Controllability and Observability Imply Exponential Decay of Sensitivity in Dynamic Optimization

Sungho Shin; Victor M. Zavala

arxiv: 2101.06350 · v5 · submitted 2021-01-16 · 🧮 math.OC

Controllability and Observability Imply Exponential Decay of Sensitivity in Dynamic Optimization

Sungho Shin , Victor M. Zavala This is my paper

Pith reviewed 2026-05-24 14:14 UTC · model grok-4.3

classification 🧮 math.OC

keywords exponential decay of sensitivitydynamic optimizationcontrollabilityobservabilitymodel predictive controlsecond order sufficiency

0 comments

The pith

Uniform controllability and observability imply exponential decay of sensitivity in dynamic optimization.

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

The paper establishes that exponential decay of sensitivity holds for dynamic optimization problems when the underlying system satisfies uniform controllability and observability. It connects these system properties to the uniform second-order sufficiency condition and uniform linear independence constraint qualification, which prior work already tied to the decay property under uniform Hessian boundedness. A reader would care because the result bounds how far a data change at one time step can influence optimal decisions at distant steps, directly aiding scalable computation over long horizons.

Core claim

Uniform controllability and observability imply the uniform second order sufficiency condition and uniform linear independence constraint qualification. Together with uniform boundedness of the Lagrangian Hessian, these conditions deliver exponential decay of sensitivity.

What carries the argument

Uniform controllability and observability of the dynamic system, which guarantee uSOSC and uLICQ and thereby imply EDS.

If this is right

Sensitivity of the solution at stage i to a perturbation at stage j decays exponentially as |i-j| grows.
Approximation and solution schemes for long-horizon problems can exploit the decay to reduce computation.
Perturbation effects along the time horizon remain localized and can be bounded explicitly.

Where Pith is reading between the lines

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

The same controllability-observability route might apply to nonlinear or infinite-horizon variants once uniformity is verified.
In applied model predictive control one could check controllability numerically as a proxy for the decay property.
The decay bound may tighten stability margins or error estimates in receding-horizon implementations.

Load-bearing premise

The prior result that uniform boundedness of the Lagrangian Hessian plus uSOSC and uLICQ already give EDS must hold; the paper only adds the controllability link to those two conditions.

What would settle it

A dynamic optimization problem that meets uniform controllability and observability yet exhibits sensitivities that fail to decay exponentially with stage separation would disprove the claimed implication.

Figures

Figures reproduced from arXiv: 2101.06350 by Sungho Shin, Victor M. Zavala.

read the original abstract

We study a property of dynamic optimization (DO) problems (as those encountered in model predictive control and moving horizon estimation) that is known as exponential decay of sensitivity (EDS). This property indicates that the sensitivity of the solution at stage $i$ against a data perturbation at stage $j$ decays exponentially with $|i-j|$. {Building upon our previous results, we show that EDS holds under uniform boundedness of the Lagrangian Hessian, a uniform second order sufficiency condition (uSOSC), and a uniform linear independence constraint qualification (uLICQ). Furthermore, we prove that uSOSC and uLICQ can be obtained under uniform controllability and observability. Hence, we have that uniform controllability and observability imply EDS.} These results provide insights into how perturbations propagate along the horizon and enable the development of approximation and solution schemes. We illustrate the developments with numerical examples.

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 shows controllability and observability yield uSOSC and uLICQ for EDS but the headline still needs separate uniform Hessian boundedness.

read the letter

The main thing to know is that this paper connects uniform controllability and observability to the uSOSC and uLICQ conditions that support exponential decay of sensitivity, building directly on the authors' prior EDS results. It does this by proving those two qualifications follow from the control properties, then notes that EDS holds when those are paired with uniform Lagrangian Hessian boundedness. The numerical examples are meant to show the decay in practice, which is the sort of concrete check that helps in MPC and MHE settings. That link is the actual new piece and it is a reasonable extension within the subfield. The soft spot is the headline implication. The abstract states EDS requires the Hessian bound plus uSOSC plus uLICQ, shows the last two come from controllability and observability, and then concludes the control properties imply EDS. The Hessian bound is not shown to follow from controllability, so it remains an extra assumption. If the paper does not establish that the bound holds automatically under the same uniformity, the stated implication is incomplete. Uniformity arguments across time-varying horizons also need to be tight for the result to apply cleanly. This work is for people already working on sensitivity in dynamic optimization who know the EDS literature. A reader in that niche gets a usable condition chain even if the step is incremental. It has enough formal structure and a clear target application to deserve referee time rather than an immediate desk reject.

Referee Report

1 major / 1 minor

Summary. The manuscript claims that uniform controllability and observability imply exponential decay of sensitivity (EDS) in dynamic optimization problems. It builds upon prior results establishing that EDS holds under uniform boundedness of the Lagrangian Hessian together with uniform second-order sufficient conditions (uSOSC) and uniform linear independence constraint qualification (uLICQ). The authors prove that uSOSC and uLICQ follow from uniform controllability and observability, and conclude that the latter therefore imply EDS. Numerical examples are provided to illustrate the developments.

Significance. If the central implication holds, the result supplies a control-theoretic route to EDS that links standard, verifiable system properties to sensitivity decay in DO problems arising in MPC and MHE. This could support the design of approximation and solution schemes that exploit the decay. The explicit reliance on previously established EDS conditions is a positive feature that keeps the contribution focused.

major comments (1)

[Abstract] Abstract: The headline claim that 'uniform controllability and observability imply EDS' omits the uniform boundedness of the Lagrangian Hessian, which the preceding sentence lists as a prerequisite for EDS (alongside uSOSC and uLICQ). The manuscript must clarify whether this boundedness condition is implied by the controllability/observability assumptions or must be retained as an independent hypothesis; without this clarification the scope of the central implication is ambiguous.

minor comments (1)

[Abstract] The abstract paragraph containing the main claim is enclosed in curly braces, which appears to be a formatting artifact and should be removed.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed reading and the constructive comment on the abstract. We address the point below and will revise the manuscript accordingly.

read point-by-point responses

Referee: [Abstract] Abstract: The headline claim that 'uniform controllability and observability imply EDS' omits the uniform boundedness of the Lagrangian Hessian, which the preceding sentence lists as a prerequisite for EDS (alongside uSOSC and uLICQ). The manuscript must clarify whether this boundedness condition is implied by the controllability/observability assumptions or must be retained as an independent hypothesis; without this clarification the scope of the central implication is ambiguous.

Authors: We agree that the abstract phrasing is ambiguous and could be misread as suggesting that controllability and observability alone suffice for EDS. In the manuscript, uniform boundedness of the Lagrangian Hessian remains an independent standing assumption (as in our prior work establishing EDS under boundedness + uSOSC + uLICQ); it is not derived from controllability/observability. We will revise the abstract to read: 'Building upon our previous results, we show that EDS holds under uniform boundedness of the Lagrangian Hessian together with uniform controllability and observability (which yield uSOSC and uLICQ).' This makes the scope explicit. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation builds on independent prior results

full rationale

The paper proves uSOSC and uLICQ from uniform controllability and observability (new contribution) and invokes prior published results only for the base case that EDS holds under those plus uniform Hessian boundedness. This self-citation is external to the present manuscript, does not reduce the target EDS implication to a definitional identity or fitted input, and satisfies the criteria for independent support. No self-definitional, fitted-prediction, or load-bearing self-citation patterns are present in the derivation chain.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 0 invented entities

The central claim rests on three uniform conditions (bounded Lagrangian Hessian, uSOSC, uLICQ) plus prior EDS theorems; no free parameters or new entities are introduced.

axioms (3)

domain assumption Uniform boundedness of the Lagrangian Hessian
Invoked as a standing assumption to obtain EDS from uSOSC and uLICQ.
domain assumption Uniform second-order sufficiency condition (uSOSC)
Derived from controllability/observability; required for the EDS proof.
domain assumption Uniform linear independence constraint qualification (uLICQ)
Derived from controllability/observability; required for the EDS proof.

pith-pipeline@v0.9.0 · 5681 in / 1303 out tokens · 19813 ms · 2026-05-24T14:14:12.529731+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

13 extracted references · 13 canonical work pages · 3 internal anchors

[1]

, " * write output.state after.block = add.period write newline

ENTRY address author booktitle chapter doi edition editor eid howpublished institution journal key month note number organization pages publisher school series title type url volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.all := #1 'mid.sent...

work page
[2]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

work page
[3]

and Rockafellar, R.T

Dontchev, A.L. and Rockafellar, R.T. (2009). Implicit functions and solution mappings, volume 543. Springer

work page 2009
[4]

Gr \"u ne, L., Schaller, M., and Schiela, A. (2019). Sensitivity analysis of optimal control for a class of parabolic P D E s motivated by model predictive control. SIAM Journal on Control and Optimization, 57(4), 2753--2774

work page 2019
[5]

Gr¨une, M

Gr \"u ne, L., Schaller, M., and Schiela, A. (2020 a ). Abstract nonlinear sensitivity and turnpike analysis and an application to semilinear parabolic P D E s. arXiv preprint arXiv:2008.13001

work page arXiv 2020
[6]

(2020 b )

Gr \"u ne, L., Schaller, M., and Schiela, A. (2020 b ). Exponential sensitivity and turnpike analysis for linear quadratic optimal control of general evolution equations. Journal of Differential Equations, 268(12), 7311--7341

work page 2020
[7]

and D'Andrea, R

Hehn, M. and D'Andrea, R. (2011). A flying inverted pendulum. In 2011 IEEE International Conference on Robotics and Automation, 763--770. IEEE

work page 2011
[8]

and Anitescu, M

Na, S. and Anitescu, M. (2020 a ). Exponential decay in the sensitivity analysis of nonlinear dynamic programming. SIAM Journal on Optimization, 30(2), 1527--1554

work page 2020
[9]

Na and M

Na, S. and Anitescu, M. (2020 b ). Superconvergence of online optimization for model predictive control. arXiv preprint arXiv:2001.03707

work page arXiv 2020
[10]

Na, S., Shin, S., Anitescu, M., and Zavala, V.M. (2020). Overlapping schwarz decomposition for nonlinear optimal control. arXiv preprint arXiv:2005.06674

work page internal anchor Pith review Pith/arXiv arXiv 2020
[11]

Robinson, S.M. (1980). Strongly regular generalized equations. Mathematics of Operations Research, 5(1), 43--62

work page 1980
[12]

Shin, S., Anitescu, M., and Zavala, V.M. (2021). Exponential decay of sensitivity in graph-structured nonlinear programs. arXiv preprint arXiv:2101.03067v1

work page internal anchor Pith review Pith/arXiv arXiv 2021
[13]

Diffusing-Horizon Model Predictive Control

Shin, S. and Zavala, V.M. (2020). Diffusing-horizon model predictive control. arXiv preprint arXiv:2002.08556

work page internal anchor Pith review Pith/arXiv arXiv 2020

[1] [1]

, " * write output.state after.block = add.period write newline

ENTRY address author booktitle chapter doi edition editor eid howpublished institution journal key month note number organization pages publisher school series title type url volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.all := #1 'mid.sent...

work page

[2] [2]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

work page

[3] [3]

and Rockafellar, R.T

Dontchev, A.L. and Rockafellar, R.T. (2009). Implicit functions and solution mappings, volume 543. Springer

work page 2009

[4] [4]

Gr \"u ne, L., Schaller, M., and Schiela, A. (2019). Sensitivity analysis of optimal control for a class of parabolic P D E s motivated by model predictive control. SIAM Journal on Control and Optimization, 57(4), 2753--2774

work page 2019

[5] [5]

Gr¨une, M

Gr \"u ne, L., Schaller, M., and Schiela, A. (2020 a ). Abstract nonlinear sensitivity and turnpike analysis and an application to semilinear parabolic P D E s. arXiv preprint arXiv:2008.13001

work page arXiv 2020

[6] [6]

(2020 b )

Gr \"u ne, L., Schaller, M., and Schiela, A. (2020 b ). Exponential sensitivity and turnpike analysis for linear quadratic optimal control of general evolution equations. Journal of Differential Equations, 268(12), 7311--7341

work page 2020

[7] [7]

and D'Andrea, R

Hehn, M. and D'Andrea, R. (2011). A flying inverted pendulum. In 2011 IEEE International Conference on Robotics and Automation, 763--770. IEEE

work page 2011

[8] [8]

and Anitescu, M

Na, S. and Anitescu, M. (2020 a ). Exponential decay in the sensitivity analysis of nonlinear dynamic programming. SIAM Journal on Optimization, 30(2), 1527--1554

work page 2020

[9] [9]

Na and M

Na, S. and Anitescu, M. (2020 b ). Superconvergence of online optimization for model predictive control. arXiv preprint arXiv:2001.03707

work page arXiv 2020

[10] [10]

Na, S., Shin, S., Anitescu, M., and Zavala, V.M. (2020). Overlapping schwarz decomposition for nonlinear optimal control. arXiv preprint arXiv:2005.06674

work page internal anchor Pith review Pith/arXiv arXiv 2020

[11] [11]

Robinson, S.M. (1980). Strongly regular generalized equations. Mathematics of Operations Research, 5(1), 43--62

work page 1980

[12] [12]

Shin, S., Anitescu, M., and Zavala, V.M. (2021). Exponential decay of sensitivity in graph-structured nonlinear programs. arXiv preprint arXiv:2101.03067v1

work page internal anchor Pith review Pith/arXiv arXiv 2021

[13] [13]

Diffusing-Horizon Model Predictive Control

Shin, S. and Zavala, V.M. (2020). Diffusing-horizon model predictive control. arXiv preprint arXiv:2002.08556

work page internal anchor Pith review Pith/arXiv arXiv 2020