\mathrm{success}}^{(T_{k^\star})}/p_{\mathrm{med}} as a function of the participation ratio for the spectrum-informed (blue circles) and spectrum-free (red triangles) strategies; the dashed line marks no gain.

In the two settings identified at the beginning of this section, degree selection can therefore be useful. The tests above select the best degree over the full window 1≤k≤2001\leq k\leq200; restricting the search to k≥kmin⁡k\geq k_{\min}, or ranking degrees to build a noncontiguous Krylov space (Sec. III C 2), applies the same predicted psuccess(Tk)p_{\mathrm{success}}^{(T_k)} to a subwindow and was not tested separately. The spectrum-informed strategy reliably locates near-optimal degrees but requires the dominant eigenpairs. The spectrum-free strategy requires only shallow-degree data and, with kfit=25k_{\mathrm{fit}}=25 and m=5m=5, consistently improves on a blind choice for the HF states studied here; its main benefit is to avoid near-node degrees at which a full circuit execution would have a near-vanishing acceptance probability.

Application to quantum Krylov method

The success-probability analysis becomes operationally concrete when the qubitized walk is used as the primitive of a quantum Krylov method. The Krylov subspace is spanned by the vectors

∣φk⟩=Tk(H~)∣ψ⟩,k=0,…,D−1,(40)|\varphi_k\rangle= T_k(\widetilde{H})|\psi\rangle,\qquad k = 0,\ldots,D-1, \tag*{(40)}

and the ground-state energy is estimated from the regularized generalized eigenvalue problem Hc=E~Sc\mathcal{H}c = \widetilde{E}Sc, with Sjk=⟨φj∣φk⟩S_{jk} = \langle\varphi_j|\varphi_k\rangle the Gram (overlap) matrix and Hjk=⟨φj∣H~∣φk⟩\mathcal{H}_{jk} = \langle\varphi_j|\widetilde{H}|\varphi_k\rangle the Hamiltonian projected onto the DD-dimensional Krylov subspace.

The Krylov vectors of Eq. (40) are not normalized and thus cannot be prepared directly as quantum states. Applying PREP†\mathrm{PREP}^{\dagger} after kk walk steps and postselecting the ancilla register on a=0a = 0 would leave ∣φk⟩/∥φk∥|\varphi_k\rangle/\|\varphi_k\| in the system register with probability psuccess(Tk)p_{\mathrm{success}}^{(T_k)}. Instead, Kirby et al. estimate only the Chebyshev moments [34, 49],

mk=⟨ψ∣Tk(H~)∣ψ⟩,(41)m_k = \langle\psi|T_k(\widetilde{H})|\psi\rangle, \tag*{(41)}

each as the expectation value of an observable, and SS and H\mathcal{H} are then assembled classically [34] as

Sjk=12(mj+k+m∣j−k∣),S_{jk} = \frac{1}{2}\left(m_{j+k} + m_{|j-k|}\right),
Hjk=14(mj+k+1+m∣j+k−1∣+m∣j−k+1∣+m∣j−k−1∣).(42)\mathcal{H}_{jk} = \frac{1}{4}\left(m_{j+k+1} + m_{|j+k-1|} + m_{|j-k+1|} + m_{|j-k-1|}\right). \tag*{(42)}

A Krylov space of dimension DD therefore requires the moments up to kmax⁡=2D−1k_{\max} = 2D - 1. In such a moment-based scheme no postselection is required. However, the success probability directly affects the resources (number of shots) needed to reach a given precision, as we now show.

The qubitization success probability of Eq. (32) can be rewritten, using Eq. (42) with m0=1m_0 = 1, as

psuccess(Tk)=∥φk∥2=Skk=1+m2k2.(43)p_{\mathrm{success}}^{(T_k)} = \|\varphi_k\|^2 = S_{kk} = \frac{1 + m_{2k}}{2}. \tag*{(43)}

Degrees with small psuccess(Tk)p_{\mathrm{success}}^{(T_k)} produce Krylov vectors of small norm, whose moments are also small, ∣mk∣=∣⟨ψ∣φk⟩∣≤∥φk∥|m_k| = |\langle\psi|\varphi_k\rangle| \le\|\varphi_k\|. As m2km_{2k} is estimated from NshotsN_{\mathrm{shots}} binary outcomes (Appendix A), its standard deviation is (1−m2k2)/Nshots\sqrt{(1-m_{2k}^2)/N_{\mathrm{shots}}}, and that of SkkS_{kk} is psuccess(Tk)(1−psuccess(Tk))/Nshots\sqrt{p_{\mathrm{success}}^{(T_k)}(1-p_{\mathrm{success}}^{(T_k)})/N_{\mathrm{shots}}}. Resolving the squared norm SkkS_{kk} of ∣φk⟩|\varphi_k\rangle to a relative precision δ\delta thus requires

Nshots≃1−psuccess(Tk)δ2psuccess(Tk).(44)N_{\mathrm{shots}} \simeq\frac{1-p_{\mathrm{success}}^{(T_k)}}{\delta^2p_{\mathrm{success}}^{(T_k)}}. \tag*{(44)}

so that the 1/psuccess(Tk)1/p_{\mathrm{success}}^{(T_k)} overhead remains even without postselection: not as discarded shots, but as the shots needed to resolve a Krylov vector of small norm with a controlled precision.

The scheme of Kirby et al. [34] is based on moment measurements, each moment being estimated without a control qubit, by preparing the walk state W⌊k/2⌋(∣PREP⟩⊗∣ψ⟩)W^{\lfloor k/2\rfloor}(|\mathrm{PREP}\rangle\otimes|\psi\rangle) and measuring RPREPR_{\mathrm{PREP}} for even kk and SELECT\mathrm{SELECT} for odd kk.

Our scheme is also based on moment measurements, rather than on postselection as in the previous sections, which makes it directly comparable with that of Kirby et al. [34]. The only difference is that we measure each moment with a Hadamard test [50, 51], as commonly done for overlaps in quantum subspace methods [8, 9, 10, 11]; the circuit is shown in Figure 4 and its readout is derived in Appendix A.

Hadamard-test circuit for the Chebyshev moment

Figure 4. Hadamard-test circuit for the Chebyshev moment mk=⟨ψ∣Tk(H~)∣ψ⟩m_k=\langle\psi|T_k(\widetilde{H})|\psi\rangle. The test qubit controls kk applications of the walk operator W=RPREP SELECTW=R_{\mathrm{PREP}}\ \mathrm{SELECT} on ∣PREP⟩⊗∣ψ⟩|\mathrm{PREP}\rangle\otimes|\psi\rangle; only the test qubit is measured, and mk=(−1)k(p0−p1)m_k=(-1)^k(p_0-p_1) from the probabilities p0,1p_{0,1} of its two outcomes, the sign (−1)k(-1)^k coming from the reflection convention of our implementation. The block-encoding ancillas are never measured, so no shot is discarded.

The two schemes offer different trade-offs between circuit depth and readout cost (Table III). In terms of depth, the scheme of Kirby et al. requires only ⌊k/2⌋\lfloor k/2\rfloor uncontrolled walk steps, i.e., a maximum depth of D−1D-1 instead of 2D−12D-1 controlled steps for our Hadamard test scheme.

ReadoutWalk stepsCounted outcomeProbabilityUseful shots
Hadamard testkk (controlled)test qubit = 0(1+mk)/2(1+m_k)/2all NN
Even kk [34]k/2k/2a=0a=0 after PREP(Tk/2)†p\mathrm{PREP}^{\dagger p}_{(T_{k/2})}success =(1+mk)/2=(1+m_k)/2all NN
Odd kk, term ii [34](k−1)/2(k-1)/2a=ia=iqk(i)q_k(i), Eq. (62)≈Niqk(i)\approx N_iq_k(i)

Table III. Readouts of the Chebyshev moment mkm_k: number of walk steps, counted outcome, its probability, and number of shots carrying information, out of NN shots in total (NiN_i for term ii). The probabilities of the two readouts of Kirby et al. are derived in Appendix A.

In our Hadamard-test scheme, every shot contributes to the estimate of mkm_k, for even and odd kk alike, with standard deviation (1−mk2)/Nshots≤1/Nshots\sqrt{(1-m_k^2)/N_{\mathrm{shots}}}\leq1/\sqrt{N_{\mathrm{shots}}} and without any discard. We prefer this trade-off here. The number of shots is already set by the success probability through Eq. (44), which can be large for degrees with small psuccess(Tk)p_{\mathrm{success}}^{(T_k)}. The term-by-term readout of the odd moments, which make up half of the moments required, never needs fewer shots than the Hadamard test for the same error, and needs at least L−1L-1 times more when qk(i)q_k(i) is spread evenly over the LL terms (Appendix A, Eq. (58)); this cost comes on top of Eq. (44). The Hadamard test avoids this additional cost at the price of a controlled walk and twice the circuit depth.

The results of Figure 5 and Figure 6 use four shot budgets per moment. For Nshots=103N_{\mathrm{shots}}=10^3 and 10610^6, the Hadamard-test circuits are sampled. The larger budgets, Nshots=1012N_{\mathrm{shots}}=10^{12} and 102010^{20}, are too costly to sample in simulation; for these, the exact moments are perturbed by Gaussian noise of standard deviation η=Nshots−1/2\eta=N_{\mathrm{shots}}^{-1/2}. In the regularized solution of the generalized eigenvalue problem, directions of SS with eigenvalue below a threshold ϵ\epsilon are discarded [6, 34, 35]. Without noise we use a fixed threshold ϵ=10−13\epsilon=10^{-13}. With noise, ϵ\epsilon is selected for each DD among 40 logarithmically spaced values between 10−1410^{-14} and 10−110^{-1}, restricted to ϵ≥10η\epsilon\geq10\eta, as the smallest value within the widest plateau of the extracted energy, which makes the result insensitive to the exact cut. Krylov vectors with psuccess(Tk)≲ϵp_{\mathrm{success}}^{(T_k)}\lesssim\epsilon therefore cannot be resolved and contribute to the ill-conditioning of the overlap matrix. For the present calculation, the energy error decreases rapidly with Krylov dimension until statistical uncertainty becomes dominant, after which the convergence reaches a shot-limited plateau. The plateau level tracks the nominal moment-noise scale η=1/Nshots\eta=1/\sqrt{N_{\mathrm{shots}}}, and increasing the shot budget lowers this plateau, as shown in the upper panel of Figure 5.

Figure 5. Finite-shot convergence and success-probability variation in the quantum Krylov calculation for the N=4N=4, U/t=2U/t=2 Hubbard model with a HF input state. Upper panel: Ground-state energy error versus Krylov dimension DD for Nshots=103N_{\mathrm{shots}}=10^{3} and 10610^{6} (sampled Hadamard-test circuits) and 101210^{12} and 102010^{20} (Gaussian moment noise η=Nshots−1/2\eta=N_{\mathrm{shots}}^{-1/2}), together with the noiseless curve. Dotted lines mark η=Nshots−1/2\eta=N_{\mathrm{shots}}^{-1/2}, in the color of the corresponding curve. Lower panel: Success probability psuccess(T2D−1)p_{\mathrm{success}}^{(T_{2D-1})} of the highest-degree moment required at each Krylov dimension.

Figure 6. A priori construction of the Krylov space from degrees with high success probability (psuccess(Tk)≥1/2p_{\mathrm{success}}^{(T_k)} \geq1/2), for N=4N=4, U/t=2U/t=2, compared with the standard contiguous construction. (a) Ground-state energy error ∣E(D)−Eref∣|E(D)-E_{\mathrm{ref}}| versus Krylov dimension DD, where E(D)E(D) is the Krylov estimate with DD basis vectors and ErefE_{\mathrm{ref}} the exact ground-state energy, for the contiguous (blue) and selected-degree (green) constructions, noiseless (dashed) and with finite-shot noise for Nshots=1020N_{\mathrm{shots}}=10^{20}, emulated by Gaussian noise of standard deviation η=Nshots−1/2\eta=N_{\mathrm{shots}}^{-1/2} on the moments (solid); the dotted line marks η\eta. (b) Condition number κ(S)=λmax⁡/λmin⁡\kappa(S)=\lambda_{\max}/\lambda_{\min} of the overlap matrix SS versus DD for each construction (noiseless). The dash-dotted lines, nearly coincident for the two constructions, mark λmax⁡(S)/ε\lambda_{\max}(S)/\varepsilon with ε=10−13\varepsilon=10^{-13}: above them, the smallest eigenvalue of SS falls below the threshold and the corresponding direction is discarded. The dotted line marks the inverse of the double-precision machine accuracy, ≈4.5×1015\approx4.5 \times10^{15}, beyond which the smallest eigenvalue of SS is at the level of round-off.

This motivates asking whether the Krylov space can be constructed using degrees with sufficiently large success probability. In Figure 6, we therefore construct the Krylov space a priori from degrees satisfying psuccess(Tk)≥1/2p_{\mathrm{success}}^{(T_k)}\geq1/2. The selected-degree construction keeps the condition number κ(S)\kappa(S) below that of the standard contiguous construction for 3≤D≤83\leq D\leq8, before both reach the round-off limit, and the improvement in convergence is already visible in the noiseless calculation. The corresponding finite-shot calculations retain the same trend. This improvement comes at the cost of a larger maximum Chebyshev degree and therefore deeper circuits. Thus, the success probability is not only a shot-overhead indicator for individual moments; it also provides a direct diagnostic for identifying Chebyshev degrees whose corresponding Krylov vectors may contribute to poor conditioning.

QSVT success probability

QSVT generalizes the qubitized walk from a single Chebyshev polynomial to a broad class of polynomial transformations P(H~)P(\widetilde{H}) satisfying the required parity and boundedness conditions [3, 20], implemented by the phase sequence of Eqs. (9) and (10). QSVT introduces a qualitatively different normalization problem: unlike LCU and qubitization block-encodings, where the scaling is inherited from the LCU 1-norm, the polynomial transformation introduces an additional scaling factor that directly affects both the success probability and the feasibility of phase synthesis.

We write the target polynomial, before scaling, in the Chebyshev basis,

P(x)=∑k=0dakTk(x),(45)P(x)=\sum_{k=0}^{d}a_kT_k(x), \tag*{(45)}

where dd is its degree (ad≠0a_d\ne0) and the real coefficients aka_k vanish whenever kk and dd have different parity, so that PP has the definite parity of dd. Let C>0C>0 denote the scaling factor used to obtain the QSVT-feasible polynomial f(x)=P(x)/Cf(x)=P(x)/C. The resulting success probability is

psuccess(P/C)(∣ψ⟩)=∥P(H~)∣ψ⟩∥2C2=1C2∑nwn∣P(E~n)∣2.(46)p_{\mathrm{success}}^{(P/C)}(|\psi\rangle)=\frac{\|P(\widetilde{H})|\psi\rangle\|^2}{C^2}=\frac{1}{C^2}\sum_n w_n\lvert P(\widetilde{E}_n)\rvert^2. \tag*{(46)}

For a given target polynomial and input state, the numerator is fixed, so the scaling factor CC directly controls the QSVT success probability. It is constrained by the feasibility condition sup⁡x∈[−1,1]∣P(x)/C∣≤1\sup_{x\in[-1,1]}|P(x)/C|\leq1 [3, 20], which

requires

C≥Copt≡sup⁡x∈[−1,1]∣P(x)∣.(47)C \ge C_{\mathrm{opt}} \equiv\sup_{x\in[-1,1]} |P(x)|. \tag*{(47)}

For a fixed target polynomial, the success probability is therefore maximized by choosing CC as close as possible to CoptC_{\mathrm{opt}} from above, so that feasibility is maintained. Estimating or certifying CoptC_{\mathrm{opt}} is thus central to the analysis below. Unless stated otherwise, the target polynomials are randomly generated in the Chebyshev basis with definite parity.

Choice of the polynomial scaling factor

Several constants bound CoptC_{\mathrm{opt}} from above, and they form an ordered chain

Cgrid≲Cdense≲Cnum≈Copt≤CMarkov≤Λ≤∥a∥1,(48)\begin{aligned} C_{\mathrm{grid}} &\lesssim C_{\mathrm{dense}} \lesssim C_{\mathrm{num}} \approx C_{\mathrm{opt}}\\ &\le C_{\mathrm{Markov}} \le\Lambda\le\lVert a\rVert_1, \tag*{(48)} \end{aligned}

where CgridC_{\mathrm{grid}} and CdenseC_{\mathrm{dense}} are the maxima of ∣P∣|P| sampled on a coarse (4000 points) and a fine (200000 points) grid, CnumC_{\mathrm{num}} denotes a numerical evaluation of CoptC_{\mathrm{opt}}, obtained by solving P′(x)=0P'(x)=0 and evaluating PP at the resulting stationary points and the interval endpoints, CMarkovC_{\mathrm{Markov}} is a certified bound derived below, Λ=max⁡∣z∣=1∣∑kakzk∣\Lambda=\max_{|z|=1}\left|\sum_k a_kz^k\right| is the sup norm of the coefficient series on the unit circle, and ∥a∥1=∑k∣ak∣\lVert a\rVert_1=\sum_k|a_k| is the coefficient ℓ1\ell_1-norm. The inequalities involving CgridC_{\mathrm{grid}} and CdenseC_{\mathrm{dense}} are empirical properties of the sampled grids, not certified bounds: they provide lower estimates of the supremum because they only evaluate PP at discrete points. Consequently, normalizing with an undershooting grid estimate can produce a polynomial that exceeds unit magnitude between grid points. The three quantities to the right of CoptC_{\mathrm{opt}} provide valid upper bounds.

The maximum of ∣P∣|P| on the closed interval [−1,1][-1,1] is attained either at a critical point, where P′(x)=0P'(x)=0, or at an endpoint x=±1x=\pm1. Using Eq. (45) and Tk′=kUk−1T'_k=kU_{k-1} with UU the second-kind Chebyshev polynomials, the derivative is

P′(x)=∑k=1dakkUk−1(x).(49)P'(x)=\sum_{k=1}^{d}a_k kU_{k-1}(x). \tag*{(49)}

The stationary points of PP in [−1,1][-1,1] are obtained numerically as the real roots of P′P', computed as the eigenvalues of its companion matrix. Evaluating ∣P(x)∣|P(x)| at these points and at the endpoints x=±1x=\pm1 gives the numerical estimate CnumC_{\mathrm{num}} of sup⁡x∈[−1,1]∣P(x)∣\sup_{x\in[-1,1]}|P(x)| without requiring a dense grid. Unlike the Markov construction below, this critical-point procedure is not a certified upper bound.

Normalization by the infinity norm is standard in QSVT, and the numerical practice is to estimate it by grid sampling. A grid maximum undershoots the true supremum, however, and so gives no guarantee of the feasibility condition sup⁡∣P/C∣≤1\sup|P/C|\le1. The grid estimate can be corrected upward into a certified bound by accounting for the largest rise the polynomial can hide between two nodes. Let x⋆x^\star be the true maximizer, ∣P(x⋆)∣=Copt|P(x^\star)|=C_{\mathrm{opt}}. With nodes spaced hh apart, x⋆x^\star lies within h/2h/2 of some node xix_i, and by the mean value theorem [52], there is a ξ\xi between xix_i and x⋆x^\star with

Copt=∣P(x⋆)∣≤∣P(xi)∣+∣P′(ξ)∣∣x⋆−xi∣≤Cgrid+h2max⁡x∈[−1,1]∣P′(x)∣.(50)\begin{aligned} C_{\mathrm{opt}}=|P(x^\star)|\le|P(x_i)|+|P'(\xi)||x^\star-x_i| \\ &\le C_{\mathrm{grid}}+\frac{h}{2}\max_{x\in[-1,1]}|P'(x)|. \tag*{(50)} \end{aligned}

where we used the triangle inequality, ∣P(xi)∣≤Cgrid|P(x_i)|\le C_{\mathrm{grid}}, which holds because xix_i is itself one of the nodes over which Cgrid=max⁡i∣P(xi)∣C_{\mathrm{grid}}=\max_i|P(x_i)| is taken (the grid includes both endpoints x=±1x=\pm1), and ∣x⋆−xi∣≤h/2|x^\star-x_i|\le h/2. The derivative is bounded by the Markov inequality [53, 54], which holds for any polynomial of degree dd on [−1,1][-1,1], with dd the degree of the target polynomial PP of Eq. (45):

max⁡x∈[−1,1]∣P′(x)∣≤d2max⁡x∈[−1,1]∣P(x)∣=d2Copt.(51)\max_{x\in[-1,1]}|P'(x)|\le d^2\max_{x\in[-1,1]}|P(x)|=d^2C_{\mathrm{opt}}. \tag*{(51)}

Substituting into Eq. (50) and solving for CoptC_{\mathrm{opt}} gives

Copt≤CMarkov≡Cgrid1−h2d2,(52)C_{\mathrm{opt}}\le C_{\mathrm{Markov}}\equiv\frac{C_{\mathrm{grid}}}{1-\frac{h}{2}d^2}, \tag*{(52)}

valid whenever hd2/2<1hd^2/2<1, which requires a grid sufficiently fine relative to the polynomial degree. Unlike CnumC_{\mathrm{num}}, Eq. (52) provides a provable upper bound and therefore guarantees feasibility.

Since the numerator of Eq. (46) is fixed by the target, the choice among the constants of Eq. (48) directly sets psuccess(P/C)p_{\mathrm{success}}^{(P/C)}, as illustrated in Figure 7(a) for a fixed degree-7 target. The certified Markov bound CMarkov=2.71C_{\mathrm{Markov}}=2.71 gives psuccess(P/C)≈0.45p_{\mathrm{success}}^{(P/C)}\approx0.45, only slightly below the tight value ≈0.46\approx0.46, whereas the loose bound Λ=4.22\Lambda=4.22 reduces the success probability to ≈0.18\approx0.18. Therefore, for this example, a looser normalization alone produces a (Λ/Cnum)2≈2.5×(\Lambda/C_{\rm num})^2 \approx2.5\times increase in the shot overhead. This ordering persists across the correlation-strength sweep considered here: Figure 7(b) shows the same separation between CnumC_{\rm num}, CMarkovC_{\rm Markov}, and the looser Λ\Lambda and ∥a∥1\lVert a\rVert_1 choices across the full U/tU/t range studied. The gap between the tight and loose constants is expected to widen with the polynomial degree. Λ\Lambda and ∥a∥1\lVert a\rVert_1 depend only on the Chebyshev coefficients of the target and accumulate their magnitudes, e.g., ∥a∥1=d+1\lVert a\rVert_1=d+1 for constant coefficients, whereas Copt=sup⁡[−1,1]∣P∣C_{\rm opt}=\sup_{[-1,1]}|P| can remain much smaller owing to cancellations between Chebyshev terms of different order at each xx (Sec. III D 3). In contrast, the classical cost of a tight estimate is independent of the system size: CnumC_{\rm num} requires the roots of P′P', i.e., the eigenvalues of a (d−1)×(d−1)(d-1)\times(d-1) companion matrix, at cost O(d3)O(d^3), and CMarkovC_{\rm Markov} requires evaluating PP on a grid with h<2/d2h<2/d^2, i.e., O(d2)O(d^2) nodes at O(d)O(d) cost each, hence O(d3)O(d^3). This one-time classical cost is negligible compared with the shot overhead (C/Copt)2(C/C_{\rm opt})^2 incurred at every execution of the circuit with a loose normalization, so the certified bound CMarkovC_{\rm Markov} should be preferred in practice.

Figure 7. QSVT success probability and the cost of the normalization constant, for the fixed degree-7 target P=−2.37T1+1.13T3+0.72T7P=-2.37T_{1}+1.13T_{3}+0.72T_{7} and an N=2N=2 HF input state. (a) Success probability at U=4tU=4t for each normalization choice, with the value of CC under each label. The unscaled numerator is ∥P(H^)∣ψHF⟩∥2=3.278\lVert P(\hat{H})\lvert\psi_{\mathrm{HF}}\rangle\rVert^{2}=3.278; CnumC_{\mathrm{num}}, CMarkovC_{\mathrm{Markov}} and Λ\Lambda (equal to ∥a∥1\lVert a\rVert_{1} for this target) give psuccess(P/C)≈0.46p_{\mathrm{success}}^{(P/C)}\approx0.46, 0.450.45 and 0.180.18, respectively. (b) Success probability as a function of U/tU/t (horizontal axis scaled as U/(U+4t)U/(U+4t); dotted line: U=4tU=4t, the value used in (a)). The curves for CgridC_{\mathrm{grid}} (dashed) and CdenseC_{\mathrm{dense}} (dotted) coincide with CnumC_{\mathrm{num}}, and the curve for ∥a∥1\lVert a\rVert_{1} (dashed) coincides with Λ\Lambda. The tight estimates CnumC_{\mathrm{num}} and CMarkovC_{\mathrm{Markov}} form the upper group, while the looser bounds Λ\Lambda and ∥a∥1\lVert a\rVert_{1} give success probabilities smaller by a factor ≈2.5\approx2.5 across the U/tU/t range studied.

Effect on angle finding

The choice of CC represents a success probability trade-off, and it also determines whether phase finding is feasible and how robustly the numerical solver converges. The QSVT phase factors implementing the scaled operator P/CP/C are found by the Newton iteration of the sym_qsp solver of pyqsp, following the phase-factor evaluation approach of [22, 23], whose numerical conditioning depends on the distance between the normalized target P/CP/C and the feasibility boundary sup⁡∣P/C∣=1\sup|P/C|=1. Three regimes emerge, separated by the sign and size of the margin C−CoptC-C_{\rm opt}.

When C<CoptC<C_{\rm opt}, the scaled polynomial violates the necessary boundedness condition sup⁡[−1,1]∣P/C∣≤1\sup_{[-1,1]}|P/C|\le1, so it is not a feasible QSVT target on the full interval. In the numerical solver, this manifests as failure to converge to the required residual; we therefore reject phase sequences whose residual exceeds 10−910^{-9}. This is the case for Csub=0.95 CdenseC_{\rm sub}=0.95\,C_{\rm dense}, about 5% below the supremum: phase finding does not converge at any tested degree (d=2d=2–88, Table IV), and for the degree-5 target the realized state keeps an infidelity 1−F∼10−11-F\sim10^{-1} whatever the solver tolerance (Figure 8). When CC is obtained from a grid estimate and chosen with only a grid-scale margin relative to CoptC_{\rm opt}, phase finding can be fragile. The grid estimates CgridC_{\rm grid} and CdenseC_{\rm dense} are not certified: although they often lie extremely close to the true supremum, they can undershoot it when a polynomial maximum falls between sampled nodes. In such cases, the nominal margin C−CoptC-C_{\rm opt} becomes negative and the solver can stall or converge to an inaccurate transformation.

ddCsubC_{\mathrm{sub}} iterCsubC_{\mathrm{sub}} residualCgridC_{\mathrm{grid}} iterCgridC_{\mathrm{grid}} residualCnumC_{\mathrm{num}} iterCnumC_{\mathrm{num}} residualCMarkovC_{\mathrm{Markov}} iterCMarkovC_{\mathrm{Markov}} residual
2–5.9×10−25.9\times10^{-2}205.5×10−135.5\times10^{-13}205.5×10−135.5\times10^{-13}95.6×10−175.6\times10^{-17}
3–1.1×10−11.1\times10^{-1}–1.3×10−71.3\times10^{-7}207.8×10−137.8\times10^{-13}88.9×10−168.9\times10^{-16}
4–7.9×10−27.9\times10^{-2}205.5×10−135.5\times10^{-13}205.5×10−135.5\times10^{-13}82.5×10−162.5\times10^{-16}
5–1.0×10−11.0\times10^{-1}–5.1×10−65.1\times10^{-6}207.3×10−137.3\times10^{-13}72.0×10−132.0\times10^{-13}
6–7.6×10−17.6\times10^{-1}206.1×10−136.1\times10^{-13}206.1×10−136.1\times10^{-13}78.9×10−168.9\times10^{-16}
7–2.32.3205.5×10−135.5\times10^{-13}205.5×10−135.5\times10^{-13}71.9×10−161.9\times10^{-16}
8–1.5×10−11.5\times10^{-1}–7.9×10−77.9\times10^{-7}205.7×10−135.7\times10^{-13}73.2×10−163.2\times10^{-16}

Table IV. Phase-finding iterations and final residual errors of sym_qsp as a function of target degree dd for the normalization constants considered (same targets as Figure 8 for d=5d=5). The infeasible Csub=0.95CdenseC_{\mathrm{sub}}=0.95C_{\mathrm{dense}} never reaches the required residual. The grid-based normalization fails to reach it at d=3d=3, 55 and 88, reflecting its lack of a certified margin above CoptC_{\mathrm{opt}}. The certified CMarkovC_{\mathrm{Markov}} converges at all tested degrees with a small positive margin.

Figure 8. State infidelity 1−F1-F of the realized QSVT circuit versus the tolerance requested from the pyqsp phase solver, for the normalization constants Csub=0.75C_{\mathrm{sub}}=0.75, Cgrid≈Cdense≈Cnum=0.79C_{\mathrm{grid}}\approx C_{\mathrm{dense}}\approx C_{\mathrm{num}}=0.79, CMarkov=0.795C_{\mathrm{Markov}}=0.795 and Λ=∥a∥1=0.898\Lambda=\lVert a\rVert_1=0.898 (the last two curves coincide). Degree-5 target P=0.13T1−0.13T3+0.64T5P=0.13T_1-0.13T_3+0.64T_5, N=2N=2, U/t=4U/t=4, HF input state. The dashed curve (CsubC_{\mathrm{sub}}) is infeasible. The horizontal dotted line marks the acceptance threshold 1−F=10−101-F=10^{-10} on the state infidelity.

If the margin is small but positive, phase finding is both fast and accurate. The certified CMarkovC_{\rm Markov} clears the supremum by about one percent and converges in 7–9 Newton iterations to a residual of order 10−1310^{-13} or below at every degree tested; the numerical estimate CnumC_{\rm num} converges in a fixed 20 iterations to ∼10−13\sim10^{-13}. Figure 8 shows how the choice of CC sets the precision of the realized state as the tolerance requested from the phase solver is tightened. Down to a tolerance of ∼10−5\sim10^{-5}, CgridC_{\rm grid}, CdenseC_{\rm dense} and CnumC_{\rm num}, which all round to 0.79, give the same state infidelity. Beyond this point the two grid estimates saturate, at 1−F∼10−61-F\sim10^{-6} for CgridC_{\rm grid} and ∼10−9\sim10^{-9} for CdenseC_{\rm dense}, consistent with a slight undershoot of CoptC_{\rm opt}, whereas CnumC_{\rm num} keeps improving and reaches ∼10−12\sim10^{-12} at a tolerance of 10−1110^{-11}. The margined constants reach a higher precision at much looser tolerances: 1−F≲10−121-F\lesssim10^{-12} once the tolerance reaches 10−410^{-4}–10−510^{-5}, and machine precision (∼10−14\sim10^{-14}) from ∼10−7\sim10^{-7}, for CMarkovC_{\rm Markov}, Λ\Lambda and ∥a∥1\lVert a\rVert_1. With an acceptance threshold of 1−F=10−61-F=10^{-6} on the state infidelity, the two grid estimates never qualify but CMarkovC_{\rm Markov} already qualifies from ∼10−5\sim10^{-5}. A smaller scaling factor therefore gives a higher success probability at the cost of precision on the phases, and hence on the prepared state. CMarkovC_{\rm Markov} resolves this trade-off favorably: for this target, its success probability is only a factor (CMarkov/Cnum)2≈1.01(C_{\rm Markov}/C_{\rm num})^2\approx1.01 below that of CnumC_{\rm num}, against ≈1.3\approx1.3 for Λ\Lambda and ∥a∥1\lVert a\rVert_1, while it reaches the same machine-precision fidelity as the looser bounds. For the targets studied here, CMarkovC_{\rm Markov} is therefore the operationally preferred choice: it is the tightest certified bound considered, preserving a high success probability while providing a small positive feasibility margin that ensures both the convergence of phase finding and an accurate state preparation. This result applies to the compilation of any QSVT circuit, since normalizing the target polynomial is a compilation step common to all QSVT applications. The Markov certificate turns a cheap grid estimate into a provable bound that guarantees the compiled phase sequence implements the intended polynomial, at a cost of a few percent in success probability for the targets studied here, while reaching machine-precision fidelity at solver tolerances several orders of magnitude looser than the tight numerical estimate.

Dependence on polynomial degree and coefficient growth

The QSVT success probability depends on the polynomial degree through the interplay between the normalization Cd=sup⁡x∈[−1,1]∣Pd(x)∣C_d=\sup_{x\in[-1,1]}|P_d(x)|, which enters as 1/Cd21/C_d^2, and the spectral response ∥Pd(H~)∣ψ⟩∥2\lVert P_d(\tilde{H})|\psi\rangle\rVert^2. We index the target by its degree dd, since we study how the success probability depends on dd, and write Eq. (45) as

Pd(x)=∑j=0⌊d/2⌋ad−2jTd−2j(x).(53)P_d(x)=\sum_{j=0}^{\lfloor d/2\rfloor}a_{d-2j}T_{d-2j}(x). \tag*{(53)}

The way psuccess(P/C)p_{\rm success}^{(P/C)} scales with dd is then governed by how fast the coefficients aka_k grow. In Krylov-type applications, the Lanczos polynomials expressed in the Chebyshev basis have coefficients that grow with degree [7]; we therefore consider polynomial and exponential growth laws to bracket this behavior. Consider first coefficients that are constant or grow polynomially, ak=(k+1)αa_k=(k+1)^\alpha with α≥0\alpha\ge0. Two sums then control the success probability. Since ak≥0a_k\ge0 and ∣Tk(x)∣≤1\lvert T_k(x)\rvert\le1, the normalization is reached at x=1x=1, where Tk(1)=1T_k(1)=1 for every kk, so that all terms of PdP_d add coherently: Cd=∑kak=O(dα+1)C_d=\sum_k a_k=O(d^{\alpha+1}). For a spectral weight spread uniformly over the angles θn\theta_n, the Chebyshev polynomials are orthogonal on average [55], so the cross terms of Pd2P_d^2 cancel and only the squared coefficients remain: ⟨Pd2⟩=a02+(1/2)∑k≥1ak2=O(d2α+1)\langle P_d^2\rangle=a_0^2+(1/2)\sum_{k\ge1}a_k^2=O(d^{2\alpha+1}). The dd terms of PdP_d thus add coherently in the normalization but

incoherently in the numerator, so that

psuccess,flat(P/C)(d)=⟨Pd2⟩Cd2=O(d−1),(54)p_{\mathrm{success,flat}}^{(P/C)}(d)=\frac{\langle P_d^2\rangle}{C_d^2}=O(d^{-1}), \tag*{(54)}

independently of α\alpha: polynomial coefficient growth changes the normalization but not the d−1d^{-1} envelope of the success probability. For the concentrated HF input state the decay is faster than this flat-spectrum estimate: in Figure 9(a) and 9(b) (N=2N=2, U=4tU=4t), a power-law fit for d≥10d\ge10 shows that the upper envelope of psuccess(P/C)p_{\mathrm{success}}^{(P/C)} decays as d−γd^{-\gamma}, with γ≈1.9\gamma\approx1.9 for constant coefficients and γ≈1.8\gamma\approx1.8 for both linear (not shown) and quadratic growth. The exponent thus depends only weakly on the coefficient profile, indicating that the decay is set by the input-state spectrum. For exponentially growing coefficients, $a_k=r^k with r>1r>1, the highest-degree terms dominate both the normalization, Cd=Θ(rd)C_d=\Theta(r^d), and the polynomial itself: the normalized polynomial Pd/CdP_d/C_d is a combination of its last few Chebyshev terms, with weights decaying as r−2jr^{-2j}. As a result, the success probability oscillates with dd around an approximately constant level, of order 10−210^{-2} for r=1.2r=1.2 (Figure 9(c)), even though CdC_d grows by about 8 orders of magnitude. Exponential coefficient growth alone thus does not drive the success probability to zero. For generic random coefficient profiles, both the normalization and the spectral response fluctuate between realizations. The median success probability over draws decreases over the first ∼20\sim20 degrees and then remains nearly constant at ≈0.1\approx0.1 (Figure 9(d)).

Figure 9. QSVT success probability versus polynomial degree dd for four coefficient profiles, N=2N = 2, U=4tU = 4t, HF input state. Upper panels: psuccess(P/C)p_{\mathrm{success}}^{(P/C)}, split by the parity of dd; lower panels: normalization Cd=sup⁡[−1,1]∣Pd∣C_d = \sup_{[-1,1]} |P_d|. All targets have the parity of dd, Pd=∑jad−2jTd−2jP_d = \sum_j a_{d-2j}T_{d-2j}. (a) Constant (ak=1a_k = 1) and (b) quadratically growing (ak=(k+1)2a_k = (k + 1)^2) coefficients: the upper envelope decays as d−γd^{-\gamma}, with γ≈1.9\gamma\approx1.9 and 1.81.8 (dashed lines, fit for d≥10d \ge10). (c) Exponential coefficients (ak=1.2ka_k = 1.2^k): CdC_d grows by about 8 orders of magnitude, while psuccess(P/C)p_{\mathrm{success}}^{(P/C)} oscillates without decaying. (d) Random targets: median over 40 draws (line) and interquartile range (band).

These observations bear on Krylov and Lanczos-based transformations, where high-degree polynomials can entail rapidly growing normalization [7]. A large normalization alone does not determine the success probability: the spectral response of the polynomial on the input-state spectrum matters equally. The precise interplay of coefficient growth, spectral structure, and Krylov convergence is problem dependent, so the success probability should be monitored alongside the Krylov dimension and polynomial degree.

Dependence on system size

The system-size dependence differs between the even and odd target families. Figure 10(a) and Figure 10(b) show the success probability across U/tU/t for system sizes N=2N=2 to 10 at fixed odd- and even-degree targets. For the odd target the success probability rises with U/tU/t and orders regularly with system size, the curves bunching for N≥4N\geq4; for the even target it falls with U/tU/t and the N=2N=2 curve sits above the rest. As for LCU (Sec. III B), the largest finite-size changes occur between the smallest systems, with no systematic decay of psuccess(P/C)p_{\mathrm{success}}^{(P/C)} for N≥4N\geq4. At small U/tU/t, the HF state is close to the ground state (participation ratio ≈1\approx1), so that psuccess(P/C)≈∣P(E~0)∣2/C2p_{\mathrm{success}}^{(P/C)}\approx|P(\tilde E_0)|^2/C^2 is set by the position of the normalized ground-state energy relative to the zeros of PP. At U=0U=0, E~0=−Ωs/λs=−pbound\tilde E_0=-\Omega_s/\lambda_s=-\sqrt{p_{\mathrm{bound}}}, which equals −1-1 for N=2N=2 but ≈−0.85\approx-0.85 for N=4N=4 (Sec. III B). As UU increases, E~0\tilde E_0 moves away from −1-1; for N=2N=2 it crosses the zero of the even target at x≈−0.92x\approx-0.92 between U=0U=0 and U=t/2U=t/2, which produces the isolated low value of the N=2N=2 curve at U=t/2U=t/2 in Figure 10(b). For larger NN, E~0\tilde E_0 starts beyond this zero and the dip is absent. The same mechanism, with the zero of the odd target at x≈−0.77x\approx-0.77, produces the low N=2N=2 value at U=tU=t in Figure 10(a).

Figure 10. System-size and interaction-strength dependence of the QSVT success probability for fixed target polynomials of odd and even parity, HF input state, N=2N = 2–1010 (color as in Figure 1(b)). (a) Odd-degree target P(x)=0.10T1−0.54T3+0.36T5P(x) = 0.10T_{1} - 0.54T_{3} + 0.36T_{5}. (b) Even-degree target P(x)=0.13T0−0.13T2+0.64T4P(x) = 0.13T_{0} - 0.13T_{2} + 0.64T_{4}; the isolated low value of the N=2N = 2 curve at U=t/2U = t/2 is discussed in the text. Lines show exact (matrix-free) values computed without sampling; triangles show sampled-circuit estimates (Nshots=20000N_{\mathrm{shots}} = 20000), which agree with them to within 0.01. The horizontal axis is scaled as U/(U+4t)U/(U + 4t). The two target families exhibit opposite U/tU/t trends.

Conclusion

We have investigated the success probability of block-encoding-based quantum algorithms and shown that, for a given encoded function ff of a Hermitian operator HH, it is jointly determined by the spectral distribution of the input state, the spectrum of HH, and the normalization. All numerical tests have been done on the Hubbard model at half filling, in which the spectral structure can be validated and the relevant quantities can be computed over a broad range of system sizes; we expect the mechanisms identified here to hold beyond this case, while the specific values are model dependent. This provides a common perspective on LCU, qubitization and QSVT block-encodings, in which the success probability can be written as psuccess(f)=∥f(H~)∣ψ⟩∥2=∑nwn∣f(E~n)∣2p_{\mathrm{success}}^{(f)}=\lVert f(\tilde H)|\psi\rangle\rVert^2=\sum_n w_n|f(\tilde E_n)|^2, a quantity that can be analyzed independently and included in resource estimation.

For LCU, the success probability is constrained by the squared ratio (Ωs/λs)2(\Omega_s/\lambda_s)^2 between the spectral norm of the shifted Hamiltonian and the LCU 1-norm of its Pauli decomposition. Within this bound, the success probability depends strongly on the input state: the half-filled ground state nearly saturates it, whereas the HF input state generally does not.

Qubitization introduces the Chebyshev degree kk as an additional control parameter, and the resulting success probability oscillates with kk. For a fixed block-encoding and input state, different degrees can therefore give substantially different shot overheads. For the HF input state, this degree dependence is largely captured by a small number of dominant spectral components: a spectrum-informed truncation identifies favorable degrees from a few dominant eigenstates, while a spectrum-free model fitted to shallow-degree data predicts favorable deeper degrees without classical diagonalization.

For QSVT, the polynomial scaling factor CC provides a different form of control. For a fixed target polynomial and input state, the success probability scales as 1/C21/C^2, so tighter normalization directly improves it. However, choosing CC close to the true supremum CoptC_{\mathrm{opt}} introduces a practical trade-off: an undershooting value is infeasible, while a barely feasible value places phase finding close to an ill-conditioned boundary. Among the estimates and bounds of CoptC_{\mathrm{opt}} compared here, the certified bound CMarkovC_{\mathrm{Markov}} that we introduce for QSVT normalization, obtained from grid samples through the Markov inequality, provides the best compromise: it stays within about one percent of the tight value while leaving a positive feasibility margin for robust phase synthesis. For the N=2N = 2 Hubbard chain studied, the QSVT success probability decays with polynomial degree (upper envelope ∝d−1.8\propto d^{-1.8}–d−1.9d^{-1.9} for the HF input state), nearly independently of the coefficient growth for constant and polynomially growing coefficients, while exponentially growing coefficients prevent it from decaying despite a normalization that grows by orders of magnitude.

Across the three methods, the numerical results show that success probability does not have a single universal dependence on system size, interaction strength, or transformation degree. Instead, for each construction, its behavior reflects the combined effect of spectral structure, normalization and input state. In particular, for the Hubbard systems studied here, no systematic decay of the success probability toward zero is observed with system size (up to 35 qubits for LCU block-encoding), and for the HF input it is quasi-independent of the input state’s fidelity to the ground state.

Finally, the application to quantum Krylov illustrates why these observations matter. The moments required by a Krylov calculation are associated with different Chebyshev degrees and therefore with different values of psuccess(Tk)p_{\mathrm{success}}^{(T_k)}. Even when the moments are measured without postselection, psuccess(Tk)p_{\mathrm{success}}^{(T_k)} is the squared norm of the corresponding Krylov vector, so degrees with small psuccess(Tk)p_{\mathrm{success}}^{(T_k)} require substantially more shots to be resolved to a given relative precision. Beyond this per-degree shot overhead, the success probability also serves as a construction criterion: selecting the Krylov space from degrees with high success probability improves the conditioning of the overlap matrix and the resulting convergence, at the cost of deeper circuits. The success probability therefore adds a cost that is not captured by gate count, circuit depth or qubit count alone, and should be reported as part of the resource estimation of block-encoding-based algorithms. Amplitude amplification [37] can reduce the number of repetitions from O(1/psuccess(f))O(1/p_{\mathrm{success}}^{(f)}) to O(1/psuccess(f))O(1/\sqrt{p_{\mathrm{success}}^{(f)}}), at the cost of O(1/psuccess(f))O(1/\sqrt{p_{\mathrm{success}}^{(f)}}) additional applications of the block-encoding per run; the success probability therefore remains the quantity that sets the overhead, now in circuit depth as well as in repetitions.

Extending our numerical analysis beyond the one-dimensional Hubbard model, e.g., to molecules and larger many-electron systems, represents a natural way forward. Another important outlook is the application to quantum Krylov subspace methods, and in particular to our recent formulation [7], which requires a QSVT polynomial transformation without definite parity. Such a polynomial is typically implemented as a linear combination of its even and odd parts, at the cost of an additional normalization factor that can reduce the success probability by up to a factor of 4; the framework developed here directly quantifies this cost. Finally, the success probability of block-encodings used for solving linear systems [1, 2] and partial differential equations [4] remains to be evaluated.

Acknowledgments

We thank TotalEnergies for permission to publish this work, Keith Gray, Donny Cooper and Harry Zong for providing and maintaining the CPU and GPU computing resources used for the simulations, Henri Calandra and Jean-Patrick Mascomère for insightful discussions and comments, and Hadi Rammal for fruitful discussions and collaboration. This work was supported in part by the Maison du Quantique de Nouvelle-Aquitaine “HybQuant”, as part of the HQI initiative and France 2030, under the French National Research Agency (ANR) grant ANR-22-PNCQ-0002, by the PEPR Quantique under Grant No. ANR-23-PETQ-0006, and by the EUR Light S&T (PIA3 Program, Grant No. ANR-17-EURE-0027). Circuit simulations were performed with the Quantum Learning Machine (myQLM) software [56]. The authors used a large language model as an assistant for proofreading and editing parts of the text.

Readout statistics of the Chebyshev moments

This appendix derives the statistics of the three readouts of the Chebyshev moments compared in Sec. III C 2 and Table III: the Hadamard test used in this work and the two readouts of Kirby et al. [34]. Throughout, an outcome of probability pp occurs about NpNp times in NN shots, with a fluctuation of about Np(1−p)\sqrt{Np(1-p)}, so that estimating pp itself to a relative precision δ\delta requires N≃(1−p)/(pδ2)N \simeq(1-p)/(p\delta^2) shots; whether this 1/p1/p cost appears depends on what is counted.

a. Hadamard test. A test qubit prepared in ∣+⟩|+\rangle controls kk applications of WW on ∣Ψ0⟩=∣PREP⟩⊗∣ψ⟩|\Psi_0\rangle= |\mathrm{PREP}\rangle\otimes|\psi\rangle, and after a second Hadamard gate, its outcome probabilities satisfy

p0−p1=Re⁡⟨Ψ0∣Wk∣Ψ0⟩=Re⁡⟨ψ∣(⟨PREP∣Wk∣PREP⟩)∣ψ⟩=mk,(55)\begin{aligned} p_0-p_1 &= \operatorname{Re}\langle\Psi_0|W^k|\Psi_0\rangle\\ &= \operatorname{Re}\langle\psi|(\langle\mathrm{PREP}|W^k|\mathrm{PREP}\rangle)|\psi\rangle= m_k, \tag*{(55)} \end{aligned}

by Eq. (8), mkm_k being real since Tk(H~)T_k(\widetilde{H}) is Hermitian. The projection onto ∣PREP⟩|\mathrm{PREP}\rangle that defines the block-encoding is therefore contained in the overlap with the initial state: the ancilla register is never measured, and every shot, with outcome ±1\pm1, contributes to the estimate of mkm_k.

b. Readouts of Kirby et al. Both start from the walk state

∣Ψj⟩=Wj(∣PREP⟩⊗∣ψ⟩),j=⌊k/2⌋,(56)|\Psi_j\rangle= W^j\left(|\mathrm{PREP}\rangle\otimes|\psi\rangle\right), \qquad j=\lfloor k/2\rfloor, \tag*{(56)}

and use ⟨Ψj∣RPREP∣Ψj⟩=m2j\langle\Psi_j|R_{\mathrm{PREP}}|\Psi_j\rangle=m_{2j} and ⟨Ψj∣SELECT∣Ψj⟩=m2j+1\langle\Psi_j|\mathrm{SELECT}|\Psi_j\rangle=m_{2j+1} [34].

c. Even kk. RPREPR_{\mathrm{PREP}} is measured by applying PREP†\mathrm{PREP}^{\dagger} and recording +1+1 if a=0a=0 and −1-1 otherwise. The probability of the +1+1 outcome is

Pr⁡(a=0)=∥(⟨PREP∣⊗I)∣Ψk/2⟩∥2=psuccess(Tk/2)=1+mk2,(57)\begin{aligned} \operatorname{Pr}(a=0)=\left\|(\langle\mathrm{PREP}|\otimes I)|\Psi_{k/2}\rangle\right\|^2 \\ &=p_{\mathrm{success}}^{(T_{k/2})}=\frac{1+m_k}{2}, \tag*{(57)} \end{aligned}

i.e., exactly the block-encoding success probability of the half-degree walk, using Tk/22=(1+Tk)/2T_{k/2}^2=(1+T_k)/2. Since mk=2Pr⁡(a=0)−1m_k=2\operatorname{Pr}(a=0)-1, failures are not discarded but carry information on mkm_k, and the error is (1−mk2)/N\sqrt{(1-m_k^2)/N}, the same as for the Hadamard test.

d. Odd kk. SELECT\mathrm{SELECT} is measured term by term [34]: for term ii, a layer of single-qubit Clifford gates maps PiP_i to a diagonal operator, the ancilla and system registers are measured in the computational basis, and the shot returns the measured eigenvalue of PiP_i if a=ia=i, and 00 otherwise. Only the shots with a=ia=i, which occur with probability qk(i)q_k(i), carry information on term ii, so the relative error on its contribution scales as 1/Niqk(i)1/\sqrt{N_iq_k(i)}, where NiN_i is the number of shots spent on that term. With the optimal allocation Ni∝qk(i)N_i\propto\sqrt{q_k(i)}, the total cost for a given error on mkm_k exceeds that of the even readout, in the worst case, by a factor (∑iqk(i))2\left(\sum_i\sqrt{q_k(i)}\right)^2, which lies between 11 (a single dominant term) and LL (equal weights); grouping qubitwise commuting Pauli strings reduces LL to the number of groups. This overhead is not an artifact of the bounds. Writing eie_i for the mean outcome of a shot on term ii, with ∣ei∣≤qk(i)|e_i|\le q_k(i) and mk=∑ieim_k=\sum_i e_i, the variance per shot on term ii is qk(i)−ei2q_k(i)-e_i^2, whereas that of the Hadamard test and of the even readout is 1−mk21-m_k^2, and

(∑iqk(i)−ei2)2≥1−mk2,(58)\left(\sum_i\sqrt{q_k(i)-e_i^2}\right)^2\ge1-m_k^2, \tag*{(58)}

since, after expanding both squares, each cross term satisfies (qk(i)−ei2)(qk(l)−el2)+eiel≥0\sqrt{(q_k(i)-e_i^2)(q_k(l)-e_l^2)}+e_ie_l\ge0, which follows from (qk(i)−ei2)(qk(l)−el2)≥∣eiel∣2(q_k(i)-e_i^2)(q_k(l)-e_l^2)\ge|e_i e_l|^2 and ∣ei∣+∣el∣≤qk(i)+qk(l)≤1|e_i|+|e_l|\le q_k(i)+q_k(l)\le1. The odd readout therefore never requires fewer shots than the Hadamard test for the same error on mkm_k, and, since qk(i)−ei2≥qk(i)(1−qk(i))q_k(i)-e_i^2\ge q_k(i)(1-q_k(i)), it requires at least L−1L-1 times more when qk(i)=1/Lq_k(i)=1/L for all ii.

The probability qk(i)q_k(i) follows from the branch structure preserved by the walk, in which each ancilla state ∣i⟩|i\rangle carries its own system state. As in SELECT\mathrm{SELECT}, the signs of the coefficients are absorbed into the Pauli strings, so that Hs=∑iαiPiH_s=\sum_i\alpha_iP_i with αi>0\alpha_i>0 and ∑iαi=λs\sum_i\alpha_i=\lambda_s. Suppose

∣Ψj⟩=∑iαiλs∣i⟩∣χi(j)⟩,∣χi(0)⟩=∣ψ⟩,(59)|\Psi_j\rangle=\sum_i\sqrt{\frac{\alpha_i}{\lambda_s}}|i\rangle|\chi_i^{(j)}\rangle,\qquad|\chi_i^{(0)}\rangle=|\psi\rangle, \tag*{(59)}

which holds at j=0j=0. SELECT\mathrm{SELECT} applies PiP_i in branch ii, and projecting onto ∣PREP⟩|\mathrm{PREP}\rangle recombines the branches,

(⟨PREP∣⊗I)SELECT∣Ψj⟩=∑lαlλsPl∣χl(j)⟩.(60)(\langle\mathrm{PREP}|\otimes I)\mathrm{SELECT}|\Psi_j\rangle=\sum_l\frac{\alpha_l}{\lambda_s}P_l|\chi_l^{(j)}\rangle. \tag*{(60)}

Applying RPREP=2∣PREP⟩⟨PREP∣−IR_{\mathrm{PREP}}=2|\mathrm{PREP}\rangle\langle\mathrm{PREP}|-I restores the form of Eq. (59), with

∣χi(j+1)⟩=2∑lαlλsPl∣χl(j)⟩−Pi∣χi(j)⟩.(61)|\chi_i^{(j+1)}\rangle=2\sum_l\frac{\alpha_l}{\lambda_s}P_l|\chi_l^{(j)}\rangle-P_i|\chi_i^{(j)}\rangle. \tag*{(61)}

By the Born rule, the probability of reading the ancilla in ∣i⟩|i\rangle for odd kk is

qk(i)=αiλs∥χi(j)∥2,j=k−12.(62)q_k(i)=\frac{\alpha_i}{\lambda_s}\left\|\chi_i^{(j)}\right\|^2,\qquad j=\frac{k-1}{2}. \tag*{(62)}

Using ∑l(αl/λs)Pl=H~\sum_l(\alpha_l/\lambda_s)P_l=\widetilde{H}, ∑lαl/λs=1\sum_l\alpha_l/\lambda_s=1 and Pl2=IP_l^2=I, Eq. (61) is solved by

∣χi(j)⟩=[Uj(H~)−PiUj−1(H~)]∣ψ⟩,(63)|\chi_i^{(j)}\rangle=\left[U_j(\widetilde{H})-P_iU_{j-1}(\widetilde{H})\right]|\psi\rangle, \tag*{(63)}

where U−1=0U_{-1}=0, U0=1U_0=1 and Uj+1(x)=2xUj(x)−Uj−1(x)U_{j+1}(x)=2xU_j(x)-U_{j-1}(x) are the Chebyshev polynomials of the second kind, as in Eq. (49). For k=1k=1, q1(i)=αi/λsq_1(i)=\alpha_i/\lambda_s exactly. For k=3k=3,

q3(i)=αiλs∥(2H~−Pi)∣ψ⟩∥2=αiλs[1+4⟨H~2⟩−2⟨H~Pi+PiH~⟩].(64)\begin{aligned} q_3(i)=\frac{\alpha_i}{\lambda_s}\left\|(2\widetilde{H}-P_i)|\psi\rangle\right\|^2 \\ &=\frac{\alpha_i}{\lambda_s}\left[1+4\langle\widetilde{H}^2\rangle-2\langle\widetilde{H}P_i+P_i\widetilde{H}\rangle\right]. \tag*{(64)} \end{aligned}

and, using ∑i(αi/λs)Pi=H~\sum_i (\alpha_i/\lambda_s)P_i = \tilde{H}, ∑iq3(i)=1+4⟨H~2⟩−4⟨H~2⟩=1\sum_i q_3(i) = 1 + 4\langle\tilde{H}^2\rangle- 4\langle\tilde{H}^2\rangle= 1. For the illustrative case H~=0.8Z+0.2X\tilde{H} = 0.8Z + 0.2X and ∣ψ⟩=∣0⟩|\psi\rangle= |0\rangle, the probability of the ZZ branch decreases from 0.80 to 0.42 and 0.14 for k=1,3,5k = 1, 3, 5, while that of the XX branch increases accordingly.

References

References

  1. [1]A. W. Harrow, A. Hassidim, and S. Lloyd, Quantum algorithm for linear systems of equations, Physical Review Letters 103, 150502 (2009).pubmed.ncbi.nlm.nih.gov/19905613
  2. [2]A. M. Childs, R. Kothari, and R. D. Somma, Quantum algorithm for systems of linear equations with exponentially improved dependence on precision, SIAM Journal on Computing 46, 1920 (2017).DOI
  3. [3]A. Gilyén, Y. Su, G. H. Low, and N. Wiebe, Quantum singular value transformation and beyond: exponential improvements for quantum matrix arithmetics, in Proceedings of the 51st Annual ACM SIGACT Symposium on Theory of Computing (ACM, 2019) pp. 193–204.DOI
  4. [4]A. M. Childs, J.-P. Liu, and A. Ostrander, High-precision quantum algorithms for partial differential equations, Quantum 5, 574 (2021).DOI
  5. [5]M. Motta, W. Kirby, I. Liepuoniute, K. J. Sung, J. Cohn, A. Mezzacapo, K. Klymko, N. Nguyen, N. Yoshioka, and J. E. Rice, Subspace methods for electronic structure simulations on quantum computers, Electronic Structure 6, 013001 (2024).arxiv.org/abs/2312.00178
  6. [6]E. N. Epperly, L. Lin, and Y. Nakatsukasa, A theory of quantum subspace diagonalization, SIAM Journal on Matrix Analysis and Applications 43, 1263 (2022).
  7. [7]H. Rammal, A. Perrin, O. Ladhari, C. Dutreix, J. Messud, and M. Saubanère, Orthogonal quantum Krylov diagonalisation (2026), arXiv:2607.09476 [quant-ph].arxiv.org/abs/2607.09476
  8. [8]R. M. Parrish and P. L. McMahon, Quantum filter diagonalization: Quantum eigendecomposition without full quantum phase estimation (2019), arXiv:1909.08925 [quant-ph].arxiv.org/abs/1909.08925
  9. [9]N. H. Stair, R. Huang, and F. A. Evangelista, A multireference quantum Krylov algorithm for strongly correlated electrons, Journal of Chemical Theory and Computation 16, 2236 (2020).DOI
  10. [10]C. L. Cortes and S. K. Gray, Quantum Krylov subspace algorithms for ground and excited-state energy estimation, Physical Review A 105, 022417 (2022).DOI
  11. [11]K. Klymko, C. Mejuto-Zaera, S. J. Cotton, F. Wudarski, M. Urbanek, D. Hait, M. Head-Gordon, K. B. Whaley, J. Moussa, N. Wiebe, W. A. de Jong, and N. M. Tubman, Real-time evolution for ultracompact Hamiltonian eigenstates on quantum hardware, PRX Quantum 3, 020323 (2022).DOI
  12. [12]N. Yoshioka, M. Amico, W. Kirby, P. Jurcevic, A. Dutt, B. Fuller, S. Garion, H. Haas, I. Hamamura, A. Ivrii, R. Majumdar, Z. Minev, M. Motta, B. Pokharel, P. Rivero, K. Sharma, C. J. Wood, A. Javadi-Abhari, and A. Mezzacapo, Krylov diagonalization of large many-body Hamiltonians on a quantum processor, Nature Communications 16, 5014 (2025).DOI
  13. [13]L. Lin and Y. Tong, Near-optimal ground state preparation, Quantum 4, 372 (2020).
  14. [14]Y. Ge, J. Tura, and J. I. Cirac, Faster ground state preparation and high-precision ground energy estimation with fewer qubits, Journal of Mathematical Physics 60, 022202 (2019).arxiv.org/abs/1712.03193
  15. [15]Y. Dong, L. Lin, and Y. Tong, Ground-state preparation and energy estimation on early fault-tolerant quantum computers via quantum eigenvalue transformation of unitary matrices, PRX Quantum 3, 040305 (2022).DOI
  16. [16]G. H. Low and I. L. Chuang, Hamiltonian simulation by qubitization, Quantum 3, 163 (2019).
  17. [17]A. M. Childs, R. Kothari, and R. D. Somma, Simulating sparse Hamiltonians with star decompositions, in Theory of Quantum Computation, Communication, and Cryptography (TQC 2010), Lecture Notes in Computer Science, Vol. 6519 (Springer, 2011) pp. 94–103.DOI
  18. [18]A. M. Childs and N. Wiebe, Hamiltonian simulation using linear combinations of unitary operations, Quantum Information & Computation 12, 901 (2012).
  19. [19]G. H. Low and I. L. Chuang, Optimal Hamiltonian simulation by quantum signal processing, Physical Review Letters 118, 010501 (2017).DOI
  20. [20]J. M. Martyn, Z. M. Rossi, A. K. Tan, and I. L. Chuang, Grand unification of quantum algorithms, PRX Quantum 2, 040203 (2021).DOI
  21. [21]J. Haah, Product decomposition of periodic functions in quantum signal processing, Quantum 3, 190 (2019).DOI
  22. [22]Y. Dong, X. Meng, K. B. Whaley, and L. Lin, Efficient phase-factor evaluation in quantum signal processing, Physical Review A 103, 024219 (2021).DOI
  23. [23]Y. Dong, L. Lin, H. Ni, and J. Wang, Robust iterative method for symmetric quantum signal processing in all parameter regimes, SIAM Journal on Scientific Computing 46, A2951 (2024).arxiv.org/abs/2307.12468
  24. [24]R. Babbush, C. Gidney, D. W. Berry, N. Wiebe, J. McClean, A. Paler, A. Fowler, and H. Neven, Encoding electronic spectra in quantum circuits with linear T complexity, Physical Review X 8, 041015 (2018).DOI
  25. [25]V. von Burg, G. H. Low, T. Häner, D. S. Steiger, M. Reiher, M. Roetteler, and M. Troyer, Quantum computing enhanced computational catalysis, Physical Review Research 3, 033055 (2021).DOI
  26. [26]J. Lee, D. W. Berry, C. Gidney, W. J. Huggins, J. R. McClean, N. Wiebe, and R. Babbush, Even more efficient quantum computations of chemistry through tensor hypercontraction, PRX Quantum 2, 030305 (2021).DOI
  27. [27]M. Reiher, N. Wiebe, K. M. Svore, D. Wecker, and M. Troyer, Elucidating reaction mechanisms on quantum computers, Proceedings of the National Academy of Sciences 114, 7555 (2017).
  28. [28]D. W. Berry, C. Gidney, M. Motta, J. R. McClean, and R. Babbush, Qubitization of arbitrary basis quantum chemistry leveraging sparsity and low rank factorization, Quantum 3, 208 (2019).DOI
  29. [29]E. Koridon, S. Yalouz, B. Senjean, F. Buda, T. E. O’Brien, and L. Visscher, Orbital transformations to reduce the 1-norm of the electronic structure Hamiltonian for quantum applications, Physical Review Research 3, 033127 (2021).DOI
  30. [30]I. Loaiza, A. Marefat Kham, N. Wiebe, and A. F. Izmaylov, Reducing molecular electronic Hamiltonian simu lation cost for linear combination of unitaries approaches, Quantum Science and Technology 8, 035019 (2023).
  31. [32]H. Weyl, Über die Gleichverteilung von Zahlen mod. Eins, Mathematische Annalen 77, 313 (1916).
  32. [33]L. Kuipers and H. Niederreiter, Uniform Distribution of Sequences (Wiley-Interscience, 1974).
  33. [34]W. Kirby, M. Motta, and A. Mezzacapo, Exact and efficient Lanczos method on a quantum computer, Quantum 7, 1018 (2023).DOI
  34. [35]G. Lee, D. D. Lee, and J. Huh, Sampling error analysis in quantum Krylov subspace diagonalization, Quantum 8, 1477 (2024).DOI
  35. [36]Z. Zhang, A. Wang, X. Xu, and Y. Li, Measurement-efficient quantum Krylov subspace diagonalisation, Quantum 8, 1438 (2024).DOI
  36. [37]G. Brassard, P. Høyer, M. Mosca, and A. Tapp, Quantum Computation and Information, Contemporary Mathematics, Vol. 305 (American Mathematical Society, 2002) pp. 53–74.DOI
  37. [38]D. W. Berry, A. M. Childs, R. Cleve, R. Kothari, and R. D. Somma, Simulating Hamiltonian dynamics with a truncated Taylor series, Physical Review Letters 114, 090502 (2015).pubmed.ncbi.nlm.nih.gov/25793789
  38. [39]T. J. Yoder, G. H. Low, and I. L. Chuang, Fixed-point quantum search with an optimal number of queries, Physical Review Letters 113, 210501 (2014).DOI
  39. [40]I. D. Kivlichan, C. Gidney, D. W. Berry, N. Wiebe, J. McClean, W. Sun, Z. Jiang, N. Rubin, A. Fowler, A. Aspuru-Guzik, H. Neven, and R. Babbush, Improved fault-tolerant quantum simulation of condensed-phase correlated electrons via Trotterization, Quantum 4, 296 (2020).
  40. [41]E. T. Campbell, Early fault-tolerant simulations of the Hubbard model, Quantum Science and Technology 7, 015007 (2022).DOI
  41. [42]J. Hubbard, Electron correlations in narrow energy bands, Proceedings of the Royal Society of London. Series A, Mathematical and Physical Sciences 276, 238 (1963).
  42. [43]E. Dagotto, Correlated electrons in high-temperature superconductors, Reviews of Modern Physics 66, 763 (1994).DOI
  43. [44]E. H. Lieb and F. Y. Wu, Absence of Mott transition in an exact solution of the short-range, one-band model in one dimension, Physical Review Letters 20, 1445 (1968).
  44. [45]F. H. L. Essler, H. Frahm, F. Göhmann, A. Klümper, and V. E. Korepin, The One-Dimensional Hubbard Model (Cambridge University Press, 2005).
  45. [46]P. Jordan and E. Wigner, Über das Paulische Äquivalenzverbot, Zeitschrift für Physik 47, 631 (1928).
  46. [47]L. Grover and T. Rudolph, Creating superpositions that correspond to efficiently integrable probability distributions (2002), arXiv:quant-ph/0208112.DOI
  47. [48]A. Szabo and N. S. Ostlund, Modern Quantum Chemistry: Introduction to Advanced Electronic Structure Theory (Dover Publications, Mineola, New York, 1996).
  48. [49]A. Weiße, G. Wellein, A. Alvermann, and H. Fehske, The kernel polynomial method, Reviews of Modern Physics 78, 275 (2006).
  49. [50]A. Y. Kitaev, Quantum measurements and the Abelian stabilizer problem (1995), arXiv:quant-ph/9511026 [quant-ph].
  50. [51]D. Aharonov, V. Jones, and Z. Landau, A polynomial quantum algorithm for approximating the Jones polynomial, Algorithmica 55, 395 (2009).
  51. [52]J. Stewart, D. K. Clegg, and S. Watson, Calculus: Early Transcendentals, 9th ed. (Cengage Learning, 2020).
  52. [53]E. W. Cheney, Introduction to Approximation Theory, 2nd ed. (AMS Chelsea Publishing, 1982).
  53. [54]P. Borwein and T. Erdélyi, Polynomials and Polynomial Inequalities, Graduate Texts in Mathematics, Vol. 161 (Springer, 1995).DOI
  54. [55]J. C. Mason and D. C. Handscomb, Chebyshev Polynomials (Chapman and Hall/CRC, 2002).
  55. [56]Eviden, Quantum learning machine (QLM) and pyAQASM documentation (2024), https://myqlm.github.io.

Paper details

vlnc:W_9x5w3d5h
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

a large language model

View statement
Looking for related papers…
Contents