Professional Documents
Culture Documents
Full Chapter Modelling and Simulation For Autonomous Systems Revised Selected Papers of The 7Th International Conference Mesas 2020 Prague Czech Republic October 21 2020 1St Edition Jan Mazal PDF
Full Chapter Modelling and Simulation For Autonomous Systems Revised Selected Papers of The 7Th International Conference Mesas 2020 Prague Czech Republic October 21 2020 1St Edition Jan Mazal PDF
https://textbookfull.com/product/agents-and-artificial-
intelligence-11th-international-conference-icaart-2019-prague-
czech-republic-february-19-21-2019-revised-selected-papers-jaap-
van-den-herik/
https://textbookfull.com/product/software-technologies-14th-
international-conference-icsoft-2019-prague-czech-republic-
july-26-28-2019-revised-selected-papers-marten-van-sinderen/
https://textbookfull.com/product/e-business-and-
telecommunications-16th-international-conference-
icete-2019-prague-czech-republic-july-26-28-2019-revised-
selected-papers-mohammad-s-obaidat/
Founding Editors
Gerhard Goos
Karlsruhe Institute of Technology, Karlsruhe, Germany
Juris Hartmanis
Cornell University, Ithaca, NY, USA
123
Editors
Jan Mazal Adriano Fagiolini
NATO M&S COE University of Palermo
Rome, Italy Palermo, Italy
Petr Vasik Michele Turi
Brno University of Technology NATO M&S COE
Brno, Czech Republic Rome, Italy
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
This volume contains selected papers presented at the MESAS 2020 Conference:
Modelling and Simulation for Autonomous Systems, held on October 21, 2020, in
Prague. The initial idea to launch the MESAS project was introduced by the NATO
Modelling and Simulation Centre of Excellence in 2013, with the intent to bring
together the Modelling and Simulation and the Autonomous Systems/Robotics com-
munities and to collect new ideas for concept development and experimentation in this
domain. From that time, the event has gathered together (in keynote, regular, poster,
and way ahead sessions) fully recognized experts from different technical communities
in the military, academia, and industry. The main topical parts of the 2020 edition of
MESAS were “Future Challenges of Advanced M&S Technology,” “M&S of Intelli-
gent Systems,” and “AxS in Context of Future Warfare and Security Environment.”
The community of interest submitted 38 papers for consideration. Each submission was
reviewed by three Technical Committee members or selected independent reviewers. In
this review process, the committee decided to accept 20 papers to be presented (in four
sessions), and 19 of these papers were accepted for inclusion in the conference
proceedings.
https://www.mscoe.org/
Organization
General Chairs
Michele Turi NATO M&S COE, Italy
Adriano Fagiolini University of Palermo, Italy
Stefan Pickl Universität der Bundeswehr München, Germany
Agostino Bruzzone University of Genoa, Italy
Alexandr Štefek University of Defence, Czech Republic
Technical Committee
Ronald C. Arkin Georgia Institute of Technology, USA
Özkan Atan Van Yüzüncü Yil University, Turkey
Richard Balogh Slovak University of Technology in Bratislava,
Slovakia
Yves Bergeon CREC St-Cyr, France
Marco Biagini NATO M&S COE, Rome, Italy
Antonio Bicchi University of Pisa, Italy
Dalibor Biolek University of Defence, Czech Republic
Arnau Carrera Viñas NATO STO CMRE, Italy
Josef Casar University of Defence, Czech Republic
Erdal Çayirci University of Stavanger, Norway
Massimo Cossentino National Research Council of Italy, Italy
Andrea D’Ambrogio University of Rome Tor Vergata, Italy
Frédéric Dalorso NATO ACT Autonomy Program, USA
Radek Doskocil University of Defence, Czech Republic
Jan Faigl Czech Technical University in Prague, Czech Republic
Luca Faramondi Università Campus Bio-Medico di Roma, Italy
Jan Farlik University of Defence, Czech Republic
Pavel Foltin University of Defence, Czech Republic
Petr Frantis University of Defence, Czech Republic
Jakub Fučík University of Defence, Czech Republic
Corrado Guarino Lo Bianco University of Parma, Italy
Karel Hajek University of Defence, Czech Republic
Kamila Hasilová University of Defence, Czech Republic
Vaclav Hlavac Czech Technical University in Prague, Czech Republic
Jan Hodicky NATO HQ SACT, USA
Jan Holub Czech Technical University in Prague, Czech Republic
x Organization
Numerical Experiment on Optimal Control for the Dubins Car Model . . . . . . 111
Stanislav Frolík, Matej Rajchl, and Marek Stodola
The GRASP Metaheuristic for the Electric Vehicle Routing Problem. . . . . . . 189
David Woller, Viktor Kozák, and Miroslav Kulich
1 Introduction
The geometric algebras have played significant role in the engineering applica-
tions. The main tool in the scientific engineering computations lies in confor-
mal geometric algebra, for more informations see the classical books [3,6,16,18]
or engineering papers [9,10,12,19]. In the context of specific applications, the
panorama of geometric algebras becomes more and more popular, for example
we can find some applications of the compass ruler algebra in the book [7] or
the geometric algebra of conics in the paper [11] or the others algebras in the
paper [13].
In the quantum computing based on classical quantum mechanics [2] a qubit
can be seen as a superposition of the bits |0 and |1 in the vector in two dimen-
sional vector space over complex numbers C2 :
|ψ = a0 |0 + a1 |1, (1)
where a0 , a1 ∈ C together with the normalization condition |a0 | +|a1 | =
2 2
a0 a∗0 +
a1 a∗1 = 1, where z ∗ is a complex conjugation of the complex number z. This is
the usual representation of a superposition of the two states |0 and |1 in the
classical notation (see, e.g., [2,8,17]). We can see |a0 |2 as the probability that
the quantum bit indicates the state |0, while |a1 |2 is the probability that the
quantum bit will be found in the state |1.
The first author was supported by the grant no. FSI-S-20-6187. The second author was
supported by solution grand FV20-31 science Fund of the FME 2020 at Brno University
of Technology.
c Springer Nature Switzerland AG 2021
J. Mazal et al. (Eds.): MESAS 2020, LNCS 12619, pp. 3–14, 2021.
https://doi.org/10.1007/978-3-030-70740-8_1
4 J. Hrdina and R. Tichý
2 Geometric Algebra
Geometric algebra is a distributive associative algebra over quadratic space Rp,q
with the quadratic form of signature (p, q) and with the neutral element 1 ∈ R.
The quadratic form defines relations for a scalar product on basis vectors, i.e.
basis elements of the quadratic space denoted by {σ1 , σ2 , . . . , σp+q } ∈ Rp,q ⊂
Gp,q , in such a way that
⎧
⎪
⎨+1, 1 ≤ (i = j) ≤ p,
σi · σj = −1, p < (i = j) ≤ p + q,
⎪
⎩
0, i = j.
The product of the algebra called geometric product satisfies the condition
a2 = a · a = Q(a), ∀a ∈ Rp,q .
The conclusion of the last case of the geometric product relations is that the
geometric algebra is a graded algebra, i.e. the product of a various number of
distinct basis vectors represents an element of the algebra. For instance σi σj ,
1 ≤ i < j ≤ p + q is a basis element of the algebra called basis bivector,
σi σj σk , 1 ≤ i < j < k ≤ p + q is basis 3-vector etc. The element of the highest
grade, i.e. the geometric product of p + q different vectors is called pseudoscalar.
This element is denoted by I.
3 Geometric Algebra G3
In the geometric algebra settings the spin states (states of quantum system) can
be written in G3 framework. Let us recall that, the algebra G3 is an algebra over
real numbers R based on three dimensional real vector space R3 equipped with
the Euclidean scalar product of signature (+, +, +). The vectors correspond
to elements σ1 , σ2 and σ3 together with identities σ12 = σ22 = σ32 = 1. The
identification between C2 and the algebra of even grade elements G+ 3 is based
on isomorphism
3
a0 + ia3
|ψ = ↔ ψ = a0 + ak Iσk , (2)
−a2 + ia1
k=1
Quantum Computing Based on Quantum Bit Algebra QBA 5
Fig. 1. Bloch’s sphere generated by Matlab. Blue circle denotes all qubits with the
same probability of states |0 and |1 (Color figure online).
The total probability of the system is ψ|ψ = 1, it follows the rule |a0 |2 +
|a1 |2 = 1 and we can write |ψ in spherical coordinates (γ, θ, ϕ) as an element
of the three dimensional Bloch sphere (see Fig. 1)
iγ θ iϕ θ
|ψ = e cos |0 + e sin |1 ,
2 2
4 Quantum Gates
Let us introduce the list of basis quantum gates, see Fig. 2. We can introduce
Pauli X-gate, Pauli Y-gate and Pauli Z-gate. We can see two dimensional com-
plex vector space C2 as a four dimensional real vector space R4 . In this context,
one can see Pauli matrices as following 4 × 4 matrices
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
0010 0 0 0 1 10 0 0
⎜0 0 0 1⎟ ⎜ 0 0 −1 0⎟ ⎜0 1 0 0 ⎟
X=⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎝1 0 0 0⎠ , Y = ⎝ 0 1 0 0⎠ , Z = ⎝0 0 −1 0 ⎠ .
0100 −1 0 0 0 0 0 0 −1
The next gates are for example Hadamart gate (H), Phase gate (S) and π8 gate
(T). These gates are represented by following matrices:
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
10 1 0 10 0 0 10 0 0
1 ⎜0 1 0 1 ⎟ ⎜ ⎟ ⎜
⎟ , S = ⎜0 1 0 0⎟ , T = ⎜0 1 0 π 0 ⎟
H=√ ⎜ ⎝ ⎠ ⎝ ⎠ ⎝
⎟
π⎠.
2 1 0 −1 0 0 0 0 1 0 0 cos 4 − sin 4
0 1 0 −1 0 0 −1 0 0 0 sin π4 cos π4
We can define the quantum equivalents to classical computing protocol, for exam-
ple one can represent the not gate by transformation
|0 → |1
|1 → |0
where ψ is a qubit, see equation (2). Finally, one can formulate the quantum
protocol as a circuit over quantum gates. For example in Fig. 4 one can see linear
transformation based on matrix
⎛ ⎞⎛ ⎞⎛ ⎞ ⎛ ⎞⎛ ⎞ ⎛ ⎞
0010 10 0 0 0010 00 0 1 0010 0 100
⎜0 0 0 1⎟ ⎜0 1 0 0⎟ ⎜0 0 0 1⎟ ⎜0 0 −1 0⎟ ⎜0 0 0 1⎟ ⎜−1 0 0 0⎟
⎜ ⎟⎜ ⎟⎜ ⎟ ⎜ ⎟⎜ ⎟ ⎜ ⎟
⎝1 0 0 0⎠ ⎝0 0 0 1⎠ ⎝1 0 0 0⎠ = ⎝1 0 0 0⎠ ⎝1 0 0 0⎠ = ⎝ 0 0 1 0⎠ .
0100 0 0 −1 0 0100 01 0 0 0100 0 001
5 QBA
A different possibility how to implement qubit in the geometric algebra frame-
work has been introduced in paper [4]. The authors proposed to model a qubit
by unit elements of Clifford algebra Cl(2, 2) with an indefinite bilinear form of
the signature (+, −, +, −). We denote the basis of the algebra by vectors
where (γ0x )2 = (γ1x )2 = 1 and (γ0t )2 = (γ1t )2 = −1. A qubit can be defined as
where cx0 , ct0 , cx1 , ct1 ∈ R. The elements cx0 γ0x + ct0 γ0t , such that (cx0 )2 − (ct0 )2 = 1,
define the state |0 and the elements cx1 γ1x + ct1 γ1t , such that (cx1 )2 − (ct1 )2 = 1,
define the state |1. It implies that qubits belong to a 3-dimensional surface
defined by the identity ψ|ψ = 1 so
θ γ0t γ0x ϕ0 x θ γ1t γ1x ϕ1 x
|ψ = cos e γ0 + sin e γ1
2 2
(5)
γ0t γ0x ϕ0 θ x γ1t γ1x ϕ1 −γ0t γ0x ϕ0 θ
=e cos γ0 + e sin γ1x
2 2
t x t x
where eγ0 γ0 ϕ0 = cosh(ϕ0 )+sinh(ϕ0 )γ0t γ0x and eγ1 γ1 ϕ1 = cosh(ϕ1 )+sinh(ϕ1 )γ1t γ1x ,
so qubits belong to hyperboloid given by (5). Algebraically we can describe a
vector ψ from (4) as an expression
ψ = cx0 γ0x + ct0 γ0t + cx1 γ1x + ct1 γ1t = (cx0 + ct0 γ0t γ0x )γ0x + (cx1 + ct1 γ1t γ1x )γ1x ,
such that (γit γix )2 = γit γix γit γix = −γit γit γix γ1x = 1, such that i ∈ {0, 1}. So we can
represent alternatively a qubit as a couple of elements from geometric algebra
G2 .
where the multiplication is a standard geometric product in Cl(2, 2). For better
readability we denote the basis elements by E1 , . . . , E6 and define them as
1 x t 1 1 x t
E1 = γ γ , E2 = γ0x γ1x , E3 = γ γ ,
2 0 0 2 2 0 1
1 1 1
E4 = γ0t γ1x , E5 = γ0t γ1t , E6 = γ1x γ1t .
2 2 2
For instance, the commutator of the elements E1 , E2 is
1 x t x x
[E1 , E2 ] = γ0 γ0 γ0 γ1 − γ0x γ1x γ0x γ0t
4
1
= −(γ0x )2 γ0t γ1x + (γ0x )2 γ1x γ0t
4
1
= −2γ0t γ1x
4
= −E4 .
This example also clarifies why we are using the factor 12 at each element. By the
same procedure for all combinations of basis elements we obtain a multiplication
table:
Quantum Computing Based on Quantum Bit Algebra QBA 9
[, ] E1 E2 E3 E4 E5 E6
E1 0 −E4 −E5 −E2 −E3 0
E2 E4 0 −E6 −E1 0 E3
E3 E5 E6 0 0 E1 E2
E4 E2 E1 0 0 E6 E5
E5 E3 0 −E1 −E6 0 E4
E6 0 −E3 −E2 −E5 −E4 0
K := E4 + E5 , L := E6 , M := E4 − E5 ,
[, ] K L M
K 0 K −2L
L −K 0 M
M 2L −M 0
Since this is the table of commutators for special linear algebra sl(2), the algebra
spanned by generators E4 , E5 , E6 is isomorphic to sl(2). The adjoint represen-
tation (the representation provided by commutator operation, see [14]) is given
by the following matrices:
X1 0 Y1 0 Z1 0
adE4 = , adE5 = , adE6 = ,
0 −Z1 0 −Y1 0 −X1
where
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
000 0 01 0 −1 0
X1 = ⎝0 0 1⎠ , Y1 = ⎝ 0 0 0⎠ , Z1 = ⎝−1 0 0⎠ .
010 −1 0 0 0 0 0
where
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
010 −1 0 0 0 0 0
X2 = ⎝0 0 1⎠ , Y2 = ⎝ 0 0 0⎠ , Z2 = ⎝−1 0 0⎠ .
000 0 01 0 −1 0
[, ] X Y Z
X 0 X Y
Y −X 0 Z
Z −Y −Z 0
which leads again to a Lie algebra isomorphic to sl(2) and adjoint representation
represents the splitting E1 , E2 , E3 ⊕ E4 , E5 , E6 .
The classification of this Lie algebra can also be determined by the eigenval-
ues of adjoint representation. Since the adjoint representation of matrices X, Y, Z
has 2 distinct real eigenvalues, it belongs to the class of Lie algebras isomorphic
to sl(2), see [1]. Thus the Lie algebra of bivectors of Cl(2, 2) can be written as
semidirect product
Cl2 (2, 2) ∼
= sl(2) sl(2).
7 Hyperbolic Qubits
Now, let us look closer to the Hyperbolic quantum formalism [15], we will see
that the qubits here belong to the same Bloch’s hyperboloid (5) as in QBA.
In detail, hyperbolic numbers belong to linear algebra over R, such that each
element can be written as the formal linear combination z = x + jy, with respect
to identity j 2 = 1. The hyperbolic conjugation is defined as (x + jy)∗ = x − jy
Quantum Computing Based on Quantum Bit Algebra QBA 11
√
and then Hyperbolic norm is defined as |z| = zz ∗ = x2 − y 2 . In the sense
of Clifford algebras, the Hyperbolic numbers are just Geometric algebra G1 . Let
us note that G1 is not an integral domain, because there are zero divisors, for
example (1 + j)(1 − j) = 0. Hyperbolic vector space is then a module H over
geometric algebra G1 .
We define Hyperbolic scalar product as a G1 -bilinear map ( , ) : H × H → G1
together with properties
1
http://www.gaalop.de/gaalopweb-for-qubits/.
12 J. Hrdina and R. Tichý
1 1
( (γ0x + γ1x )(γ0t + γ1t ))(α0 γ0x + α1 γ0t + β0 γ1x + β1 γ1t )( (γ0t + γ1t )(γ0x + γ1x ))
2 2
= (β0 γ0x + β1 γ0t + α0 γ1x + α1 γ1t ).
10 Conclusions
References
1. Bowers, A.: Classification of three-dimensional real liealgebras, April 2005. http://
www.math.ucsd.edu/∼abowers/downloads/survey/3d Lie alg classify.pdf
2. de Lima Marquezino, F., Portugal, R., Lavor, C.: A Primer on Quantum Comput-
ing. SCS. Springer, Cham (2019). https://doi.org/10.1007/978-3-030-19066-8
3. Dorst, L., Fontijne, D., Mann, S.: Geometric Algebra for Computer Science. An
Object-Oriented Approach to Geometry. Morgan Kaufmann, Burlington (2007)
4. Hildenbrand, D., Steinmetz, C., Alves, R., Hrdina, J., Lavor, C.: An online calcu-
lator for Qubits based on geometric algebra. In: Magnenat-Thalmann, N., et al.
(eds.) CGI 2020. LNCS, vol. 12221, pp. 526–537. Springer, Cham (2020). https://
doi.org/10.1007/978-3-030-61864-3 45
14 J. Hrdina and R. Tichý
5. Hildenbrand, D., Steinmetz, C., Tichý, R.: Gaalopweb for matlab: an easy to han-
dle solution for industrial geometric algebra implementations. Adv. Appl. Clifford
Algebr. 30(4) (2020)
6. Hildenbrand, D.: Foundations of Geometric Algebra Computing. Springer, Heidel-
berg (2013). https://doi.org/10.1007/978-3-642-31794-1
7. Hildenbrand, D.: Introduction to Geometric Algebra Computing. Taylor & Francis
Group, Boca Raton (2019)
8. Homeister, M.: Quantum Computing verstehen. Grundlagen - Anwendungen - Per-
spektiven. Friedrich Vieweg & Sohn Verlag (2018). https://doi.org/10.1007/978-
3-658-22884-2
9. Hrdina, J., Návrat, A.: Binocular computer vision based on conformal geometric
algebra. Adv. Appl. Clifford Algebras 27, 1945–1959 (2017)
10. Hrdina, J., Návrat, A., Vašı́k, P.: Control of 3-link robotic snake based on conformal
geometric algebra. Adv. Appl. Clifford Algebr. 26(3) (2016)
11. Hrdina, J., Návrat, A., Vašı́k, P.: Geometric algebra for conics. Adv. Appl. Clifford
Algebras 28, 66 (2018)
12. Hrdina, J., Návrat, A., Vašı́k, P., Matoušek, R.: CGA-based robotic snake con-
trol. Adv. Appl. Clifford Algebras 27(1), 621–632 (2016). https://doi.org/10.1007/
s00006-016-0695-5
13. Hrdina, J., Vašı́k, P., Matoušek, R., Návrat, A.: Geometric algebras for uniform
colour spaces. Math. Methods Appl. Sci. 41, 4117–4130 (2018)
14. K. Erdmann, M.J.W.: Introduction to Lie Algebras. Springer, London
(2011). https://www.ebook.de/de/product/4018299/k erdmann mark j wildon
introduction to lie algebras.html
15. Khrennikov, A.: Hyperbolic quantum mechanics. Adv. Appl. Clifford Algebras
13(1), 1–9 (2003)
16. Lounesto, P.: Clifford Algebra and Spinors. CUP, Cambridge (2006)
17. McMahon, D.: Quantum Computing Explained. Wiley, New York (2008)
18. Perwass, C.: Geometric Algebra with Applications in Engineering. Springer, Hei-
delberg (2009). https://doi.org/10.1007/978-3-540-89068-3
19. Stodola, M.: Monocular kinematics based on geometric algebras. In: Mazal, J. (ed.)
MESAS 2018. LNCS, vol. 11472, pp. 121–129. Springer, Cham (2019). https://doi.
org/10.1007/978-3-030-14984-0 10
On Building Communication Maps
in Subterranean Environments
1 Introduction
on a GP regressor [3]. The reported results support the feasibility of the pro-
posed approach and show significant improvements in the prediction accuracy
in the extrapolation tasks.
The rest of the paper is organized as follows. A brief overview of the related
work with the focus on multi-robot wireless communication is presented in
Sect. 2. The proposed approach for communication map building is introduced
in Sect. 3. Results of the experimental evaluation are reported and discussed in
Sect. 4, and the work is concluded in Sect. 5.
2 Related Work
Signal attenuation model [25] accounts for natural signal strength decay that
occurs with the growing distance from the signal source. Since obstacles are
not modeled, the model is called Free-Space Path Loss (FSPL). Its predictions
of relative signal decay Lp are computed given the distance between receiver
and transmitter d, their respective antennas directivities Dr , Dt , and the signal
wavelength λ: 2
λ
Lp = 10 log10 Dt Dr . (1)
4πd
Unlike the constant-range and line-of-sight methods, the signal attenuation
model returns a continuous value of signal strength instead of the binary predic-
tions. An extension of the attenuation modeling by considering obstacles present
in the environment is presented in [14]. However, explicit knowledge about sig-
nal and material properties must be known beforehand. Thus both explorative
deployments and the ability of on-line adaption to environment change are
limited.
Another research path is led by Malmirchegini et al. [13], who proposed a
purely probabilistic model for estimation of signal characteristics with the notion
of some spatial dependence. Their model also provides mathematical reasoning
about the predictability of the communication parameters. However, since their
model lacks an explicit environment model, it cannot infer the attenuation there-
upon; it always needs some prior measurements in the area of interest.
More sophisticated models are based on raytracing the signal propagation
from the source by casting rays through the environment [22]. These models need
to comprehend phenomena such as multi-path signal propagation, waveguide
effect, or reflections. Hence, the raytracing-based approaches require detailed
geometric environment models that include surface normals and carry informa-
tion about obstacle material properties. Further, such approaches are computa-
tionally very demanding and sensitive to inaccurate or biased data about the
environment that make these approaches unsuitable for on-board deployments
with limited resources and field measurement devices.
The methods mentioned above tend to yield pessimistic but guaranteed
results when used under particular assumptions in an austere environment. How-
ever, in sites that suffer from waveguide effect, non-trivial reflections, or non-
trivial attenuation, these basic methods can return wrong predictions even in
the close vicinity of the queried points. Only one group of existing approaches
for communication map building is, to the best of our knowledge, based on pure
machine learning without relying on any strong prior knowledge of the physi-
cal properties of the environment. In [3], a communication map is built using
a spatial GP, where the core idea is to interpolate samples of communication
link quality gathered in-vivo. The GP provides the most likely regression of the
initially unknown signal strength function, and it interpolates the input samples
together with the variance of the predicted estimates. It can thus be used to
determine the most probable link quality with the confidence estimate. Con-
sequently, the GP-based model can be utilized in active perception tasks such
as [11] because information about predictive variance can be used as a valuation
On Building Communication Maps in Subterranean Environments 19
of the model uncertainty for the given configurations. If the whole space of all
spatial pairs is sampled with sufficiently high resolution, it is possible to predict
the signal characteristics perfectly. Even though the efforts are being made to
sample the whole space in [3,18], such sampling is a problem with the complexity
that can be bounded by O(n4 ) because we need to sample all (two-dimensional)
coordinate pairs, which can be too costly or unfeasible. Therefore, we consider
GP-based modeling to address this drawback and develop a more general model
that scales better with the environment size without dense sampling before a
particular deployment in the same environment.
We use GP regressors in each model to infer the signal strength from samples
transformed by respective descriptors. In this section, we briefly describe the GP
regression to make the paper self-contained. First, let the function of interest
f (x) be observed with Gaussian noise