Pith. sign in

REVIEW 3 major objections 4 minor 92 references

Bridging the Prediction Error Method and Subspace Identification: A Weighted Null Space Fitting Method

T0 review · 3 major / 4 minor · reviewed 2026-08-04 · deepseek-v4-flash

Pith's one-line read This paper claims that state-space models can be estimated by a five-step weighted least-squares procedure that is consistent and asymptotically efficient, matching the maximum-likelihood bound, whenever the chosen canonical parameterizatio

desk verdict SISO efficiency result is a solid contribution; the MIMO theorem overreaches — the proof never handles joint covariance. read the letter →

arxiv 2510.02529 v2 pith:OOXPFJYY submitted 2025-10-02 stat.ME cs.SYeess.SY

classification stat.MEcs.SYeess.SY MSC 62F1262M1093E12
keywords systemidentificationsubspacepredictionerrormethodasymptoticefficiencyCramér-Raolowerboundweightednullspacefittingstate-spacemodelsmulti-stepleast-squares
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to close a long-standing gap: subspace identification methods are numerically robust but rarely reach maximum-likelihood efficiency, while prediction-error methods are efficient but require non-convex optimization. It proposes WNSF-SS, which starts from a least-squares estimate of a high-order ARX model and reduces it to a state-space model in canonical form through two ordinary least-squares steps and two weighted least-squares refinements. For single-output systems, open or closed loop, the estimates are proven consistent and asymptotically efficient, meaning their asymptotic covariance reaches the Cramér-Rao lower bound. For multi-output systems, the same conclusion holds provided the canonical form selected by a Kronecker index is admissible; the paper explicitly states that structure selection lies beyond its scope and assumes the state order is known in advance.

What carries the argument

The key object is the left null space of the extended observability/Hankel matrix. In SISO, Cayley-Hamilton gives the identity aH+ + H- = 0, where a contains the characteristic-polynomial coefficients, so null-space fitting becomes a linear least-squares problem. In MIMO, a similar relation, AK H+(ν̄) = H-(ν̄), holds for a submatrix selected by a Kronecker index ν̄, a set of integers that picks which rows of the Hankel matrix form a basis. The estimation runs in five steps - HOARX OLS, OLS for AK, WLS for AK, OLS for B and K, WLS for B and K - where the WLS weights are constructed from the asymptotic covariance of the HOARX parameter errors.

What would settle it

Run WNSF-SS on closed-loop data from a multi-output system whose chosen Kronecker index is not admissible: the selected rows of the Hankel matrix do not span the row space. With the HOARX order growing as in Assumption 5.1, the parameter estimates should fail to converge to the true values, and the empirical covariance should fail to reach the Cramér-Rao bound of the true model. Alternatively, for a SISO system, compute the empirical MSE and compare it to the CRLB across increasing sample sizes; a persistent gap would contradict Theorem 5.3.

Watch

Extended reading notes

Core claim

The central discovery is that the null space of the Hankel matrix built from Markov parameters is linearly parameterized by the coefficients of the system's characteristic polynomial in the single-output case, and by the free rows of the canonical state matrix in the multi-output case. This linear structure turns null-space estimation into a sequence of least-squares problems, avoiding the non-convex cost of the prediction-error method. Using the asymptotic covariance of the high-order ARX estimates as an optimal weighting, the weighted least-squares steps deliver estimates whose asymptotic covariance equals the Cramér-Rao lower bound of the chosen parameterization. Theorems 5.1-5.3 prove co

Load-bearing premise

The estimates are only guaranteed consistent and efficient if the chosen canonical parameterization is admissible - the selected rows of the Hankel matrix must span the whole row space - and if the true state order is known in advance.

Editorial extensions

If this is right

  • For SISO systems, WNSF-SS attains the Cramér-Rao lower bound for both open- and closed-loop data without local nonlinear optimization.
  • The method provides a finite, five-step least-squares counterpart to PEM that can serve as an asymptotically efficient benchmark for evaluating subspace identification methods.
  • For MIMO systems with an admissible canonical parameterization, the same consistency and efficiency guarantees hold; when several parameterizations are admissible, the determinants of the asymptotic error covariances are identical.
  • Because the estimate of A is formed as AK + CK, efficiency of AK, B, and K transfers to the original state matrix by the invariance principle.
  • On numerical and practical benchmark problems, WNSF-SS shows competitive finite-sample accuracy and often lower validation errors than several standard subspace methods.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The null-space formulation suggests why SVD-based subspace methods are hard to make efficient: they estimate the range space, where an explicit optimal weighting is difficult to derive, whereas efficiency is naturally expressed in the null-space parameterization.
  • Because admissibility is a generic property, a randomly chosen Kronecker index should yield the MIMO efficiency result with probability one; the practical bottleneck is therefore structure selection and order selection rather than the estimator itself.
  • The same two-step weighted least-squares logic could be extended to order estimation, for example by scanning HOARX orders and Kronecker indices and selecting via validation error, which the paper leaves to future work.
  • Replacing the homoskedastic Gaussian assumption by heteroskedastic or non-Gaussian innovations is a natural testable extension; the current proof confines the analysis to the homoskedastic case.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper proposes WNSF_SS, a multi-step least-squares method for estimating state-space models in observer canonical form. A high-order ARX (HOARX) model is first estimated by ordinary least squares to obtain Markov parameters; a weighted null-space fitting step then estimates the characteristic-polynomial coefficients of A_K (and hence A_K itself); finally B and K are estimated by OLS/WLS. The authors claim consistency and asymptotic efficiency for SISO systems in both open- and closed-loop operation, and for MIMO systems whenever an admissible Kronecker-index canonical parameterization is specified. Numerical experiments compare WNSF_SS with N4SID, SSARX, PBSID, WNSF for ARMAX models, and PEM on synthetic and DaISy benchmark data.

Significance. If fully established, the result would be a notable step toward resolving the long-standing question of whether an asymptotically efficient subspace-type method exists. The paper's strengths include a concrete algorithm that avoids SVD, a serious appendix with consistency proofs for the SISO case, machine-checkable-looking derivations for the variance calculations, and extensive simulations including CRLB comparisons and practical benchmark data. However, the central MIMO efficiency claim is not supported by the provided proof, and the SISO efficiency proof relies on unquantified first-order approximations. The contribution is therefore significant but needs a substantial repair before the claimed efficiency result can be accepted.

major comments (3)
  1. [§4.2, Theorem 5.4, Appendix D.3] The MIMO efficiency proof is incomplete. In Step 3 the rows a_1 and a_2 are estimated by separate WLS problems with weights ar\Lambda_n(a_i)=K_n(a_i)^\top\bar R_n^{-1}K_n(a_i). These weights include only the marginal covariance of each row residual and ignore the cross-covariance K_n(a_i)^\top\bar R_n^{-1}K_n(a_j), i≠j. Joint asymptotic efficiency requires the asymptotic covariance of the stacked estimator to equal the inverse of the full Fisher information matrix; matching each diagonal block is not sufficient. The paper itself notes that the n_y parallel ARMAX models are “not independent,” so the cross blocks are generally nonzero. Appendix D.3 says that “a similar proof as in Appendix C can be derived,” but it does not derive the joint distribution or show that the information matrix is block diagonal. Appendix G provides a general framework but does not show that the per-row WLS step
  2. [§3, Eq. (30) and (33), Appendix C] The SISO efficiency proof for B and K relies on first-order perturbation expansions marked with “≃,” specifically C(\hat A_K^k-A_K^k) ≃ ... in Eq. (30) and vec(\tilde O_{n-1}) ≃ \tilde a_\mathrm{wls}S_n(a) in Eq. (31). No explicit bound on the higher-order terms in \tilde A_K is provided. To conclude that \sqrt{N}(\hat\eta_\mathrm{wls}-\eta) is asymptotically normal with covariance σ_e^2\bar\Lambda_n(a,\eta), one needs to show that the remainder terms are o_p(N^{-1/2}) and that replacing consistent estimates in K_n(a,\eta) is asymptotically negligible. The remark that higher-order terms “decay much faster” is not a proof. This is load-bearing for Theorem 5.3.
  3. [Appendix C.1, Eq. (C.8)] The equality M_\mathrm{CR,a}=M(g_n,a) is the crux of the SISO efficiency result, but its proof is only sketched. The text introduces \Omega_n, \Sigma_n, and \gamma, states the inner-product relations, and then invokes “a similar way to [26, Th. 2]” with the geometric approach of [34]. Since this equality is used to identify the asymptotic covariance of the WLS estimator with the CRLB, the authors should either provide a complete proof of the limit or state the exact result from [26] and verify all its conditions in the present closed-loop setting.
minor comments (4)
  1. [Abstract, §4.1, §6.1] Typographical errors: “canoncial” should be “canonical”; “Monte Corlo” should be “Monte Carlo”; “excacerbated” should be “exacerbated”; “mush faster” should be “much faster.”
  2. [Assumption 2.2] For MIMO systems, the conditional covariance should be σ_e^2 I_{n_y}, not σ_e^2 I_1. As written, the assumption appears to be stated only for a scalar innovation.
  3. [§6.3] The definitions of the identification and validation errors state “N_I=0.7N and N_I=0.3N”; the second should be N_V=0.3N.
  4. [§4.1, after Eq. (42)] The sentence “rank(H_{n x n}(\bar\nu))=n_x” is correct because all rows lie in the n_x-dimensional row space of H_{n x n}, but it may confuse readers since the displayed matrix has 2n_x rows; a clarifying remark would help.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the derivation is self-contained and benchmarked against an external CRLB; the MIMO efficiency proof gap is a correctness issue, not circularity.

full rationale

The claimed derivation is not circular. Step 1 fits a HOARX model and uses its asymptotic covariance; Steps 3 and 5 are feasible WLS estimators whose optimal weights are consistent estimates of the inverse residual covariance, which is standard feasible-GLS logic. The SISO efficiency claim (Theorem 5.3, Appendix C) is proved by showing that the WLS asymptotic covariance equals M^{-1}(g_n,a), and then showing via the geometric argument that this limit coincides with the externally derived CRLB M_CR,a. No fitted parameter is later relabeled as a prediction. The unknown optimal weighting matrix is replaced by consistent estimates from earlier steps, which is the usual device that does not affect asymptotic optimality; it is not a case of fitting a quantity and then reporting the same quantity as a prediction. The paper does cite the authors' prior work for technical lemmas (e.g., Lemma 4 and Lemma 6 from [25], the geometric CRLB technique from [26] and [34]) and for the WNSF origin ([25], [31], [32]). These are published, parameter-free results whose stated assumptions do not include the target efficiency theorem, so under the review rules they count as independent support rather than a load-bearing self-citation chain. Assumption 5.1 is imported from [25] as an order-growth condition, not as an assumption of the theorem being proved. The weakest point is Theorem 5.4 for MIMO systems: Appendix D.3 analyzes the rows a_i separately and says 'a similar proof as in Appendix C can be derived,' while the same appendix states that the system can be rewritten as 'n_y parallel but not independent single-output ARMAX models.' This leaves the joint, cross-row efficiency claim insufficiently demonstrated. That is a correctness/completeness gap, not circularity: the claimed efficiency is not equivalent by construction to an input or to a self-citation. Therefore the circularity score is 0.

Assumptions & free parameters 2 free parameters · 7 assumptions · 0 invented entities

The method introduces no new physical entities or conserved quantities. The free parameters are the algorithmic tunables (HOARX order, Kronecker index) and the ordinary model parameters estimated by least squares. The axioms are the standard assumptions of linear system identification, plus the admissibility condition for MIMO canonical forms, which is the structurally weakest premise.

free parameters (2)
  • HOARX order n(N) = selected via prediction-error minimization over a candidate set (e.g., n ∈ {10:1:150} in Section 6.3)
    The asymptotic efficiency proof requires n to grow with the sample size N so that n^{4+δ}(N)/N→0 and √N·d(N)→0 (Assumption 5.1). In practice n is tuned on data, and the theoretical claims depend on this growth behavior.
  • Kronecker index ν̄ for MIMO = chosen or enumerated; e.g., ν̄1 = {1,3} in Section 6.2.2, or selected by prediction error among {1,2} and {2,1} in Secti
    The MIMO consistency and efficiency results require the selected canonical parameterization to be admissible; the structure selection is an input to the method and is acknowledged as beyond scope.
assumptions (7)
  • domain assumption Assumption 2.1: system is stable, minimal, and the order n_x is known
    Required for the Cayley-Hamilton null-space construction, the rank properties of the Hankel matrix, and for the consistency proofs in Appendix A.
  • domain assumption Assumption 2.2: innovations form a homoskedastic martingale difference sequence with E(|e|^10)<∞
    Used to establish the asymptotic normality of the HOARX estimate in (10) and the residual distributions in (21), (34).
  • domain assumption Assumption 2.3: reference and noise spectral density is coercive; closed-loop is f_N-stable; input is f_N-quasi-stationary
    Support the consistency of the regressor covariance R_n and the closed-loop analysis; needed for the CRLB derivation under feedback.
  • domain assumption Assumption 5.1: HOARX order growth conditions (n^{4+δ}/N→0, √N·d(N)→0)
    Ensures the truncation bias is negligible and the estimated HOARX parameters form an asymptotically sufficient statistic.
  • domain assumption Lemmas 1 and 2 (Gevers-Wertz and standard canonical-form theory): the canonical state-space form (38) describes almost all systems and is admissible when the selected rows of H_{n x n} span the row space
    The MIMO extension and Theorem 5.4 rely on this; without admissibility there is model mismatch and consistency fails.
  • standard math Geometric limit identity from [34] and [26, Th. 2] used to prove M_CR,a = M(g_n,a)
    Appendix C uses this external result to equate the limiting WLS covariance with the CRLB; it is cited but not re-derived.
  • standard math Cayley-Hamilton theorem and rank properties of Hankel matrices for minimal systems
    Basis for the null-space parameterization by the characteristic polynomial coefficients.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Bridging the Prediction Error Method and Subspace Identification: A Weighted Null Space Fitting Method." pith.science (2026). https://pith.science/paper/OOXPFJYY

@misc{pith2026251002529,
  author       = {Pith},
  title        = {Pith review of: Bridging the Prediction Error Method and Subspace Identification: A Weighted Null Space Fitting Method},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OOXPFJYY}},
  note         = {Machine review of arXiv:2510.02529}
}
read the original abstract

Subspace identification methods (SIMs) have proven to be very useful and numerically robust for building state-space models. While most SIMs are consistent, few if any can achieve the efficiency of the maximum likelihood estimate (MLE). Conversely, the prediction error method (PEM) with a quadratic criteria is equivalent to MLE, but it comes with non-convex optimization problems and requires good initialization points. This contribution proposes a weighted null space fitting (WNSF) approach for estimating state-space models, combining some key advantages of the two aforementioned mainstream approaches. It starts with a least-squares estimate of a high-order ARX model, and then a multi-step least-squares procedure reduces the model to a state-space model on canoncial form. It is demonstrated through statistical analysis that when a canonical parameterization is admissible, the proposed method is consistent and asymptotically efficient, thereby making progress on the long-standing open problem about the existence of an asymptotically efficient SIM. Numerical and practical examples are provided to illustrate that the proposed method performs favorable in comparison with SIMs.

Figures

Figures reproduced from arXiv: 2510.02529 by the authors.

Figure 1
Figure 1. Average MSE of θˆ from 1000 Monte Carlo trials (SISO system): Open-loop (OL) and Closed-loop (CL) Cases. 6.1 Illustration of Asymptotic Properties In this subsection, we use a single-output system and multi￾output system to illustrate that WNSFSS is asymptotically efficient. 6.1.1 A SISO System Consider the following ARMAX model: G◦(q) = b1q −1 + b2q −1 1 + f1q−1 + f2q−1 , H◦(q) = 1 + a1q −1 + a2q −1 1 + f1q−1 + f2q… view at source ↗
Figure 2
Figure 2. Average MSE of θˆ from 100 Monte Carlo trials (SIMO system): Open-loop (OL) and Closed-loop (CL) Cases. 6.1.2 A SIMO System Consider the following three order state-space model: AK =     0.4 0.1 0 0 0 1 0.5 0.2 0.6     , B =     1 0.2 0.5     , K =     0.5 0.1 0 0.6 −0.5 −0.56     , C = " 1 0 0 0 1 0# . The innovations {ek} consist of independent Gaussian white sequences with unit variance. F… view at source ↗
Figure 3
Figure 3. FITs from 100 Monte Carlo trials: Open-loop. [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: Average transfer functions of the identified MIMO systems [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]
Figure 6
Figure 6. Figure 6: Joint FIT distribution from 50 Monte Carlo trials (10-order [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]
Figure 7
Figure 7. Figure 7: Joint FIT distribution from 50 Monte Carlo trials (3-order [PITH_FULL_IMAGE:figures/full_fig_p016_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

92 extracted references · 2 linked inside Pith

  1. [34]

    A geometric approach to variance analysis in system identification.IEEE Trans

    Håkan Hjalmarsson and Jonas Martensson. A geometric approach to variance analysis in system identification.IEEE Trans. Autom. Control, 56(5):983–997, 2010

  2. [26]

    Estimating models with high-order noise dynamics using semi- parametric weighted null-space fitting.Automatica, 102:45–57, 2019

    Miguel Galrinho, Cristian R Rojas, and Håkan Hjalmarsson. Estimating models with high-order noise dynamics using semi- parametric weighted null-space fitting.Automatica, 102:45–57, 2019

  3. [1]

    Numerical identification of linear dynamic systems from normal operating records

    Karl Johan Åström and Bohlin Torsten. Numerical identification of linear dynamic systems from normal operating records. InProc. 2nd IFAC Symp. Theory Self-Adapt. Control Syst., Teddington, UK, September 14-17, 1965, volume 2, pages 96–111, 1965

  4. [2]

    A new approach to learning linear dynamical systems

    Ainesh Bakshi, Allen Liu, Ankur Moitra, and Morris Yau. A new approach to learning linear dynamical systems. InProc. 55th Annu. ACM Symp. Theory Comput. (STOC), pages 335–348, 2023

  5. [3]

    Asymptotic properties of subspace estimators

    Dietmar Bauer. Asymptotic properties of subspace estimators. Automatica, 41(3):359–376, 2005

  6. [4]

    Comparing the cca subspace method to pseudo maximum likelihood methods in the case of no exogenous inputs

    Dietmar Bauer. Comparing the cca subspace method to pseudo maximum likelihood methods in the case of no exogenous inputs. J. Time Ser. Anal., 26(5):631–668, 2005

  7. [5]

    Consistency and asymptotic normality of some subspace algorithms for systems without observed inputs.Automatica, 35(7):1243–1254, 1999

    Dietmar Bauer, Manfred Deistler, and Wolfgang Scherrer. Consistency and asymptotic normality of some subspace algorithms for systems without observed inputs.Automatica, 35(7):1243–1254, 1999

  8. [6]

    Analysis of the asymptotic properties of the MOESP type of subspace algorithms.Automatica, 36(4):497–509, 2000

    Dietmar Bauer and Magnus Jansson. Analysis of the asymptotic properties of the MOESP type of subspace algorithms.Automatica, 36(4):497–509, 2000

Show all 92 references
  1. [7]

    Some facts about the choice of the weighting matrices in Larimore type of subspace algorithms

    Dietmar Bauer and Lennart Ljung. Some facts about the choice of the weighting matrices in Larimore type of subspace algorithms. Automatica, 38(5):763–773, 2002

  2. [8]

    Prediction error estimators: Asymptotic normality and accuracy

    Peter E Caines and Lennart Ljung. Prediction error estimators: Asymptotic normality and accuracy. InProc. IEEE Conf. Decis. Control & 15th Symp. Adapt. Process., pages 652–658, 1976

  3. [9]

    On the relation between CCA and predictor-based subspace identification.IEEE Trans

    Alessandro Chiuso. On the relation between CCA and predictor-based subspace identification.IEEE Trans. Autom. Control, 52(10):1795– 1812, 2007

  4. [10]

    The role of vector autoregressive modeling in predictor-based subspace identification.Automatica, 43(6):1034– 1048, 2007

    Alessandro Chiuso. The role of vector autoregressive modeling in predictor-based subspace identification.Automatica, 43(6):1034– 1048, 2007

  5. [11]

    The asymptotic variance of subspace estimates.J

    Alessandro Chiuso and Giorgio Picci. The asymptotic variance of subspace estimates.J. Econom., 118(1-2):257–291, 2004

  6. [12]

    Consistency analysis of some closed-loop subspace identification methods.Automatica, 41(3):377– 391, 2005

    Alessandro Chiuso and Giorgio Picci. Consistency analysis of some closed-loop subspace identification methods.Automatica, 41(3):377– 391, 2005

  7. [13]

    B. De Moor. DaISy: Database for the Identification of Systems.http://homes.esat.kuleuven.be/~smc/ daisy/, Jan 2010. Dept. Electr. Eng., ESAT/SISTA, K.U.Leuven, Leuven, Belgium (Online)

  8. [14]

    Structured total least squares and L2 approximation problems.Linear Algebra Appl., 188:163–205, 1993

    Bart De Moor. Structured total least squares and L2 approximation problems.Linear Algebra Appl., 188:163–205, 1993

  9. [15]

    Least squares realization of lti models is an eigenvalue problem

    Bart De Moor. Least squares realization of lti models is an eigenvalue problem. In2019 18th European Control Conference (ECC), pages 2270–2275, 2019

  10. [16]

    Least squares optimal realisation of autonomous LTI systems is an eigenvalue problem.Commun

    Bart De Moor. Least squares optimal realisation of autonomous LTI systems is an eigenvalue problem.Commun. Inf. Syst., 20(2):163– 207, 2020

  11. [17]

    Consistency and relative efficiency of subspace methods.Automatica, 31(12):1865–1875, 1995

    Manfred Deistler, K Peternell, and Wolfgang Scherrer. Consistency and relative efficiency of subspace methods.Automatica, 31(12):1865–1875, 1995

  12. [18]

    Critical point behavior of objective functions defined on spaces of multivariable systems

    David F Delchamps and Christopher I Byrnes. Critical point behavior of objective functions defined on spaces of multivariable systems. In Proc. 21st IEEE Conf. Decis. Control (CDC), pages 937–943, 1982

  13. [19]

    J. M. Dufour and T. Jouini. Asymptotic distributions for quasi- efficient estimators in echelon V ARMA models.Comput. Stat. Data Anal., 73:69–86, 2014

  14. [20]

    Efficient estimation of parameters in moving-average models.Biometrika, 46(3/4):306–316, 1959

    James Durbin. Efficient estimation of parameters in moving-average models.Biometrika, 46(3/4):306–316, 1959

  15. [21]

    The fitting of time-series models.Revue de l’Institut International de Statistique, pages 233–244, 1960

    James Durbin. The fitting of time-series models.Revue de l’Institut International de Statistique, pages 233–244, 1960

  16. [22]

    Optimal least squares time- domain synthesis of recursive digital filters.IEEE Trans

    A Evans and Robert Fischl. Optimal least squares time- domain synthesis of recursive digital filters.IEEE Trans. Audio Electroacoust., 21(1):61–65, 1973

  17. [23]

    PhD thesis, KTH Royal Institute of Technology, 2018

    Miguel Galrinho.System identification with multi-step least-squares methods. PhD thesis, KTH Royal Institute of Technology, 2018

  18. [24]

    A weighted least-squares method for parameter estimation in structured models

    Miguel Galrinho, Cristian Rojas, and Håkan Hjalmarsson. A weighted least-squares method for parameter estimation in structured models. InProc. IEEE Conf. Decis. Control, Los Angeles, California, USA, 2014

  19. [25]

    Parametric identification using weighted null-space fitting.IEEE Trans

    Miguel Galrinho, Cristian R Rojas, and Håkan Hjalmarsson. Parametric identification using weighted null-space fitting.IEEE Trans. Autom. Control, 64(7):2798–2813, 2018

  20. [27]

    Uniquely identifiable state- space and arma parametrizations for multivariable linear systems

    Michel Gevers and Vincent Wertz. Uniquely identifiable state- space and arma parametrizations for multivariable linear systems. Automatica, 20(3):333–347, 1984

  21. [28]

    Subspace-based system identification: weighting and pre-filtering of instruments.Automatica, 38(3):433–443, 2002

    Tony Gustafsson. Subspace-based system identification: weighting and pre-filtering of instruments.Automatica, 38(3):433–443, 2002

  22. [29]

    E. J. Hannan and L. Kavalieris. Multivariate linear time series models.Adv. in App. Probability, 16(3):492–561, 1984

  23. [30]

    SIAM, 2012

    Edward James Hannan and Manfred Deistler.The Statistical Theory of Linear Systems. SIAM, 2012

  24. [31]

    Weighted null space fitting (WNSF): A link between the prediction error method and subspace identification.arXiv preprint arXiv:2411.00506, 2024

    Jiabao He and Håkan Hjalmarsson. Weighted null space fitting (WNSF): A link between the prediction error method and subspace identification.arXiv preprint arXiv:2411.00506, 2024

  25. [32]

    Range space or null space: Least-squares methods for the realization problem.arXiv preprint arXiv:2505.19639, 2025

    Jiabao He, Yueyue Xu, Yue Ju, Cristian R Rojas, and Håkan Hjalmarsson. Range space or null space: Least-squares methods for the realization problem.arXiv preprint arXiv:2505.19639, 2025

  26. [33]

    Finite sample analysis of subspace identification methods.arXiv preprint arXiv:2501.16639, 2025

    Jiabao He, Ingvar Ziemann, Cristian R Rojas, S Joe Qin, and Håkan Hjalmarsson. Finite sample analysis of subspace identification methods.arXiv preprint arXiv:2501.16639, 2025

  27. [35]

    Effective construction of linear state-variable models from input/output functions

    B L Ho and Rudolf E Kálmán. Effective construction of linear state-variable models from input/output functions. Automatisierungstechnik, 14(1-12):545–548, 1966

  28. [36]

    Cambridge university press, 2012

    Roger A Horn and Charles R Johnson.Matrix analysis. Cambridge university press, 2012

  29. [37]

    Subspace identification and ARX modeling

    Magnus Jansson. Subspace identification and ARX modeling. In Proc. 13th IFAC Symp. Syst. Identification, Netherlands, 2003

  30. [38]

    A linear regression approach to state-space subspace system identification.Signal Process., 52(2):103–129, 1996

    Magnus Jansson and Bo Wahlberg. A linear regression approach to state-space subspace system identification.Signal Process., 52(2):103–129, 1996. 19

  31. [39]

    On consistency of subspace methods for system identification.Automatica, 34(12):1507–1519, 1998

    Magnus Jansson and Bo Wahlberg. On consistency of subspace methods for system identification.Automatica, 34(12):1507–1519, 1998

  32. [40]

    Prentice-Hall Englewood Cliffs, NJ, 1980

    Thomas Kailath.Linear systems, volume 156. Prentice-Hall Englewood Cliffs, NJ, 1980

  33. [41]

    Springer, 2005

    Tohru Katayama.Subspace methods for system identification. Springer, 2005

  34. [42]

    Consistency analysis of subspace identification methods based on a linear regression approach.Automatica, 37(1):81–89, 2001

    Torben Knudsen. Consistency analysis of subspace identification methods based on a linear regression approach.Automatica, 37(1):81–89, 2001

  35. [43]

    A new identification and model reduction algorithm via singular value decomposition

    Sun-Yuan Kung. A new identification and model reduction algorithm via singular value decomposition. InProc. Asilomar Conf. Circuits, Syst. Comput., Pacific Grove, USA, 1978

  36. [44]

    Larimore

    Wallace E. Larimore. Canonical variate analysis in identification, filtering and adaptive control. InProc. IEEE Conf. Decis. Control, Honolulu, HI, USA, 1990

  37. [45]

    Statistical optimality and canonical variate analysis system identification.Signal Process., 52(2):131–144, 1996

    Wallace E Larimore. Statistical optimality and canonical variate analysis system identification.Signal Process., 52(2):131–144, 1996

  38. [46]

    IQML-like algorithms for solving structured total least squares problems: a unified view.Signal Process., 81(9):1935– 1945, 2001

    Philippe Lemmerling, Leentje Vanhamme, Sabine Van Huffel, and Bart De Moor. IQML-like algorithms for solving structured total least squares problems: a unified view.Signal Process., 81(9):1935– 1945, 2001

  39. [47]

    Interior-point method for nuclear norm approximation with application to system identification

    Zhang Liu and Lieven Vandenberghe. Interior-point method for nuclear norm approximation with application to system identification. SIAM J. Matrix Anal. Appl., 31(3):1235–1256, 2010

  40. [48]

    On the consistency of prediction error identification methods

    Lennart Ljung. On the consistency of prediction error identification methods. InMath. Sci. Eng., volume 126, pages 121–164. 1976

  41. [49]

    Citeseer, 1995

    Lennart Ljung.System identification toolbox: User’s guide. Citeseer, 1995

  42. [50]

    Prentice Hall information and system sciences series, Prentice Hall PTR, 1999

    Lennart Ljung.System identification: Theory for the user. Prentice Hall information and system sciences series, Prentice Hall PTR, 1999

  43. [51]

    Subspace identification from closed loop data.Signal Process., 52(2):209–215, 1996

    Lennart Ljung and Tomas McKelvey. Subspace identification from closed loop data.Signal Process., 52(2):209–215, 1996

  44. [52]

    Asymptotic properties of the least- squares method for estimating transfer functions and disturbance spectra.Adv

    Lennart Ljung and Bo Wahlberg. Asymptotic properties of the least- squares method for estimating transfer functions and disturbance spectra.Adv. Appl. Probab., 24(2):412–440, 1992

  45. [53]

    Overview of total least- squares methods.Signal Process., 87(10):2283–2302, 2007

    Ivan Markovsky and Sabine Van Huffel. Overview of total least- squares methods.Signal Process., 87(10):2283–2302, 2007

  46. [54]

    Application of structured total least squares for system identification and model reduction.IEEE Trans

    Ivan Markovsky, Jan C Willems, Sabine Van Huffel, Bart De Moor, and Rik Pintelon. Application of structured total least squares for system identification and model reduction.IEEE Trans. Auto. Control, 50(10):1490–1500, 2005

  47. [55]

    Probabilistic reduced-dimensional vector autoregressive modeling with oblique projections.Automatica, 180:112476, 2025

    Yanfang Mo and S Joe Qin. Probabilistic reduced-dimensional vector autoregressive modeling with oblique projections.Automatica, 180:112476, 2025

  48. [56]

    Realizing lti models by identifying characteristic parameters using least squares optimization

    Tim Nicolai, Mark Haring, Esten I Grøtli, Jan T Gravdahl, and Johann Reger. Realizing lti models by identifying characteristic parameters using least squares optimization. InEuropean Control Conf. (ECC), pages 1–6, 2023

  49. [57]

    Revisiting Ho-Kalman-based system identification: Robustness and finite-sample analysis.IEEE Trans

    Samet Oymak and Necmiye Ozay. Revisiting Ho-Kalman-based system identification: Robustness and finite-sample analysis.IEEE Trans. Autom. Control, 67(4):1914–1928, 2021

  50. [58]

    Statistical analysis of novel subspace identification methods.Signal Processing, 52(2):161–177, 1996

    Klaus Peternell, Wolfgang Scherrer, and Manfred Deistler. Statistical analysis of novel subspace identification methods.Signal Processing, 52(2):161–177, 1996

  51. [59]

    Poskitt and M

    D. Poskitt and M. Salau. On the relationship between generalized least squares and Gaussian estimation of vector ARMA models.J. of Time Series Anal., 16(6):617–645, 1995

  52. [60]

    D.S. Poskitt. A method for the estimation and identification of transfer function models.J. Roy. Statist. Soc. B, 50:304–315, 1989

  53. [61]

    D.S. Poskitt. Estimation and structure determination of multivariate input output systems.J. Multivar. Anal., 33:157–182, 1990

  54. [62]

    An overview of subspace identification.Comput

    S Joe Qin. An overview of subspace identification.Comput. Chem. Eng., 30(10-12):1502–1513, 2006

  55. [63]

    A novel subspace identification approach with enforced causal models.Automatica, 41(12):2043–2053, 2005

    S Joe Qin, Weilu Lin, and Lennart Ljung. A novel subspace identification approach with enforced causal models.Automatica, 41(12):2043–2053, 2005

  56. [64]

    Closed-loop subspace identification with innovation estimation

    S Joe Qin and Lennart Ljung. Closed-loop subspace identification with innovation estimation. InProc. 13th IFAC Symp. Syst. Identification, Netherlands, 2003

  57. [65]

    Reinsel, S

    G. Reinsel, S. Basu, and S. Yap. Maximum likelihood estimators in the multivariate autoregressive moving-average model from a generalized least squares viewpoint.J. Time Series Anal., 13(2):133– 145, 1992

  58. [66]

    Analyzing iterations in identification with application to nonparametric H∞-norm estimation.Automatica, 48(11):2776– 2790, 2012

    Cristian R Rojas, Tom Oomen, Håkan Hjalmarsson, and Bo Wahlberg. Analyzing iterations in identification with application to nonparametric H∞-norm estimation.Automatica, 48(11):2776– 2790, 2012

  59. [67]

    Transfer function synthesis as a ratio of two complex polynomials.IEEE Trans

    Cok Sanathanan and Judith Koerner. Transfer function synthesis as a ratio of two complex polynomials.IEEE Trans. Autom. Control, 8(1):56–58, 1963

  60. [68]

    Optimal identification of discrete-time systems from impulse response data.IEEE Trans

    Arnab K Shaw. Optimal identification of discrete-time systems from impulse response data.IEEE Trans. Signal Process., 42(1):113–120, 1994

  61. [69]

    On computing the cramer-rao bound and covariance matrices for pem estimates in linear state space models

    Torsten Söderström. On computing the cramer-rao bound and covariance matrices for pem estimates in linear state space models. IFAC Proc. Volumes, 39(1):600–605, 2006

  62. [70]

    Prentice Hall, 1989

    Torsten Söderström and Petre Stoica.System identification. Prentice Hall, 1989

  63. [71]

    Instrumental variable methods for system identification.Circuits Syst

    Torsten Söderström and Petre Stoica. Instrumental variable methods for system identification.Circuits Syst. Signal Process., 21(1):1–9, 2002

  64. [72]

    An indirect prediction error method for system identification

    Torsten Söderström, Petre Stoica, and Benjamin Friedlander. An indirect prediction error method for system identification. Automatica, 27(1):183–188, 1991

  65. [73]

    A technique for the identification of linear systems.IEEE Trans

    K Steiglitz and L McBride. A technique for the identification of linear systems.IEEE Trans. Autom. Control, 10(4):461–464, 1965

  66. [74]

    Optimal instrumental variable estimation and approximate implementations.IEEE Trans

    Petre Stoica and Torsten Söderström. Optimal instrumental variable estimation and approximate implementations.IEEE Trans. Autom. Control, 28(7):757–772, 1983

  67. [75]

    Weighted ls and tls approaches yield asymptotically equivalent results.Signal Process., 45(2):255–259, 1995

    Petre Stoica and Mats Viberg. Weighted ls and tls approaches yield asymptotically equivalent results.Signal Process., 45(2):255–259, 1995

  68. [76]

    A subspace fitting method for identification of linear state-space models

    A Swindlehust, R Roy, Björn Ottersten, and Thomas Kailath. A subspace fitting method for identification of linear state-space models. IEEE Trans. Auto. Control, 40(2):311–316, 1995

  69. [77]

    Finite sample analysis of stochastic system identification

    Anastasios Tsiamis and George J Pappas. Finite sample analysis of stochastic system identification. InIEEE Conf. Decis. Control, pages 3648–3654, 2019

  70. [78]

    Closed-loop subspace identification methods: An overview.IET Control Theory Appl., 7(10):1339–1358, 2013

    Gijs Van der Veen, Jan-Willem van Wingerden, Marco Bergamasco, Marco Lovera, and Michel Verhaegen. Closed-loop subspace identification methods: An overview.IET Control Theory Appl., 7(10):1339–1358, 2013

  71. [79]

    On-line structure selection for multivariable state-space models.Automatica, 18(5):529–543, 1982

    AJM Van Overbeek and Lennart Ljung. On-line structure selection for multivariable state-space models.Automatica, 18(5):529–543, 1982

  72. [80]

    N4SID: Subspace algorithms for the identification of combined deterministic-stochastic systems

    Peter Van Overschee and Bart De Moor. N4SID: Subspace algorithms for the identification of combined deterministic-stochastic systems. Automatica, 30(1):75–93, 1994

  73. [81]

    A unifying theorem for three subspace system identification algorithms.Automatica, 31(12):1853– 1864, 1995

    Peter Van Overschee and Bart De Moor. A unifying theorem for three subspace system identification algorithms.Automatica, 31(12):1853– 1864, 1995. 20

  74. [82]

    Springer, 2012

    Peter Van Overschee and Bart De Moor.Subspace identification for linear systems: Theory-Implementation- Applications. Springer, 2012

  75. [83]

    Subspace model identification part I: the output-error state-space model identification class of algorithm.Int

    Michel Verahegen and Patrick Dewilde. Subspace model identification part I: the output-error state-space model identification class of algorithm.Int. J. Control, 56:1187–1210, 1992

  76. [84]

    Application of a subspace model identification technique to identify LTI systems operating in closed-loop

    Michel Verhaegen. Application of a subspace model identification technique to identify LTI systems operating in closed-loop. Automatica, 29(4):1027–1040, 1993

  77. [85]

    Analysis of state space system identification methods based on instrumental variables and subspace fitting.Automatica, 33(9):1603–1616, 1997

    Mats Viberg, Bo Wahlberg, and Björn Ottersten. Analysis of state space system identification methods based on instrumental variables and subspace fitting.Automatica, 33(9):1603–1616, 1997

  78. [86]

    Model reductions of high-order estimated models: the asymptotic ml approach.Int

    Bo Wahlberg. Model reductions of high-order estimated models: the asymptotic ml approach.Int. J. Control, 49(1):169–192, 1989

  79. [87]

    Perturbation theory for pseudo-inverses.BIT Numerical Mathematics, 13:217–232, 1973

    Per-Åke Wedin. Perturbation theory for pseudo-inverses.BIT Numerical Mathematics, 13:217–232, 1973

  80. [88]

    The determination of optimum structures for the state space representation of multivariate stochastic processes.IEEE Trans

    VINCENT Wertz, MICHEL Gevers, and E Hannan. The determination of optimum structures for the state space representation of multivariate stochastic processes.IEEE Trans. Auto. Control, 27(6):1200–1211, 1982

  81. [89]

    The refined instrumental variable method.Journal Européen des Systemes Automatisés, 42(2-3):149–179, 2008

    Peter C Young. The refined instrumental variable method.Journal Européen des Systemes Automatisés, 42(2-3):149–179, 2008

  82. [90]

    Constrained subspace method for the identification of structured state-space models (COSMOS).IEEE Trans

    Chengpu Yu, Lennart Ljung, Adrian Wills, and Michel Verhaegen. Constrained subspace method for the identification of structured state-space models (COSMOS).IEEE Trans. Autom. Control, 65(10):4201–4214, 2019

  83. [91]

    Wiley, 1971

    Shelemyahu Zacks.The theory of statistical inference. Wiley, 1971

  84. [92]

    uk ek # =   − L(θ) F(θ)A(θ) Vnx − 1 F(θ) Vnx 1 A(θ) Vnx 0 0 1 A(θ) Vnx  

    Yucai Zhu and Håkan Hjalmarsson. The Box–Jenkins Steiglitz– Mcbride algorithm.Automatica, 65:170–182, 2016. A Consistency of Steps 2 and 3 A.1 Auxiliary Results To prove Theorem 5.1, we introduce some auxiliary results. (1)∥ ˜gn∥ →0,asN→ ∞w.p.1: For the firstntrue Markov param...

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.