tensorkelechi/arxiv-cstext · Datasets at Hugging Face

text stringlengths 0 89.3k
simulation that the openloop system is unstable
Figure 4 shows the control inputs corresponding to the
continuum kernels computed using the fullorder in y
10power series of order N 610152025and the solu
tion12to then1kernel equations 1 2 The controls
start to diverge for N 6 implying that the controls fail
to stabilize the closedloop system This is expected based
on the numerical experiments of Section 62 as the accu
racy obtained with such a loworder approximation is not
adequate As the approximation order Nincreases one
can see that the controls tend to zero implying that for
larger values of N the controls are stabilizing For the
lower orders N 10andN 15the controls are distin
guishable and the decay rate is relatively slow whereas
forN 20andN 25the controls are virtually identi
cal and converge to zero after two seconds implying that
all the state components have converged to zero by that
time as well This is consistent with the control gains
displayed in Figure 2 which are distinctly diﬀerent for
N 51015and virtually identical for N 202530
Finally we see that the approximate controls for N 20
andN 25are also very close to the controls computed
based on the solution to the n 1kernel equations 1
2
0 2 4 6 8 1005005
Figure 4 The controls obtained with approximate continuum ker
nels computed with fullorder in y power series 9 of order N
610152025and by solving 1 2 for the exact control
Figure 5 shows the corresponding results as Figure 4
but for the reducedorder in y power series approxima
tion 13 for the continuum kernel The same comments
apply to Figure 5 as to Figure 4 A slight diﬀerence can
be observed for the loworder approximations where the
caseN 6diverges at a slower rate than in Figure 4
the closedloop system is still unstable while the case
N 10appears to converge to zero at a faster rate Re
gardless for larger values of N the controls are virtually
indistinguishable from one another as well as from the
ones displayed in Figure 4 This is consistent with the
control gains in Figure 3 We note here that as 17 Thm
41 predicts for suﬃciently large nn 10in the present
case the corresponding feedback law obtained employing
the continuum control kernel k1ξyis stabilizing pro
vided that k1ξyis approximated at a suﬃcient accu
racy via the power series In turn the latter is established
12The solution is computed based on a ﬁnitediﬀerence approxi
mation of 1 2in Theorem 1 In conclusion stabilization with perfor
mance close to the one corresponding to exact control ker
nels is achieved with computational complexity of order
ON2that does not grow with n in contrast to the com
putation of the exact control kernels see Table 1 We
note that the computational complexity of solving 1
2 using a ﬁnitediﬀerence approximation grows roughly
linearly with n as shown in 17 Figure 6
0 2 4 6 8 10005
Figure 5 The controls obtained with approximate continuum ker
nels computed with reducedorder in y power series 13 of order
N 610152025andNy 2and by solving 1 2 for the exact
control
7 Conclusions and Future Work
In this paper we provided computational tools for con
struction of backsteppingbased stabilizing kernels for c on
tinua of hyperbolic PDEs and thus also for largescale
PDE systems These tools include both explicit kernels
construction even though the class of systems for which
this is possible may be restrictive and approximation via
power series We demonstrated the accuracy and eﬃciency
of these tools on two numerical examples In the ﬁrst
we derived the closedform solution to continuum kernel
equations and demonstrated the convergence of the power
series approximation to the closedform solution as Nin
creases In the second numerical example we could not
derive the solution in closed form and thus we employed
only the power series approach and reducedorder power
series approach to solve the continuum kernel equations
We further demonstrated that the computed continuum
kernels provided a stabilizing feedback law for the cor
responding largescale system of linear PDEs with per
formance comparable to the case of employing the exact
largescale control kernels
For future work the eﬃcacy of Algorithm 1 could po
tentially be substantially improved by adopting the doubl e
vectormatrix framework from 20 Sect IIB However
this has to be modiﬁed to ﬁt the triple power series em
ployed here potentially requiring a triple vectormatrix
framework which may not be straightforward to imple
ment Another topic for future research is to derive ana
lytically the convergence rate of the power series represen
tation for the solution to the continuum kernel equations
thus providing explicit although potentially conservati ve
11estimates of the lowest power required for an accurate
power seriesbased approximation
Appendix
Appendix A Proof of Proposition 1
The solution stated in Proposition 1 can be constructed
by looking for a separable solution to 3 4 ie kxξy
kxxkξξkyyandkxξ kxxkξξ After dividing
the equations 3 by the solution we get
µk
xx

simulation that the openloop system is unstable

Figure 4 shows the control inputs corresponding to the

continuum kernels computed using the fullorder in y

10power series of order N 610152025and the solu

tion12to then1kernel equations 1 2 The controls

start to diverge for N 6 implying that the controls fail

to stabilize the closedloop system This is expected based

on the numerical experiments of Section 62 as the accu

racy obtained with such a loworder approximation is not

adequate As the approximation order Nincreases one

can see that the controls tend to zero implying that for

larger values of N the controls are stabilizing For the

lower orders N 10andN 15the controls are distin

guishable and the decay rate is relatively slow whereas

forN 20andN 25the controls are virtually identi

cal and converge to zero after two seconds implying that

all the state components have converged to zero by that

time as well This is consistent with the control gains

displayed in Figure 2 which are distinctly diﬀerent for

N 51015and virtually identical for N 202530

Finally we see that the approximate controls for N 20

andN 25are also very close to the controls computed

based on the solution to the n 1kernel equations 1

2

0 2 4 6 8 1005005

Figure 4 The controls obtained with approximate continuum ker

nels computed with fullorder in y power series 9 of order N

610152025and by solving 1 2 for the exact control

Figure 5 shows the corresponding results as Figure 4

but for the reducedorder in y power series approxima

tion 13 for the continuum kernel The same comments

apply to Figure 5 as to Figure 4 A slight diﬀerence can

be observed for the loworder approximations where the

caseN 6diverges at a slower rate than in Figure 4

the closedloop system is still unstable while the case

N 10appears to converge to zero at a faster rate Re

gardless for larger values of N the controls are virtually

indistinguishable from one another as well as from the

ones displayed in Figure 4 This is consistent with the

control gains in Figure 3 We note here that as 17 Thm

41 predicts for suﬃciently large nn 10in the present

case the corresponding feedback law obtained employing

the continuum control kernel k1ξyis stabilizing pro

vided that k1ξyis approximated at a suﬃcient accu

racy via the power series In turn the latter is established

12The solution is computed based on a ﬁnitediﬀerence approxi

mation of 1 2in Theorem 1 In conclusion stabilization with perfor

mance close to the one corresponding to exact control ker

nels is achieved with computational complexity of order

ON2that does not grow with n in contrast to the com

putation of the exact control kernels see Table 1 We

note that the computational complexity of solving 1

2 using a ﬁnitediﬀerence approximation grows roughly

linearly with n as shown in 17 Figure 6

0 2 4 6 8 10005

Figure 5 The controls obtained with approximate continuum ker

nels computed with reducedorder in y power series 13 of order

N 610152025andNy 2and by solving 1 2 for the exact

control

7 Conclusions and Future Work

In this paper we provided computational tools for con

struction of backsteppingbased stabilizing kernels for c on

tinua of hyperbolic PDEs and thus also for largescale

PDE systems These tools include both explicit kernels

construction even though the class of systems for which

this is possible may be restrictive and approximation via

power series We demonstrated the accuracy and eﬃciency

of these tools on two numerical examples In the ﬁrst

we derived the closedform solution to continuum kernel

equations and demonstrated the convergence of the power

series approximation to the closedform solution as Nin

creases In the second numerical example we could not

derive the solution in closed form and thus we employed

only the power series approach and reducedorder power

series approach to solve the continuum kernel equations

We further demonstrated that the computed continuum

kernels provided a stabilizing feedback law for the cor

responding largescale system of linear PDEs with per

formance comparable to the case of employing the exact

largescale control kernels

For future work the eﬃcacy of Algorithm 1 could po

tentially be substantially improved by adopting the doubl e

vectormatrix framework from 20 Sect IIB However

this has to be modiﬁed to ﬁt the triple power series em

ployed here potentially requiring a triple vectormatrix

framework which may not be straightforward to imple

ment Another topic for future research is to derive ana

lytically the convergence rate of the power series represen

tation for the solution to the continuum kernel equations

thus providing explicit although potentially conservati ve

11estimates of the lowest power required for an accurate

power seriesbased approximation

Appendix

Appendix A Proof of Proposition 1

The solution stated in Proposition 1 can be constructed

by looking for a separable solution to 3 4 ie kxξy

kxxkξξkyyandkxξ kxxkξξ After dividing

the equations 3 by the solution we get

µk

xx