This action might not be possible to undo. Are you sure you want to continue?
A new construction of bent functions based
Zbent functions
Sugata Gangopadhyayl, Anand , Gregor Leander
3
, and Rajendra Kumar
Sharma
2
Department of Mathematics, Indian Institnte of Technology Roorkee
Roorkee 247667 INDIA
gsugata@gmail.com
2 Department of Mathematics, Indian Institute of Delhi
New Delhi 110016 INDIA
{anadjoshi111, rksharmaiitd}@gmail.com
3 Department of lVlathematics, Technical of Denmark
DENMARK
g.leander@dtu.mat.dk
Abstract. Dobbertin has embedded the problem of construction of bent
functions in a recursive framework by using a of bent func:
lions called Zbent functions, his ideas. we generalize COll
struction of bent functions to partial spreads Zbent func
tions of arbitrary leveL Furthermore, ,ve show how these partial spreads
Zbent functions rise to a new construction of bent func
tions. \Ve underline the variety given by this construction
that all bent function in 6 variables can be constructed in this way.
Introduction
Bent functions. first, introduced in 9]. have maximal distance to t he set of an
affine function. This property, with connection [0 coding and
cryptography, makes them to study. Since the introduction
of bent functions substantial efforts have been directed towards their in
the last three decades. Many primary and secondary constructions are known
but a general understanding of bent functions is still missing. Even the set of all
8 variable bent functions could not be completely classified so far. vVe refer the
reader to the excellent survey chapter by Cadet for detailed information on
what is known about bent functions in general.
In the problem of constructing bent functions has been embedded into
a recursive framework by generalizing the notion of bent functions to integer
valued functions 'with certain properties.
MatheIllatics Subject Classification: 06E30, 94CIO
Key Words: Boolean functions; Zbent functions; Fourier transform.
2
Those functions are referred to as Zbent functions for level r. where r can be
any nonnegative integer. The union of all such functions is said to be the set of
Zbent functions and the classical bent function correspond to Zbent functions
of level 0.
Most importantly. Zbent functions of level r on n variables can be used to
construct all Zbent functions of level r  1 on n + 2 variables by a ··gluing"
technique. Continuing in this ,vay eventually all Zbent functions of level 0 on
n+2r variables are obtained are same as classical bent functions on n+2r
variables) .
The motivation for our work is thaI, one can to find new
construction of bent functions following this gluing process. For this, one could
first construct Zbent functions of level 1 and than demonstrate that those
can be glued together to classical bent functions.
After our notations and recalling some results on Zbent functions
in Section 2, we generalize the construction of partial spreads bent functions
to partial spreads Zbent functions of level T for any 2: 1 Section 3).
Based on a suitable subclass of those Zbent functions we give a new primary
construction of bent functions in Section 4. To demonstrate the
of bent functions that can be constructed this way, we argue that all
bent functions on 6 affine can be obtained our
construction (see Section 4.1
Preliminaries
Let lF
2
be the finite field with two elements and be the ndimensional vec
torspace over Any function from lFR into is said to be a Boolean function
on n variables. The set of all Boolean functions on n variables is denoted 3".
Let us denote the set of integers by Z. Suppose FE 3
n
.
Throughout this paper n = 2k is a positive integer.
The Walsh of a Boolean function F at a point a E: is defined
as
= L( 1
:rElF'!i
where is the canonical inner product on , that is :x:) = L ai:X:i. Sup
pose E is a vector subspace of lFR. Then E' := E lFR : y) = 0, for all y E
E}.
A Boolean function is called bent if for all a E l F ~ it holds that F
W
The identity
FW(a)
2
provides the link between the vValshtransforffi and the distance of the function
F to the linear function It follows from Parseval's identity
that 2
k
and therefore bent functions are exactly those
Boolean functions an even number of variables) that have the maximal dis
tance to the set of all affine functions.
2.1 From Bent to Zbent functions and back
In order to generalize bent functions to Zbent functions it is most convenient to
replace Boolean, that is {O. I} valued functions ± 1 valued functions. Given a
Boolean function F we consider the function
f: +{1.1}C;;Z
= (J
there is a onetoone correspondence between the functions defined in
these two ways. Throughout this paper we denote F, G Boolean functions and
./. g the integervalued function associated
with F and G. By abuse of tenninology we shall refer to these functions from
to L I} as Boolean functions as well.
For an integervalued more general real or complexvalued) function f
the Fourzer defined
1 (a.x)
is an important tool. Note the close relation to the \:Valshtransform given
vVith this notation at hand a function f is bent if and if both .f and j are
{1. 1}valued. hom this viewpoint, the generalization to Zbent function, to
be discussed below. is almost naturaL
In order to put the construction of bent functions in a recursive framework.
Dobbertin generalized the notion of bent functions to Zbent functions.
Consider the following sequence of subsets of Z
W
o
={l,l}
2
r 1
lV'r = E ZI  
1
:s; w :s; 2
r
 } for T > O.
Definition 1. A f : 7 W,. is said to be a Zbent size
k on n and level if and if j is also a tu''iCif,wn into
WT' The set all Zbent functions of size k and le1Jel T is denoted BF:. Arty
function belonging to Ur?oBF;c is said to be a Zbent fU:'iCl;w'n.
Next. we recall how Zbent function of size k and level r can be decomposed
into four Zbent function of size k  1 and level T + 1 Proposition 2 of
Suppose f E B F , ~ ,
and
yE
Define functions f" '2 as follows:
Case 1 For r :::: 1:
f10 \
(2\
(
j.
. )
f01 fll )
/
Case 2 For r = 0:
1 (1 1 \ (
h
iO
1
(
fll
)
=:2 i 1)
1: 11 /
. (3)
Proposition 2 in states that the functions '2 are Zbent functions of
size k 1 and level r + 1 is E BF:il
l
). for all (1. (2 C
Conversely. under certain conditions. it is to COllstrUe! bent functions
of size k: and level r from bent functions of size k1 and level r ' 1. This is referred
to as glu.ing in Theorem 3 of The theorem summarizes these
result in the special case of level 1 functions being glued together to a classical
hem function. Sate that our notation differs from the one in as we
avoid to introduce so called bent squares.
Theorem 1 ([4J, Theorem 3). Let fou.r Zocnt fU??CtW11S
level 1 and size k be gwen such that
+ 1 mod 2
== fOl
flO(.r) == fll (x) + 1 mod 2
+ 1 mod 2
+ 1 mod 2.
== fll
Then the .t'unctwn
h : !F
2
X x ! F ~ + {l.l}
z.
where
(h
Ol
)
ho)
IS a bent func/.ion (of feud 0).
Especialiy in light of this gluing process. the construction of Zbent functions
becomes an interesting problem.
3 Generalization of partial spreads bent functions to
partial spreads Zbent functions of arbitrary level
Let E be any subset of . The function (j) E defined
= {l if:c EO E
o ifxt/'E
is the indicator function of E. Suppose {E, :i = 1,2, ... ,s} is a set of mutually
disjoint kdimensional subspaces of JF2'. Here disjoint means E
i
n E
J
=
whenever i =F j. The partial spreads class of bent functions (PS) consists of
two subclasses PS" and PS+. These functions were first constructed Dillon
. The elements of PS' are those functions whose supports are unions of I
disjoint kdimensional subspaces of JF2' excluding 0, whereas the elements of PS+
are those whose supports are unions of 2
k
.. ] + 1 disjoint kdimensional subspaces
of . A function F EO En belonging to the class PS can be expressed as
=L
for all EO
1,=1
where s = ] if F E PS" and s = 2"] + 1 if F E PS+ and the sum is taken
over the integers.
Note in this construction is not efficient in sense that
is difficult to find those disjoint kdimensional subspaces. For n = 8 a complete
classification of all spreads bent funetior]s has been obtained in . How
ever, there is a special choice of subspaces where it is easy to ensure that all
subspaces are pairwise disjoint. This leads to the class of PS
ap
bent functions
and is explained next.
Here, we consider the function from to instead, that is we consider
the finite field with 2
n
elements instead of the vector space only. Furthermore,
let = iF2" , subfield of order 2
k
of , and Vi = ('JF2" for all i = I, , 2
k
,
where ( is a primitive element of iF
2
". the set S = : i = 0, , 2
k
}
consists of kdimensional subspaces of . A subclass of PS
type bent functions, called PS
ap
, is obtained constructing functions whose
supports are union of any 2
k
] subspaces belonging to S excluding O.
Based on the parial spread class of bent function we propose a construction
technique of Zbent functions of level T 2: 1. on variables. However it should
be noted that the functions considered so far in this section are from iF2' into
, whereas the functions considered below are integer valued, that is functions
from JF2' into Z, the set of integers.
Theorem 2. Let 'In], 'ln2, ' .. ,'In, be integers and E], E
2
, ... ,E
s
be kdimensional
subspaces of iF2', then the
=L
,=1
is a Zbent and its dual is
Proof. Ali we have to show is that both f and .l are integer valued. For .f this
is dear definition. The Fourier transform of f at a E is as f()lJows
1
1=1
Thus, i is also an integer valued function and the result follows.
\Vhile the theorem is it is difficult to exact
level of the Zbent functions constructed. In order to be able to construct bent
functions of a specific level the following techl1icallemma is usefuL Note that,
we skip the proof in this extended abstract.
Lemma 1. Let U and V be kdirnension S'uOSj')al;eS such that
Un V = {O}
U'nV' =
\Vith this lemma at hand we now can ('onstrucl bent functions of any
level.
Corollary 1. Suppose {El: i = 1,2", . ,s} is a set of kdirnensional SUi/sp/lee
of with the pToperty that E
i
n E
j
= {O} wheneveTi io j. The
s
= LTni/PE, all x E (8)
i=1
wheTe Tni E faT all 'i = 1,2, .... s, is a Zbent level . any
r ::;, 1. 'if and only if Tni E Wr .
Proof. From Theorem 2 we already know that f is a Zbent function. To prove
that it is a bent function oflevel r we have to show that E W
r
and /(a) E W
r
for all er, a E \Vhile the first part follows immediately from the conslnIction
of.f we elaborate a bit on the second part.
4
Applying Theorem 2 yields that = Z:=1 and Lemma 1 im
plies that the kdimensional spaces Et are pairwise disjoint. Thus for a 1= 0 we
conclude
E {m, I 1 :s. i :s. U{O} :;;
and 1(0) = Z;'=I Tn, E W, by construction.
Remark 1. It is to be noted that in the binary case we obtain bent functions if
and only if s E pkI, 2
k

1
+ I} whik the imeger valued case is :;:ore flexible.
\Ve refer to these integer valued functions as PS type 2bent functions of level
/'.
Remark 2. It is possible to construct the analogue of P Sap type bent function
by considering the functions of the form
2"
for all x E IF2'
= LmL 0 V,
.
i=O
where V
t
= for all = 0..... 2'". E W, for all i = 1. 2.... ,2'" and
1n
L
E vVr . We shall refer to these functions as PS
ap
Zbent functions of
level r. In the next section we use those functions together with Theorem 1 to give
a new general construction of bent functions. In particular, we demonstrate that
all the bent functions on 6 variables. up co affine ca.n be constructed
by "gluing" P Sop type 2bent functions of level 1 on 4 variables.
Remark 3. It is to be noted that for each i E {O. 1..... 2
k
} there exists j E
{O.• 1,
'
. . . .
2k }." r ,. ,
= v'J'
T' ,. I" d· . f" DS'
ap type jL,
b
ent SUCll tllat
T •
lleretore t le ual 0, a
'17
L
function of level r is also a PSap type 2bent function of level r.
A New Construction of Bent functions
In this section ,'le describe a new construction of (classical) bent functions based
on the bent functions of level 1 presented in Corollary 1 and the gluing Theorem
1.
'While, in general. the conditions given in Theorem 1 on the
functions fij or on their duals is easy. all four conditions at the same
time scems diflicult. However, bem functions of level 1 described in Corollary I,
lead to a special case where things are a lot easier.
We start letting S = {Si} be a spread, i.e. a collection of 2
k
tl subspaces
of dimension k with the condition that
Sj r'i Si = {O} and U, S, = .
Next, we partition this spread S into two pans, A and [3, .e. A,, [3 = I{; and
AuB = S and select four collections of coefficienTs. each in 1,1}
such that L mA E {I.O.I}
such that L m ~ E {l. CL I}
IBEE such that LlIBE{1.0.1}
(n'a)BEE such that Ln'a E {1.0.i}.
Giver: those coefficients. we are to construct our four Z bent functions of
level 1 as f01l0ws
=L
AEA
= L nBQB
BEE
=L
BEE
=L
AEA
In order to Theorem I we have to verify that the four conditions to
are fulfilled. To the first condition. let :£ E be given. vVe compute
+ = L rn.'l C).A.
AEA
(mod 2)
AeA
(mod 2)
S,ES
If x cl 0 then. as S is a spread, there exists exactly one 8ubspace Sk such that
x EO Sk and
= 9s,. = I (mod
+
S,ES
On the other hand. if ,1: = 0 than
foo(O) + fcn = L 1 = 2
k
+ 1 = 1 (mod 2).
S,ES
The other conomons follow in a verv similar way. In particular conditions
(6) and (7) follow from the fact that, due to Corollary 1. the duals Z; of fi; are
again of the same type so the condition carries over to the duals nicely.
4.1 Construction of all 6 variable bents up to affine equivalence
In this section we consider PS
ap
type Zbent functions of level 1 on 4 variables.
Let ( be a root of the primitive polynomial + x + 1 on The finite field
lF
2
4 = {C : i = O. L .... 14} u {O}.
In this case the elements of S can be explicitly written as follows.
vc) = {0.1,(5, V
{O ( ;6 } Tl  [0 ,2
1 i,,''''' V2l''':l'
V3 = {O. (3. (8
= {O ;4 ;9. (14}
V
4 ~ 'l 1'" . ..., .
\Ve generate all the PS
ap
type Zbent functions of level 1 on 4 variables and
"glue" them to construct bent functions on 6 variables. Then by exhaustive
search we find seven PSap type Zbent functions of level 1 on 4 variables such
that all the 6variable bent functions up to affine equivaJence can be generated by
gluing them. We denote these functions by gO,.91,g2 . .93 •.94•.95.g6 and list them
below.
=0.
+ + QV.
+
,
+
T
+ +
+ +
+ 0 , ~ T
The truth tables of the the above functions are as follows.
Igol 0 0 () 0 () 0 () 0 0 () () 0 0 0 0 01
1.911 1 1 0 0 0 0 1 1 0 0 0 0 0 0 0 0
1
1.921 ~ 1 1 1 1 1 1 I l 1 1 I I I 1 11
1.931 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 11
.941 0 0 1 1 1 1 0 0 1 1 1 1 1 1 1 11
.95 !1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1/
.961 0 0 1 1 1 1 0 0 I 1 1 1 1 1 1 11
By "gluing" them we obtain four bent functions on 6 variables which can be
demonstrate to be affine nonequivalent. by using the weight distributions of the
second derivative spectrum introduced in . Since by [9] it is known there are
four bent functions on 6variables. up to afIine equivalence. this proves that
all the bent functions up to affine equivalence are obtained our construction.
Let us denote the four functions obtained is this way
Define
, E2. yE for i = L 2. 3. 4.
The functions are defined as follows:
(
(
) Cl )
/
= (1 1)\ (g0.93)
) 1 1 g2 go
5
(1 1 \ (gO 095 \
( ) \ 1 1) .92 gO)
\) Ch.96 \ .
( )=c
.94 g1)
The truth tables of the functions are as follows
= 0110001100001010100111001111010101100011000010100110001100001010
= 0110001100001010100111001111010101001011000100010100101100010001
= 0110001100001010100111001111010111101011000110111110101100011011
=0010100000011011000101001110010000100100100011101110011110001110
It can be directly checked that H{i" for i = 1, 2, 3, 4 are bent functions. Moreover,
it is not hard to show (for example using the techniques presented in tha1, all
those functions are pairwise inequivalent. Thus we obtain all the bent functions
on 6 variables up to affine equivalence.
Conclusion
III this paper we generalize the PS type bents to PS type :2bent func1 ions of
level r for any r ?: 1. vVe also identify the natural analogue of the class P Sap.
Finally we demonstrate that all the 6variable bent functions can be constructed
by "gluing" P Sap type bent functions of level 1 on 4 variables.
References
1. C. Cadet, "Generalized partial spreads", IEEE Tmns. Inforrn. Theory 41 (1995),
14821487.
2. C. CarJet, P. Guillot, "A characterization of binary bent functions", J. Comb.
Theory, Ser. A 76 328·335.
3. C. Cadet, Boolean functions for cryptography and error correcting codes. In: Y.
Crama, P. Hammer Boolean Methods and Models, Cambridge Univ. Press,
Cambridge. Available: http://wwwroc.inria.fr/secre1/Claude.Carlet/pubs.htmL
4. H. Dobbertin, G. Leander.. "Bent functions embedded into the recursive framework
of Zbent functions", Des. Codes Cryptogr. 19 (2008). 3·22.
.:3. J. F. Di11on. "Elemcmary Hadarnard difference sets", Proceedings of Sixth S. E.
Conference of Com6ma/oT'lcs, Graph Theory, and Computing, Utility Mathematics,
Winnipeg, (1975). 237249.
6. S. Gangopadhyay, D. Sharma, S. Sarkc.r and S. r..laitm, "On c.ffine (!l(}!l.iequi'>'alen,ce
of Boolean functions", Computing 85 (2009) :1755.
7. P. Langevin and X.D. Hou, "Counting Partial Spread Functions in Eight Vari
ables" , preprint (2010)
8. F. J. MacWiIliams, N. J, A. Sloane, The theory of cl'Torc07Tecting codes. North
Holland, Amsterdam. New York, Oxford, 1977.
9. 0, S. Rothaus, "On bent functions", J. Comb. Theory, Ser. A 20 (1976), 300305.