arXiv reader

Finding compositional inverses of permutations from the AGW criterion

Tailin Niu, Kangquan Li, Longjiang Qu and Qiang Wang

arXiv
2004.12552v2 · PDF · 1 April 2021
Categories
cs.IT, math.IT, math.NT
DOI
10.1109/TIT.2021.3089145
Licence
CC BY SA 4.0 The site shows the text.

Abstract

Permutation polynomials and their compositional inverses have wide applications in cryptography, coding theory, and combinatorial designs. Motivated by several previous results on finding compositional inverses of permutation polynomials of different forms, we propose a general method for finding these inverses of permutation polynomials constructed by the AGW criterion. As a result, we have reduced the problem of finding the compositional inverse of such a permutation polynomial over a finite field to that of finding the inverse of a bijection over a smaller set. We demonstrate our method by interpreting several recent known results, as well as by providing new explicit results on more classes of permutation polynomials in different types. In addition, we give new criteria for these permutation polynomials being involutions. Explicit constructions are also provided for all involutory criteria.

Items 44

Lemma

Lemma 2.1

JIRX

([akbary2011constructing, Lemma 1.2], the AGW Criterion) Let A, S, and \overline{S} be finite sets with \# S=\# \overline{S}, and let f: A\to A, g: S\to \overline{S}, \lambda: A\to S and \overline{\lambda}: A\to\overline{S} be maps such that \bar{\lambda}\circ f=g\circ \lambda. If both \lambda and \bar{\lambda} are surjective, then the following statements are equivalent:

  1. f is a bijection and

  2. g is a bijection from S to \overline{S} and f is injective on \lambda^{-1}(s) for each s\in S.

Theorem

Theorem 2.2

2KM7

Let A be a finite set, f: A\to A, and let \phi=(\lambda, \eta) and \overline{\phi}=(\overline{\lambda}, \overline{\eta}) be two bijective mappings from A to some subsets of A \times A, and denote by {\phi }^{-1}, {\bar{\phi}}^{-1} their compositional inverses respectively. Let \psi=(g,\tau): \phi(A)\to \overline{\phi}(A) be a mapping such that \bar{\phi}\circ f=\psi\circ \phi. Then f is bijective if and only if \psi is bijective. Furthermore, if \psi is bijective and its compositional inverse is denoted by \psi^{-1}, then

f^{-1}= {\phi }^{-1} \circ \psi^{-1} \circ {\bar{\phi}}

is the compositional inverse of f on A.

Used by Example 2.3, Example 2.4, Example 2.5, Example 2.6, Proof, Proof, Proof, Proof, Proof, Proof
Example

Example 2.3

HNT6

[li2019compositional, Theorem 2.3] We take

  1. A={\mathbb F}_q;

  2. f(x)=x^rh\left(x^s\right) permutes {\mathbb F}_q, where s \mid (q-1), \gcd(r,q-1) =1 and h(0) \ne 0;

  3. \phi(x) =\overline{\phi}(x)= \left( x^{q-s}, x^s \right);

  4. \phi(A)=\overline{\phi}(A);

  5. \psi(y,z)=\left(y^{r} h(z)^{q-s}, z^{r} h(z)^{s}\right); and

  6. Let l(x) be the compositional inverse of g(x)=x^rh(x)^s over \mu_{\frac{q-1}{s}} =\left\{x\in{{\mathbb F}}_{q}^*: x^\frac{q-1}{s}=1\right\} and r' be an integer which satisfies rr'\equiv1\pmod {q-1}.

Then it follows from Theorem 2.2 that

f^{-1}(x)=\left(\alpha(x) h(l(\beta(x)))^{s-1}\right)^{r^{'}}l(\beta(x))

is the compositional inverse of f(x), where \alpha(x)=x^{q-s} and \beta(x)=x^s.

Uses Theorem 2.2
Example

Example 2.4

BG5W

[niu2019new, Theorem 3.7] Assume that

  1. A={\mathbb F}_{q^m};

  2. f(x) = g\left(x^{q^i} - x + \delta\right) + cx \in {\mathbb F}_{q^m}[x] permutes {\mathbb F}_{q^m}, where q is a prime power, m,i be positive integers with 1 \le i \le m-1, c \in {\mathbb F}_{q^{\gcd(i,m)}}^* and g(x) \in {\mathbb F}_{q^m}[x]. Then h(x) = g(x)^{q^i} - g(x) + cx +(1-c)\delta \in {\mathbb F}_{q^m}[x] permutes {\mathbb F}_{q^m} (see [zheng2019two, Proposition 3]), where \delta\in {\mathbb F}_{q^m}.

  3. \phi(x) =\overline{\phi}(x)=\left( -x^{q^i}, x^{q^i}-x+\delta \right);

  4. \phi(A)=\overline{\phi}(A);

  5. \psi(y,z)=\left(c^{q^i}y-g(z)^{q^i}, h(z) \right); and

  6. Assume H(x) is the compositional inverse of h(x).

Then it follows from Theorem 2.2 that, for any \delta \in {\mathbb F}_{q^m}, the compositional inverse of f(x) is

f^{-1}(x)= c^{-1}x^{q^i} - c^{-1}g\left(H(x^{q^i}-x+\delta)\right)^{q^i} -H(x^{q^i}-x+\delta)+\delta.
Uses Theorem 2.2
Example

Example 2.5

5GCW

[tuxanidy2014inverses, Theorem 1.2] Let

  1. A={\mathbb F}_{q^n};

  2. f(x)=h(\psi_0(x)) \varphi_0(x)+g_0(\psi_0(x)) permute {\mathbb F}_{q^n}, where {\varphi_0}, \psi_0 \in {\mathbb F}_{q^n}[x] are additive polynomials, q-polynomial \overline{\psi_0} satisfies \varphi_0 \circ \psi_0=\overline{\psi_0} \circ \varphi_0 and \left|\psi_0\left({{\mathbb F}_{q^{n}}}\right)\right|=\left|\overline{\psi_0}\left( {{\mathbb F}_{q^{n}}}\right)\right|, and polynomial h \in {\mathbb F}_{q^n}[x] such that h\left(\psi_0\left(\mathbb{F}_{q^{n}}\right)\right) \subseteq \mathbb{F}_{q} \backslash\{0\};

  3. \phi(x) =\phi_{\psi_0}\left( x \right) =\left( \psi_0(x), x-\psi_0(x) \right), \overline{\phi}=\phi_{\overline{\psi_0}}\left( x \right) = \left( \overline{\psi_0}(x), x-\overline{\psi_0}(x) \right) in [tuxanidy2014inverses, Lemma 2.10];

  4. \phi(A)=\phi_{\psi_0}\left(\mathbb{F}_{q^{n}}\right) ,\overline{\phi}(A)=\phi_{\overline{\psi_0}}\left(\mathbb{F}_{q^{n}}\right);

  5. \psi(y,z)=\left(h(y) \varphi_0(y)+\overline{\psi_0}(g_0(y)), h(y) \varphi_0(z)+g_0(y)-\overline{\psi_0}(g_0(y)) \right); and

  6. Assume that \left|S_{\psi_0}\right|=\left|S_{\overline{\psi_0}}\right| and \ker(\varphi_0) \cap \psi_0\left(S_{\psi_0}\right)=\{0\}. Then \varphi_0 induces a bijection from S_{\overline{\psi_0}} to S_{\overline{\psi_0}}. Let \overline{f}^{-1} and \left.\varphi_0^{-1}\right|_{S_{\overline{\psi_0}}} \in {\mathbb F}_{q^{n}}[x] induce the inverses of \left.\overline{f}\right|_{\psi_0\left(\mathbb{F}_{q^n}\right)} (x) =h(x) \varphi_0(x)+\overline{\psi_0}(g_0(x)) and \varphi_0 |_{S_{\psi_0}} respectively,

Then it follows from Theorem 2.2 that the compositional inverse of f(x) is given by

f^{-1}(x)=\overline{f}^{-1}(\overline{\psi_0}(x))+\left.\varphi_0^{-1}\right|_{S_{\overline{\psi_0}}}\left(\frac{x-\overline{\psi_0}(x)-g_0\left(\overline{f}^{-1}(\overline{\psi_0}(x))\right)+\overline{\psi_0}\left(g_0\left(\overline{f}^{-1}(\overline{\psi_0}(x))\right)\right)}{h\left(\overline{f}^{-1}(\overline{\psi_0}(x))\right)}\right) .
Uses Theorem 2.2
Example

Example 2.6

7X5T

[wu2013compositional, Theorem 2.3] Let

  1. A={\mathbb F}_{q^n}, where q is even and n is odd;

  2. f(x)=x(L({\rm Tr}_{q^n/q} (x))+a {\rm Tr}_{q^n/q} (x)+a x) permute {\mathbb F}_{q^n}, where xL(x) is a bilinear PP over {\mathbb F}_q for a linearized polynomial L(x) \in {\mathbb F}_q [x], a \in {\mathbb F}_q^*, and the trace function from {\mathbb F}_{q^n} to {\mathbb F}_{q} is denoted by {\rm Tr}_{q^n/q}(\cdot): x \to \sum_{i=0}^{n-1}x^{q^i};

  3. \phi(x) =\overline{\phi}(x)=\left( {\rm Tr}_{q^n/q}(x), x + {\rm Tr}_{q^n/q}(x) \right);

  4. \phi(A)=\overline{\phi}(A)={\mathbb F}_{q} \oplus { \ker( {\rm Tr}_{q^n/q}) };

  5. \psi(y,z)=\left(y L(y), a z^{2}+(L(y)+a y) z\right); and

  6. Let q = 2^m for a positive integer m. Assume the compositional inverse of xL ( x ) is g_0( x ) \in {\mathbb F}_q [ x ].

Then it follows from Theorem 2.2 that the compositional inverse of f(x) is

f^{-1}(x)=a^{2^{m-1}-1} x^{2^{n m-1}}+\left(g_0({\rm Tr}_{q^n/q}(x))+a^{2^{m-1}-1} \sum_{k=1}^{\frac{n-1}{2}} x^{2^{(2 k-1) m-1}}\right)
{\times} \left(\frac{{\rm Tr}_{q^n/q}(x)}{g_0({\rm Tr}_{q^n/q}(x))}+ag_0({\rm Tr}_{q^n/q}(x))\right)^{q-1}
+\sum_{j=0}^{m-2} a^{2^{j}-1}\left(\frac{{\rm Tr}_{q^n/q}(x)}{g_0({\rm Tr}_{q^n/q}(x))}+ag_0({\rm Tr}_{q^n/q}(x))\right)^{2^{m}-2^{j+1}}\left(\sum_{k=0}^{\frac{n-1}{2}} x^{q^{2 k}}\right)^{2^{j}}.
Uses Theorem 2.2
Corollary

Corollary 2.7

5D6M

Let A be a finite set and f, g, \lambda, \overline{\lambda} be mappings satisfying the assumption of Lemma 2.1 (e.g., satisfy the commutative diagram in Fig. 1). We assume \eta(x)=P(x)-\lambda(x) , \overline{\eta}(x)=P(x) - \overline{\lambda}(x), and \tau are mappings such that \bar{\eta}\circ f=\tau \circ \eta, where P(x) permutes A. Then both \phi=(\lambda, \eta) and \overline{\phi}=(\overline{\lambda}, \overline{\eta}) are bijective mappings from A to some subsets of A \times A and {\phi }^{-1}(y,z)=P^{-1}(y+z), {\bar{\phi}}^{-1}(\alpha,\beta) =P^{-1}(\alpha+\beta) are their compositional inverses respectively. Moreover, f is bijective if and only if \psi =(g, \tau) is bijective. Furthermore, if both g, \psi are bijective and their compositional inverses are denoted by g^{-1}(\alpha) and \psi^{-1}(\alpha,\beta)=\left(g^{-1}(\alpha) , M(\alpha,\beta) \right) respectively, where M(\alpha,\beta): {\rm Im}(\overline{\phi}) \to {\rm Im}(\eta), then

f^{-1}(x)= P^{-1} \left(g^{-1}(\overline{\lambda}(x)) + M(\overline{\lambda}(x), P(x) - \overline{\lambda}(x) ) \right)

is the compositional inverse of f on A.

Proof of Corollary 2.7

Assume \phi(x) = \phi(x^\prime). Then \lambda(x) = \lambda(x^\prime) and P(x) - \lambda(x) =\eta(x) = \eta(x^\prime) = P(x^\prime) - \lambda(x^\prime). Hence P(x) = P(x^\prime). Because P(x) permutes A, we must have x = x^\prime and thus \phi is a bijection. Plug y=\lambda(x), z=P(x)-\lambda(x) into P^{-1}(y+z), one can obtain P^{-1}(\lambda(x)+P(x)-\lambda(x) ) = x. Thus {\phi }^{-1}(y,z)=P^{-1}(y+z). Similarly, \bar{\phi} is bijective and {\bar{\phi}}^{-1}(\alpha,\beta) =P^{-1}(\alpha+\beta). The rest of proof follows from Theorem 2.2.

Uses Theorem 2.2
Corollary

Corollary 2.8

V5OA

Let A be a finite set and f, g, \lambda, \overline{\lambda} be mappings satisfying the assumption of Lemma 2.1 (e.g., satisfy the commutative diagram in Fig. 1). Let \lambda(x), \bar{\lambda}(x) \ne 0 for any x \in A. We assume \eta(x)=P\left( \frac{x}{\lambda(x)} \right), \overline{\eta}(x)=P\left( \frac{x}{\overline{\lambda}(x)} \right), and \tau are mappings such that \bar{\eta}\circ f=\tau \circ \eta, where P(x) permutes A. Then both \phi=(\lambda, \eta) and \overline{\phi}=(\overline{\lambda}, \overline{\eta}) are bijective mappings from A to some subsets of A \times A and {\phi }^{-1}(y,z)=yP^{-1}(z), {\bar{\phi}}^{-1}(\alpha,\beta) =\alpha P^{-1}(\beta) are their compositional inverses respectively. Moreover, f is bijective if and only if \psi =(g, \tau) is bijective. Furthermore, if both g, \psi are bijective and their compositional inverses are respectively denoted by g^{-1}(\alpha), \psi^{-1}(\alpha,\beta)=\left(g^{-1}(\alpha) , M(\alpha,\beta) \right), where M(\alpha,\beta): {\rm Im}(\overline{\phi}) \to {\rm Im}(\eta), then

f^{-1}(x)= g^{-1} \left( \overline{\lambda}(x) \right) P^{-1} \left( M\left( \overline{\lambda}(x),P\left( \frac{x}{\overline{\lambda}(x)} \right) \right) \right)

is the compositional inverse of f on A.

Proof of Corollary 2.8

Assume \phi(x) = \phi(x^\prime). Then \lambda(x) = \lambda(x^\prime) and P(\frac{x}{ \lambda(x)}) =\eta(x) = \eta(x^\prime) = P(\frac{x^\prime }{\lambda(x^\prime)}). Because P(x) permutes A, the latter implies that \frac{x}{ \lambda(x) }= \frac{x^\prime }{\lambda(x^\prime)}. Since \lambda(x) = \lambda(x^\prime), we must have x = x^\prime and thus \phi is a bijection. Plug y=\lambda(x), z=P\left( \frac{x}{\lambda(x)} \right) into yP^{-1}(z), one can obtain \lambda(x)P^{-1}(P\left( \frac{x}{\lambda(x)} \right))= x. Thus {\phi }^{-1}(y,z)=yP^{-1}(z). Similarly, \bar{\phi} is bijective and {\bar{\phi}}^{-1}(\alpha,\beta) =\alpha P^{-1}(\beta). The rest of proof follows from Theorem 2.2.

Uses Theorem 2.2
Lemma

Lemma 2.9

CAF9

Let A and S be finite sets, and let f:A \to A, g: S \to S, \lambda:A \to S be maps such that \lambda is surjective and \lambda \circ f = g \circ \lambda. Assume f is an involution on A. Then g is an involution on S.

Used by Proof, Proof, Proof
Proof

Proof

9M4H

We obtain \lambda = \lambda \circ f \circ f = g \circ \lambda \circ f = g \circ g \circ \lambda. Since \lambda is surjective, g is an involution on S according to \lambda = g \circ g \circ \lambda.

Lemma

Lemma 3.1

858Y

[park2001permutation, Theorem 2.3] [wang2007cyclotomic, Theorem 1] [zieve2009some, Lemma 2.1] Let q be a prime power and f(x)=x^rh\left(x^s\right) \in{\mathbb F}_q[x], where s=\frac{q-1}{\ell} and \ell is an integer. Then f(x) permutes {\mathbb F}_q if and only if

  1. \gcd\left(r,s\right)=1 and

  2. g(x)=x^rh(x)^s permutes \mu_{\ell}, where \mu_{\ell}=\left\{x\in{{\mathbb F}}_{q}^* : ~~ x^{\ell}=1\right\}.

Theorem

Theorem 3.2

50PD

Let f(x)=x^rh\left(x^s\right) defined as in Lemma 3.1 be a permutation over {\mathbb F}_q and g^{-1}(x) be the compositional inverse of g(x)=x^rh(x)^s over \mu_{\frac{q-1}{s}}. Suppose a and b are two integers satisfying as+br=1. Then the compositional inverse of f(x) in {\mathbb F}_q[x] is given by

f^{-1}(x)=g^{-1}(x^s)^a x^b h\left(g^{-1}(x^s)\right)^{-b}.
Used by Proof
Proof of Theorem 3.2

Let \phi be a map defined by

\phi: {\mathbb F}_q \to \phi({\mathbb F}_q),
x\mapsto \left( \lambda(x), \eta (x) \right) = \left(x^{s},x^r\right),

where \phi({\mathbb F}_q)=\left\{ \left( x^{s}, x^{r} \right): ~~ x\in {\mathbb F}_{q} \right\} .

Given one element \left( y,z\right)\in \phi({\mathbb F}_q), there exists x_0\in{\mathbb F}_{q} such that y=x_0^s and z=x_0^r. Furthermore, it is clear that y^az^b = x_0^{as+br}=x_0. Thus \phi is a bijection and \phi^{-1}(y,z)=y^az^b.

Next, we compute the expression of \psi from the relation \psi \circ \phi(x) =\phi \circ f(x). From a simple computation, we get

\phi \circ f(x) = \left( x^{sr}h(x^s)^s , x^{r^2}h(x^s)^r \right).

Substituting x^s and x^r in Eq. (?) with y and z respectively, we obtain

\psi(y,z): \phi({\mathbb F}_q) \to \phi({\mathbb F}_q),
(y,z) \mapsto\left(y^rh(y)^s, z^rh(y)^r \right).

Now we compute the compositional inverse of \psi.

For (y,z), (\alpha,\beta)\in \phi({\mathbb F}_q) with \psi(y,z)=(\alpha,\beta), we have

\left\{ \begin{aligned} y^rh(y)^s &=& \alpha, \\ z^rh(y)^r &=& \beta. \\ \end{aligned} \right.

Then we get

y=g^{-1}(\alpha).

Moreover,

z^r=\beta h(y)^{-r} = \beta h\left( g^{-1}(\alpha)\right)^{-r}.

Clearly, for any given (\alpha, \beta)\in\phi({\mathbb F}_q), there exists a unique element denoted by {x_{(\alpha,\beta)}}\in{\mathbb F}_q such that \alpha={x_{(\alpha,\beta)}}^s and \beta={x_{(\alpha,\beta)}}^r. Therefore, z^r = \beta h\left( g^{-1}(\alpha)\right)^{-r} = {x_{(\alpha,\beta)}}^rh\left( g^{-1}({x_{(\alpha,\beta)}}^s) \right)^{-r}. To show that z= x_{(\alpha,\beta)}h\left( g^{-1}({x_{(\alpha,\beta)}}^s) \right)^{-1} is the unique solution to Eq. (?), it suffices to prove that x_{(\alpha,\beta)}h\left( g^{-1}({x_{(\alpha,\beta)}}^s) \right)^{-1} \in \{x^r: ~~ x \in {\mathbb F}_q\}.

Since f is a PP, there exists a unique x_0\in{\mathbb F}_q such that {x_{(\alpha,\beta)}}=f(x_0)=x_0^r h\left(x_0^s\right). Furthermore, we have

g^{-1}(f(x_0)^s) =g^{-1}(x_0^{rs}h(x_0^s)^s) = g^{-1}(g(x_0^s)) = x_0^s.

Plugging {x_{(\alpha,\beta)}}=x_0^r h\left(x_0^s\right) into {x_{(\alpha,\beta)}}h\left( g^{-1}({x_{(\alpha,\beta)}}^s) \right)^{-1}, we obtain

{x_{(\alpha,\beta)}}h\left( g^{-1}({x_{(\alpha,\beta)}}^s) \right)^{-1} = x_0^{r} h\left(x_0^s\right) h(g^{-1}(f(x_0)^s))^{-1} = x_0^r h\left(x_0^s\right) h\left(x_0^s\right)^{-1} =x_0^r,

which belongs to \{x^r: ~~ x \in {\mathbb F}_q\}. Thus z={x_{(\alpha,\beta)}}h\left( g^{-1}({x_{(\alpha,\beta)}}^s) \right)^{-1} and then

\psi^{-1}\left(\alpha,\beta \right)=\left( g^{-1}(\alpha), {x_{(\alpha,\beta)}}h\left( g^{-1}({x_{(\alpha,\beta)}}^s) \right)^{-1} \right).

According to Theorem 2.2, the compositional inverse of f(x) is

f^{-1}(x)={\phi}^{-1} \circ \psi^{-1} \circ {\phi}(x)= g^{-1}(\alpha)^a z^b = g^{-1}(x^s)^a x^b h\left(g^{-1}(x^s)\right)^{-b}.
Uses Theorem 2.2
Corollary

Corollary 3.3

8Q4Y

Let s \mid(q-1), h(x) \in \mathbb{F}_{q}[x] satisfies that h(\zeta)^{s}=\zeta^{n} for every \zeta \in \mu_{(q-1) / s}. Suppose \gcd(r+n,(q-1) / s)=1 and t , a,b be integers that satisfy (r+n) t \equiv 1(\bmod (q-1) / s) and ar+bs=1. Then f(x)=x^{r} h\left(x^{s}\right) permutes \mathbb{F}_{q}, g(\zeta)=\zeta^{r} h(\zeta)^{s}=\zeta^{r+n}, and g^{-1}(\zeta)=\zeta^t. Moreover, the compositional inverse of f(x) in \mathbb{F}_{q}[x] is

f^{-1}(x)=x^{ast+b} h\left(x^{st}\right)^{-b}.
Corollary

Corollary 3.4

29VW

The PP f(x) = x^rh(x^s) over {\mathbb F}_q defined as in Lemma 3.1 is an involution if and only if

  1. g(x)=x^rh(x)^s is involutory on \mu_{\frac{q-1}{s}} and

  2. \varphi(x)=g(x^s)^a x^{b-r} h\left(g(x^s)\right)^{-b}h(x^s)^{-1}= 1 holds for any x \in {\mathbb F}_q^*, where integers a and b satisfy as+br=1.

Used by Proof
Proof of Corollary 3.4

Since g(x) being involutory is necessary for f(x) being involutory by Lemma 2.9, we have g^{-1}(x) = g(x) on \mu_{\frac{q-1}{s}}. In this case, by Theorem 3.2, f^{-1} (x) = f(x) if and only if

g(x^s)^a x^{b-r} h\left(g(x^s)\right)^{-b}h(x^s)^{-1}= 1.
Uses Lemma 2.9, Theorem 3.2
Example

Example 3.5

3OTK

[niu2019new, Corollary 2.17] Let q be a power of 2 and k be a positive integer such that \gcd(k,q+1)=1. Then for any \gamma , \beta \in {\mathbb F}_q^* such that {\rm Tr}_{q/2}(\beta)=0, the polynomial f(x)=x^{q^2-2}h\left(x^{q-1}\right) is an involution on {\mathbb F}_{q^2}, where h(x)=\gamma\left( x^{-1}+\beta x^{-k-1}+\beta x^{k-1}\right).

Proof of Example 3.5

According to the proof of [niu2019new, Corollary 2.17], we obtain h(x) \ne 0 and g(x)=x^{-1}h(x)^{q-1}=x, for any x \in \mu_{q+1}. Let a= -1, b= -q, r=-1, s=q-1 in Corollary 3.4. Then \varphi(x)=x^{1-q} x^{-q+1} h\left(x^s\right)^{q}h(x^s)^{-1}= x^{1-q} (x^{-q+1} h(x^{q-1})^{q-1}) = x^{1-q} x^{q-1} = 1. Thus f is an involution.

Uses Corollary 3.4
Lemma

Lemma 4.1

CWAX

[yuan2011permutation, Theorem 6.1] Assume that F is a finite field and S, \overline{S} are finite subsets of F with \#S = \#\overline{S} such that the maps \lambda : F \rightarrow S and \overline{\lambda}: F \rightarrow \overline{S} are surjective and \overline{\lambda} is additive, i.e.,

\overline{\lambda}( x + y) = \overline{\lambda}(x) + \overline{\lambda}(y),\ \ \ \ \ x, y \in F.

Let g_0: S \rightarrow F, and g: F \rightarrow F be maps such that

\overline{\lambda} \circ ( g+g_0 \circ \lambda )= g \circ \lambda ,

g(S)=\bar{S} and \overline{\lambda}( g_0(\lambda(x))) = 0 for every x \in F. Then the map f(x) = g (x) + g_0(\lambda(x)) permutes F if and only if g permutes F.

Theorem

Theorem 4.2

R77P

Let the symbols be defined as in Lemma 4.1. Let f(x) = g (x) + g_0(\lambda(x)) be a permutation over F and g^{-1}(x) be the compositional inverse of g(x) over S. Then the compositional inverse of f(x) is given by

f^{-1}(x)= g^{-1}\left(x-g_0(g^{-1}(\overline{\lambda}(x))) \right) .
Used by Proof
Proof of Theorem 4.2

Let L(x) be a linearized PP over F and \phi,\overline{\phi} be maps defined by

\phi: F \to \phi(F)
x \mapsto \left( \lambda(x), L(x)-\lambda(x) \right) ,

and

\overline{\phi}: F \to \overline{\phi}(F)
x \mapsto \left( \overline{\lambda}(x), L(x)-\overline{\lambda}(x) \right) .

Then \phi and \overline{\phi} are bijections, and for (y,z)\in \phi(F), we have \phi^{-1}(y,z)=L^{-1}(y+z), where \phi^{-1} and L^{-1} denote the compositional inverses of \phi and L respectively. Here is the commutative diagram as stated in Theorem 2.2.

\xymatrix{ F \ar[rr]^{f=g + g_0\circ\lambda }\ar[d]_{\phi=(\lambda, \eta)} & & F \ar[d]^{\overline{\phi}=(\overline{\lambda}, \overline{\eta})} \\ \phi(F) \ar[rr]^{\psi=(g,\tau)} & & \overline{\phi}(F) }

Next, we find the map \psi such that \psi \circ \phi =\overline{\phi} \circ f. After direct calculation, we have

\overline{\phi}\circ f(x)=\left( g(\lambda(x)), L \left( g(x) \right) + L\left( g_0(\lambda(x)) \right) -g(\lambda(x)) \right).

To compute \psi(y,z) accordingly such that \bar{\phi}\circ f(x)=\psi\circ \phi (x), we substitute \lambda(x) and L(x)-\lambda(x) in Eq. (?) with y and z respectively. We obtain

\psi(y,z): \phi(F) \to \overline{\phi}(F),
(y,z) \mapsto\left( g(y), L(g(L^{-1}(y+z)))+ L( g_0(y) ) - g(y) \right).

Clearly \psi is a bijection, since \psi \circ \phi(x) =\phi \circ f(x). Now we compute the compositional inverse of \psi. Since f permutes F, g permutes F, and we assume that g^{-1} is the compositional inverse of g. For any (y,z)\in\phi(F) with \psi(y,z)=(\alpha,\beta) \in \overline{\phi}(F), we have

\left\{ \begin{aligned} g(y)&=& \alpha, \\ L(g(L^{-1}(y+z)))+ L( g_0(y) ) - g(y)&=& \beta. \\ \end{aligned} \right.

Clearly,

y=g^{-1}(\alpha).

Moreover,

L(g(L^{-1}(y+z)))=\alpha+ \beta-L( g_0(y) ).

Composing L \circ g^{-1} \circ L^{-1} on Eq. (?) and simplifying it, we have

z=L \left( g^{-1} \left( L^{-1} ( \alpha+ \beta )- g_0(g^{-1}(\alpha)) \right) \right) - g^{-1}(\alpha).

Hence we have

\psi^{-1}\left(\alpha,\beta \right)=\left( g^{-1}(\alpha) , L \left( g^{-1} \left( L^{-1} ( \alpha+ \beta )- g_0(g^{-1}(\alpha)) \right) \right) - g^{-1}(\alpha) \right) .

Finally we compute the compositional inverse of f(x). According to Theorem 2.2, together with \alpha = \overline{\lambda}(x) and \beta=L(x) -\overline{\lambda}(x), we obtain the compositional inverse of f(x) is

f^{-1}(x)={\phi}^{-1} \circ \psi^{-1} \circ \overline{\phi}(x)= g^{-1} \left( L^{-1} ( \alpha+ \beta )- g_0(g^{-1}(\alpha)) \right) = g^{-1}\left(x-g_0(g^{-1}(\overline{\lambda}(x))) \right) .
Uses Theorem 2.2
Corollary

Corollary 4.3

AR9U

Let n and k be positive integers such that \gcd(n, k)=d>1, let s be any positive integer with s\left(q^{k}-1\right) \equiv 0 \pmod {q^{n}-1} . Let g(x) \in \mathbb{F}_{q}[x] be a linearized polynomial permuting \mathbb{F}_{q^{n}}, \lambda(x)=\overline{\lambda}(x) be a q^{d}-polynomial with \lambda(1)=0 and g_1(x) \in \mathbb{F}_{q^{n}}[x]. Then the compositional inverse of f(x)=\left(g_1(\lambda(x))\right)^s+g(x) over \mathbb{F}_{q^{n}} is

f^{-1}(x)= g^{-1}\left(x-g_1(g^{-1}({\lambda}(x)))^s \right) .
Corollary

Corollary 4.4

2LRK

In Lemma 4.1, let \lambda=\bar{\lambda}, S=\bar{S} be a finite subset of F such that \lambda(F)=S, and let g be a bijection from S to S. Then, the PP f(x) = g (x) + g_0(\lambda(x)) over F defined as in Lemma 4.1 is an involution if and only if \varphi(x)=g^{-1}\left(x-g_0(g({\lambda}(x))) \right) - g (x) -g_0(\lambda(x)) = 0 holds for any x\in F.

Proof of Corollary 4.4

Its proof can be easily derived by applying Theorem 4.2 and thus we omit the details.

Uses Theorem 4.2
Corollary

Corollary 4.5

I65B

Let q be an even prime power and \lambda( g_0(\lambda(x))) = 0 hold for any x \in F={\mathbb F}_{ q^n } defined as in Lemma 4.1, with \overline{\lambda} = \lambda being additive. Then the PP f (x)= x + g_0(\lambda(x)) is an involution on {\mathbb F}_{ q^n }.

Example

Example 4.6

N240

Let q be a power of 2, S={\mathbb F}_q, and g_0 be any polynomial such that g_0({\mathbb F}_q) \subseteq {\mathbb F}_q. Assume n is an even integer, \lambda(x)= {\rm Tr}_{q^n/q}(x). Thus we have {\rm Tr}_{q^n/q}(1)=0 and {\rm Tr}_{q^n/q}( g_0({\rm Tr}_{q^n/q}(x))) ={\rm Tr}_{q^n/q}( 1 ) g_0({\rm Tr}_{q^n/q}(x))= 0 holds for any x \in F={\mathbb F}_{ q^n }. Then f (x)= x + g_0({\rm Tr}_{q^n/q}(x)) is an involution on {\mathbb F}_{ q^n }.

Lemma

Lemma 5.1

TON4

[akbary2011constructing, Theorem 6.3] Let q be any power of the prime number p, let n be any positive integer, and let S be any subset of {\mathbb F}_{q^n} containing 0. Let h,k \in {\mathbb F}_{q^n} be any polynomials such that h(0) \ne 0 and k(0) = 0, and let {\lambda(x)} \in {\mathbb F}_{q^n}[x] be any polynomial satisfying

  1. h(\lambda({\mathbb F}_{q^n})) \subseteq S; and

  2. \lambda(a\alpha) = k(a) \lambda(\alpha) for all a \in S and all \alpha \in {\mathbb F}_{q^n}.

Then the polynomial f (x) = xh(\lambda(x)) is a permutation polynomial for {\mathbb F}_{q^n} if and only if g(x) = xk(h(x)) induces a permutation of \lambda({\mathbb F}_{q^n}).

Used by Proof
Theorem

Theorem 5.2

0YEO

Let the symbols be defined as in Lemma 5.1. Let f(x) = xh(\lambda(x)) permute {\mathbb F}_{q^n} and g^{-1}(x) be the compositional inverse of g(x)=xk(h(x)) over \lambda({\mathbb F}_{q^n}). Then the compositional inverse of f (x) is given by

f^{-1}(x)= \frac{x-\lambda(x) + k\left(h\left(g^{-1}\left(\lambda(x)\right)\right)\right)g^{-1}\left(\lambda(x)\right)}{h\left(g^{-1}\left(\lambda(x)\right)\right)} .
Used by Proof
Proof of Theorem 5.2

Let \eta(x) = x-\lambda(x) and \phi be a map defined by

\phi: {\mathbb F}_{q^n} \to \phi({\mathbb F}_{q^n}),
x \mapsto \left( \lambda(x), x-\lambda(x) \right) .

Then \phi is a bijection and for (y,z)\in \phi({\mathbb F}_{q^n}),\ \phi^{-1}(y,z)=y+z. Let us consider the following commutative diagram.

\xymatrix{ {\mathbb F}_{q^n} \ar[rr]^{f(x)=xh(\lambda(x))}\ar[d]_{\phi=(\lambda, \eta)} & & {\mathbb F}_{q^n} \ar[d]^{\phi=(\lambda, \eta)} \\ \phi({\mathbb F}_{q^n}) \ar[rr]^{\psi=(g,\tau)} & & \phi({\mathbb F}_{q^n}) }

First we determine the expression of \psi to establish the connection {\phi}\circ f(x)=\psi\circ \phi (x). After direct computation, we have

\phi \circ f(x)=\left(\lambda(x)k(h(\lambda(x))), xh(\lambda(x))-k(h(\lambda(x)))\lambda(x) \right) .

Then, substituting \lambda(x) and x-\lambda(x) in Eq. (?) with y and z respectively, we obtain

\psi(y,z): \phi({\mathbb F}_{q^n}) \to \phi({\mathbb F}_{q^n}),
(y,z) \mapsto\left(yk(h(y)), (y+z)h(y)-k(h(y))y \right).

Since f(x)=xh(\lambda(x)) permutes {\mathbb F}_{q^n}, we have that both \psi and g(x) = xk(h(x)) are bijective. Recall that g^{-1} denotes the compositional inverse of g(y) = yk(h(y)) over \lambda({\mathbb F}_{q^n}). In the following, we compute the compositional inverse of \psi. Let (y,z), (\alpha,\beta)\in \phi({\mathbb F}_{q^n}), satisfy \psi(y,z)=(\alpha,\beta), i.e.,

\left\{ \begin{aligned} yk(h(y))&=& \alpha, \\ (y+z)h(y)-k(h(y))y&=& \beta. \\ \end{aligned} \right.

Then, we have

y=g^{-1}(\alpha).

Moreover, for z, we firstly explain h(y) \ne 0 for any y \in \lambda({\mathbb F}_{q^n}). Assume that there exists some y_0 \in \lambda({\mathbb F}_{q^n}) such that h(y_0) = 0. We have g(y_0) = y_0k(h(y_0)) =0 and g(0) =0. Since g(y) is bijective, we have y_0 =0, which is conflict with h(0) \ne 0. Thus h(y) \ne 0 for any y \in \lambda({\mathbb F}_{q^n}). Then it follows from Eq.(?) that

z= \frac{\beta + k\left(h\left( y \right)\right) y}{h\left(y \right)}-y.

Hence, we have

\psi^{-1}\left(\alpha,\beta \right)=\left( g^{-1}(\alpha) , \frac{\beta + k\left(h\left(g^{-1}(\alpha)\right)\right)g^{-1}(\alpha)}{h\left(g^{-1}(\alpha)\right)}-g^{-1}(\alpha) \right).

Finally, we compute the compositional inverse of f(x). From Theorem 2.2, together with the compositional inverses of \phi, \psi and \alpha = \lambda(x), \beta=x-\lambda(x), the compositional inverse of f(x) is

\begin{aligned} f^{-1}(x)&= {\phi}^{-1} \circ \psi^{-1} \circ {\phi}(x) \\ &= g^{-1}(\alpha) + \frac{\beta + k\left(h\left(g^{-1}(\alpha)\right)\right)g^{-1}(\alpha)}{h\left(g^{-1}(\alpha)\right)} - g^{-1}(\alpha) \\ &= \frac{x-\lambda(x) + k\left(h\left(g^{-1}\left(\lambda(x)\right)\right)\right)g^{-1}\left(\lambda(x)\right)}{h\left(g^{-1}\left(\lambda(x)\right)\right)}. \end{aligned}
Uses Theorem 2.2
Corollary

Corollary 5.3

L2B8

Let \lambda(x) be either \lambda_2(x) =\sum\limits_{0\le i < j\le n - 1} {{x^{{p^i} + {p^j}}}} or T_2(x)={\rm Tr}_{p^n/p}(x^2). Let h(x)\in {\mathbb F}_p[x] such that h(0) \ne 0. If the polynomial g(x)=x(h(x))^2 permutes {\mathbb F}_{p}, then the polynomial f(x)=xh(\lambda(x)) permutes {\mathbb F}_{p^n}, and the compositional inverse of f (x) is given by

f^{-1}(x)= \frac{x-\lambda(x) + \left(h\left(g^{-1}\left(\lambda(x)\right)\right)\right)^2 g^{-1}\left(\lambda(x)\right)}{h\left(g^{-1}\left(\lambda(x)\right)\right)} .
Corollary

Corollary 5.4

JMWH

Let f(x)=xh(\lambda(x)), g(x) = xk(h(x)) and h, k, \lambda defined as in Lemma 5.1. Let \theta(x)=k(h(x)) for any x \in {\mathbb F}_{q^n}. Then f is an involution over {\mathbb F}_{q^n} if and only if

  1. \theta \left( \theta(y) y \right) \theta(y) =1 holds for any y \in \lambda({\mathbb F}_{q^n}^*), and

  2. \varphi(y)=h\left( g(y) \right) h(y) -1 =0 holds for any y \in \lambda({\mathbb F}_{q^n}^*).

Used by Proof
Proof of Corollary 5.4

Since f(0)=0, we only need to consider the nonzero situation. Assume f is an involution. By Lemma 2.9, we have g is an involution, which is equivalent to \theta \left( \theta(y) y \right) \theta(y) = k\left(h\left( k(h(y)) y \right)\right) k(h(y)) =1 for y \in \lambda({\mathbb F}_{q^n}^*). From now on, we assume (1) hold and prove that f is an involution if and only if \varphi(x)=0. Plugging

k\left(h\left( k(h(y)) y \right)\right) k(h(y)) =1

for y \in \lambda({\mathbb F}_{q^n}^*), g^{-1}(x)= g(x) and f^{-1}(x) by Theorem 5.2 into \varphi(x), we have

\begin{aligned} x\varphi(x)&= h\left( k(h(\lambda(x))) \lambda(x) \right) xh(\lambda(x)) - x \\ &= h\left( k(h(\lambda(x))) \lambda(x) \right) xh(\lambda(x)) - x +\lambda(x) - \lambda(x) k\left(h\left( k(h(\lambda(x))) \lambda(x) \right)\right) k(h(\lambda(x))) \\ &= h\left(g\left(\lambda(x)\right)\right) \left( xh(\lambda(x)) -\frac{x-\lambda(x) + k\left(h\left(g^{-1}\left(\lambda(x)\right)\right)\right)g^{-1}\left(\lambda(x)\right)}{h\left(g^{-1}\left(\lambda(x)\right)\right)} \right) \\ &= h\left(g\left(\lambda(x)\right)\right) \left( f(x)-f^{-1}(x) \right), \end{aligned}

for x \in {\mathbb F}_{q^n}^*. Note h(y) \ne 0 because f(x)=xh(\lambda(x)) permutes {\mathbb F}_{q^n} and f(0)=0. Hence f(x) is an involution if and only if \varphi=0.

Uses Lemma 2.9, Theorem 5.2
Example

Example 5.5

8IKG

Let q=3^n. Then, f(x) = x (\lambda(x)^2 + 1) is an involution over {\mathbb F}_q, where \lambda (x)=\sum\limits_{0\le i < j\le n - 1} {{x^{{3^i} + {3^j}}}}.

Proof.

Let h(x) = x^2 + 1, S= {\mathbb F}_3 in Lemma 5.1. We have k(x) = x^2 and \lambda({\mathbb F}_q^*)={\mathbb F}_3. One can obtain g(y)=y(y^2 + 1)^2=y and \theta(y)= (y^2+1)^2=1 for y \in {\mathbb F}_3. Hence \theta( \theta(y)y) \theta(y)= \theta( g(y) ) \theta(y) = \theta(y)^2 =1 and \varphi(y)=h\left( g(y) \right) h(y) -1= h(y)^2-1=0. By Corollary 5.4, f(x) is an involution.

Proof

Proof

F97N

Let h(x) = x^2 + 1, S= {\mathbb F}_3 in Lemma 5.1. We have k(x) = x^2 and \lambda({\mathbb F}_q^*)={\mathbb F}_3. One can obtain g(y)=y(y^2 + 1)^2=y and \theta(y)= (y^2+1)^2=1 for y \in {\mathbb F}_3. Hence \theta( \theta(y)y) \theta(y)= \theta( g(y) ) \theta(y) = \theta(y)^2 =1 and \varphi(y)=h\left( g(y) \right) h(y) -1= h(y)^2-1=0. By Corollary 5.4, f(x) is an involution.

Uses Lemma 5.1, Corollary 5.4
Lemma

Lemma 5.6

G382

[akbary2011constructing, Theorem 6.4] Let S \subseteq {\mathbb F}_q and \lambda : {\mathbb F}_q \rightarrow S be a surjective map. Let \gamma \in {\mathbb F}_q^* be a b-linear translator with respect to S for the map \lambda. Then for any G \in {\mathbb F}_q[x] which maps S into S, we have that f(x)= x + \gamma G(\lambda(x)) is a permutation polynomial of {\mathbb F}_q if and only if g(x)= x + bG(x) permutes S.

Used by Proof
Theorem

Theorem 5.7

6UAI

Let f(x)= x + \gamma G(\lambda(x)) defined as in Lemma 5.6 be a PP on {\mathbb F}_q and g^{-1}(x) be the compositional inverse of g(x)=x+bG(x). Then the compositional inverse of f (x) is given by

f^{-1}(x)=(b-\gamma)G\left(g^{-1}(\lambda(x))\right)+g^{-1}(\lambda(x)) - \lambda(x) +x .
Used by Proof
Proof of Theorem 5.7

Using the same notation and assumptions of Lemma 5.6, let \phi be a map defined by

\phi: {\mathbb F}_{q^n} \to \phi({\mathbb F}_{q^n}),
x \mapsto \left( \lambda(x), x-\lambda(x) \right) .

Then \psi is a bijection and for (y,z)\in \phi({\mathbb F}_{q^n}),\ \psi^{-1}(y,z)=y+z.

Next, we determine the expression of \psi from the equation {\phi}\circ f(x)=\psi\circ \phi (x). After direct computation, we have

{\phi}\circ f(x)=\left( \lambda(x) + bG(\lambda(x)) , x-\lambda(x)+ (\gamma-b)G(\lambda(x)) \right).

Substituting \lambda(x) and x-\lambda(x) in Eq. (?) with y and z respectively, we obtain

\psi(y,z): \phi({\mathbb F}_{q^n}) \to \phi({\mathbb F}_{q^n}),
(y,z) \mapsto\left( y + bG(y) , z+ (\gamma-b)G(y) \right).

Since f(x)=x + \gamma G(\lambda(x)) is a PP on {\mathbb F}_{q}, \psi and g(x)= x + bG(x) are both bijective. We assume that g^{-1} denotes the compositional inverse of g(x)= x + bG(x) over S. In the following, we compute the compositional inverse of \psi. Let (y,z), (\alpha,\beta)\in \phi({\mathbb F}_{q^n}), satisfy \psi(y,z)=(\alpha,\beta), i.e.,

\left\{ \begin{aligned} y + bG(y)&=& \alpha, \\ z+ (\gamma-b)G(y)&=& \beta. \\ \end{aligned} \right.

We have

y=g^{-1}(\alpha).

Moreover,

z=\beta + (b-\gamma)G\left(g^{-1}(\alpha)\right).

Hence, we obtain

\psi^{-1}\left(\alpha,\beta \right)=\left( g^{-1}(\alpha) , \beta + (b-\gamma)G\left(g^{-1}(\alpha)\right) \right).

Finally, we compute the compositional inverse of f(x). From Theorem 2.2, together with the compositional inverses of \phi, \psi and \alpha = \lambda(x), \beta=x-\lambda(x), the compositional inverse of f(x) is

\begin{aligned} f^{-1}(x)&= {\phi}^{-1} \circ \psi^{-1} \circ {\phi}(x) \\ &= g^{-1}(\alpha) + \beta + (b-\gamma)G\left(g^{-1}(\alpha)\right) \\ &= (b-\gamma)G\left(g^{-1}(\lambda(x))\right)+g^{-1}(\lambda(x)) - \lambda(x) +x . \end{aligned}
Uses Lemma 5.6, Theorem 2.2
Corollary

Corollary 5.8

AGCD

[kyureghyan2011constructing, Theorem 3] If b \ne -1, then the compositional inverse of the permutation x + \gamma \lambda(x) on {\mathbb F}_q is \frac{-\gamma}{b+1}\lambda(x) +x.

Corollary

Corollary 5.9

4F2T

Suppose \gamma \in {\mathbb F}_q^*. Then the PP f(x) = x + \gamma G(\lambda(x)) over {\mathbb F}_{q} defined as in Lemma 5.6 is an involution if and only if

  1. bG(y) + bG(y + bG(y))=0, for y \in \lambda({\mathbb F}_{q}) and

  2. anyone of the following holds:

    1. (i) b\neq 0;

    2. (ii) q is even;

    3. (iii) G(\lambda(x)) = 0 when q is odd and b = 0.

Proof of Corollary 5.9

Assume f is an involution. By Lemma 2.9, we have bG(y) + bG(y + bG(y))=0 for y \in \lambda({\mathbb F}_{q}), where g(x)= x + bG(x) is defined as in Lemma 5.6. From now on, we assume Condition (1) holds. Then, it suffices to prove that f is an involution if and only if \varphi(x)= 0. Let \varphi(x) = \gamma G(\lambda(x)) +\gamma G\left( \lambda(x) + bG(\lambda(x)) \right). Plugging bG(y) + bG(y + bG(y))=0 for y \in \lambda({\mathbb F}_{q}), g^{-1}(x)= g(x) and f^{-1}(x) by Theorem 5.7 into \varphi(x), we have

f(x) is an involution if and only if \varphi(x)=0 holds for any x \in {\mathbb F}_q. If b\neq 0, then Condition (?) implies that G(\lambda(x))=-G(\lambda(x)+b G(\lambda(x))) and thus \varphi(x)= 0 for any x \in {\mathbb F}_q. If b=0 and q is even, then \varphi(x)=2 \gamma G(\lambda(x))=0 as well. If b=0 and q is odd, then

\varphi(x)=\gamma G(\lambda(x))+\gamma G(\lambda(x)+b G(\lambda(x)))=2 \gamma G(\lambda(x)).

In this case, \varphi(x) =0 if and only if G(\lambda(x)) =0.

Uses Lemma 2.9, Theorem 5.7
Corollary

Corollary 5.10

0MK4

Let q be a power of 2. Assume S \subseteq {\mathbb F}_q and \lambda : {\mathbb F}_q \rightarrow S is a surjective map. Let \gamma \in {\mathbb F}_q be a 0-linear translator with respect to S for the map \lambda. Then for any G \in {\mathbb F}_q[x] which maps S into S, we have that f(x)= x + \gamma G(\lambda(x)) is an involution on {\mathbb F}_q.

Example

Example 5.11

W66W

Let q be a power of 2 and n>2 be any integer. Let S={\mathbb F}_q and \lambda(x) = \sum\limits_{1 \le i < j \le n}{ \beta_i( x^{q^i} + x^{q^j} ) } \in \mathbb{F}_{q^n}[x]. Clearly, each \gamma \in {\mathbb F}_{q} is a 0-linear translator with respect to {\mathbb F}_q for the map \lambda(x). Then for any G \in {\mathbb F}_{q^n}[x] which maps {\mathbb F}_q into {\mathbb F}_q, we have that f(x)= x + \gamma G \left( \sum\limits_{1 \le i < j \le n}{ \beta_i( x^{q^i} +x^{q^j} ) } \right) is an involution on {\mathbb F}_{q^n}.

Source and licence

The text on this page is from arXiv:2004.12552v2 by Tailin Niu, Kangquan Li, Longjiang Qu and Qiang Wang, under the licence CC BY SA 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY SA 4.0.

The record of each item is in analysis/objects/2004/2004.12552.jsonl.