the curse of linearity and time-invariance
TRANSCRIPT
![Page 1: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/1.jpg)
The curse of
linearity and time-invariance
Alessandro Astolfi
Imperial College London
and
University of Rome “Tor Vergata”
CDC 2019 -- Nice
![Page 2: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/2.jpg)
Motivations
A semi-plenary at the CDC provides a unique opportunity to reflect on the
past and to look at the future …
I spent 20+ years understanding linear objects from a nonlinear
perspective and I now believe that
linearity and time-invariance are a curse
They confuse our intuition and delay our understanding ….
pretty much like Euclidean geometry and the standard notion of
orthogonality obfuscate our understanding of space
Before spending another 20+ years in trying to (mis-)understand more
linear objects I would like to reflect on some of the lessons I have learnt.
![Page 3: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/3.jpg)
Motivations
A semi-plenary at the CDC provides a unique opportunity to reflect on the
past and to look at the future …
I spent 20+ years understanding linear objects from a nonlinear
perspective and I now believe that
linearity and time-invariance are a curse
They confuse our intuition and delay our understanding ….
pretty much like Euclidean geometry and the standard notion of
orthogonality obfuscate our understanding of space
Before spending another 20+ years in trying to (mis-)understand more
linear objects I would like to reflect on some of the lessons I have learnt.
L TI
![Page 4: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/4.jpg)
The rules
How to break the curse of linearity and time-invariance
1. Study linear, time-invariant systems
2. Avoid matrix multiplication/inversion
3. While differentiating, keep track of constant terms
4. Never use frequency or Laplace transforms
5. If need use, matrices always act on something
6. Replace linear algebra with interconnection, invariance, pde’s,
coordinates transformations, the principle of optimality, dynamic
programming, trajectories, differential operators, graph theory
![Page 5: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/5.jpg)
The plan
1. Moments and phasors (Giordano Scarciotti)
2. The Loewner functions(Joel Simard)
3. Persistence of excitation (Alberto Padoan)
4. Adaptive control (Kaiwen Chen)
5. Optimal control (Mario Sassano)
Analysis
Design
![Page 6: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/6.jpg)
Moments are ubiquitous in maths/physics/biology ….
Moments in probability: 1, mean, variance, skewness, ....
Moment of a force/torque: first order moment
Electrical dipole: first order moment
Moment of inertia: second order moment
Phasors: first order moments
Biomass concentration: first order moment
Cell density: zero-th order moment
T. Stieltjes, Recherches sur les fractions continues. Annales de la Faculté des Sciences de l'Université de Toulouse pour les Sciences Mathématiques et les Sciences Physiques, 1984.
![Page 7: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/7.jpg)
… and systems theory
h is the impulse response of a linear time-invariant (SISO) system
0-moment at :
k-moment at :L
L
![Page 8: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/8.jpg)
Moments: from transfer functions to state space
h is the impulse response of a linear time-invariant (SISO) system
0-moment at :
L
![Page 9: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/9.jpg)
Moments: from transfer functions to state space
h is the impulse response of a linear time-invariant (SISO) system
k-moment at :
L
![Page 10: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/10.jpg)
The notion of moment – Linear systems
The interpolation point The system
Steady state response
Moments
Asymptotic stabilityExcitability
Observability
![Page 11: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/11.jpg)
The notion of moment – Linear systems
The interpolation point The system
Steady state response
Moments
Alternatively
![Page 12: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/12.jpg)
The notion of moment – Linear systems – Swapped
The systemThe interpolation point
![Page 13: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/13.jpg)
The notion of moment – Linear systems – Swapped
The system The interpolation point
Asymptotic stability Controllability
Excitability
Steady state response
Moments
Non
-obs
erva
bili
ty
Du
ali
ty
L
![Page 14: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/14.jpg)
The notion of moment – Linear systems – Summary
The systemThe generator
(interpolation points)
Steady state response
Moments
Steady state response
Moments
Krylov projectors
A. N. Krylov, On the numerical solution of the equation by which, in technical questions, frequencies of small oscillations of material systems are determined, Izv. Akad. Nauk SSSR, ser. fis.-mat., 1931
![Page 15: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/15.jpg)
The notion of moment – Linear systems – Questions
Is the interconnection approach adequate to extend the notion of moment
to nonlinear systems?
How do we interpret the non-observability condition?
Can we provide an intrinsic interpretation of the swapped interconnection
without using duality?
Can we simultaneously interconnect generators left and right?
How do we interpret the excitability condition?
![Page 16: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/16.jpg)
The notion of moment – Nonlinear systems
The interpolation point The system
Moments
Steady state response
Asymptotic stabilityExcitability
Observability
Poisson stability
![Page 17: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/17.jpg)
The notion of moment – Nonlinear systems
The interpolation point The system
Moments
Steady state response
![Page 18: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/18.jpg)
The notion of moment – Nonlinear systems
The signal generator captures the requirement that one is interested
in studying the behaviour of the system only in specific circumstances
The interconnected system possesses an invariant manifold
and the dynamics restricted to the manifold are a copy of the
dynamics of the signal generator
The interpolation point The system
is by definition the moment of the nonlinear system at
![Page 19: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/19.jpg)
The systemThe interpolation point
Asymptotic stability
Moments
Non-observability
Excitability
Controllability
CIBS
The notion of moment – Nonlinear systems – Swapped
Poisson stability
![Page 20: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/20.jpg)
The system The interpolation points
The notion of moment – Nonlinear systems – Swapped
With special
“care”
Non-observability
Moments
![Page 21: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/21.jpg)
The notion of moment – Linear systems – Questions
Is the interconnection approach adequate to extend the notion of moment
to nonlinear systems?
How do we interpret the non-observability condition?
Can we provide an intrinsic interpretation of the swapped interconnection
without using duality?
Can we simultaneously interconnect generators left and right?
How do we interpret the excitability condition?
Before answering these questions we take a detour into circuits theory
![Page 22: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/22.jpg)
Phasor transform and moments
The phasor transform coincides with the Sylvester equation
defining the moment (at the phasor angular frequency)
The component of the matrix P are the phasors of the currents
and of the integrals of the currents
The phasor of the output response is the moment of the system
(at the phasor angular frequency)
K.A.R. Steinmetz & E.J. Berg , "Complex Quantities and Their Use in Electrical Engineering". Proceedings of the International Electrical Congress, Chicago, American Institute of Electrical Engineers. 1893
A link between the phasor transform and moments:
to be used for nonlinear circuits and switched mode electronics
![Page 23: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/23.jpg)
Phasor transform and basic circuit elements
Phasor
transform
It contains
information on
the “regime”
Discontinuous
phasor
transform
L
![Page 24: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/24.jpg)
Phasor transform and circuit elements/properties
Time domain Phasor domain
Average power Reactive power
Switched mode operation with 50% duty cycle
½ in the
sinusoidal case
It contains information on the “regime”
![Page 25: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/25.jpg)
… back to the plan …
1. Moments and phasors
2. The Loewner functions
3. Persistence of excitation
4. Adaptive control
5. Optimal control
![Page 26: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/26.jpg)
The notion of moment – Double-sided interconnection
The Loewner (divided difference) matrix
C. Loewner, Über monotone Matrixfunktionen, Mathematische Zeitschrift, 1938.
L
![Page 27: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/27.jpg)
The notion of moment – Double-sided interconnection
Moment at li Moment at mi
The right (!) interpolation points The left (!) interpolation points
![Page 28: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/28.jpg)
The notion of moment – Double-sided interconnection
The right (!) interpolation points The left (!) interpolation points
Moment at mi
![Page 29: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/29.jpg)
The notion of moment – Double-sided interconnection
Moment at li
The right (!) interpolation points The left (!) interpolation points
![Page 30: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/30.jpg)
The notion of moment – Double-sided interconnection
The right (!) interpolation points The left (!) interpolation points
Generalized reachability
and observability
matrices
![Page 31: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/31.jpg)
The notion of moment – Double-sided interconnection
The Loewner matrix allows double-sided interpolation, but heavily relies on
the transfer function and on linearity
The associated Sylvester equation is not related to any invariance condition
L
![Page 32: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/32.jpg)
The notion of moment – Double-sided interconnection
The Loewner matrix allows double-sided interpolation, but heavily relies on
the transfer function and on linearity
The associated Sylvester equation is not related to any invariance condition
Two invariance-like
equations
![Page 33: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/33.jpg)
The notion of moment – Double-sided interconnection
The shifted left and shifted right Loewner matrices
as Lie derivatives “along the interpolation points”
![Page 34: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/34.jpg)
The notion of moment – Double-sided interconnection
The left- and right-
Loewner matrices
transform
the cascade into a
parallel
interconnection
![Page 35: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/35.jpg)
The notion of moment – Double-sided interconnection
The Loewner matrices
define the interpolating system
Moment at li
Moment at mi
![Page 36: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/36.jpg)
The notion of moment – Double-sided interconnection
The Loewner matrices
define the interpolating system
The autonomous behaviour of the interpolating system is such that
the shift and the time differentiation commute TI
![Page 37: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/37.jpg)
The notion of moment – Double-sided interconnection
The shifted left and shifted right Loewner matrices
define the interpolating system
The autonomous behaviour of the interpolating system is such that
the shift and the time differentiation commute with some correction
![Page 38: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/38.jpg)
The notion of moment – Double-sided interconnection
The left- and right- Loewner “functions” transform
the cascade into a parallel interconnection
The shifted left and shifted right Loewner “functions” are Lie derivatives “along the interpolation points”
Two (?) invariance-like equations
The autonomous behaviour of the interpolating system is such that
the shift and the time differentiation commute with some correction
![Page 39: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/39.jpg)
The notion of moment – Double-sided interconnection
The left- and right- Loewner “functions” transform
the cascade into a parallel interconnection
The shifted left and shifted right Loewner “functions” are Lie derivatives “along the interpolation points”
The autonomous behaviour of the interpolating system is such that
the shift and the time differentiation commute with some correction
Three invariance-like equations
![Page 40: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/40.jpg)
… making progress …
1. Moments and phasors
2. The Loewner functions
3. Persistence of excitation
4. Adaptive control
5. Optimal control
![Page 41: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/41.jpg)
Excitability
Moments
Steady state response
Asymptotic stabilityExcitability
Observability
Poisson stability
ExcitabilityReachability
L
![Page 42: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/42.jpg)
Excitability – Linear systems
The excitation distribution The excitation rank condition
The excitation distribution and the excitation rank condition do not
rely on linear concepts (such as the Cayley-Hamilton theorem)
ExcitabilityReachability
L
![Page 43: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/43.jpg)
Excitability – Nonlinear systems
Excitation distribution ≠ Strong accessibility distribution
The excitation distribution
![Page 44: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/44.jpg)
Excitability – Nonlinear systems
The excitation distribution
The excitation rank condition
Excitation rank condition ≠ Strong accessibility rank condition
![Page 45: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/45.jpg)
Excitability – A geometric characterization
Suppose solutions are analytic
Solution
is PE
PD for all t and some T
The excitation
rank condition
holds at every
![Page 46: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/46.jpg)
Excitability – A geometric characterization
Suppose solutions are analytic
and is almost periodic
Solution
is PE
PD for all t and some T
The excitation
rank condition
holds at
H. Bohr, "Zur Theorie der fastperiodischen Funktionen I" Acta Math., 1925
![Page 47: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/47.jpg)
Excitability – Applications
The excitability rank condition plays a fundamental role in model
reduction problems from data
Time
![Page 48: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/48.jpg)
Excitability – Applications
The excitability rank condition, hence the PE condition, plays a
fundamental role stability analysis of skew-symmetric systems
Almost-periodicity, excitability rank condition, minimality
imply uniformly globally exponential stability
A. Cayley, "Sur les determinants gauches" , Crelle's Journal, 1847.
![Page 49: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/49.jpg)
... more progress …
1. Moments and phasors
2. The Loewner functions
3. Persistence of excitation
4. Adaptive control
5. Optimal control
![Page 50: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/50.jpg)
Adaptive control
Excitability and PE naturally lead to adaptive control
Where is the curse of linearity/time-invariance in adaptive control?
Classical adaptive control I&I adaptive control
Design the update law to
create a passive
interconnection
Design the update law to
create an L2 stable
cascade
TI
![Page 51: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/51.jpg)
Adaptive control
Excitability and PE naturally lead to adaptive control
Where is the curse of linearity/time-invariance in adaptive control?Classical adaptive control I&I adaptive control
Adaptation
gain
Passivity?
TI
![Page 52: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/52.jpg)
Adaptive control – Congelation of variables
Excitability and PE naturally lead to adaptive control
Where is the curse of linearity/time-invariance in adaptive control?
Passivity?
A new passive
interconnection
S. Bochner, Vorlesungen über Fouriersche Integrale, 1932
![Page 53: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/53.jpg)
Adaptive control – Congelation of variables
A new passive
interconnection
A new passive interconnection
![Page 54: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/54.jpg)
Adaptive control – Congelation of variables
The congelation of variables allows to recover, in simple cases, passive
interconnections.
How do we extend this idea to more general systems?
Is time-invariance exploited in other steps of classical adaptive design?
![Page 55: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/55.jpg)
Adaptive control – Congelation of variables
Kreisselmeier filters
(input, output, regressor)
A congealed change of
coordinates
![Page 56: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/56.jpg)
Adaptive control – Congelation of variables
The re-parameterized system
The inverse dynamics
![Page 57: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/57.jpg)
Adaptive control – Congelation of variables
The inverse dynamics
A change of coordinates, the chain rule and standard backstepping …
![Page 58: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/58.jpg)
Adaptive control – Congelation of variables
The time-invariant case does
not reveal all interconnections
The congelation of variables
allows constructing a new
interconnection
TI
The interconnections are
however present whenever
parameters vary
![Page 59: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/59.jpg)
Adaptive control – Congelation of variables
I&I adaptive control A general stabilization result
TIL. Euler, Solutio problematis ad geometriam situs pertinentis, 1741
![Page 60: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/60.jpg)
… and to conclude
1. Moments and phasors
2. The Loewner functions
3. Persistence of excitation
4. Adaptive control
5. Optimal control
![Page 61: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/61.jpg)
Optimal control
Timaeus of Taormina, „Queen Dido and the isoperimetric problem“, c. 345 BC – c. 250 BC
Optimal control problems can be solved using dynamic programming or
Pontryagin’s minimum principle
Combining both approaches may yield a new perspective and new
optimality conditions
The basic linear ingredients
L
![Page 62: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/62.jpg)
Optimal control – A graphical interpretation
For any initial condition and any t>0 the
composition has a fixed point
Flow along the
optimal closed-
loop system
Lifting on the
invariant subspace
of the Hamiltonian
Flow backward
along the
Hamiltonian system
Projection
onto x
![Page 63: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/63.jpg)
Optimal control – A graphical interpretation
… and equivalently
Flow backward
along the optimal
closed-loop system
Flow along the
Hamiltonian system
Projection
onto x
Optimal
path
Non-optimal
path
![Page 64: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/64.jpg)
Optimal control – A graphical interpretation
The optimal feedback gain and the solution
of the Riccati equation are such that
Stability
Fixed point
![Page 65: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/65.jpg)
Optimal control – Linear systems
The optimal feedback gain and the solution
of the Riccati equation are such that
This relies on Cayley-Hamilton theorem L
Polynomial
in K*
Linear in P*
![Page 66: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/66.jpg)
Optimal control – Linear systems
The optimal feedback gain and the solution
of the Riccati equation are such that
Positive semi-definite, also for
robust control problemsPositive semi-definite
This is amenable to the use of optimization algorithms on the
manifold of positive definite matrices with cost
![Page 67: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/67.jpg)
Optimal control – Nonlinear systems
The basic nonlinear ingredients
Skew-symmetric
![Page 68: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/68.jpg)
Optimal control – A graphical interpretation
Exactly the same interpretation as for linear systems
Flow backward
along the optimal
closed-loop system
Flow along the
Hamiltonian system
Projection
onto x
Optimal
path
Non-optimal
path
![Page 69: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/69.jpg)
Optimal control – A graphical interpretation
The optimal feedback gain and the solution
of the optimal costate equation are such that
Stability
Fixed point
This is amenable to compute arbitrarily accurate approximations
of the optimal control law without solving any PDE
![Page 70: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/70.jpg)
Summary
1. Moments and phasors
2. The Loewner functions
3. Persistence of excitation
4. Adaptive control
5. Optimal control
Analysis
Design
![Page 71: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/71.jpg)
Some take-away messages
1. Linearity and time-invariance provide a very powerful structure which
may be misleading
2. A careful study of linear, time-invariant systems with abstract tools
allows
• developing nonlinear/time-varying enhancement of analysis
and design tools
• improving our understanding of essential features and
interactions
• ….
• breaking the curse of linearity and time-invariance
![Page 72: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/72.jpg)
Thanks!
R. Antonelli
P. Di Franco
T. Mylvaganam
P. Ascencio
A. Giannari
K. Nunna
S. Bardi
J. Jiang
A. Padoan
D. Carnevale
A. Jones
M. Sassano
H. Chang
G. Kaliora
G. Scarciotti
K. Chen
M.C. Laiou
J. Simard
R. Ortega L. Praly P. Colaneri A.C. Antoulas
A. Isidori
![Page 73: The curse of linearity and time-invariance](https://reader030.vdocument.in/reader030/viewer/2022012703/61a529b3c2b12242f849bf55/html5/thumbnails/73.jpg)
The curse of
linearity and time-invariance
Alessandro Astolfi
Imperial College London
and
University of Rome “Tor Vergata”
CDC 2019 -- Nice