pith. sign in

arxiv: 2606.01733 · v1 · pith:RCKKZH5Fnew · submitted 2026-06-01 · 🪐 quant-ph · cs.NA· math-ph· math.MP· math.NA

Pauli-structured preconditioning for quantum linear system solvers

Hantao Nie , Zhijian Lai , Dong An This is my paper

Pith reviewed 2026-06-28 14:19 UTC · model grok-4.3

classification 🪐 quant-ph cs.NAmath-phmath.MPmath.NA
keywords Pauli-structured preconditioningquantum linear systemsblock-encodingPauli expansionsnormalization overheadrandomized quantum solverspreconditioned operator
0
0 comments X

The pith

Pauli regrouping of structured matrices and preconditioners lowers the coefficient weight of the combined operator in quantum linear solvers.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that when a linear system matrix and its preconditioner both admit Pauli-structured representations, algebraic regrouping of their Pauli products can produce a preconditioned operator whose total Pauli coefficient weight is smaller than the sum of the separate encodings. This reduction directly affects the normalization constants that appear in block-encoding constructions and in randomized Pauli-based linear-system algorithms. A sympathetic reader would care because the normalization overhead has been shown in prior work to erase the classical benefits of preconditioning; any algebraic technique that shrinks those constants therefore changes when preconditioning yields a net resource saving in quantum access models. The authors supply explicit bounds on the size and weight of the regrouped expansion and demonstrate the effect on both direct block-encoding diagnostics and a per-sample depth proxy for randomized solvers. Numerical tests on a synthetic finite-dimensional benchmark confirm measurable drops in those diagnostics under the Pauli-structured regime.

Core claim

We show that Pauli-structured representations of both the system matrix and the preconditioner allow the preconditioned operator to be accessed through regrouped Pauli expansions. In this setting, algebraic regrouping of Pauli products can reduce the Pauli coefficient weight of the preconditioned operator, thereby altering the normalization parameters relevant to quantum algorithms. We derive explicit size and coefficient-weight bounds for the regrouped Pauli representations, and we trace their consequences for both direct block-encoding constructions and randomized Pauli linear system solvers. These results identify when Pauli-structured preconditioning can reduce the effective complexity p

What carries the argument

Algebraic regrouping of Pauli products from the separate expansions of the system matrix and preconditioner, which produces a single Pauli expansion of the preconditioned operator with provably lower total coefficient weight.

If this is right

  • The effective normalization factor for a block-encoding of the preconditioned operator can be smaller than the product of the individual normalization factors.
  • Randomized Pauli linear-system solvers see a lower per-sample depth proxy when the regrouped expansion is used.
  • Preconditioning changes the quantum complexity parameters precisely when the algebraic weight reduction outweighs any increase in the number of terms.
  • Norm-aware direct block-encoding diagnostics improve on synthetic benchmarks that satisfy the Pauli-structured assumption.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same regrouping technique could be applied to any quantum algorithm whose cost is dominated by the normalization of a composed block-encoding of two Pauli-sum operators.
  • Preconditioners could be deliberately chosen or constructed to maximize the weight reduction under regrouping rather than solely to minimize the classical condition number.
  • The bounds derived here supply a concrete criterion for deciding, on a given instance, whether to invest in a Pauli-structured preconditioner or to solve the unpreconditioned system directly.

Load-bearing premise

The system matrix and preconditioner must admit Pauli-structured representations that permit algebraic regrouping to reduce the coefficient weight of the preconditioned operator.

What would settle it

An explicit pair of Pauli-structured matrices A and P for which every regrouping of their product yields a Pauli expansion whose total coefficient weight is strictly larger than the sum of the separate weights.

read the original abstract

Preconditioning is a fundamental technique for accelerating classical linear system solvers, and understanding when its benefits persist in quantum linear system (QLS) solvers is important for assessing the practical resource requirements of quantum linear algebra. In QLS algorithms, however, the potential advantage of preconditioning may be offset by the normalization overhead incurred by composing separate block-encodings of the system matrix and the preconditioner, as observed in recent work. This limitation leaves open whether additional algebraic structure can make preconditioning effective in quantum access models. Motivated by this question, we show that Pauli-structured representations of both the system matrix and the preconditioner allow the preconditioned operator to be accessed through regrouped Pauli expansions. In this setting, algebraic regrouping of Pauli products can reduce the Pauli coefficient weight of the preconditioned operator, thereby altering the normalization parameters relevant to quantum algorithms. We derive explicit size and coefficient-weight bounds for the regrouped Pauli representations, and we trace their consequences for both direct block-encoding constructions and randomized Pauli linear system solvers. These results identify when Pauli-structured preconditioning can reduce the effective complexity parameters of QLS algorithms, rather than merely improving the classical condition number. Numerical experiments on a finite-dimensional synthetic benchmark show reductions in norm-aware direct block-encoding diagnostics and in the randomized QLS per-sample depth proxy.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper claims that Pauli-structured representations of the system matrix and preconditioner in quantum linear system solvers permit the preconditioned operator to be accessed via regrouped Pauli expansions; algebraic regrouping can reduce the Pauli coefficient weight (and thus alter normalization parameters) relative to separate block-encodings. Explicit bounds on the size and coefficient weight of the regrouped form are derived, their consequences for direct block-encoding and randomized Pauli solvers are traced, and numerical experiments on a finite-dimensional synthetic benchmark demonstrate reductions in norm-aware block-encoding diagnostics and randomized QLS per-sample depth proxies.

Significance. If the claimed weight reductions hold under the stated Pauli structure and produce net improvement in normalization factors, the work would identify a concrete algebraic route to making classical-style preconditioning resource-efficient inside quantum access models, rather than merely improving the classical condition number. The provision of explicit bounds and the use of concrete QLS-relevant proxies (norm-aware diagnostics, per-sample depth) are strengths that allow falsifiable assessment.

major comments (2)
  1. [Abstract / bounds derivation] Abstract and the section deriving explicit size and coefficient-weight bounds: the central claim that regrouping 'alters the normalization parameters relevant to quantum algorithms' requires that the post-regrouping weight (or its induced block-encoding norm) is strictly smaller than the product of the separate weights/norms in a manner that survives the cost of block-encoding the preconditioner itself. The bounds appear to be stated as O(w_A * w_P) in the worst case; without an explicit comparison showing when cancellations produce a net saving below the separate-encoding baseline, the alteration of QLS parameters does not follow in general.
  2. [Numerical experiments] Numerical experiments section: the reported reductions in 'norm-aware direct block-encoding diagnostics' and 'randomized QLS per-sample depth proxy' are obtained on a synthetic benchmark, but the experiments do not appear to construct or cost the block-encoding of the preconditioner itself. This leaves open whether the observed improvement survives when both operators must be accessed through quantum oracles, which is load-bearing for the claim that Pauli-structured preconditioning reduces effective QLS complexity.
minor comments (2)
  1. Notation for Pauli coefficient weight (l1-norm versus term count) should be defined once at first use and used consistently when stating the bounds.
  2. The synthetic benchmark should be described with enough detail (dimension, sparsity pattern, how the preconditioner is chosen) to allow reproduction of the reported reductions.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful review and constructive feedback on our manuscript. We address each major comment below, providing clarifications and indicating planned revisions where appropriate.

read point-by-point responses

  1. Referee: [Abstract / bounds derivation] Abstract and the section deriving explicit size and coefficient-weight bounds: the central claim that regrouping 'alters the normalization parameters relevant to quantum algorithms' requires that the post-regrouping weight (or its induced block-encoding norm) is strictly smaller than the product of the separate weights/norms in a manner that survives the cost of block-encoding the preconditioner itself. The bounds appear to be stated as O(w_A * w_P) in the worst case; without an explicit comparison showing when cancellations produce a net saving below the separate-encoding baseline, the alteration of QLS parameters does not follow in general.

    Authors: The derived bounds establish that the regrouped Pauli representation has cardinality and coefficient weight bounded by O(w_A w_P) in the worst case, matching the product of separate weights. However, the manuscript explicitly notes that algebraic regrouping of Pauli products permits cancellations in the coefficients of the preconditioned operator that are unavailable when block-encoding A and P separately; these cancellations reduce the effective weight below the product baseline. The consequences section traces how such reductions alter the normalization factor α for direct block-encodings and the per-sample depth proxy for randomized Pauli solvers. We will revise the bounds derivation section to add an explicit comparison paragraph stating the condition (nonzero cancellation in overlapping Pauli supports) under which the regrouped weight is strictly smaller than the separate-encoding product, thereby confirming the alteration of QLS parameters. revision: yes

  2. Referee: [Numerical experiments] Numerical experiments section: the reported reductions in 'norm-aware direct block-encoding diagnostics' and 'randomized QLS per-sample depth proxy' are obtained on a synthetic benchmark, but the experiments do not appear to construct or cost the block-encoding of the preconditioner itself. This leaves open whether the observed improvement survives when both operators must be accessed through quantum oracles, which is load-bearing for the claim that Pauli-structured preconditioning reduces effective QLS complexity.

    Authors: The synthetic benchmark constructs the preconditioned operator directly via the regrouped Pauli expansion of the product, which is the quantity whose reduced coefficient weight determines the reported diagnostics and depth proxies. In the Pauli-structured access model, both A and P are given by their Pauli oracles; the regrouping step combines them algebraically before any block-encoding, so the measured reductions already reflect the effective parameters after accounting for the structured representations. We will add clarifying text in the numerical experiments section stating that the proxies are computed on the regrouped form and that separate oracle costs for P are not multiplied into the combined normalization precisely because of the regrouping. revision: partial

Circularity Check

0 steps flagged

No significant circularity; algebraic regrouping bounds are independent of inputs

full rationale

The derivation rests on explicit size and coefficient-weight bounds obtained by collecting like terms in the product of two Pauli expansions. These bounds are standard consequences of the multiplication rule for Pauli operators and do not reduce to a fitted parameter, a self-referential definition, or a load-bearing self-citation. The paper states that regrouping “can reduce” the weight (via cancellations) and supplies worst-case O(w_A w_P) expressions; the improvement is therefore conditional on support overlap rather than guaranteed by construction. No step equates a claimed prediction to its own input data or prior result by definition. The synthetic numerics are presented only as illustration, not as the source of the theoretical claims.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Based on abstract only. The claim rests on the standard fact that Pauli operators form a basis for matrix representations and on the modeling choice that matrices admit structured Pauli expansions permitting regrouping. No free parameters or invented entities are mentioned.

axioms (1)
  • standard math Pauli operators form a basis for the space of Hermitian matrices and allow expansions whose products can be regrouped algebraically.
    This is invoked implicitly when the abstract discusses regrouped Pauli expansions of the preconditioned operator.

pith-pipeline@v0.9.1-grok · 5768 in / 1255 out tokens · 35546 ms · 2026-06-28T14:19:25.044337+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

53 extracted references · 1 canonical work pages

  1. [1]

    Its Pauli coefficient weight satisfiesw(P A)≤w(P)w(A)

    IfP Ais Hermitian, then after regrouping identi- cal Pauli words,P A= PJL j=1 µ(L) j Q(L) j , JL ≤LM, withQ (L) j ∈ V n andµ (L) j ∈R. Its Pauli coefficient weight satisfiesw(P A)≤w(P)w(A)

  2. [2]

    Its Pauli coefficient weight satis- fiesw(P AP †)≤w(P) 2w(A)

    The symmetric product admits a Pauli expansion P AP † =PJS j=1 µ(S) j Q(S) j , JS ≤LM 2,withQ (S) j ∈ Vn andµ (S) j ∈R. Its Pauli coefficient weight satis- fiesw(P AP †)≤w(P) 2w(A). Proof.The product of Pauli words is a phase times a Pauli word. Thus every termP mAℓ and every term PmAℓP † m′ can be rewritten as a scalar multiple of a Pauli word. Regroupin...

  3. [3]

    For a real time parametert∈R, define the unitary U(t) :=e itP = exp it MX m=1 βmPm ! .(B1) SetH m :=iβ mPm so that eachH m is anti-Hermitian

    Simulation ofU(t) =e itP and its inverse Without loss of generality, we assume that theMlisted coefficients are nonzero. For a real time parametert∈R, define the unitary U(t) :=e itP = exp it MX m=1 βmPm ! .(B1) SetH m :=iβ mPm so that eachH m is anti-Hermitian. Define the (p+ 1)-fold commutator scaling ˜αcomm := MX m1,...,mp+1=1 Hmp+1 ,[· · ·[H m2 , Hm1]...

  4. [4]

    ReF n ImF n # β≈

    Matrix-logarithm construction First, we state the following proposition, which im- proves the norm assumption of Corollary 71 of [7]. Proposition 15Letδ, ε∈ 0, 1 2 . Suppose thatU= eiH, whereHis a Hamiltonian of norm at most π 2 (1−δ). Then, under our block-encoding normalization conven- tion, we can implement a π 2 ,2, ε -block-encoding ofH usingO 1 δ2 l...

  5. [5]

    Saad,Iterative Methods for Sparse Linear Systems, 2nd ed

    Y. Saad,Iterative Methods for Sparse Linear Systems, 2nd ed. (SIAM, Philadelphia, 2003)

    2003

  6. [6]

    Benzi, Preconditioning techniques for large linear sys- tems: A survey, Journal of Computational Physics182, 418 (2002)

    M. Benzi, Preconditioning techniques for large linear sys- tems: A survey, Journal of Computational Physics182, 418 (2002)

    2002

  7. [7]

    M. J. Grote and T. Huckle, Parallel preconditioning with sparse approximate inverses, SIAM Journal on Scientific Computing18, 838 (1997)

    1997

  8. [8]

    A. W. Harrow, A. Hassidim, and S. Lloyd, Quantum al- gorithm for linear systems of equations, Physical Review Letters103, 150502 (2009)

    2009

  9. [9]

    A. M. Childs, R. Kothari, and R. D. Somma, Quantum algorithm for systems of linear equations with exponen- tially improved dependence on precision, SIAM Journal on Computing46, 1920 (2017)

    1920

  10. [10]

    G. H. Low and I. L. Chuang, Optimal Hamiltonian sim- ulation by quantum signal processing, Physical Review Letters118, 010501 (2017)

    2017

  11. [11]

    Gily´ en, Y

    A. Gily´ en, Y. Su, G. H. Low, and N. Wiebe, Quantum singular value transformation and beyond: Exponential improvements for quantum matrix arithmetics, inPro- ceedings of the 51st Annual ACM SIGACT Symposium on Theory of Computing(2019) pp. 193–204

    2019

  12. [12]

    Lin and Y

    L. Lin and Y. Tong, Optimal polynomial based quantum eigenstate filtering with application to solving quantum linear systems, Quantum4, 361 (2020)

    2020

  13. [13]

    P. C. S. Costa, D. An, Y. R. Sanders, Y. Su, R. Bab- bush, and D. W. Berry, Optimal scaling quantum linear- systems solver via discrete adiabatic theorem, PRX Quantum3, 040303 (2022)

    2022

  14. [14]

    M. E. S. Morales, L. Pira, P. Schleich, K. Koor, P. Costa, D. An, A. Aspuru-Guzik, L. Lin, P. Rebentrost, and D. W. Berry, Quantum linear system solvers: A survey of algorithms and applications, Reviews of Modern Physics 10.1103/x6gh-d8gh (2026)

    work page doi:10.1103/x6gh-d8gh 2026

  15. [15]

    B. D. Clader, B. C. Jacobs, and C. R. Sprouse, Precon- ditioned quantum linear system algorithm, Physical Re- view Letters110, 250504 (2013)

    2013

  16. [16]

    Shao and H

    C. Shao and H. Xiang, Quantum circulant preconditioner for a linear system of equations, Physical Review A98, 062321 (2018)

    2018

  17. [17]

    Wan, C.-H

    L.-C. Wan, C.-H. Yu, S.-J. Pan, F. Gao, Q.-Y. Wen, 15 and S.-J. Qin, Asymptotic quantum algorithm for the Toeplitz systems, Physical Review A97, 062322 (2018)

    2018

  18. [18]

    J. K. Golden, D. O’Malley, and H. S. Viswanathan, Quantum computing and preconditioners for hydrolog- ical linear systems, Scientific Reports12, 22285 (2022)

    2022

  19. [19]

    Y. Tong, D. An, N. Wiebe, and L. Lin, Fast inversion, preconditioned quantum linear system solvers, and fast evaluation of matrix functions, Physical Review A104, 032422 (2021)

    2021

  20. [20]

    Hosaka, K

    A. Hosaka, K. Yanagisawa, S. Koshikawa, I. Kudo, X. Al- ifu, and T. Yoshida, Preconditioning for a variational quantum linear solver (2023), arXiv:2312.15657

    arXiv 2023

  21. [21]

    Z. Wu, X. Yang, and T. Terlaky, A preconditioned inex- act infeasible quantum interior point method for linear optimization, Computational Optimization and Applica- tions93, 1225 (2026)

    2026

  22. [22]

    Suresh and K

    S. Suresh and K. Suresh, Computing a sparse approxi- mate inverse on quantum annealing machines, Journal of Computing and Information Science in Engineering26, 011008 (2026)

    2026

  23. [23]

    C. D. Phillips, H. Akbari, and V. I. Okhmatovski, QW- HHL: A quantum computer amenable general matrix equation solver, IEEE Journal on Multiscale and Mul- tiphysics Computational Techniques11, 156 (2026)

    2026

  24. [24]

    S. Jin, N. Liu, C. Ma, and Y. Yu, Quantum pre- conditioning method for linear systems problems via Schr¨ odingerization (2025), arXiv:2505.06866

    Pith/arXiv arXiv 2025

  25. [25]

    A. H. Salehi Shayegan and L. Dejam, Quantum linear system algorithm for solving an ill-posed quasi-linear el- liptic problem by preconditioning operator, The Euro- pean Physical Journal Plus140, 467 (2025)

    2025

  26. [26]

    O. M. Raisuddin and S. De, A review of quantum sci- entific computing algorithms relevant to computational mechanics, Archives of Computational Methods in Engi- neering33, 745 (2026)

    2026

  27. [27]

    Lapworth and C

    L. Lapworth and C. S¨ underhauf, Preconditioned block encodings for quantum linear systems, Quantum Science and Technology10, 045064 (2025)

    2025

  28. [28]

    Pechan, J

    A. Pechan, J. Golden, and D. O’Malley, Block encoding of the three-dimensional heterogeneous Poisson equation with application to fracture flow, Physical Review Ap- plied25, 044038 (2026)

    2026

  29. [29]

    S¨ underhauf, E

    C. S¨ underhauf, E. Campbell, and J. Camps, Block- encoding structured matrices for data input in quantum computing, Quantum8, 1226 (2024)

    2024

  30. [30]

    Camps, L

    D. Camps, L. Lin, R. Van Beeumen, and C. Yang, Ex- plicit quantum circuits for block encodings of certain sparse matrices, SIAM Journal on Matrix Analysis and Applications45, 801 (2024)

    2024

  31. [31]

    Zhang and X

    X.-M. Zhang and X. Yuan, Circuit complexity of quan- tum access models for encoding classical data, npj Quan- tum Information10, 42 (2024)

    2024

  32. [32]

    A. M. Childs, J.-P. Liu, and A. Ostrander, High-precision quantum algorithms for partial differential equations, Quantum5, 574 (2021)

    2021

  33. [33]

    Y. Sato, R. Kondo, I. Hamamura, T. Onodera, and N. Yamamoto, Hamiltonian simulation for hyperbolic partial differential equations by scalable quantum cir- cuits, Physical Review Research6, 033246 (2024)

    2024

  34. [34]

    Sturm and N

    A. Sturm and N. Schillo, Efficient and explicit block en- coding of finite difference discretizations of the Laplacian (2025), arXiv:2509.02429

    arXiv 2025

  35. [35]

    Lloyd, Universal quantum simulators, Science273, 1073 (1996)

    S. Lloyd, Universal quantum simulators, Science273, 1073 (1996)

    1996

  36. [36]

    Somma, G

    R. Somma, G. Ortiz, J. E. Gubernatis, E. Knill, and R. Laflamme, Simulating physical phenomena by quan- tum networks, Physical Review A65, 042323 (2002)

    2002

  37. [37]

    Jordan and E

    P. Jordan and E. Wigner, ¨Uber das Paulische ¨Aquivalenzverbot, Zeitschrift f¨ ur Physik47, 631 (1928)

    1928

  38. [38]

    McArdle, S

    S. McArdle, S. Endo, A. Aspuru-Guzik, S. C. Benjamin, and X. Yuan, Quantum computational chemistry, Re- views of Modern Physics92, 015003 (2020)

    2020

  39. [39]

    Y. Cao, J. Romero, J. P. Olson, M. Degroote, P. D. John- son, M. Kieferov´ a, I. D. Kivlichan, T. Menke, B. Per- opadre, N. P. D. Sawaya, S. Sim, L. Veis, and A. Aspuru- Guzik, Quantum chemistry in the age of quantum com- puting, Chemical Reviews119, 10856 (2019)

    2019

  40. [40]

    Babbush, D

    R. Babbush, D. W. Berry, J. R. McClean, and H. Neven, Quantum simulation of chemistry with sublinear scaling in basis size, npj Quantum Information5, 92 (2019)

    2019

  41. [41]

    M. A. Nielsen and I. L. Chuang,Quantum Computation and Quantum Information, 10th ed. (Cambridge Univer- sity Press, Cambridge, 2010)

    2010

  42. [42]

    J. M. Martyn, Z. M. Rossi, A. K. Tan, and I. L. Chuang, Grand unification of quantum algorithms, PRX Quan- tum2, 040203 (2021)

    2021

  43. [43]

    D. W. Berry, A. M. Childs, R. Cleve, R. Kothari, and R. D. Somma, Simulating Hamiltonian dynamics with a truncated Taylor series, Physical Review Letters114, 090502 (2015)

    2015

  44. [44]

    G. H. Low and I. L. Chuang, Hamiltonian simulation by qubitization, Quantum3, 163 (2019)

    2019

  45. [45]

    Schillo, A

    N. Schillo, A. Sturm, and R. Quay, TARE: Block encod- ing linear combinations of Pauli strings without ancilla state preparation (2026), arXiv:2601.05740

    Pith/arXiv arXiv 2026

  46. [46]

    S. Wang, S. McArdle, and M. Berta, Qubit-efficient ran- domized quantum algorithms for linear algebra, PRX Quantum5, 020324 (2024)

    2024

  47. [47]

    Chakraborty, Implementing any Linear Combination of Unitaries on intermediate-term quantum computers, Quantum8, 1496 (2024), arXiv:2302.13555 [quant-ph]

    S. Chakraborty, Implementing any Linear Combination of Unitaries on intermediate-term quantum computers, Quantum8, 1496 (2024), arXiv:2302.13555 [quant-ph]

    arXiv 2024

  48. [48]

    Chakraborty, S

    S. Chakraborty, S. Hazra, T. Li, C. Shao, X. Wang, and Y. Zhang, Quantum Singular Value Transformation with- out block encodings: Near-optimal complexity with min- imal ancilla (2025), arXiv:2504.02385 [quant-ph]

    arXiv 2025

  49. [49]

    X. Wang, Y. Zhang, S. Hazra, T. Li, C. Shao, and S. Chakraborty, Randomized Quantum Singular Value Transformation (2025), arXiv:2510.06851 [quant-ph]

    arXiv 2025

  50. [50]

    Hariprakash, R

    S. Hariprakash, R. Van Beeumen, K. Klymko, and D. Camps, Are randomized quantum linear systems solvers practical? (2025), arXiv:2510.13766

    arXiv 2025

  51. [51]

    A. M. Childs, Y. Su, M. C. Tran, N. Wiebe, and S. Zhu, Theory of Trotter error with commutator scaling, Phys- ical Review X11, 011020 (2021)

    2021

  52. [52]

    Tibshirani, Regression shrinkage and selection via the LASSO, Journal of the Royal Statistical Society

    R. Tibshirani, Regression shrinkage and selection via the LASSO, Journal of the Royal Statistical Society. Series B58, 267 (1996)

    1996

  53. [53]

    Beck and M

    A. Beck and M. Teboulle, A fast iterative shrinkage- thresholding algorithm for linear inverse problems, SIAM Journal on Imaging Sciences2, 183 (2009)

    2009