=Paper= {{Paper |id=Vol-2923/paper1 |storemode=property |title=Computing of Odd Degree Isogenies on Supersingular Twisted Edwards Curves |pdfUrl=https://ceur-ws.org/Vol-2923/paper1.pdf |volume=Vol-2923 |authors=Anatoly Bessalov,Volodymyr Sokolov,Pavlo Skladannyi,Oleksii Zhyltsov |dblpUrl=https://dblp.org/rec/conf/cpits/BessalovSSZ21 }} ==Computing of Odd Degree Isogenies on Supersingular Twisted Edwards Curves== https://ceur-ws.org/Vol-2923/paper1.pdf
Computing of Odd Degree Isogenies
on Supersingular Twisted Edwards Curves
Anatoly Bessalova, Volodymyr Sokolova, Pavlo Skladannyia, and Oleksii Zhyltsova
a
    Borys Grinchenko Kyiv University, 18/2 Bulvarno-Kudriavska str., Kyiv, 04053, Ukraine

                Abstract
                An overview of the properties of three classes of curves in generalized Edwards form Ea,d with
                two parameters is given. The known formulas for the odd degree isogenies on curves Ed with
                one parameter are generalized to all classes of curves in Edwards form, and Theorem 1 on the
                isogenic mapping of the points of these curves is proved. The analysis of the known effective
                method for computing isogenies in Farashahi-Hosseini w-coordinates, justified for the curve
                Ed, is given. Theorem 2 proves the applicability of this method to the class of twisted Edwards
                curves. Examples of 3- and 5-isogenies of twisted Edwards curves are given. Methods for
                bypassing the exceptional points of such curves in PQC cryptosystems like CSIDH are
                proposed.

                Keywords1
                Generalized Edwards form curve, complete Edwards curve, twisted Edwards curve, quadratic
                Edwards curve, curve order, points order, isomorphism, isogeny, w-coordinate, quadratic
                residue, quadratic nonresidue.

1. Introduction
   Recently, there has been significant progress in the prospects for post-quantum cryptography (PQC)
on isogenies of supersingular elliptic curves. An effective alternative to the well-known Supersingular
Isogeny Diffie-Hellman (SIDH) [1] protocol is the new faster algorithm with a very short key lengthβ€”
Commutative SIDH (CSIDH) [2]. It offers a non-interactive key exchange protocol based on Alice and
Bob’s secret keys. Instead of the extended field 𝐹𝑝2 in SIDH, operations in CSIDH are performed over
a prime field 𝐹𝑝 , which for the given 𝑝 halves the length of the field elements and key sizes. Instead of
the acyclic curve in SIDH with subgroups of 2i and 3k of higher orders in CSIDH, the elliptic curve
contains cyclic subgroups of simple odd-order l1,l2,..,lmax, where lmax is specified by the security level.
   The implementation of SIDH and CSIDH algorithms was mainly based on the fastest arithmetic of
isogenies of curves in Montgomery form or mixed arithmetic of curves in Montgomery and Edwards
form. In [3], a new effective method for computing odd degree isogenies on Edwards curves based on
Farashahi-Hosseini w-coordinates [4] was proposed. This work, in turn, is based on Montgomery’s
method of differential points addition and adapts it to Edwards curves. The formulae for computing of
odd degree isogenies on Edwards curves [5] also contain components of differential points addition,
which allowed in [3, 4] with the help of w-coordinates to align the speeds of performing the
corresponding operations on the curves in the Montgomery and Edwards forms. The results of the
implementation of the CSIDH algorithm on Edwards curves [3] are already ahead of the closest
competitor.
   To work of the authors [3] was preceded their article [6], in which, in particular, an efficient
algorithm for computing 3-isogenies in projective coordinates was developed with the minimal
computation cost for today. However, for 5-isogenies, as our analysis showed [7], computations in
classical projective coordinates became almost three times more complicated. There are reasons to


Cybersecurity Providing in Information and Telecommunication Systems, January 28, 2021, Kyiv, Ukraine
EMAIL: a.bessalov@kubg.edu.ua (A.1); v.sokolov@kubg.edu.ua (A.2); p.skladannyi@kubg.edu.ua (A.3); o.zhyltsov@kubg.edu.ua (A.4)
ORCID: 0000-0002-6967-5001 (A.1); 0000-0002-9349-7946 (A.2); 0000-0002-7775-6039 (A.3); 0000-0002-7253-5990 (A.4)
             ©️ 2021 Copyright for this paper by its authors.
             Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
             CEUR Workshop Proceedings (CEUR-WS.org)



                                                                                                                                 1
consider the use of Farashahi-Hosseini w-coordinates and the method [3] as a method for optimal
computation of odd degree isogenies on Edwards curves.
   Complete Edwards curves 𝐸𝑑 with one parameter 𝑑, defined in [8] (πœ’(𝑑) = βˆ’1), have well-known
advantages: maximum exponentiation rate of a point, the universality of points addition law, affine
coordinates of a neutral element of a group. The introduction of the second parameter π‘Ž on the curve
πΈπ‘Ž,𝑑 in [9] expanded the class of curves in the Edwards form and gave rise, according to the
classification adopted in [10], to two new classes: twisted and quadratic Edwards curves. The last class,
together with complete ones in terminology [9], is called Edwards curves 𝐸𝑑 .
   The computing of odd degree isogenies on Edwards curves 𝐸𝑑 is carried out using the formulas
defined by Theorems 2–4 in [5]. Although Theorem 3 of this paper is formulated for normalized
Edwards curves πΈπ‘Ž,𝑑 β†’ 𝐸1,𝑑/π‘Ž , its existence condition for βˆšπ‘Ž is not satisfied in the class of twisted
Edwards curves over a prime field 𝐹𝑝 . In other words, isogenic mapping πΈπ‘Ž,𝑑 β†’ πΈβ€²π‘Ž,𝑑 remained
unknown for this class. One of the goals of this paper is to fill this gap and to prove Theorem 1 with a
generalization of results known for curves 𝐸𝑑 to curves πΈπ‘Ž,𝑑 over a prime field 𝐹𝑝 .
   Further, in the work [3], based on Farashahi-Hosseini w-coordinates, a method for computing of odd
degree isogenies for Edwards curves 𝐸𝑑 were developed and implemented, and Theorems were proved
for isogenic mappings of these curves. But the question remained whether this method works in the
existing conditions of twisted Edwards curves πΈπ‘Ž,𝑑 . Theorem 2 in this article puts an end to this question
as well.
   Our analysis in this paper is based on the properties of twisted and quadratic Edwards curves
connected as pairs of quadratic twists [10, 12]. Supersingular curves of these classes with a similar
order 𝑁𝐸 = 𝑝 + 1 = 2π‘š 𝑛, π‘š β‰₯ 3 (n is odd) exist only for 𝑝 = 3 mod 4 [11]. The minimum even
                                                                                          𝑖max
cofactor of the order of such curves is 8, then for CSIDH algorithm with odd 𝑛 = βˆπ‘–=1           𝑙𝑖 the field
modulus 𝐹𝑝 should be chosen as 𝑝 = 8𝑛 βˆ’ 1. To adapt the definitions for the arithmetic of isogenies on
Edwards curves and curves in Weierstrass form, we use a modified points addition law [10, 13].
   Sect. 2 gives a brief overview of the properties of three classes of Edwards curves according to the
classification [10]. In Sect. 3, we consider the properties of odd degree isogenies and prove Theorem 1
for a rational mapping πΈπ‘Ž,𝑑 β†’ πΈβ€²π‘Žβ€² ,𝑑′ expressed by functions of two and one variables, and give
examples of isogenies on twisted Edwards curves. In Sect. 4, based on Theorem 1, Theorem 2 is
formulated and proved for the isogenic mapping of the curve πΈπ‘Ž,𝑑 in Farashahi-Hosseini w-coordinates.
Estimates of the cost of computing isogenies in projective coordinates (π‘Š: 𝑍) [3] are given. Examples
are considered for classes of quadratic and twisted Edwards curves and methods are proposed to bypass
the exceptional points of the 2nd order on twisted Edwards curves.

2. Classes of Curves in the Generalized Edwards Form

    The elliptic curve in the generalized Edwards form [10] is determined by the equation
                 πΈπ‘Ž,𝑑 : π‘₯ 2 + π‘Žπ‘¦ 2 = 1 + 𝑑π‘₯ 2 𝑦 2 , π‘Ž, 𝑑 ∈ πΉπ‘βˆ— , 𝑑 β‰  1, π‘Ž β‰  𝑑, 𝑝 β‰  2.                (1)
   In contrast to the equation of this curve in [9], here we multiply the parameter π‘Ž by 𝑦 2
instead of π‘₯ 2 . With the quadratic character πœ’(π‘Žπ‘‘) = βˆ’1, the curve (1) is isomorphic to the
complete Edwards curve [8] with one parameter 𝑑
                           𝐸𝑑 : π‘₯ 2 + 𝑦 2 = 1 + 𝑑π‘₯ 2 𝑦 2 , πœ’(𝑑) = βˆ’1, 𝑑 β‰  0,1.                  (2)
    In the case πœ’(π‘Ž) = πœ’(𝑑) = 1, there is an isomorphism of the curve (1) with quadratic Edwards curve
[10]
                            𝐸𝑑 : π‘₯ 2 + 𝑦 2 = 1 + 𝑑π‘₯ 2 𝑦 2 , πœ’(𝑑) = 1, 𝑑 β‰  0,1,                  (3)
which, in contrast to (2), has the parameter 𝑑 defined as a square. This difference leads to radically
different properties of curves (2) and (3) [10], which are summarized below. Despite this, in world
literature, these classes of curves are often combined by the general term Edwards curves [9].
    In our paper [13], we proposed to change places of π‘₯ and 𝑦 coordinates in the form (1) of the
Edwards curve. The modified universal law of addition of the points of the curve (1) has the form:
                                                  π‘₯1 π‘₯2 βˆ’ π‘Žπ‘¦1 𝑦2 π‘₯1 𝑦2 + π‘₯2 𝑦1
                     (π‘₯1 , 𝑦1 ) + (π‘₯2 , 𝑦2 ) = (                 ,                 ).           (4)
                                                 1 βˆ’ 𝑑π‘₯1 π‘₯2 𝑦1 𝑦2 1 + 𝑑π‘₯1 π‘₯2 𝑦1 𝑦2


2
   If two points from (4) coincide, we have
                                              π‘₯12 βˆ’ π‘Žπ‘¦12   2π‘₯1 𝑦1
                             2(π‘₯1 , 𝑦1 ) = (         2 2,            ).                          (5)
                                             1 βˆ’ 𝑑π‘₯1 𝑦1 1 + 𝑑π‘₯12 𝑦12
   Determining the inverse point as βˆ’π‘ƒ = (π‘₯1 , βˆ’π‘¦1 ) we obtain according to (4) the coordinates of the
neutral element 𝑂 = (1,0) of the group of points. In addition to the neutral element 𝑂, the axis 𝑋 also
always contains the point 𝐷0 = (βˆ’1,0) of the 2nd order, such that 2𝐷0 = (1,0) = 𝑂. Depending on the
properties of the parameters π‘Ž and 𝑑, we can get two exceptional points of the 2nd order and two
exceptional points of 4th order with the coordinates:
                                           π‘Ž                      1
                             𝐷1,2 = (±√ , ∞) , ±𝐹1 = (∞, ± ),                                    (6)
                                           𝑑                    βˆšπ‘‘
where we put the sign ∞ when dividing by 0. They arise in the cases πœ’(π‘Žπ‘‘) = 1 and πœ’(𝑑) = 1,
respectively.
   Depending on the properties of the parameters π‘Ž and 𝑑, the curves in the generalized Edwards form
(1) are divided into 3 disjoint classes [10]:
   ο‚·     Complete Edwards curves with the conditions C1: πœ’(π‘Žπ‘‘) = βˆ’1.
   ο‚·     Twisted Edwards curves with the conditions C2.1: πœ’(π‘Ž) = πœ’(𝑑) = βˆ’1.
   ο‚·     Quadratic Edwards curves with the conditions C2.2: πœ’(π‘Ž) = πœ’(𝑑) = 1 [14–16].
   Basic properties of curves of these classes:
   1. For the points of the 2nd order, the class of complete Edwards curves over a prime field is the
   class of cyclic curves (with one point of the 2nd order), while twisted and quadratic Edwards curves
   form classes of acyclic curves (three points of the second-order each). The maximum order of points
                                                  𝑁
   of curves of the two last classes is equal to 𝐸⁄2.
   2. The class of complete Edwards curves does not contain exceptional points.
                                                                                              π‘Ž
   3.   Twisted Edwards curves contain two exceptional points of the 2nd order 𝐷1,2 = (±√ , ∞), and
                                                                                              𝑑
   quadratic Edwards curves, besides them, contain two more exceptional points of the 4th order ±𝐹1 =
            1
   (∞, ± ).
           βˆšπ‘‘
   4. Twisted and quadratic Edwards curves form quadratic twist pairs based on parameters
   transformations: π‘ŽΜƒ = π‘π‘Ž, 𝑑̃ = 𝑐𝑑, πœ’(𝑐) = βˆ’1.
   5. In the classes of twisted and quadratic Edwards curves, the replacement π‘Ž ↔ 𝑑 gives the
   isomorphism of curves πΈπ‘Ž,𝑑 ~𝐸𝑑,π‘Ž .
   6. Complete and quadratic Edwards curves are isomorphic to the curves with the parameter
   π‘Ž = 1: πΈπ‘Ž,𝑑 ~𝐸1,𝑑/π‘Ž = 𝐸𝑑̃ . The introduction of the new parameter π‘Ž into the equation of the curve
   (1) is necessary only for the class of twisted Edwards curves.
   For the curve (1) J-invariant is equal to [17]
                                  16(π‘Ž2 + 𝑑2 + 14π‘Žπ‘‘)3
                       𝐽(π‘Ž, 𝑑) =                        , π‘Žπ‘‘(π‘Ž βˆ’ 𝑑) β‰  0.
                                        π‘Žπ‘‘(π‘Ž βˆ’ 𝑑)4
   This parameter distinguishes between isogenic (with different J-invariants) and isomorphic (with
equal J-invariants) curves.

3. Odd Degree Isogenies on Twisted Edward Curves
   The isogeny from the elliptic curve 𝐸(𝐾) over the field 𝐾 to the curve 𝐸′(𝐾) is a homomorphism
πœ™: 𝐸(𝐾) β†’ 𝐸′(𝐾) that is given by rational functions. This means that for all 𝑃, 𝑄 ∈ 𝐸(𝐾), πœ™(𝑃 + 𝑄) =
πœ™(𝑃) + πœ™(𝑄) and that there are rational functions [17]
                                        𝑝(π‘₯) 𝑓(π‘₯)
                            πœ™(π‘₯, 𝑦) = (      ,𝑦       ) = (π‘₯β€², 𝑦′),                               (7)
                                        π‘ž(π‘₯) 𝑔(π‘₯)
mapping the points of curve 𝐸 to the points of the curve 𝐸′. The maximum of the degrees
𝑙 = deg πœ™(π‘₯, 𝑦) = max{deg 𝑝(π‘₯), deg π‘ž(π‘₯)} is called the degree of isogeny and its kernel
ker πœ™ = 𝐺 is the subgroup 𝐺 βŠ† 𝐸, the points of which are mapped by the function πœ™(π‘₯, 𝑦) into the
neutral element 𝑂 of the group 𝐸′. The degree of separable isogeny is equal to the order 𝑙 of its kernel.


                                                                                                       3
Isogeny compresses the set of the curve 𝐸 points by a factor 𝑙 (𝑙 points of the curve 𝐸 are mapped to
one point of the curve 𝐸′). At 𝐺 = 𝑂 isogeny becomes the isomorphism with the degree 1.
    The construction of odd degree isogenies on Edwards curves is based on Theorem 2 [5]. Let’s
formulate it taking into account the modification (4) of the points addition law of the curve (1)
at π‘Ž = 1.
    Theorem 2 [5]. Let 𝐺 = {(1,0), ±𝑄1 , ±𝑄2 , . . , ±𝑄𝑠 } the subgroup of odd order 𝑙 = 2𝑠 + 1 of the
points ±𝑄𝑖 = (𝛼𝑖 , ±𝛽𝑖 ) on the curve 𝐸𝑑 .
    We define
                                              π‘₯𝑃+𝑄𝑖 π‘₯π‘ƒβˆ’π‘„π‘–    𝑦𝑃+𝑄
                            πœ™(𝑃) = (∏                     ,∏      ).
                                               π‘₯𝑄𝑖 π‘₯βˆ’π‘„π‘–       π‘₯𝑄
                                        π‘„βˆˆπΊ               π‘„βˆˆπΊ
   Then πœ™(π‘₯, 𝑦) is 𝑙-isogeny with the kernel 𝐺 from the curve 𝐸𝑑 to the curve 𝐸′𝑑′ with the parameter
𝑑 = 𝐴8 𝑑𝑙 and the mapping function
 β€²
                                    𝑆                       𝑆
                           π‘₯    (𝛼𝑖 π‘₯)2 βˆ’ (𝛽𝑖 𝑦)2 𝑦   (𝛼𝑖 𝑦)2 βˆ’ (𝛽𝑖 π‘₯)2
                πœ™(π‘₯, 𝑦) = ( 2 ∏                  ,  ∏                   ).                          (8)
                           𝐴    1 βˆ’ (𝑑𝛼𝑖 𝛽𝑖 π‘₯𝑦)2 𝐴2   1 βˆ’ (𝑑𝛼𝑖 𝛽𝑖 π‘₯𝑦)2
                                𝑖=1                        𝑖=1
    Its proof is given in [5]. Its important consequence is that isogenic curves are in the same classes as
curves 𝐸𝑑 (i.e., complete Edwards curves are mapped to complete curves, twisted curvesβ€”to twisted
ones, and quadratic curvesβ€”to quadratic ones). This essentially distinguishes odd degree isogenies
from 2-isogenies (for them, the complete Edwards curves are mapped to quadratic ones).
    The formula (8) for the function πœ™(π‘₯, 𝑦) directly follows from the definition πœ™(𝑃) in the statement
of the Theorem and the addition law (4) for the arbitrary point (π‘₯𝑃 , 𝑦𝑃 ) = (π‘₯, 𝑦) with the kernel points
±𝑄𝑖 = (𝛼𝑖 , ±𝛽𝑖 ), so for coordinates pairs we have
          π‘₯𝑃+𝑄𝑖 π‘₯π‘ƒβˆ’π‘„π‘–       1 (𝛼𝑖 π‘₯)2 βˆ’ (𝛽𝑖 𝑦)2      𝑦𝑃+𝑄𝑖 π‘¦π‘ƒβˆ’π‘„π‘–      1 (𝛼𝑖 𝑦)2 βˆ’ (𝛽𝑖 π‘₯)2
                        = 2                     ,                  =                      .
           π‘₯𝑄𝑖 π‘₯βˆ’π‘„π‘–        𝛼𝑖 1 βˆ’ (𝑑𝛼𝑖 𝛽𝑖 π‘₯𝑦)2        𝑦𝑄𝑖 π‘¦βˆ’π‘„π‘–       𝛼𝑖2 1 βˆ’ (𝑑𝛼𝑖 𝛽𝑖 π‘₯𝑦)2
    The factors x and y before the products in the coordinates of the function πœ™(π‘₯, 𝑦) take into account
the neutral element 𝑂 = (1,0) of the isogeny kernel. It is obvious from (8) that the property
πœ™(1,0) = (1,0) holds, i.e. the neutral element is mapped into itself. For all points of the kernel
πœ™(±𝑄𝑖 ) = πœ™(𝛼𝑖 , ±𝛽𝑖 ) = (1,0) is also true.
    Theorem 2[5] and the mapping (8) are valid only for the classes of complete and quadratic Edwards
curves with the parameter π‘Ž = 1. The authors of [5] further formulated and proved Theorem 3 for
curves πΈπ‘Ž,𝑑 in the form (1), relying on the property of normalization of this curve with isomorphism
                                                        𝑦
πΈπ‘Ž,𝑑 ~𝐸1,𝑑/π‘Ž with the change of the coordinate 𝑦 β†’ ⁄ . For the class of twisted Edwards curves
                                                           βˆšπ‘Ž
(πœ’(π‘Ž) = πœ’(𝑑) = βˆ’1) over a prime field 𝐹𝑝 such a change does not exist, and the results of Theorem
3[5] are applicable only for the curves 𝐸𝑑 over the extended fields πΉπ‘π‘š , π‘š β‰₯ 2. For the PQC protocol
SIDH [1] with implementation on the curves 𝐸𝑑 over the field𝐹𝑝2 Theorem 3 [5] may be useful. But
for the CSIDH protocol [2] with curves over the field 𝐹𝑝 , this theorem does not give results for the
whole class of twisted Edwards curves. In this paper, we fill this gap and for the first time present
mapping formulas πœ™(𝑃) for the curve (1) that depends on two parameters π‘Ž and 𝑑.

  Theorem 1. Let 𝐺 = {(1,0), ±𝑄1 , ±𝑄2 , . . , ±𝑄𝑠 } is a subgroup of odd order 𝑙 = 2𝑠 + 1 of the points
±𝑄𝑖 = (𝛼𝑖 , ±𝛽𝑖 ) of the curve 𝐸𝑑 over the field 𝐹𝑝 .
  We define
                                               π‘₯𝑃+𝑄𝑖 π‘₯π‘ƒβˆ’π‘„π‘–    𝑦𝑃+𝑄𝑖 π‘¦π‘ƒβˆ’π‘„π‘–
                   πœ™(𝑃) = (π‘₯β€², 𝑦′) = (∏                    ,∏             ).
                                                π‘₯𝑄𝑖 π‘₯βˆ’π‘„π‘–       𝑦𝑄𝑖 π‘¦βˆ’π‘„π‘–
                                         π‘„βˆˆπΊ               π‘„βˆˆπΊ
  Then πœ™(π‘₯, 𝑦) is 𝑙-isogeny with the kernel 𝐺 from the curve πΈπ‘Ž,𝑑 to the curve πΈβ€²π‘Žβ€² ,𝑑′ with parameters
π‘Ž = π‘Žπ‘™ , 𝑑′ = 𝐴8 𝑑𝑙 , 𝐴 = βˆπ‘ π‘–=1 𝛼𝑖 and the mapping function
    β€²
                                𝑆                               𝑆
                         π‘₯    (𝛼𝑖 π‘₯)2 βˆ’ π‘Ž2 (𝛽𝑖 𝑦)2 𝑦   (𝛼𝑖 𝑦)2 βˆ’ (𝛽𝑖 π‘₯)2
              πœ™(π‘₯, 𝑦) = ( 2 ∏                     ,  ∏                   ),                         (9)
                         𝐴     1 βˆ’ (𝑑𝛼𝑖 𝛽𝑖 π‘₯𝑦)2 𝐴2     1 βˆ’ (𝑑𝛼𝑖 𝛽𝑖 π‘₯𝑦)2
                               𝑖=1                          𝑖=1




4
or
                                        𝑆                    𝑆
                                 π‘₯    π‘₯ 2 βˆ’ π‘Žπ›½π‘–2 βˆ’π‘¦     π‘₯ 2 βˆ’ 𝛼𝑖2
                      πœ™(π‘₯, 𝑦) = ( 2 ∏            ,   ∏             ).                                (10)
                                 𝐴    1 βˆ’ 𝑑𝛽𝑖 π‘₯ 2 𝐴2   π‘Ž βˆ’ 𝑑𝛼𝑖 π‘₯ 2
                                       𝑖=1                  𝑖=1
    Proof. The formula (9) follows directly from the definition πœ™(𝑃) and the points addition law (4) of
the curve (1).
    From (1) it is true that 𝑦 2 = (1 βˆ’ π‘₯ 2 )/(π‘Ž βˆ’ 𝑑π‘₯ 2 ), 𝛼𝑖2 + π‘Žπ›½π‘–2 = 1 + 𝑑𝛼𝑖2 𝛽𝑖2. Then, in the numerator
of the first coordinate π‘₯β€² in (9), each factor is transformed as:
                                                1 βˆ’ π‘₯2      π‘Ž(𝛼𝑖2 + π‘Žπ›½π‘–2 )π‘₯ 2 βˆ’ π‘Ž2 𝛽𝑖2 βˆ’ 𝑑𝛼𝑖2 π‘₯ 4
     π‘ˆπ‘– = 𝛼𝑖2 π‘₯ 2 βˆ’ 𝛽𝑖2 𝑦 2 = 𝛼𝑖2 π‘₯ 2 βˆ’ π‘Ž2 𝛽𝑖2           =                                           =
                                               π‘Ž βˆ’ 𝑑π‘₯ 2                     π‘Ž βˆ’ 𝑑π‘₯ 2
           π‘Ž(1 + 𝑑𝛼𝑖2 𝛽𝑖2 )π‘₯ 2 βˆ’ π‘Ž2 𝛽𝑖2 βˆ’ 𝑑𝛼𝑖2 π‘₯ 4 π‘Ž(π‘₯ 2 βˆ’ π‘Žπ›½π‘–2 ) βˆ’ 𝑑(𝛼𝑖2 π‘₯ 4 βˆ’ 𝛼𝑖2 𝛽𝑖2 π‘₯ 2 )
        =                                            =                                            =
                          π‘Ž βˆ’ 𝑑π‘₯ 2                                      π‘Ž βˆ’ 𝑑π‘₯ 2
                                           2       2             2 2
                                        (π‘₯ βˆ’ π‘Žπ›½π‘– )(π‘Ž βˆ’ 𝑑𝛼𝑖 π‘₯ )
                                      =                               .
                                                  π‘Ž βˆ’ 𝑑π‘₯ 2
    Similarly, we transform the factors of the common denominator of coordinates π‘₯β€² and 𝑦′ into (9):
                         2            2 2 2 2
                                                 1 βˆ’ π‘₯2       π‘Ž βˆ’ 𝑑π‘₯ 2 βˆ’ 𝑑2 𝛼𝑖2 𝛽𝑖2 π‘₯ 2 + 𝑑2 𝛼𝑖2 𝛽𝑖2 π‘₯ 4
  𝑍𝑖 = 1 βˆ’ (𝑑𝛼𝑖 𝛽𝑖 π‘₯𝑦) = 1 βˆ’ 𝑑 𝛼𝑖 𝛽𝑖 π‘₯                    =                                              =
                                                π‘Ž βˆ’ 𝑑π‘₯ 2                       π‘Ž βˆ’ 𝑑π‘₯ 2
                            2        2 2       2 2 2 4                  2 2            2 2
                  π‘Ž βˆ’ 𝑑(𝛼𝑖 + π‘Žπ›½π‘– )π‘₯ + 𝑑 𝛼𝑖 𝛽𝑖 π‘₯              (1 βˆ’ π‘Žπ›½π‘– π‘₯ )(π‘Ž βˆ’ 𝑑𝛼𝑖 π‘₯ )
               =                         2
                                                          =                                .
                                 π‘Ž βˆ’ 𝑑π‘₯                                  π‘Ž βˆ’ 𝑑π‘₯ 2
    After reducing the common factors for the π‘₯β€²-coordinate, we obtain
                               π‘ˆπ‘– (𝛼𝑖 π‘₯)2 βˆ’ π‘Ž2 (𝛽𝑖 𝑦)2           π‘₯ 2 βˆ’ π‘Žπ›½π‘–2
                                   =                        =                .
                               𝑍𝑖      1 βˆ’ (𝑑𝛼𝑖 𝛽𝑖 π‘₯𝑦)2         1 βˆ’ 𝑑𝛽𝑖2 π‘₯ 2
    The second coordinate 𝑦′ in (9) as factors in the numerator
                                                 𝛼𝑖2 (1 βˆ’ π‘₯ 2 ) βˆ’ 𝛽𝑖2 π‘₯ 2 (π‘Ž βˆ’ 𝑑π‘₯ 2 )
                    𝑉𝑖 = (𝛼𝑖 𝑦)2 βˆ’ (𝛽𝑖 π‘₯)2 =                                          =
                                                              π‘Ž βˆ’ 𝑑π‘₯ 2
             βˆ’π‘₯ 2 (𝛼𝑖2 + π‘Žπ›½π‘–2 π‘₯ 2 ) + 𝑑𝛽𝑖2 π‘₯ 4 + 𝛼𝑖2 βˆ’π‘₯ 2 (1 + 𝑑𝛼𝑖2 𝛽𝑖2 ) + 𝑑𝛽𝑖2 π‘₯ 4 + 𝛼𝑖2
          =                                           =                                        =
                            π‘Ž βˆ’ 𝑑π‘₯ 2                                    π‘Ž βˆ’ 𝑑π‘₯ 2
                                                    2            2
                                        βˆ’(π‘₯ 2 βˆ’ 𝛼𝑖 )(1 βˆ’ 𝑑𝛽𝑖 π‘₯ 2 )
                                      =                               .
                                                  π‘Ž βˆ’ 𝑑π‘₯ 2
    Then
                               𝑉𝑖 (𝛼𝑖 𝑦)2 βˆ’ (𝛽𝑖 π‘₯)2              π‘₯ 2 βˆ’ 𝛼𝑖2
                                   =                     =βˆ’                  .
                               𝑍𝑖 1 βˆ’ (𝑑𝛼𝑖 𝛽𝑖 π‘₯𝑦)2             π‘Ž βˆ’ 𝑑𝛼𝑖2 π‘₯ 2
    As a result, the function (9) can be written in the equivalent form (10)
                                       𝑆                     𝑆
                                π‘₯     π‘₯ 2 βˆ’ π‘Žπ›½π‘–2 βˆ’π‘¦     π‘₯ 2 βˆ’ 𝛼𝑖2
                     πœ™(π‘₯, 𝑦) = ( 2 ∏             ,   ∏              ),
                                𝐴    1 βˆ’ 𝑑𝛽𝑖2 π‘₯ 2 𝐴2   π‘Ž βˆ’ 𝑑𝛼𝑖2 π‘₯ 2
                                      𝑖=1                   𝑖=1
depending on the parameters π‘Ž and 𝑑. Formulas for the parameters of the isogenic curve
π‘Žβ€² = π‘Žπ‘™ , 𝑑′ = 𝐴8 𝑑𝑙 , 𝐴 = βˆπ‘ π‘–=1 𝛼𝑖 are proved in Theorem 3 [5]. The theorem is proved.
    For the curves 𝐸𝑑 with the parameter π‘Ž = 1 the formula (10) is given in Theorem 4 [5]. In this paper,
we generalized it for the curves πΈπ‘Ž,𝑑 (1) with the arbitrary value π‘Ž β‰  𝑑, which allows us to compute the
isogenies of twisted Edwards curves (πœ’(π‘Ž) = πœ’(𝑑) = βˆ’1).
    Let’s note that the rational function (10) corresponds to the classical form (7). Its obvious advantage
over (9) is its simplicity and minimal computational complexity in affine coordinates. Also, the degree
of isogeny as the maximum degree of the polynomial 𝑝(π‘₯) in (7) is immediately defined as 𝑙 = 2𝑠 + 1.
The form (1) of the Edwards curve with the addition law (4) adapts it to the definitions of isogenies in
the Weierstrass form [17].
    Let’s consider examples of isogenies of the supersingular twisted Edwards curve (STEC). Such
curves exist only at 𝑝 ≑ βˆ’1 mod 8 and have the order 𝑁𝐸 = 𝑝 + 1 ∈ {8𝑛, 16𝑛, … } (𝑛 is odd). The
curve, for example, contains kernels of the 3rd and 5th order at the smallest value 𝑛 = 15, then the
minimum prime number is 𝑝 = 239 and the order of such curve is 𝑁𝐸 = 16𝑛 = 240. The parameters
of the entire family of 118 twisted Edwards curves can be taken as quadratic non-residues




                                                                                                            5
π‘Ž = βˆ’1, 𝑑 = βˆ’π‘š2 mod 𝑝, π‘š = 2. .119. Of these, there are 30 STEC’s with the parameters (βˆ’π‘‘), given
in Table 1. They are written as squares in ascending order π‘š = 5. .118.
Table 1
Values of the parameter (βˆ’π’…) of STEC at 𝒑 = πŸπŸ‘πŸ—, 𝒂 = βˆ’πŸ, and 𝑡𝑬 = πŸπŸ’πŸŽ
   25         64      121       196       50      183       4       10                    87       176
   24        153       11       110       48      187      120     193                    27       160
   213        44        2       201       61       3       206     192                    80       62
                                        (0)
    For the first curve (1) πΈβˆ’1,βˆ’25 = πΈπ‘Ž,𝑑 from Table 1 we can construct 3- and 5-isogenies and find
                              (𝑖)                                   (πœ‹)   (0)
chains of isogenic curves πΈβˆ’1,𝑑𝑖 , 𝑖 = 1,2, …, πœ‹ such that πΈπ‘Ž,𝑑 = πΈπ‘Ž,𝑑 . Then the chain of mappings
                                              (1)         (0)
πœ™ (2) ∘ πœ™ (3) ∘ … ∘ πœ™ (πœ‹) gives dual isogeny πΈβˆ’1,𝑑1 β†’ πΈβˆ’1,𝑑0. The parameter π‘Ž of all isogenic STEC’s can
                                                                                                   𝑙
be fixed as the quadratic non-residue π‘Ž = βˆ’1, since according to Theorem 3 [5] π‘Ž(𝑖+1) = (π‘Ž(𝑖) ) = βˆ’1
for all odd degrees 𝑙.
    The curve πΈβˆ’1,βˆ’25 contains the point of the 3rd order 𝑄1 = (149,64), then according to
Theorem 3[5], 𝐴 = βˆπ‘ π‘˜=1 π›Όπ‘˜ = 149, 𝐴8 = 8, 𝑑(1) = 𝐴8 (𝑑(0) )3 = 𝐴8 𝑑3 = βˆ’3. The calculated
parameters βˆ’π‘‘(𝑖) , 𝐽(𝑑(𝑖) ) of the chain of 3-isogenous curves with the starting value 𝑑 = βˆ’25 are given
in Table 2. The period of the chain πœ‹ = 5 divides the number of all STEC’s equal to 30. Specifying the
starting value 𝑑 = βˆ’2 not included in Table 2, it is possible to obtain a different sequence βˆ’π‘‘(𝑖) ∈
{2,61,62,193,5,2} with period 5 with the elements from Table 1 for all STEC’s. For 3-isogenies, we
can calculate 6 tables similar to Table 2 with disjoint values βˆ’π‘‘(𝑖) Table 1. We note that all J-invariants
𝐽(𝑑(𝑖) ) of adjacent 3-isogenic curves (except the last pair) are different, i.e. they are not isomorphic.
However, inside the chain, there may be isomorphic curves with equal J-invariants.
    The kernel of 5-isogeny on the curve πΈβˆ’1,βˆ’25 is the subgroup of points of the 5th order
±𝑄1 = (𝛼1 , ±𝛽1 ) = (βˆ’95, Β±28), ±𝑄2 = (𝛼2 , ±𝛽2 ) = (βˆ’72, Β±119), and 5𝑄1 = 𝑂 = (1,0). It is
uniquely determined by the coordinates 𝛼1 , 𝛼2 of two points and equation (1). Then for each 5-isogenic
                              (𝑖) (𝑖)                 8         5
curve, we compute 𝐴(𝑖) = 𝛼1 𝛼2 , 𝑑(𝑖+1) = (𝐴(𝑖) ) (𝑑(𝑖) ) , 𝑖 = 0,1, …. The results of calculations of
the parameters of the chain of 5-isogenic curves are given in Table 3. The period of this chain is πœ‹ =
15, so we can build one more similar table (up to a cyclic shift) with the other half of the parameters of
Table 1. The data of Tables 2 and 3 are used to construct the graphs of isogenies.
Table 2
Values of parameters of the chain of 3-isogenous SSCE at 𝒑 = πŸπŸ‘πŸ—, 𝒂 = βˆ’πŸ
        𝑖              0                 1            2           3                  4             5
        (𝑖)           149              227          152         174                 179           149
      𝛼
     βˆ’π‘‘(𝑖)            25                 3           10          50                 110           25
         (𝑖)          225              105           55         105                 225           225
    𝐽(𝑑 )

Table 3
Values of parameters of the chain of 5-isogenous SSCE at 𝒑 = πŸπŸ‘πŸ—, 𝒂 = βˆ’πŸ
      𝑖         0          1          2         3         4         5     6   7
   (𝑖) (𝑖)   –95,–72 69, βˆ’53 57,–8 103, βˆ’102 107,– 118, βˆ’55 25, βˆ’18 41, βˆ’52
 𝛼1 , 𝛼2
                                                          34
   βˆ’π‘‘(𝑖)       25          2         11        50        193       187    3  61
  𝐽(𝑑(𝑖) )     225        218       215        105       215       215   105 235
      𝑖         8          9         10        11         12       13    14  15
   (𝑖) (𝑖) 103, βˆ’88 79, βˆ’91 51,108 βˆ’68, βˆ’43 13, βˆ’ βˆ’103, βˆ’46 βˆ’48,βˆ’8 βˆ’93, βˆ’72
 𝛼1 , 𝛼2
   βˆ’π‘‘(𝑖)       183        110         5        121        10       62    201 25
       (𝑖)     218        225       113        327        55       217   113 225
  𝐽(𝑑 )




6
    The mapping (10) of the points 𝑃 = (π‘₯, 𝑦) of the curve πΈβˆ’1,βˆ’25 with the kernel
𝐺 = {(1,0), ±𝑄1 = (βˆ’90, Β±64)} of 3-isogeny has the form:
                                      π‘₯ π‘₯ 2 βˆ’ π‘Žπ›½ 2 βˆ’π‘¦ π‘₯ 2 βˆ’ 𝛼 2
                        πœ™3 (π‘₯, 𝑦) = ( 2             ,                )=
                                     𝐴 1 βˆ’ 𝑑𝛽 2 π‘₯ 2 𝐴2 1 + 𝑑𝛼 2 π‘₯ 2
                               π‘₯    π‘₯ 2 + 642     βˆ’π‘¦ π‘₯ 2 βˆ’ 902
                          =( 2                  ,                     ).
                              90 1 + 252 642 π‘₯ 2 902 1 βˆ’ 252 902 π‘₯ 2
    The point of the maximum 120th order 𝑃 = (3,75) of the curve πΈβˆ’1,βˆ’25 is mapped by this function
to the point 𝑃′ = (βˆ’116,94) of the 40th order of the curve πΈβ€²βˆ’1,βˆ’3 . The point of the maximum odd 15th
order 𝑃 = (βˆ’44, βˆ’12) is mapped to the point 𝑃′ = (βˆ’18, βˆ’114) of the 5th order, the point 𝑃 =
(βˆ’95,28) of the 5th order is mapped to the point 𝑃′ = (25, βˆ’66) of the 5th order, and the point 𝑃 =
(βˆ’90,64) of the 3rd order is mapped to the point 𝑃′ = (1,0) = 𝑂. As we can see, the function πœ™3 (π‘₯, 𝑦)
reduces by 3 times the orders of the domain points, which are multiples of 3, and does not change the
orders of the other points. In subgroups of orders multiples of 3, three points are mapped into one
(surjection property).
    For the same curve πΈβˆ’1,βˆ’25 with the kernel of the 5th order 𝐺 = {(1,0), ±𝑄1 = (βˆ’95, Β±28), ±𝑄2 =
(βˆ’72, Β±119)} 5-isogeny in the form (10) is written as
                    π‘₯    π‘₯ 2 + 282      π‘₯ 2 βˆ’ 952      βˆ’π‘¦ π‘₯ 2 + 1192          π‘₯ 2 βˆ’ 722
         (π‘₯,
     πœ™5 𝑦) = ( 2                                     ,                                   ),
                   𝐴 1 + 252 642 π‘₯ 2 1 βˆ’ 252 952 π‘₯ 2 𝐴2 1 + 252 1192 π‘₯ 2 1 βˆ’ 252 722 π‘₯ 2
                                    𝐴2 = (95 βˆ™ 72)2 = 155.
                        th
    The point of the 120 order 𝑃 = (3,75) of the curve πΈβˆ’1,βˆ’25 is mapped by this function to the point
𝑃′ = (βˆ’116,94) of the 24th order of the curve πΈβ€²βˆ’1,βˆ’2 . The point 𝑃 = (8, βˆ’16) of the 30th order is
mapped to the point 𝑃′ = (18, βˆ’7) of the 6th order. The point of the 5th order 𝑃 = (βˆ’95,28) is mapped
to the point 𝑃′ = (1,0) = 𝑂, Here, too, in subgroups of orders multiples of 5, five points are mapped
into one.

4. The Computing of Isogenies on Supersingular Twisted Edwards Curves
   in Projective Coordinates of Farashai-Hosseini
        Significant progress has been made in the efficiency of computing odd degree isogenies on
Edwards curves in paper [3]. It is based on the idea of the method of differential addition (i.e., the
addition of two points with the known difference) on the curve in the Farashahi-Hosseini projective
coordinates [4]. Since the formulae of isogenies [5] contain the coordinates of the point pairs 𝑃 Β± 𝑄 as
multipliers, it is possible to obtain results for the isogenies similar to the results of the differential
addition on the curve.
        In the paper [3], Theorem 1 was proved, which determines the odd degree isogenic mapping from
Edwards curve 𝐸𝑑 to the curve 𝐸′𝑑 in Farashahi-Hosseini coordinates 𝑀(π‘₯, 𝑦) = 𝑑π‘₯ 2 𝑦 2 (or 𝑀(π‘₯, 𝑦) =
π‘₯ 2 /𝑦 2). As in the paper [5], it is proved only for the Edwards curve 𝐸𝑑 (π‘Ž = 1), and it is not known
whether its results are applicable in the class of twisted Edwards curves πΈπ‘Ž,𝑑 (πœ’(π‘Ž) = πœ’(𝑑) = βˆ’1).
Below we prove this theorem for all curves in the generalized form πΈπ‘Ž,𝑑 (1). Instead of the formula (8)
taken as a basis in [3], we proceed from the more laconic formula (10) obtained above in Theorem 1.
    Theorem 2. Let 𝐺 = {(1,0), ±𝑄1 , ±𝑄2 , … , ±𝑄𝑠 } is the subgroup of odd order 𝑙 = 2𝑠 + 1 of the
points ±𝑄𝑖 = (𝛼𝑖 , ±𝛽𝑖 ) of the curve πΈπ‘Ž,𝑑 (1). Let 𝑀𝑖 = 𝑑𝛼𝑖2 𝛽𝑖2 , 𝑀 = 𝑑π‘₯ 2 𝑦 2 , 𝑃 = (π‘₯, 𝑦) ∈ πΈπ‘Ž,𝑑 . Then
𝑀(πœ™(π‘₯, 𝑦)) = 𝑀(π‘₯β€², 𝑦′) is 𝑙-isogeny with the kernel 𝐺 from the curve πΈπ‘Ž,𝑑 to the curve πΈβ€²π‘Žβ€² ,𝑑′ with the
parameters π‘Žβ€² = π‘Žπ‘™ , 𝑑′ = 𝐴8 𝑑𝑙 , 𝐴 = βˆπ‘ π‘–=1 𝛼𝑖 , and the mapping function
                                                  𝑠
                                                      (𝑀 βˆ’ 𝑀𝑖 )2
                                      𝑀(πœ™) = 𝑀 ∏                 .                                     (11)
                                                     (1 βˆ’ 𝑀𝑀𝑖 )2
                                               𝑖=1
   Proof. From the equation of the curve (1) we have 𝛼𝑖2 + π‘Žπ›½π‘–2 = 1 + 𝑑𝛼𝑖2 π‘Žπ›½π‘–2. The factors under the
sign of the product of isogeny (10) have the form:
                           π‘ˆπ‘–     π‘₯ 2 βˆ’ π‘Žπ›½π‘–2       𝑉𝑖    π‘₯ 2 βˆ’ 𝛼𝑖2
                               =             ,        =             .
                           𝑍π‘₯𝑖 1 βˆ’ 𝑑𝛽𝑖2 π‘₯ 2        𝑍𝑦𝑖 π‘Ž βˆ’ 𝑑𝛼𝑖2 π‘₯ 2


                                                                                                         7
    Let’s denote the products of numerators and denominators:
       𝑆𝑖 = π‘ˆπ‘– 𝑉𝑖 = (π‘₯ 2 βˆ’ π‘Žπ›½π‘–2 )(π‘₯ 2 βˆ’ 𝛼𝑖2 ),    𝑅𝑖 = 𝑍π‘₯𝑖 𝑍𝑦𝑖 = (1 βˆ’ 𝑑𝛽𝑖2 π‘₯ 2 )(π‘Ž βˆ’ 𝑑𝛼𝑖2 π‘₯ 2 ).
    Then, taking into account 𝛼𝑖2 + π‘Žπ›½π‘–2 = 1 + 𝑀𝑖 and the multiplying of these equations by 𝑑𝑦 2 and
𝑦 2 respectively, we obtain:
   𝑆𝑖 𝑑𝑦 2 = βˆ’π‘ˆπ‘– 𝑉𝑖 = 𝑑𝑦 2 (π‘₯ 2 βˆ’ π‘Žπ›½π‘–2 )(π‘₯ 2 βˆ’ 𝛼𝑖2 ) = π‘₯ 2 𝑀 βˆ’ π‘₯ 2 𝑦 2 (𝛼𝑖2 + π‘Žπ›½π‘–2 ) + π‘Žπ‘‘π›Όπ‘–2 𝛽𝑖2 𝑦 2 =
                                   = 𝑀π‘₯ 2 βˆ’ 𝑀(1 + 𝑀𝑖 ) + π‘Žπ‘€π‘– 𝑦 2 ,
         𝑅𝑖 𝑦 = 𝑦 (1 βˆ’ 𝑑𝛽𝑖 π‘₯ )(π‘Ž βˆ’ 𝑑𝛼𝑖2 π‘₯ 2 ) = π‘Žπ‘¦ 2 βˆ’ 𝑑π‘₯ 2 𝑦 2 (𝛼𝑖2 + π‘Žπ›½π‘–2 ) + 𝑀𝑀𝑖 π‘₯ 2 =
              2     2          2 2

                                   = π‘Žπ‘¦ 2 βˆ’ 𝑀(1 + 𝑀𝑖 ) + 𝑀𝑀𝑖 π‘₯ 2 .
    Substitution in the last equations π‘Žπ‘¦ 2 = 1 + 𝑀 βˆ’ π‘₯ 2 gives:
         𝑆𝑖 𝑑𝑦 2 = 𝑀π‘₯ 2 βˆ’ 𝑀(1 + 𝑀𝑖 ) + π‘Žπ‘€π‘– 𝑦 2 = 𝑀π‘₯ 2 βˆ’ 𝑀 βˆ’ 𝑀𝑀𝑖 + 𝑀𝑖 (1 + 𝑀 βˆ’ π‘₯ 2 ) =
                                       = (𝑀 βˆ’ 𝑀𝑖 )(π‘₯ 2 βˆ’ 1),
          𝑅𝑖 𝑦 2 = π‘Žπ‘¦ 2 βˆ’ 𝑀(1 + 𝑀𝑖 ) + 𝑀𝑀𝑖 π‘₯ 2 = (1 + 𝑀 βˆ’ π‘₯ 2 ) βˆ’ 𝑀 βˆ’ 𝑀𝑀𝑖 + 𝑀𝑀𝑖 π‘₯ 2 =
                                       = (1 βˆ’ 𝑀𝑀𝑖 )(π‘₯ 2 βˆ’ 1).
Then
                                    𝑠                     𝑠                    𝑠
         𝑆𝑖      βˆ’1
                    𝑀 βˆ’ 𝑀𝑖               𝑆𝑖 2       βˆ’2𝑠
                                                              𝑆𝑖 2        βˆ’2𝑠
                                                                                   (𝑀 βˆ’ 𝑀𝑖 )2
            =𝑑               ,     ∏(        ) =𝑑       ∏( ) = 𝑑              ∏                 .
         𝑅𝑖         1 βˆ’ 𝑀𝑀𝑖             𝑑𝑅𝑖                   𝑅𝑖                  (1 βˆ’ 𝑀𝑀𝑖 )2
                                    𝑖=1               𝑖=1                 𝑖=1
    As a result, for 𝑙-isogeny (10) taking into account the value of the parameter of the isogenic curve
                                                                    𝑠
                                  β€² (π‘₯ β€²   β€² )2   8 2𝑠+1 2 2 βˆ’8
                                                                         𝑆𝑖 2
                   𝑀(πœ™(𝑃)) = 𝑑           βˆ™π‘¦ =𝐴 𝑑         π‘₯ 𝑦 𝐴 ∏(           ) =
                                                                        𝑑𝑅𝑖
                                                                  𝑖=1
                                           𝑠           𝑠
                               2 2
                                     𝑆𝑖 2      (𝑀 βˆ’ 𝑀𝑖 )2
                           = 𝑑π‘₯ 𝑦 ∏ ( ) = 𝑀 ∏             .
                                     𝑅𝑖       (1 βˆ’ 𝑀𝑀𝑖 )2
                                          𝑖=1         𝑖=1
    The theorem is proved.
    We emphasize that isogeny (11) for w-coordinate πΈπ‘Ž,𝑑 (1) does not depend on the parameter a and
is equally valid for quadratic and twisted Edwards curves forming quadratic twist pairs [10]. In other
words, function (11) maps the curve points of one of these two classes to the curve points of the same
class.
    Let’s take as an example the 3-isogeny of the twisted curve πΈβˆ’1,βˆ’25 of the previous section and its
point 𝑃 = (3,75) of the 120th order. For it, we will receive the coordinate 𝑀 = βˆ’25 βˆ™ 32 βˆ™ 752 = 119.
For the kernel point 𝑄1 = (149,64), respectively, 𝑀1 = βˆ’25 βˆ™ 1492 βˆ™ 642 = βˆ’60. According to the
formula (11) 𝑀(πœ™(𝑃)) = 78 the point of the isogenic curve πΈβ€²βˆ’1,βˆ’3 , calculated by the formula (10), is
the point of the 40th order 𝑃′ = (βˆ’116,94). For it, the coordinate 𝑀(πœ™(𝑃)) = 𝑑 β€² (π‘₯′𝑦′)2 = 78 coincides
with the calculations by formula (11).
    Let’s now turn to the quadratic curve 𝐸1,25 = 𝐸25 as a pair of quadratic torsion of the curve πΈβˆ’1,βˆ’25 .
All points of this pair of curves have different coordinates (except for the points (Β±1,0)) and,
accordingly, the curve 𝐸25 has the different kernel of the 3rd degree 𝐺 = {(1,0), ±𝑄1 = (97, Β±14)}.
Characteristically, the parameter of the isogenic curve 𝑑′ = 𝑑(1) = 𝐴8 𝑑(0)3 = 978 253 = 110 also
changes. For the curve 𝐸25 the mapping (10) of the point of the 120th 𝑃 = (20,108) is the point of the
40th order 𝑃′ = (βˆ’16,57) on the isogenic curve 𝐸110 . For point 𝑃 we obtain the coordinate 𝑀 = 25 βˆ™
202 βˆ™ 1082 = 113, for the point of the kernel 𝑀1 = 25 βˆ™ 972 βˆ™ 142 = 44, respectively. According to
formula (11) 𝑀(πœ™(𝑃)) = 100. For the point 𝑃′ = (βˆ’24,57) the w-coordinate is 𝑀(πœ™(𝑃)) = 110 βˆ™
242 βˆ™ 572 = 100. This corresponds to Theorem 2 and the formula (11).
    The implementation of computing of isogenies (11) is given in [3]. To calculate the parameters 𝑑(𝑖)
of the chain of isogenies in projective coordinates, an additional parameter 𝐢 is introduced into the
equation of the curve 𝐸𝑑 (2) or (3). For STEC (1) at 𝑝 ≑ 3 mod 4, we accept π‘Ž = βˆ’1,
𝑑(𝑖=π‘š) = βˆ’π‘š2 mod 𝑝, π‘š ∈ {2. . (𝑝 βˆ’ 1)/2} and define the curve:
                       𝐸𝐢,𝐷 : 𝐢π‘₯ 2 βˆ’ 𝐢𝑦 2 = 𝐢 + 𝐷π‘₯ 2 𝑦 2 , 𝐷 = 𝑑𝐢, πœ’(𝑑) = βˆ’1.                      (12)
    To calculate the parameter 𝑑′ of the isogenic curve the formula [5] is used:
                                                 𝑠
                                β€²         8 𝑙
                              𝑑 = 𝐴 𝑑 , 𝐴 = ∏ 𝛼𝑖 , 𝑙 = 2𝑠 + 1.                                      (13)
                                                𝑖=1



8
   To express the parameter 𝐴 with the replacement 𝛼𝑖 β†’ 𝑀𝑖 in [3], the idea of doubling the kernel
points is proposed, which does not change the points of subgroups 𝐺 of odd order 𝑙. From the law of
doubling (5) we have
                                                𝛼𝑖2 βˆ’ π‘Žπ›½π‘–2     2𝛼𝑖 𝛽𝑖
                               2(𝛼𝑖 , 𝛽𝑖 ) = (         2 2,             ).
                                               1 βˆ’ 𝑑𝛼𝑖 𝛽𝑖 1 + 𝑑𝛼𝑖2 𝛽𝑖2
   By squaring the second coordinate, we obtain
             2𝛼𝑖 𝛽𝑖 2       4π‘‘βˆ’1 𝑀𝑖         2         2 2
                                                              4π›Όπ‘˜2 𝑀𝑖        2   (1 + 𝑀𝑖 )2
           (        ) =                =  𝛽 π‘˜ ⟹    𝑑𝛼  𝛽
                                                      π‘˜ π‘˜ =             ⟹  𝛼𝑖  =            .
            1 + 𝑀𝑖         (1 + 𝑀𝑖 )2                        (1 + 𝑀𝑖 )2              4
   Here we take into account that for each point of the kernel (𝛼𝑖 , 𝛽𝑖 ) there exists the reverse point
(𝛼𝑖 , βˆ’π›½π‘– ) = (π›Όπ‘˜ , π›½π‘˜ ) and for such a pair of points 𝑀𝑖 = π‘€π‘˜ . Then the formula (13) takes the form
                                                    𝑠
                                                 𝑙
                                                       (1 + 𝑀𝑖 )8
                                         𝑑′ = 𝑑 ∏                 .                                 (14)
                                                           44
                                                 𝑖=1
   Transition to projective coordinates (π‘Š: 𝑍) allows avoiding inversions in the formula (11), thus for
the curve (12)
                                  𝑠                               𝑠

                   π‘Š = π‘Š ∏(π‘Šπ‘π‘– βˆ’ π‘Šπ‘– 𝑍) , 𝑍 = 𝑍 ∏(𝑍𝑍𝑖 βˆ’ π‘Šπ‘Šπ‘– )2 .
                      β€²                           2      β€²

                                 𝑖=1                             𝑖=1
   Here 4𝑠𝑀 + 2𝑀 + 2𝑆 operations in the field are performed for every 𝑠 (𝑀 is multiplication, 𝑆 is
squaring). If we enter intermediate formulas:
                                    𝐻 = (π‘Š + 𝑍)(π‘Šπ‘– βˆ’ 𝑍𝑖 ),
                                    𝐽 = (π‘Š βˆ’ 𝑍)(π‘Šπ‘– + 𝑍𝑖 ),
then
                                            𝑠                         𝑠
                            β€²                    )2
                          2π‘Š = π‘Š ∏(𝐻𝑖 βˆ’ 𝐽𝑖             , 2𝑍 = 𝑍 ∏(𝐻𝑖 βˆ’ 𝐽𝑖 )2 .
                                                             β€²

                                           𝑖=1                       𝑖=1
   And we need only 4𝑠𝑀 + 2𝑆 operations when calculating one isogeny (11).
   For calculating the parameter 𝑑′ = 𝐷′⁄ of the isogenic curve (12) in projective coordinates
                                         𝐢′
according to (14) we obtain
                                            𝑠                          𝑠

                           𝐷 = 𝐷 ∏(𝑍𝑖 + π‘Šπ‘– , 𝐢 = 𝐢 ∏(2𝑍𝑖 )8 .
                             β€²         𝑙           )8        β€²   𝑙

                                           𝑖=1                        𝑖=1
    Therefore, for the small degrees of isogenies 𝑙 = 2𝑠 + 1 ≀ 9(𝑠 ≀ 4) in each of these formulae, it is
enough to perform three squares, within which to substitute values 𝐷 and 𝐢 at different steps. Herewith
the cost of calculations is determined by the linear function 2(𝑠 + 1)𝑀 + 6𝑆. With the degree of
isogeny 𝑙 > 9 additional operations 𝑀 and 𝑆 are required, the number of which needs an estimate.
    Within the limits of the linear trend (lower value), the total cost of calculating 𝑙-isogenies in
Farashahi-Hosseini coordinates [3] is 2(3𝑠 + 1)𝑀 + 8𝑆 (the formula of the trend is obtained in this
paper). For example, at 𝑙 = 3 and 𝑙 = 5 we obtain 8𝑀 + 8𝑆 and 14𝑀 + 8𝑆, respectively. The
calculation of these isogenies in the classical projective coordinates (𝑋: 𝑍) gives in [6] the best result
6𝑀 + 5𝑆 for 3-isogenies and the worst result 21𝑀 + 12𝑆 for 5-isogenies in [7]. With the growth of 𝑙
the calculations of isogenies in coordinates (𝑋: 𝑍) become significantly more complicated.
    Supersingular twisted and quadratic Edwards curves with the same order 𝑁𝐸 = 𝑝 + 1, as follows
from Sect. 2, have different structures [10]. STEC has only two exceptional points of the 2nd order, and
in the class of quadratic curves, two more exceptional points of the 4th order (6) are added to them as
well as two non- exceptional points of the 4th order ±𝐹0 = (0, ±1). It is easier to bypass exceptional
points in the STEC class, which makes them preferable to quadratic ones in cryptosystems.
    For this purpose, for STEC with the order of the curve with the minimum even cofactor 𝑁𝐸 = 8𝑛 (𝑛
is odd), it is sufficient to select the base point 𝑃 of Alice and Bob as the point of order n or maximum
order 4𝑛 at the stage of selection of system-wide parameters. In the latter case, quadratic curves are not
acceptable, because the mappings to points of the 4th order are possible. At the same time, it is clear for
the twisted Edwards curves that with such a choice no chain of isogenies of odd degrees will generate
the point of the 2nd order. Another solution to this problem may be to select the point 𝑃 from one of the



                                                                                                         9
three subgroups of the curve that don`t contain exceptional points (they are replaced in this subgroup
by the ordinary point of the 2nd order 𝐷0 = (βˆ’1,0)).
   The results of the implementation of the Edwards-CSIDH model [3] in projective coordinates
(π‘Š: 𝑍) claim that it is 20% faster than the Montgomery-CSIDH model in coordinates (𝑋: 𝑍). It should
be noted that this model is built on complete Edwards curves with the order 𝑁𝐸 = 𝑝 + 1 = 4𝑛 With the
same success, it can be realized on supersingular twisted Edwards curves with the order 𝑁𝐸 = 8𝑛.
Quadratic Edwards curves with redundant exceptional points of the 4th order are outside the
recommended range.

5. Conclusions
   It can be concluded that the method of computing odd degree isogenies in coordinates (π‘Š: 𝑍),
proposed in [3], using supersingular complete and twisted Edwards curves, allows implementing the
fastest calculations to date in the construction of the PQC of protocol CSIDH and the like. The theorems
proved in this paper open a class of twisted Edwards curves for their implementation.

6. References
[1] D. Jao, L. de Feo, Towards quantum-resistant cryptosystems from supersingular elliptic curve
     isogenies, Lect. Notes Comput. Sci. 7071 (2011) 19–34. doi:10.1007/978-3-642-25405-5_2.
[2] W. Castryck, et al., CSIDH: An efficient post-quantum commutative group action, in: Advances
     in Cryptologyβ€”ASIACRYPT, 2018, pp. 395–427. doi:10.1007/978-3-030-03332-3_15.
[3] S. Kim, et al., Optimized method for computing odd-degree isogenies on Edwards curves, in:
     Advances in Cryptologyβ€”ASIACRYPT, 2019, pp. 273–292. doi:10.1007/978-3-030-34621-
     8_10.
[4] R. R. Farashahi, S. G. Hosseini, Differential addition on twisted Edwards curves, Lect. Notes
     Comput. Sci. 10343 (2017) 366–378. doi:10.1007/978-3-319-59870-3_21.
[5] D. Moody, D. Shumow, Analogues of Velu’s formulas for isogenies on alternate models of elliptic
     curves, Math. Computation 85(300) (2015) 1929–1951. doi:10.1090/mcom/3036.
[6] S. Kim, et al., Efficient isogeny computations on twisted Edwards curves, Secur. Commun. Netw.
     2018 (2018). 1–11. doi:10.1155/2018/5747642.
[7] A. Bessalov, V. Sokolov, P. Skladannyi, Modeling of 3- and 5-isogenies of supersingular Edwards
     curves, in: Proceedings of the 2nd International Workshop on Modern Machine Learning
     Technologies and Data Science, June 2–3, 2020, no. I, vol. 2631, pp. 30–39.
[8] D. J. Bernstein, T. Lange, Faster addition and doubling on elliptic curves, Lect. Notes Comput.
     Sci. 4833 (2007) 29–50. doi:10.1007/978-3-540-76900-2_3.
[9] D. J. Bernstein, et al., Twisted Edwards curves, Lect. Notes Comput. Sci. 5023 (2008) 389–405.
     doi:10.1007/978-3-540-68164-9_26.
[10] A. V. Bessalov, Edwards Elliptic Curves and Cryptography, 2017. [Publication in Russian].
[11] A. V. Bessalov, L. V. Kovalchuk, Supersingular twisted Edwards curves over prime fields. II.
     Supersingular twisted Edwards curves with the J-invariant equal to 663, Cybern. Syst. Anal. 55(5)
     (2019) 731–741. doi:10.1007/s10559-019-00183-y.
[12] A. V. Bessalov, O. V. Tsygankova, Number of curves in the generalized Edwards form with
     minimal even cofactor of the curve order, Probl. Inf. Transm. 53(1) (2017) 92–101.
     doi:10.1134/s0032946017010082. [Publication in Russian].
[13] A. V. Bessalov, O. V. Tsygankova, Interrelation of families of points of high order on the Edwards
     curve over a prime field, Probl. Inf. Transm. 51(4) (2015) 391–397.
     doi:10.1134/s0032946015040080. [Publication in Russian].
[14] A. Bessalov, et al., Analysis of 2-isogeny properties of generalized form Edwards curves, in:
     Proceedings of the Workshop on Cybersecurity Providing in Information and Telecommunication
     Systems, July 7, 2020, vol. 2746, pp. 1–13.




10
[15] A. V. Bessalov, Calculation of parameters of cryptic curves Edwards over the fields of 5 th and 7th
     characteristic, Cybersecur. Educ. Sci. Tech. 1 (2018) 94–104. doi:10.28925/2663-4023.2018.
     1.94104. [Publication in Ukrainian].
[16] A. Bessalov, et al., 3- and 5-isogenies of supersingular Edwards curves, Cybersecur. Educ. Sci.
     Tech. 4(8) (2020) 6–21. doi:10.28925/2663-4023.2020.8.621.
[17] L. Washington, Elliptic Curves. Discrete Mathematics and Its Applications, 2008.
     doi:10.1201/9781420071474.




                                                                                                     11