An $L^2_T$-error bound for time-limited balanced truncation

Martin Redmann

arxiv: 1907.05478 · v1 · pith:EJZEFYH7new · submitted 2019-07-11 · 🧮 math.OC · cs.NA· math.NA

An L²_T-error bound for time-limited balanced truncation

Martin Redmann This is my paper

Pith reviewed 2026-05-24 22:45 UTC · model grok-4.3

classification 🧮 math.OC cs.NAmath.NA

keywords model order reductionbalanced truncationtime-limitederror boundGramiansingular valuesL2 normH-infinity

0 comments

The pith

An L²_T-error bound for time-limited balanced truncation is proved using its truncated singular values and recovers the classical H∞ bound as T tends to infinity.

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

The paper studies time-limited balanced truncation as a method to obtain reduced models that approximate the original system accurately only on a finite interval [0, T]. It establishes an explicit bound on the L²_T approximation error in terms of the sum of the neglected time-limited singular values. A sympathetic reader would care because this supplies a concrete certificate of accuracy for reduced models of discretized PDEs when only finite-time behavior matters. The bound is shown to approach the standard H∞ error bound of unrestricted balanced truncation in the limit, and the same proof ideas also shorten the derivation of that classical result.

Core claim

The central claim is that the L²_T norm of the output error between the original and reduced systems after time-limited balanced truncation is bounded by twice the sum of the discarded time-limited singular values. This bound converges to the well-known H∞ error bound of standard balanced truncation when the time horizon T goes to infinity. The techniques in the proofs also deliver a relatively short time-domain derivation of the H∞ bound for the unrestricted case.

What carries the argument

The time-limited controllability and observability Gramians whose singular values directly determine the L²_T error bound.

If this is right

Reduced models obtained by time-limited balanced truncation can be certified with an a-priori L²_T error guarantee on any chosen finite interval.
The same reduced model becomes certified by the classical H∞ bound once the time horizon is taken sufficiently large.
The proof methods supply an alternative, shorter derivation of the H∞ error bound for ordinary balanced truncation.
The technique applies directly to spatially discretized partial differential equations whose order must be reduced while preserving accuracy only up to time T.

Where Pith is reading between the lines

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

The finite-time bound may be especially useful in control or simulation settings that only require accurate behavior on short horizons rather than forever.
Numerical checks on simple linear systems could reveal how tight the bound is in practice compared with the actual error.
The convergence to the infinite-time bound indicates that time-limited reduction is consistent with the optimal long-term reduction as the horizon lengthens.

Load-bearing premise

The time-limited controllability and observability Gramians exist, are positive definite, and admit a singular-value decomposition that controls the approximation error.

What would settle it

For a low-order linear system with explicitly computable solution, truncate after the first few time-limited singular values, evaluate the actual L²_T output error on [0, T], and check whether that error remains at most twice the sum of the neglected singular values; any T where the actual error exceeds the bound would falsify the claim.

read the original abstract

Model order reduction (MOR) is often applied to spatially-discretized partial differential equations to reduce their order and hence decrease computational complexity. A reduced system can be obtained, e.g., by time-limited balanced truncation, a method that aims to construct an accurate reduced order model on a given finite time interval $[0, T]$. This particular balancing related MOR technique is studied in this paper. An $L^2_T$-error bound based on the truncated time-limited singular values is proved and is the main result of this paper. The derived error bound converges (as $T\rightarrow \infty$) to the well-known $\mathcal H_\infty$-error bound of unrestricted balanced truncation, a scheme that is used to construct a good reduced system on the entire time line. The techniques within the proofs of this paper can also be applied to unrestricted balanced truncation so that a relatively short time domain proof of the $\mathcal H_\infty$-error bound is found here.

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.

Desk Editor's Note private letter to a colleague

The paper supplies a concrete L²_T error bound for time-limited balanced truncation that recovers the classical H∞ bound as T grows, plus a short time-domain proof of the unrestricted case.

read the letter

The main contribution here is an explicit L²_T error bound stated in terms of the neglected time-limited singular values from the finite-horizon Gramians. That bound is new and directly useful for finite-time model reduction on [0,T]. The paper also gives a compact time-domain derivation of the usual H∞ bound that had been obtained earlier via frequency-domain arguments; the new proof is shorter and stays inside the time domain, which is a modest but real improvement for readers who prefer that setting. Both results rest on the standard construction of the time-limited controllability and observability Gramians and the usual SVD balancing step, so the technical machinery is familiar. The limit statement as T→∞ is shown under the usual stability assumption that makes the infinite Gramians finite, and the argument does not appear to introduce hidden circularity or unjustified limit swaps. The derivation steps are presented explicitly enough that a reader can check them without external references. One minor soft spot is that the paper assumes the time-limited Gramians are positive definite so that the singular values are well-defined and the balancing transformation exists; this is standard but worth noting when the underlying system is only marginally stable or the interval is very short. No other load-bearing gaps stand out. The work is aimed at people already working on balanced truncation and finite-horizon MOR for large-scale linear systems; it will be of interest to that group but is unlikely to change practice outside it. The math is clean and the claims are modest, so the paper deserves a serious referee rather than a desk rejection.

Referee Report

0 major / 3 minor

Summary. The paper proves an L²_T-error bound for time-limited balanced truncation of LTI systems, expressed in terms of the sum of the neglected time-limited Hankel singular values obtained from the finite-horizon controllability and observability Gramians on [0,T]. It further shows that the bound converges to the classical H_∞ error bound of unrestricted balanced truncation as T→∞ (under the usual stability assumption), and re-uses the same techniques to supply a short time-domain proof of the standard infinite-horizon result.

Significance. If the derivation holds, the result supplies a rigorous, computable a-priori error certificate for finite-time model reduction, directly relevant to spatially discretized PDEs where only a finite interval matters. The explicit recovery of the classical bound as T→∞ provides a consistency check, and the alternative short proof of the H_∞ bound is a useful side contribution. The work is grounded in the standard Gramian construction and SVD balancing, with no free parameters or ad-hoc fitting.

minor comments (3)

[§2] §2 (Preliminaries): the integral expressions for the time-limited Gramians W_c^T and W_o^T are stated but the precise lower integration limit (0 or -∞) and the handling of the input/output operators could be written out once more explicitly for readers who skip the references.
[Theorem 3.1] Theorem 3.1 (main bound): the statement that the bound holds for any stable LTI system is correct, but a one-sentence remark on the case of marginally stable poles (where the infinite-horizon Gramians diverge) would clarify the domain of the T→∞ limit statement.
Notation: the symbol σ_i^T for the time-limited singular values is introduced without an explicit reminder that they are the square roots of the eigenvalues of the product of the two finite-horizon Gramians; a parenthetical cross-reference to the SVD step would help.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive assessment and the recommendation to accept the manuscript. The report correctly identifies the main contributions: the L²_T-error bound in terms of neglected time-limited Hankel singular values, its convergence to the classical H∞ bound as T→∞, and the short time-domain proof of the unrestricted case.

Circularity Check

0 steps flagged

No significant circularity; direct proof from Gramian definitions

full rationale

The paper's central result is an explicit mathematical proof of an L²_T-error bound expressed in terms of the neglected singular values of the time-limited Gramians. These Gramians are defined directly as finite integrals over [0,T] (standard construction, independent of the bound), admit an SVD by construction, and the error estimate follows from the usual balancing transformation and integral estimates on the impulse response. The T→∞ recovery of the classical H∞ bound is obtained by taking the limit under the stability assumption that makes the infinite-horizon integrals converge; this limit interchange is justified by monotone convergence of the integrals and does not presuppose the target bound. No parameter is fitted to data, no quantity is renamed as a prediction, and no load-bearing step reduces to a self-citation or self-definition. The derivation is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, invented entities, or non-standard axioms are visible. The derivation is presumed to rest on standard LTI system theory and the existence of time-limited Gramians.

axioms (1)

domain assumption The underlying system is linear time-invariant and obtained by spatial discretization of a PDE.
Stated in the first sentence of the abstract as the setting for the MOR application.

pith-pipeline@v0.9.0 · 5692 in / 1153 out tokens · 19620 ms · 2026-05-24T22:45:34.127502+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

13 extracted references · 13 canonical work pages · 2 internal anchors

[1]

A. C. Antoulas. Approximation of large-scale dynamical systems. Advances in Design and Control 6. Philadelphia, PA: Society for Industrial and App lied Mathematics (SIAM), 2005

work page 2005
[2]

I. P. Duﬀ and P. K¨ urschner. Numerical computation and ne w output bounds for time-limited balanced truncation of discrete-time systems. arXiv preprint:1902.01652, 2019

work page internal anchor Pith review Pith/arXiv arXiv 1902
[3]

E. Emmrich. Discrete versions of Gronwall’s lemma and th eir application to the numerical analysis of parabolic problems. Preprint No. 637, TU Berlin , 1999

work page 1999
[4]

D. F. Enns. Model reduction with balanced realizations: An error bound and a frequency weighted generalization. Proceedings of the 23rd IEEE Conference on Decision and Control, pages 127–132, 1984

work page 1984
[5]

Gawronski and J

W. Gawronski and J. Juang. Model reduction in limited tim e and frequency intervals. Int. J. Syst. Sci. , 21(2):349–376, 1990

work page 1990
[6]

K. Glover. All optimal Hankel-norm approximations of li near multivariable systems and their L∞-error norms. 39(6):1115–1193, 1984

work page 1984
[7]

Gugercin and A

S. Gugercin and A. C. Antoulas. A Survey of Model Reductio n by Balanced Truncation and Some New Results. International Journal of Control , 77(8):748–766, 2004

work page 2004
[8]

K¨ urschner

P. K¨ urschner. Balanced truncation model order reducti on in limited time intervals for large systems. Advances in Computational Mathematics , pages 1–24, 2018

work page 2018
[9]

B. C. Moore. Principal component analysis in linear syst ems: Controllability, observability, and model reduction. IEEE Trans. Autom. Control , 26:17–32, 1981

work page 1981
[10]

Pernebo and L

L. Pernebo and L. M. Silverman. Model reduction via bala nced state space representations. IEEE Transactions on Automatic Control , 27(2):382–387, 1982

work page 1982
[11]

M. Redmann. Energy estimates and model order reduction for stochastic bilinear systems. International Journal of Control , 2018

work page 2018
[12]

M. Redmann. A new type of singular perturbation approxi mation for stochastic bilinear systems. arXiv preprint:1903.11600, 2019

work page internal anchor Pith review Pith/arXiv arXiv 1903
[13]

Redmann and P

M. Redmann and P. K¨ urschner. An output error bound for t ime-limited balanced truncation. Systems & Control Letters , 121:1–6, 2018. 7

work page 2018

[1] [1]

A. C. Antoulas. Approximation of large-scale dynamical systems. Advances in Design and Control 6. Philadelphia, PA: Society for Industrial and App lied Mathematics (SIAM), 2005

work page 2005

[2] [2]

I. P. Duﬀ and P. K¨ urschner. Numerical computation and ne w output bounds for time-limited balanced truncation of discrete-time systems. arXiv preprint:1902.01652, 2019

work page internal anchor Pith review Pith/arXiv arXiv 1902

[3] [3]

E. Emmrich. Discrete versions of Gronwall’s lemma and th eir application to the numerical analysis of parabolic problems. Preprint No. 637, TU Berlin , 1999

work page 1999

[4] [4]

D. F. Enns. Model reduction with balanced realizations: An error bound and a frequency weighted generalization. Proceedings of the 23rd IEEE Conference on Decision and Control, pages 127–132, 1984

work page 1984

[5] [5]

Gawronski and J

W. Gawronski and J. Juang. Model reduction in limited tim e and frequency intervals. Int. J. Syst. Sci. , 21(2):349–376, 1990

work page 1990

[6] [6]

K. Glover. All optimal Hankel-norm approximations of li near multivariable systems and their L∞-error norms. 39(6):1115–1193, 1984

work page 1984

[7] [7]

Gugercin and A

S. Gugercin and A. C. Antoulas. A Survey of Model Reductio n by Balanced Truncation and Some New Results. International Journal of Control , 77(8):748–766, 2004

work page 2004

[8] [8]

K¨ urschner

P. K¨ urschner. Balanced truncation model order reducti on in limited time intervals for large systems. Advances in Computational Mathematics , pages 1–24, 2018

work page 2018

[9] [9]

B. C. Moore. Principal component analysis in linear syst ems: Controllability, observability, and model reduction. IEEE Trans. Autom. Control , 26:17–32, 1981

work page 1981

[10] [10]

Pernebo and L

L. Pernebo and L. M. Silverman. Model reduction via bala nced state space representations. IEEE Transactions on Automatic Control , 27(2):382–387, 1982

work page 1982

[11] [11]

M. Redmann. Energy estimates and model order reduction for stochastic bilinear systems. International Journal of Control , 2018

work page 2018

[12] [12]

M. Redmann. A new type of singular perturbation approxi mation for stochastic bilinear systems. arXiv preprint:1903.11600, 2019

work page internal anchor Pith review Pith/arXiv arXiv 1903

[13] [13]

Redmann and P

M. Redmann and P. K¨ urschner. An output error bound for t ime-limited balanced truncation. Systems & Control Letters , 121:1–6, 2018. 7

work page 2018