1

Binary Polar Codes Based on Bit Error Probability

Jun Muramatsu Senior Member, IEEE

Abstract
This paper revisits the theory of binary polar source/channel codes based on the bit error probability of successive-cancellation
decoding. The polarization theorem is introduced based on the bit error probability and then techniques to compute the bit error
probability are introduced. These techniques can be applied to the construction of polar codes and the computation of an upper
bound of the block decoding error probability.

Index Terms
binary polar codes, source coding with decoder side information, channel coding, successive-cancellation decoding, bit error
probability

I. I NTRODUCTION
arXiv:2101.08943v1 [cs.IT] 22 Jan 2021

Polar source/channel codes were introduced by Arıkan [2], [3], [4]. When these codes are applied to source coding with
decoder side information for joint stationary memoryless sources, the coding rate achieves a fundamental limit called the
conditional entropy. When applied to the channel coding for a symmetric channel, the coding rate achieves a fundamental limit
called the channel capacity. Arıkan introduced successive-cancellation decoding, which can be implemented with computational
complexity O(N log2 N ) with respect to block length N .
One theoretical element of polar codes is the notion of polarization. The polarization theorem is usually introduced using
mutual information, conditional entropy, and the Battacharyya parameters [2], [3], [14].
In this paper, we reconstruct the polarization theorem using the bit error probability of successive-cancellation decoding
instead of the Battacharyya parameter. A similar approach [1, Chapter 5] was introduced for the analysis of the likelihood
ratio of the polar channel code. In contrast, we directly analyze the bit error probability to simplify the analysis of the block
decoding error probability and introduce techniques to compute the bit error probability. With these techniques, we can directly
evaluate the upper bound of the block decoding error probability, while conventional techniques used conditional entropy,
mutual information, or the Battacharyya parameter. Furthermore, we introduce a novel loss function for a greedy degrading-
/upgrading-merge algorithm and evaluate the degrading/upgrading loss after a single polar transform. With this technique, we
can reduce the computational complexity of the construction of polar codes. When the joint probability distribution consists of
rational numbers, the computation can also be precisely performed using the arithmetic of rational numbers if sufficient memory
is available. In contrast, we had to focused on the precision of real numbers to compute mutual information in conventional
techniques.
The rest of this paper is organized as follows. Section II introduces the definitions and notations. Section III reviews polar
codes based on bit error probability. Section IV introduces their construction based on bit error probability, where a novel loss
function for the greedy degrading-/upgrading-merge algorithm is introduced. Section V introduces experimental results on the
computation of bit error probability. Theorems and lemmas are proved in Section VI. Some lemmas are shown in Appendix.

II. D EFINITIONS AND N OTATIONS

Throughout this paper, we use the following definitions and notations.
For given n, let 2n denote the block length. We use the bit-indexing approach introduced in [2]. The indexes of a 2n -
n
dimensional vector are represented by n-bit sequences as X 2 ≡ (X0n , . . . , X1n ), where 0n /1n is the n-bit all zero/one
n
sequence. For given I ⊂ {0, 1}n, we define the sub-sequences of X 2 as
XI ≡ {Xbn }bn ∈I .
To represent an interval of integers, we use notations
[0n : bn ] ≡ {0n , . . . , bn }
[0n : bn ) ≡ [0n : bn ] \ {bn }
[bn : 1n ] ≡ [0n : 1n ] \ [0n : bn )
(bn : 1n ] ≡ [0n : 1n ] \ [0n : bn ].

J. Muramatsu is with NTT Communication Science Laboratories, NTT Corporation, 2-4, Hikaridai, Seika-cho, Soraku-gun, Kyoto 619-0237, Japan (E-mail:
jun.muramatsu@ieee.org).
 2

Let bbk , bk b ∈ {0, 1}k+1 be the concatenations of b ∈ {0, 1} and bk ∈ {0, 1}k . For given b ∈ {0, 1} and bk ∈ {0, 1}k , define
b[0 : bk ], b[0 : bk ) ⊂ {0, 1}k+1 as
b[0 : bk ] ≡ {bb′k : b′k ∈ [0k : bk ]}
b[0 : bk ) ≡ {bb′k : b′k ∈ [0k : bk )}.
Bipolar-binary conversions ±b and ∓b of b ∈ {0, 1} are defined as
(
+ if b = 1
±b ≡
− if b = 0
(
− if b = 1
∓b ≡
+ if b = 0.

III. B INARY P OLAR C ODES

In this section, we revisit the binary polar source/channel codes introduced in previous works [2], [3], [4], [14].
Assume that X = {0, 1} is the binary finite field. For given positive integer n, polar transform G is defined as
 ⊗n
1 0
G≡ ΠBR ,
1 1
where ⊗n denotes the n-th Kronecker power and ΠBR is the previously defined bit-reversal permutation matrix [2]. Then
n n
vector u ∈ X 2 is defined as u ≡ xG for given vector x ∈ X 2 . Let {I0 , I1 } be a partition of {0, 1}n, satisfying I0 ∩ I1 = ∅
and I0 ∪ I1 = {0, 1}n. We will define {I0 , I1 } in Section III-C.
Let X ≡ (X0n , . . . , X1n ) and Y ≡ (Y0n , . . . , Y1n ) be random variables and let U ≡ (U0n , . . . , U1n ) be a random variable
defined as U ≡ XG. Then joint distribution PUI0 UI1 Y of (UI0 , UI1 , Y ) is defined using joint distribution PXY of (X, Y )
as
PUI0 UI1 Y (uI0 , uI1 , y) ≡ PXY ((uI1 , uI0 )G−1 , y),
where the elements in (uI1 , uI0 ) are sorted in index order before operating G−1 .
Let PUbn |U[0n :bn ) Y be the conditional probability distribution, defined as
P
u(bn :1n ] PUI0 UI1 Y (uI0 , uI1 , y)
PUbn |U[0n :bn ) Y (ubn |u[0n :bn ) , y) ≡ P . (1)
u[bn :1n ] PUI0 UI1 Y (uI0 , uI1 , y)
n
For vector uI1 and side information y ∈ Y 2 , output u
b ≡ f (uI1 , y) of successive-cancellation (SC) decoder f is defined
recursively (
u[0n :bn ) , y) if bn ∈ I0
fbn (b
bbn ≡
u
ubn if bn ∈ I1
using functions {fbn }bn ∈I0 defined as
fbn (u[0n :bn ) , y) ≡ arg max PUbn |U[0n :bn ) Y (ubn |u[0n :bn ) , y),
ubn ∈U

which is the maximum a posteriori decision rule after observation (u[0n :bn ) , y).
n n
For a polar source code (with decoder side information), x ∈ X 2 is a source output, uI1 is a codeword, and y ∈ Y 2 is
a side information output. The decoder reproduces source output x b ≡ f (uI1 , y)G−1 from codeword uI1 and y. The (block)
−1
decoding error probability is given as Prob(f (UI1 , Y )G 6= X).
For a systematic polar channel code [4], let I0′ and I1′ be defined as
I0′ ≡ {b0 b1 · · · bn−1 : bn−1 · · · b1 b0 ∈ I0 } (2)
I1′ ≡ {b0 b1 · · · bn−1 : bn−1 · · · b1 b0 ∈ I1 } (3)
for given (I0 , I1 ). We assume that encoder and decoder share a vector uI1 . The encoder computes (xI1′ , uI0 ) from a message
xI0′ and a shared vector uI1 so that (xI1′ , xI0′ ) = (uI1 , uI0 )G−1 , where the elements in (xI1′ , xI0′ ) and (uI1 , uI0 ) are sorted in
index order before operating G−1 . Then the encoder generates channel input x ≡ (xI0′ , xI1′ ), where the elements in (xI1′ , xI0′ )
n
are sorted in index order. The decoder reproduces a pair of vectors (uI1 , u bI0 ) ≡ f (uI1 , y) from channel output y ∈ Y 2
and shared vector uI1 . Then x b≡u b G−1 is reproduced and x bI0 is a reproduction of the message. The (block) decoding error
probability is also given as Prob(f (UI1 , Y )G−1 6= X).
For a non-systematic polar channel code, uI0 is a message and vector uI1 is shared by the encoder and decoder. The
n
encoder generates channel input x ∈ X 2 as x ≡ (uI1 , uI0 )G−1 , where the elements in (uI1 , uI0 ) are sorted in index
order before operating G−1 . The decoder reproduces a pair of vectors as (uI1 , u bI0 ) ≡ f (uI1 , y) from channel output y ∈
 3

n
Y 2 and shared vector uI1 . Then u bI0 is a reproduction of the message. The (block) decoding error probability is given as
Prob (f (UI1 , Y ) 6= (UI0 , UI1 )).
We have the following lemma, which can be shown similarly to a previous proof [2].
Lemma 1 ([11, Lemma 2],[15, Eq. (1)]):
Prob(f (UI1 , Y )G−1 6= X) = Prob (f (UI1 , Y ) 6= (UI0 , UI1 ))
X
≤ Prob(fbn (U[0n :bn ) , Y ) 6= Ubn ).
bn ∈I0

In the following, we call Prob(fbn (U[0n :bn ) , Y ) 6= Ubn ) the bit error probability.
In [2], [3], [14], the decoding error probability is upper bounded in terms of Battacharyya parameter Z, where relation
Prob(fbn (U[0n :bn ) , Y ) 6= Ubn ) ≤ Z(Ubn |U[0n :bn ) , Y )
is implicitly used in the proof. The above lemma provides a tighter upper bound than the Battacharyya parameter.
From the above lemma, we can make a decoding error probability close to zero by specifying index set I0 so that
Prob(fbn (U[0n :bn ) , Y ) 6= Ubn ) is close to zero for all bn ∈ I0 . One of our goals is computing Prob(fbn (U[0n :bn ) , Y ) 6= Ubn )
for every bn ∈ {0, 1}n to improve the upper bound of the decoding error probability.

A. Symmetric Parametrization of Joint Sources

This section introduces the symmetric parametrization of joint sources to simplify the analysis of the bit error probability.
This approach is introduced in [1, Chapter 5] to analyze the likelihood ratio of the polar channel code. Constructions of the
successive-cancellation decoder and the successive-cancellation list decoder are introduced based on this approach [12].
Let (U, V ) be a pair of random variables on U × V, where U ≡ {0, 1}. Let PV , and PU|V be the distribution of V and the
conditional distribution of U for given V . We define a list P(U |V ) ≡ {(µv , θv )}v∈V as
µv ≡ PV (v) (4)
θv ≡ PU|V (0|v) − PU|V (1|v), (5)
where θv ∈ [−1, 1]. We have relation
1 ∓ u θv
PU|V (u|v) = (6)
2
from the definition of θv and relation PU|V (0|v) + PU|V (1|v) = 1, where ∓u is the bipolar-binary conversion of u. Then list
P(U |V ) represents the joint probability distribution of (U, V ). Note that expression {·} does not represent a set; it represents
a list. That is, we treat (µv , θv ) and (µv′ , θv′ ) differently when v 6= v ′ . Furthermore, when bijection π : V → V exists, we
identify list P(U |π(V )) ≡ {(µπ(v) , θπ(v) )}v∈V with P(U |V ) ≡ {(µv , θv )}v∈V . For this reason, we denote P ≡ P(U |V ) to
represent an identical list when the corresponding random variables are not essential. We denote1 P ′ = P when P ′ is identical
with P in the sense mentioned above. Note such an equivalence was previously introduced [7, Definition 3], [10, Definition
4].
When we guess the outcome of U by observing that for V ’s outcome, the decision error probability is defined as
XX
Error(f ) ≡ PU|V (u|v)PV (v)χ(f (v) 6= u)
u∈U v∈V

for given decision rule f , where χ is the support function defined as

(
1 if the statement is true
χ(statement) ≡
0 if the statement is false.
The maximum a posteriori (MAP) decision rule
fMAP (u|v) ≡ arg max PU|V (u|v)
u∈U

minimizes the decision error probability. The MAP decision error probability Error(U |V ) is given as
Error(U |V ) ≡ Error(fMAP )
X
= PV (v) min{PU|V (0|v), PU|V (1|v)}
v∈V
X µv [1 − |θv |]
=
2
v∈V

P
1 In [10], [7], such equivalence is denoted by P ′ ∼ P. We use notation P ′ = P because P and P ′ are identical as multisets.
 4

" #
1 X
= 1− µv |θv |
2
v∈V
1
= [1 − M (P(U |V ))] , (7)
2
where X
M (P) ≡ µv |θv |, (8)
v∈V

and the second equality comes from (6). Note that M (P) is invariant under any permutation π : V → V. Symmetric
parametrization is a binary version of the Fourier transform of the a posteriori probability [10, Definitions 24 and 25].
In the following, we focus on value M (P).
Remark 1: Value M (P(U |V )) equals the variational distance between joint distribution PUV and distribution PUe × PV ,
where Ue is the uniform distribution on U. To show this fact, we assume PU|V (0|v) ≥ 1/2 ≥ PU|V (1|v) without loss of
generality. Then we have
1X  
M (P(U |V )) = PV (v) PU|V (0|v) − PU|V (1|v)
2
v∈V
1X  
= PV (v) PU|V (0|v) − PUe (0) + PUe (1) − PU|V (1|v)
2
v∈V
1XX  
= PV (v) PU|V (u|v) − PUe (u)
2
v∈V u∈U
1 X X  
= PUV (u, v) − PUe × PV (u, v) , (9)
2
v∈V u∈U

where the first equality comes from (4), (5), (8), and the assumption, the second and third equalities come from the assumption
and the fact that PUe (0) = PUe (1) = 1/2, and the right hand side of the last inequality is the definition of the variational
distance. Note that M (P(U |V )) is also relevant to the polar lossy source codes and the polar channel codes for asymmetric
channels [8], [9].

B. Plus and Minus Operations

In this section, we define plus and minus operations based on symmetric parametrization.
In the basic polar transform, a pair of binary random variables (U0 , U1 ) is transformed into B0 ≡ U0 ⊕ U1 and B1 ≡ U1 ,
where ⊕ denotes the addition on the binary finite field. Assume that random variables U0 , U1 , V0 , V1 satisfy that U0 , U1 ∈ {0, 1},
V0 , V1 ∈ V, and (U0 , V0 ) and (U1 , V1 ) are independent, but Ub and Vb are allowed to be correlated for each b ∈ {0, 1}.
Regarding probability distribution PV 2 and conditional probability distribution PB0 |V 2 , we have
PV 2 (v0 , v1 ) = µv0 µv1 (10)
1 ∓b0 θv0 θv1
PB0 |V 2 (b0 |v0 , v1 ) = , (11)
2
where ∓b0 is bipolar-binary conversion of b0 . To describe the joint probability distribution of (B0 , V 2 ), we introduce minus
operation P − ≡ {(µ− −
v0 ,v1 , θv0 ,v1 )}(v0 ,v1 )∈V×V as

µ−
v0 ,v1 ≡ µv0 µv1 (12)
θv−0 ,v1 ≡ θv0 θv1 . (13)
Then we have relation
P − = P(U0 ⊕ U1 |V0 , V1 ). (14)
Regarding probability distribution PB0 V 2 and conditional probability distribution PB1 |B0 V 2 , we have
µv0 µv1 [1 ∓b0 θv0 θv1 ]
PB0 V 2 (b0 , v0 , v1 ) =
2
1 ∓b1 [θv1 ∓b0 θv0 ]/[1 ∓b0 θv0 θv1 ]
PB1 |B0 V 2 (b1 |b0 , v0 , v1 ) = ,
2
where ∓b0 and ∓b1 are the bipolar-binary conversions of b0 and b1 . To describe the joint probability distribution of (B1 , (B0 , V 2 )),
we introduce plus operation P + ≡ {(µ+ +
u,v0 ,v1 , θv,v0 ,v1 )}(v,v0 ,v1 )∈U ×V×V as

µv0 µv1 [1 ∓u θv0 θv1 ]

µ+
u,v0 ,v1 ≡ (15)
2
 5

+ θv1 ∓u θv0
θu,v0 ,v1
≡ , (16)
1 ∓u θv0 θv1
where we define (µ+ +
u,v0 ,v1 , θu,v0 ,v1 ) ≡ (0, 0) when 1 ∓u θv0 θv1 = 0. Then we have relation

P + = P(U1 |U0 ⊕ U1 , V0 , V1 ). (17)

(n)
Recursively define Ubn as
U (0) ≡ X (18)
(k+1) (k) (k)
Ubk 0 ≡ U0bk ⊕ U1bk (19)
(k+1) (k)
Ubk 1 ≡ U1bk , (20)
(k) (k) (k) (n) n
where U0bk and U1bk are two independent copies of Ubk . Then we have U[0n :1n ] = X 2 G, which is the polar transform of
n
X 2 . Recursively define Y[0k :bk ] as
Y (0) ≡ Y
 
(k+1) (k) (k)
Y[0k+1 :1k+1 ] ≡ Y0[0k :1k ] , Y1[0k :1k ] ,
(k) (k) (k)
where Y (0) ≡ Y is correlated with U (0) ≡ X, and Y0[0k :1k ] and Y1[0k :1k ] are two independent copies of Y[0k :1k ] . Then we have
     
(k+1)  (k+1) (k+1) (k+1)  (k) (k) (k) (k)
P Ubk 0  U[0k+1 :bk 0) , Y[0k+1 :1k+1 ] = P Ubk 0  U0[0k :bk ) , Y0[0k :1k ] , U1[0k :bk ) , Y0[0k :1k ]
  
(k) (k)  (k) (k) (k) (k)
= P U0bk ⊕ U1bk  U0[0k :bk ) , Y0[0k :1k ] , U1[0k :bk ) , Y0[0k :1k ]
  −
(k)  (k) (k)
= P Ubk  U[0k :bk ) , Y[0k :1k ] , (21)
where the first equality comes from the fact that we can construct a bijection
   
(k) (k) (k) (k) (k+1) (k+1)
π U0[0k :bk ) , Y0[0k :1k ] , U1[0k :bk ) , Y0[0k :1k ] ≡ U[0k+1 :bk 0) , Y[0k+1 :1k+1 ]
using relations (19) and (20), the second equality comes from (19), and the last one comes from (14). Similarly, we have
     
(k+1)  (k+1) (k+1) (k+1)  (k+1) (k) (k) (k) (k)
P Ubk 1  U[0k+1 :bk 1) , Y[0k+1 :1k+1 ] = P Ubk 1  Ubk 0 , U0[0k :bk ) , Y0[0k :1k ] , U1[0k :bk ) , Y0[0k :1k ]
  
(k)  (k) (k) (k) (k) (k) (k)
= P U1bk  U0bk ⊕ U1bk , U0[0k :bk ) , Y0[0k :1k ] , U1[0k :bk ) , Y0[0k :1k ]
  +
(k)  (k) (k)
= P Ubk  U[0k :bk ) , Y[0k :1k ] , (22)
where the second equality comes from (19) and (20), and the last one comes from (17). From (21) and (22), we have
     ±b
(k+1)  (k+1) (k+1) (k)  (k) (k)
P Ubk b  U[0k+1 :bk b) , Y[0k+1 :1k+1 ] = P Ubk  U[0k :bk ) , Y[0k :1k ] , (23)

where ±b is the bipolar-binary conversion of b.

C. Polarization
In this section, we revisit the polarization theorem. We prepare some lemmas, where proofs are given in Sections VI-A–VI-C.
Lemma 2: Let P ≡ {(µv , θv )}v∈V . Then we have
X X
M (P − ) = µv0 µv1 |θv0 θv1 | (24)
v0 ∈V v1 ∈V
X X
+
M (P ) = µv0 µv1 max{|θv0 |, |θv1 |}. (25)
v0 ∈V v1 ∈V

Lemma 3: Let P ≡ {(µv , θv )}v∈V . Then we have

M (P − ) ≤ M (P) ≤ M (P + ) (26)
− 2
M (P ) = M (P) (27)
− +
M (P ) + M (P ) ≤ 2M (P). (28)
Lemma 4: For P(U |V ) ≡ {(µv , θv )}v∈V , define h(ξ) and H(U |V ) as
h(ξ) ≡ −ξ log2 (ξ) − [1 − ξ] log2 (1 − ξ)
 6

X  
1 − |θv |
H(U |V ) ≡ µv h .
2
v∈V

Then  
1 − M (P(U |V ))
0 ≤ 1 − M (P(U |V )) ≤ H(U |V ) ≤ h ≤ 1.
2
From this inequality, we have
H(U |V ) → 1 ⇔ M (P(U |V )) → 0 (29)
H(U |V ) → 0 ⇔ M (P(U |V )) → 1. (30)
Recursively define joint distribution P ±1 ···±n as
  ±n
P ±1 ···±n ≡ P ±1 ···±n−1 ,
where (±1 , . . . , ±n ) is a sequence of ‘+’ or ‘−‘ operators. Then from (7) and (23), we have
1 
Prob(fbn (U[0n :bn ) , Y ) 6= Ubn ) = 1 − M (P(Ubn |U[0n :bn ) , Y[0n :1n ] ))
2
1 
= 1 − M (P ±b1 ···±bn ) (31)
2
by assuming that joint source {(Xbk , Ybk )}bk ∈{0,1}k is stationary memoryless subject to joint distribution PXY , where M is
defined by (8), P ≡ {(µy , θy )}y∈Y is defined as
µy ≡ PY (y)
θy ≡ PX|Y (0|y) − PX|Y (1|y),
and ±bi is the bipolar-binary conversion of bi for each i ∈ {1, . . . , n}. Then we have
   
Prob lim H(P ±1 ···±n ) = 1 = 1 − Prob lim H(P ±1 ···±n ) = 0 = H(X|Y )
n→∞ n→∞
2
from [14, Theorem 2.4] , where (±1 , . . . , ±n ) is the sequence of the stationary memoryless random variables subject to uniform
distribution on {−, +}. Furthermore, from (29) and (30), we have
   
Prob lim M (P ±1 ···±n ) = 0 = 1 − Prob lim M (P ±1 ···±n ) = 1 = H(X|Y ). (32)
n→∞ n→∞

Then from the above equalities, (27), (28), (32), and [5, Theorem 1],[14, Lemma 2.10], we have the following polarization
theorem, where the Battacharyya parameter in [14, Theorem 4.10] is replaced by M , which is related directly to the bit error
probability. Note that it is unnecessary to introduce the Battacharyya parameter for the proof of this theorem.
Theorem 1: Define I0 and I1 as
n n nβ
o
I0 ≡ bn ∈ {0, 1}n : M (P ± ) ≥ 1 − 2−2
n n nβ
o
I1 ≡ bn ∈ {0, 1}n : M (P ± ) < 1 − 2−2 ,

where ±n ≡ (±b1 , . . . , ±bn ). Then we have

|I0 |
lim = 1 − H(X|Y ) (33)
n→∞ 2n
|I1 |
lim = H(X|Y ) (34)
n→∞ 2n
n
for any β ∈ [0, 1/2). Note that I0 corresponds to the low entropy set (H(P ± ) → 0) and I1 corresponds to the high entropy
n
set (H(P ± ) → 1).
For the polar source code, (34) implies that encoding rate |I1 |/2n approaches H(X|Y ), suggesting that the fundamental
limit is achievable with the polar source code. For the polar channel code, (33) implies that encoding rate |I0 |/2n approaches
1 − H(X|Y ) = I(X; Y ) by assuming that PX is the uniform distribution on {0, 1}. This implies that symmetric capacity can
be achieved using the polar channel code. From Lemma 1 and (31), we can evaluate the upper bound of the decoding error
probability:
Prob(f (UI1 , Y )G−1 6= X) = Prob (f (UI1 , Y ) 6= (UI0 , UI1 ))
X
≤ Prob(fbn (U[0n :bn ) , Y ) 6= Ubn )
bn ∈I0

2 When Hn ≡ H(P ±1 ···±n ) and H ± is defined by [14, Eq. (2.9)], we have the relation [14, Eq. (2.11)] from (14), (17), and the definition of H(P).
 7

X 1 
= 1 − M (P ±b1 ···±bn ) , (35)
n
2
b ∈I0

which goes to zero as n goes to infinity by letting I0 and I1 be defined as those in Theorem 1.

IV. C ONSTRUCTION OF P OLAR C ODES BASED ON B IT E RROR P ROBABILITY

PBased on (35), we expect that the decoding error probability is minimized by establishing set I0 so that it maximizes value
±n n
n
b ∈I0 M (P ), where ±n ≡ (±b1 , . . . , ±bn ). To this end, it is sufficient to compute value M (P ± ) for all bn ∈ {0, 1}n .
n
However, the computation seems difficult because the cardinality of P ± grows rapidly with respect to block length 2n . In
fact, when P is given, we have |P − | = |P|2 and |P + | = 2|P|2 from the definition of P ± .
In the following, we introduce two techniques to reduce the computational complexity. The first one reduces the cardinality
n n
of set P ± , where M (P ± ) is invariant under this reduction. In the second, we introduce a novel quantization loss function
n
to obtain an approximation of M (P ± ).

A. Canonical Representation of Joint Probability Distribution

n
In this section, we introduce the reduction of list P, where M (P ± ) is invariant under this reduction.
First, we focus on the fact that M is an even function. We introduce operator A applied to list P ≡ {(µv , θv )}v∈V . The
proof is given in VI-D.
Lemma 5: Define P A as
P A ≡ {(µv , |θv |) : v ∈ V} .
Then we have
M (P) = M (P A ).
Next we introduce the following lemma. The proof is given in VI-E.
Lemma 6: Let P A±A and P ±A be defined as
P A±A ≡ [[P A ]± ]A
P ±A ≡ [P ± ]A ,
where ± denotes a ‘+’ or ‘−’ operator and the same operator appears on both sides. Then we have
P A±A = P ±A .
Next we introduce another operator, Σ, which is applied to P to reduce set V. The proof is given in VI-F.
Lemma 7: Define P Σ as ( ! )
X
Σ
P ≡ µv , θ : θ ∈ Θ , (36)
v:θv =θ

where
Θ ≡ {θv : v ∈ V s.t. µv > 0}. (37)
Then we have
M (P) = M (P Σ ).
Next we show the following lemma. The proof is given in VI-G.
Lemma 8: Let P Σ±Σ and P ±Σ be defined as
P Σ±Σ ≡ [[P Σ ]± ]Σ
P ±Σ ≡ [P ± ]Σ ,
where ± denotes a ‘+’ or ‘−’ operator and the same operator appears on both sides. Then we have
P Σ±Σ = P ±Σ .
Next we introduce the following lemma regarding operations A and Σ. The proof is given in VI-H.
Lemma 9:
P ΣAΣ = P AΣ .
Here we introduce the following theorem. The proof is given in VI-I.
Theorem 2: Define operator C as
C ≡ AΣ.
 8

Then we have
M (P C ) = M (P) (38)
C±C ±C
P =P (39)
C± C±C ±
M (P ) = M (P ) = M (P ), (40)
where ± denotes a ‘+’ or ‘−’ operator and the same operator appears on both sides.
By repeatedly applying the above theorem, we have
P C±1 C±2 C···C±n C = P ±1 C±2 C···C±n C
= P ±1 ±2 C···C±n C
..
.
= P ±1 ±2 ···±n C (41)
and
M (P C±1 C±2 C···C±n C ) = M (P ±1 ±2 ···±n C )
= M (P ±1 ±2 ···±n ), (42)
where we can reduce parameter set V to Θ by applying operation Σ. This implies that computing M (P C±1 C±2 C···C±n C ) might
be easier than directly computing M (P ±1 ±2 ···±n ).
Based on (39), call C the canonicalizing operator and P C the canonical3 representation of P. Then the left hand side of
(42) represents the computation of M (P ±1 ±2 ···±n ) based on the canonical representation.
Remark 2: Let
X  1 + θv 2 1 − θv 2

H(U |V ) ≡ µv log2 + log2
2 1 + θv 2 1 − θv
v∈V
X X  
1 + |θ| 2 1 − |θ| 2
= µv log2 + log2
2 1 + |θ| 2 1 − |θ|
θ v:θv =θ
r
X 1 + θv 1 − θv
Z(U |V ) ≡ 2 µv ·
2 2
v∈V
X X p
= µv 1 − |θ|2
θ v:θv =θ
X 1 + θv 1 − θv
HQ (U |V ) ≡ 2 µv · ·
2 2
v∈V
" #
1 X X
= 1− µv |θ|2 ,
2
θ v:θv =θ

where H(U |V ) is the conditional entropy, Z(U |V ) is the source of the Battacharyya parameters, and HQ (U |V ) is the quadratic
entropy. We have similar results regarding these quantities.

B. Computation of M (P ± )
In this section, we introduce techniques for the computation of M (P ± ). When P ≡ {(µv , θv )}v∈V is given, we have
|P − | = |P|2 and |P + | = 2|P|2 from the definition of P ± . It seems that the computational complexity for the computation of
M (P ± ) is O(|P|2 ). However, we can reduce it as follows.
For given P, we can compute M (P) with complexity O(|P|) and then
M (P − ) = M (P)2
by applying (27). In the following, we introduce the algorithm for the computation of M (P + ).
First, we apply operator C introduced in the last section to obtain canonical list P C ≡ {(µv , θv )}C
v=1 , satisfying

µv > 0 for all v ∈ {1, . . . , C} (43)

C
X
µv = 1 (44)
v=1

3 ‘Canonical’ reflects our conjecture that P C is the smallest representation of P satisfying (39).
 9

0 ≤ θ1 < θ2 < · · · < θC ≤ 1. (45)

|V|
We introduce Algorithm 1 to compute P C for given P ≡ {(µv , θv )}v=1 . The computational complexity of Algorithm 1 is
O(|P| log2 |P|), which corresponds to that of the sorting elements of list P.

Algorithm 1 Computation of P C
|V|
1: Sort list P A ≡ {(µv , |θv |)}v=1 in ascending order of |θv |.
2: v ← 1, C ← 0.
3: while v ≤ |V| do
4: if µv > 0 then
5: C ← C + 1.
6: (µC , θC ) ← (µv , θv ).
7: v ← v + 1.
8: while θv = θC do
9: µC ← µC + µv .
10: v ← v + 1.
11: end while
12: else
13: v ← v + 1.
14: end if
15: end while
16: Output P C ≡ {(µv , θv )}C v=1 .

Next we introduce the following lemma. The proof is given in Section VI-J.
Lemma 10: Assume that P C ≡ {(µv , θv )}C
v=1 satisfies (44) and (45). Then we have

M (P + ) = M (P C+ )
C
" v−1
#
X X
= µv θ v µv + 2 µv ′ . (46)
v=1 v ′ =1

Based on this lemma, we introduce Algorithm 2, which computes M (P C+ ) for given P C . In Algorithm 2, we have
i−1
X
τ =2 µv ′
v ′ =1

just before line 4. At line 4, we have

i−1
X
τ = µv + 2 µv ′ .
v ′ =1

At line 5, we have " #

i
X v−1
X
M= µv θ v µv + 2 µv ′
v=1 v ′ =1

and finally we have output " #

C
X v−1
X
M= µv θ v µv + 2 µv ′ .
v=1 v ′ =1

From (40), we have

M (P + ) = M (P C+ ).
The computational complexity of Algorithm 2 is O(|P|).
n
To obtain value M (P ± ) for given ±n ≡ (±1 , . . . , ±n ), we compute
P C±1 C±2 ···C±n−1 C
using (12)–(16), and Algorithm 1, and then compute
n
M (P ± ) = M ([P C±1 C±2 ···C±n−1 C ]±n ) (47)
using (27) when ±n = − and using Algorithm 2 when ±n = +.
 10

Algorithm 2 Computation of M (P C+ )
1: Start from list P C ≡ {(µv , θv )}C
v=1 .
2: M ← 0, τ ← 0.
3: for i ∈ {1, . . . , C} do
4: τ ← τ + µi .
5: M ← M + µi θ i τ .
6: τ ← τ + µi .
7: end for
8: Output M .

n
C. Approximation of M (P ± )
n
In the last section, we introduced techniques for the exact computation of M (P ± ) based on the canonical expression
of P to reduce the computational complexity. However, they are not sufficient for a large n. In this section, we introduce
n
techniques to obtain an approximation of M (P ± ) by the quantization of P. The quantization is based on the previously
introduced greedy degrading-/upgrading-merge algorithm [13][15], where we introduce a novel loss function instead of the
difference of mutual information. We can compute this loss function using the four fundamental rules of arithmetic, which
is much easier than computing the difference of the mutual information, including logarithmic functions. In addition, when
initial value P ≡ {(µy , θy )}y∈Y consists of rational numbers, the computation can be exactly performed using the arithmetic
of rational numbers, assuming sufficient memory. In contrast, we had to focus on the precision of the mutual information in
conventional techniqes.
First we introduce the notions of degradation and upgradation in terms of expression P(U |V ) ≡ {(µv , θv )}v∈V . Let PUV
be the joint distribution of (U, V ). When U − V − Vb forms a Markov chain, there is a conditional probability distribution
PVb |V such that the joint distribution of (U, V, Vb ) is given as
v |v)PUV (u, v)
PUV Vb (u, v, vb) = PVb |V (b
b where U, V, and V
for each (u, v, vb) ∈ U × V × V, b are the alphabets of respective random variables U , V , and Vb . The channel
b
between U and V is (physically) degraded with respect to the channel between U and V . Conversely, the channel between
U and V is (physically) upgraded with respect to the channel between U and Vb . Note that physical degradation/upgradation
implies the stochastic degradation/upgradation [13][15]; conditional probability distribution PVb |V exists, such that
X
v |u) =
PVb |U (b v |v)PV |U (v|u)
PVb |V (b
v∈V

b
for all (u, vb) ∈ U × V.
Here we consider a particular case of degradation/upgradation where Vb ⊂ V and P b (b b
V |V v |v) ∈ {0, 1} for all (v, v
b) ∈ V × V.
b b b
In this case, V is a quantization of V, and V is a extension of V. We define surjection q : V → V called quantization map as
q(v) ≡ vb if PVb |V (b
v |v) = 1,

where for all v ∈ V there is a unique vb ∈ V b such that P b (b

V |V v |v) = 1 because PV b |V (·|v) is a probability distribution. The
conditional probability distribution PVb |V can be represented in terms of q, and {q −1 (b
v )}vb∈Vb forms a partition of V. Probability
distributions PU Vb , PVb and PU|Vb are given as
X
PU Vb (u, vb) = v |v)PUV (u, v)
PVb |V (b
v∈V
X
= PUV (u, v)
v∈q−1 (b
v)
X 1 ∓ u θv
= µv ·
2
v∈q−1 (b
v)
 
1 X X
= µv ∓ u µv θ v  (48)
2 −1 −1
v∈q (bv) v∈q (bv)
XX
PVb (b
v) = v |v)PUV (u, v)
PVb |V (b
u∈U v∈V
X X
= PUV (u, v)
u∈U v∈q−1 (b
v)
 11

Q
(µ1 , θ1 )

(µ0 + µ1 , θ)

(µ0 , θ0 )
P b
P

µ0 θ0 +µ1 θ1 b
Fig. 1. Degrading-merge operation: We define µ ≡ µ0 + µ1 and θ ≡ µ0 +µ1
, where (µ, θ) is a member of P.

X
= PV (v)
v∈q−1 (b
v)
X
= µv (49)
v∈q−1 (b
v)
PU Vb (u, vb)
PU|Vb (u|b
v) =
P b (b
v)
"V P #
1 v ) µv θ v
v∈q−1 (b
= 1 ∓u P , (50)
2 v∈q−1 (bv ) µv

where ∓u is the bipolar-binary conversion of u. Then the joint distribution is represented as

 P 
 X −1 µ v θ v

v∈q (b v)
Pb ≡  µv , P  . (51)
 −1 v ) µv
v∈q−1 (b 
v∈q (b
v) b
b∈V
v

Let Q denote a quantize operator that is applied to P, defined as

P Q , ≡ Pb
where Q depends on quantization map q. Then we call P Q a degrading merger of P. Let E denote an extend operator applied
b defined as
to P,
b E ≡ P,
P
where Pb is a quantization of P that satisfies (51). Note that E depends on quantization map q. Then upgrading merger P
b Q of
b is defined as
P
bQ ≡ P
P b EΣ , (52)
b Q is reduced to that of Θ defined by (37) when |Θ| < |P|,
where Σ is defined by (36). Note that the cardinality of the list P b
b
but it may be increased when |Θ| > |P|.
Particular cases of degrading- and upgrading-merge operations are shown in Figs. 1 and 2, and details will be presented in
Sections IV-D and IV-E.
We introduce the following lemma, which asserts that M is invariant of any quantize operations. The proof is given in
Section VI-K.
Lemma 11: Assume that P ≡ {(µv , θv )}v∈V satisfies that µv > 0 and θv ≥ 0 for all v ∈ V. Define P b by (51). Then
b = M (P)
M (P) (53)
Q
M (P ) = M (P) (54)
M (P b
b Q ) = M (P). (55)
The above lemma implies that M is inappropriate for the quantization loss function because M is unchanged by any
degrading/upgrading operations. Here we introduce a lemma that motivates our quantization loss function. The proof is given
in Section VI-L.
Lemma 12: For given P ≡ {(µv , θv )}v∈V , we have
k k
M (P − ) = M (P)2 (56)
+ −+ +−
M (P)M (P ) ≤ M (P ) ≤ M (P ) (57)
 12

E Σ
(µ1 , θ1 )
(µ1 , θ1 )
(µ1 + µ′1 , θ1 )
(µ′1 , θ1 )

(µ, θ)

(µ′0 , θ0 )
(µ0 + µ′0 , θ0 )
(µ0 , θ0 )
(µ0 , θ0 )
b
P P Θ

µ′0 θ0 +µ′1 θ1
Fig. 2. Upgrading-merge operation: For given (µ, θ) and (θ0 , θ1 ), a pair (µ′0 , µ′1 ) is defined so that µ = µ′0 + µ′1 and θ = µ′0 +µ′1
.

k k k
M (P)2 −1
M (P + ) ≤ M (P − +
) ≤ M (P + )2 . (58)
k k
Assume that P is canonical. From (56), we can compute value M (P − ) = M (P)2 . This implies that the quantization is
k
unnecessary when P is given. Thus, it is enough to consider the computation of M (P − + ).
k k k
+
Here assume that M (P − + ) > M (P Q−
), which corresponds to the case of degradation. Then quantization loss M (P − + )−
k k
M (P Q− + ) can be minimized by maximizing M (P Q− + ). Furthermore, we have
k k
M (P)2 −1
M (P Q+ ) ≤ M (P Q− +
)
Q−k +
from (54) and (58). This inequality suggests that the lower bound of M (P ) can be maximized by maximizing M (P Q+ ).
k k k
Similarly, assume that M (P − + ) < M (P Q− + ), which is the case of upgradation. Then quantization loss M (P Q− + ) −
k k
M (P − + ) can be minimized by minimizing M (P Q− + ). Furthermore, we have
k k
+
M (P Q− ) ≤ M (P Q+ )2
k
from (55) and (58). This inequality suggests that the upper bound of M (P Q− + ) can be minimized by minimizing M (P Q+ ).
In the following sections, we introduce the greedy degrading-/upgrading-merge algorithm to reduce quantization loss |M (P Q+ )−
M (P + )|, where Q ∈ {Q, Q}. Although the computational complexity of the algorithms introduced below is O(|P|2 ), it can
be reduced to O(|P| log2 |P|) using a doubly linked list and a heap, as in [15, Sec. IV-A].
To state the theorems in the next sections, we assume that P ≡ {(µv , θv )}Cv=1 is canonical, that is, it satisfies (43)–(45).
We define
1
Error(P) ≡ [1 − M (P)] (59)
2
θC ≡ max θv , (60)
v∈V

where (59) comes from (7). For given lists P and P , P \ P denotes the deletion of elements in P ′ from list P, and P ∪ P ′
′ ′

denotes the addition of elements in P ′ to P.

Remark 3: Let P ≡ {(1, θ)} be the list of the initial joint distribution, which corresponds to the case of source coding
n
without side information and channel coding for symmetric channel. In this case, computing θC of a canonical list P ± C ≡
+ −
{(µv , θv )}C n n
v=1 for given ± ≡ (±1 , . . . ±n ) ∈ {−, +} is easy. Define t and t as

t− (x) ≡ x2
2x
t+ (x) ≡ ,
1 + x2
where the first equality comes from  2
max θv0 θv1 = max θv
v0 ∈V,v1 ∈V v∈V
 13

and the second one comes from

   
 θv + θv0   θv1 − θv0  θv1 + θv0
max max  1 ,
  1 − θv θv
 = max

v0 ∈V,v1 ∈V: 1 + θv θv 0 1 0 1
v0 ∈V,v1 ∈V: 1 + θv θv 0 1

2 maxv∈V θv
= 2. (61)
1 + [maxv∈V θv ]
Then we can compute θC ≡ xn with recursion:
x0 = θ
xi = t±i (xi−1 ).
Note that θC ∈ [0, 1) if θ ∈ [0, 1).

D. Greedy Degrading-Merge Algorithm

This section introduces a greedy degrading-merge algorithm based on the one introduced in [13].
Assume that P ≡ {(µv , θv )}C v=1 is canonical satisfying (43)–(45). The algorithm iterates the merging of two elements into
one (Fig. 1) until the cardinality of P is reduced to desired number Q.
For i ∈ {C, C − 1, . . . , Q}, let Pi be the list obtained by the algorithm, where the cardinality of the list is i, PC ≡ P, and
PQ ≡ P Q . Assume that two elements, (µj , θj ) and (µj+1 , θj+1 ), are merged into the element
 
µj θj + µj+1 θj+1
µj + µj+1 ,
µj + µj+1
for given j ∈ {1, . . . i − 1}, that is, Pi−1 is defined as
 
µj θj + µj+1 θj+1
Pi−1 ≡ [Pi \ {(µj , θj ), (µj+1 , θj+1 )}] ∪ µj + µj+1 , .
µj + µj+1
From Lemma 10, we have
i
" v−1
#
X X
M (Pi+ ) = µv θ v µv + 2 µv ′
v=1 v ′ =1
and
j−1
" v−1
# " j−1
#
X X µj θj + µj+1 θj+1 X
+
M (Pi−1 ) = µv θ v µv + 2 µv′ + [µj + µj+1 ] · · [µj + µj+1 ] + 2 µv ′
µj + µj+1
v=1 v ′ =1 v ′ =1
  
X i j−1
X v−1
X
+ µv θv µv + 2  µv′ + [µj + µj+1 ] + µv ′  
v=j+2 v ′ =1 v ′ =j+2
j−1
" v−1
# " j−1
#
X X X
= µv θ v µv + 2 µv′ + [µj θj + µj+1 θj+1 ] µj + µj+1 + 2 µv ′
v=1 v ′ =1 v ′ =1
i
" v−1
#
X X
+ µv θ v µv + 2 µv ′ . (62)
v=j+2 v ′ =1

Then we have
M (Pi+ ) − M (Pi−1
+
)
" j−1
# " j
# " j−1
#
X X X
= µj θ j µj + 2 µv′ + µj+1 θj+1 µj+1 + 2 µv′ − [µj θj + µj+1 θj+1 ] µj + µj+1 + 2 µv ′
v ′ =1 v ′ =1 v ′ =1
= µj µj+1 [θj+1 − θj ] . (63)
Using the condition (45), we have M (Pi+ )−M (Pi−1
+
> 0 for every j ∈ {1, . . . , i−1}, which implies that the degrading-merge
)
operation reduces the M -value after the plus operation, that is,
M (P + ) > M (P Q+ ). (64)
Then (63) introduces a greedy degrading-merge algorithm as Algorithm 3.
Regarding the quantization loss, we have the following theorem, where (65) and (66) show the absolute precision and (67)
and (68) show the relative precision. The relative precision is relevant when the value is close to 0 or 1. The proof is given
in Section VI-M.
 14

Algorithm 3 Greedy Degrading-Merge Algorithm

1: Start from a canonical list P ≡ {(µv , θv )}Cv=1 .
2: i ← C.
3: while i > Q do
4: Find
j ≡ arg min µv µv+1 [θv+1 − θv ].
v∈{1,...,i−1}

5: Merge two pairs (µj , θj ) and (µj+1 , θj+1 ) into a single pair (µj , θj ):
µj ← µj + µj+1
µj θj + µj+1 θj+1
θj ← .
µj + µj+1
6: Obtain a new list {(µv , θv )}i−1
v=1 :

µv ← µv+1
θv ← θv+1
for v ∈ {j + 1, . . . i − 1}, where we do nothing when j = i − 1.
7: i ← i − 1.
8: end while
9: Output P Q ≡ {(µv , θv )}Q v=1 .

Theorem 3: Let M be defined by (8), let θC be defined by (60), and Q ≡ |P Q+ |. Then we have
θC +
M (P + ) − 2
≤ M (P Q ) < M (P + ) (65)
Q
θC
1 − M (P + ) < 1 − M (P Q+ ) ≤ 1 − M (P + ) + 2 (66)
Q
 
+ θC
M (P ) 1 − ≤ M (P Q+ ) < M (P + ) (67)
M (P)Q2
 
+ Q+
 +
 θC
1 − M (P ) < 1 − M (P ) ≤ 1 − M (P ) 1 + . (68)
[1 − θC ]Q2
Remark 4: When θC = 1, the right hand side inequality of (68) is trivial because the right hand side is ∞. In this case,
C−1
we can modify the algorithm as follows: apply it to {(µv , θv )}v=1 , where (µC , θC ) is excluded and return (µC , θC ) to the
Q
output. Then we have P , which has Q + 1 elements, satisfying
 
+   θC−1
1 − M (P Q ) ≤ 1 − M (P + ) 1 − .
[1 − θC−1 ]Q2
From the above theorem, we have the following corollary.
Corollary 1: Define Error(P) by (59). Then we have
θC
Error(P + ) < Error(P Q+ ) ≤ Error(P + ) +
2Q2
1 θC 1 1
− Error(P + ) − 2
≤ − Error(P Q+ ) < − Error(P + )
2 2Q 2 2
 
+ Q+ + θC
Error(P ) < Error(P ) ≤ Error(P ) 1 +
[1 − θC ]Q2
  
1 θC 1 1
− Error(P + ) 1 − ≤ − Error(P Q+ ) < − Error(P + ).
2 M (P)Q2 2 2

E. Greedy Upgrading-Merge Algorithm

This section introduces a greedy upgrading-merge algorithm based on the one introduced in [13].
Assume that P ≡ {(µv , θv )}C v=1 is canonical satisfying (43)–(45). The algorithm iterates the merging of three elements into
two elements (Fig. 2) until the cardinality of P is reduced to desired number Q.
For i ∈ {C, C − 1, . . . , Q}, let Pi be the list obtained by the algorithm, where the cardinality of the list is i, PC ≡ P, and
PQ ≡ P Q . Here assume that, for given j ∈ {2, . . . , i − 1}, an element (µj , θj ) is extended to two elements (µ′j−1 , θj−1 ) and
 15

(µ′j+1 , θj+1 ), and then four elements (µj−1 , θj−1 ), (µ′j−1 , θj−1 ), (µ′j+1 , θj+1 ), and (µj+1 , θj+1 ) are merged into two elements
(µj−1 + µ′j−1 , θj−1 ) and (µj+1 + µ′j+1 , θj+1 ), that is, Pi−1 is defined as

Pi−1 ≡ [Pi \ {(µj−1 , θj−1 ), (µj , θj ), (µj+1 , θj+1 )}] ∪ (µj−1 + µ′j−1 , θj−1 ), (µj+1 + µ′j+1 , θj+1 ) .
Then µ′j−1 and µ′j+1 should satisfy
µj = µ′j−1 + µ′j+1 (69)
µ′j−1 θj−1 + µ′j+1 θj+1
θj = . (70)
µ′j−1 + µ′j+1
From the above two equalities, we have
µj [θj − θj−1 ]
µ′j−1 = (71)
θj+1 − θj−1
µj [θj+1 − θj ]
µ′j+1 = . (72)
θj+1 − θj−1
From Lemma 10, we have
j−2
" v−1
# " j−2
#
X X   X
+ ′ ′
M (Pi−1 ) = µv θ v µv + 2 µv′ + µj−1 + µj−1 θj−1 [µj−1 + µj−1 ] + 2 µv ′
v=1 v ′ =1 v ′ =1
" " j−2 ## C
" v−1
#
  X X X
+ µj+1 + µ′j+1 θj+1 [µj+1 + µ′j+1 ] + 2 µv′ + [µj−1 + µ′j−1 ] + µv θ v µv + 2 µv ′
v ′ =1 v=j+2 v ′ =1
j−1
" v−1
# " j−1
#
X X X
= µv θ v µv + 2 µv′ + µ′j−1 θj−1 µj − µ′j+1 + 2 µv ′
v=1 v ′ =1 v ′ =1
" j−1
# C
" v−1
#
X X X
+ µ′j+1 θj+1 µj + µ′j−1 +2 µv ′ + µv θ v µv + 2 µv ′ , (73)
v ′ =1 v=j+1 v ′ =1

where we use (69) in the last equality. Then we have

" j−1
# " j−1
# " j−1
#
X X X
+
M (Pi−1 ) − M (Pi+ ) = µ′j−1 θj−1 µj − µ′j+1 + 2 µv′ + µ′j+1 θj+1 µj + µ′j−1 + 2 µv ′ − µj θ j µj + 2 µv ′
v ′ =1 v ′ =1 v ′ =1
= µ′j−1 µ′j+1 [θj+1 − θj−1 ], (74)
+
where we use (69) and (70) in the last equality. By using the condition (45), we have − M (Pi−1
> 0 for every ) M (Pi+ )
j ∈ {2, . . . , i − 1}, which implies that the upgrading-merge operation increases the M -value after the plus operation, that is,
M (P + ) < M (P Q+ ). (75)
Then (74) introduces a greedy upgrading-merge algorithm as Algorithm 4.
Regarding the quantization loss, we have the following theorem, where (77) and (78) show the absolute precision and (79)
and (80) show the relative precision. The relative precision is relevant when the value is close to 0 or 1. The proof is given
in Section VI-N.
Theorem 4: Let M be defined by (8) and let θC be defined by (60). Then we have
θC
M (P + ) < M (P Q+ ) ≤ M (P + ) + (77)
[Q − 1]2
θC
1 − M (P + ) − ≤ 1 − M (P Q+ ) < 1 − M (P + ) (78)
[Q − 1]2
 
θC
M (P + ) < M (P Q+ ) ≤ M (P + ) 1 + (79)
M (P)[Q − 1]2
 
  θC
1 − M (P + ) 1 − ≤ 1 − M (P Q+ ) < 1 − M (P + ). (80)
[1 − θC ][Q − 1]2
Remark 5: When θC = 1, the left hand side inequality of (80) is trivial because the left hand side is -∞. In this case, we
C−1
can modify the algorithm: apply it to {(µv , θv )}v=1 , where (µC , θC ) is excluded, and return (µC , θC ) to the output. Then we
have P Q , which has Q + 1 elements, satisfying
 
  θC−1
1 − M (P Q+ ) ≥ 1 − M (P + ) 1 + .
[1 − θC−1 ][Q − 1]2
 16

Algorithm 4 Greedy Upgrading-Merge Algorithm

1: Start from list P ≡ {(µv , θv )}C
v=1 .
2: i ← C.
3: while i > Q do
4: Find
j ≡ arg min µ′v−1 µ′v+1 [θv+1 − θv−1 ], (76)
v∈{2,...,i−1}

where µ′v−1 and µ′v+1 are respectively defined by (71) and (72).
5: Split the pair (µj , θj ) into two pairs (µ′j−1 , θj−1 ) and (µ′j+1 , θj+1 ) and merge into the respective pairs
(µj−1 , θj−1 ) and (µj+1 , θj+1 ):
µj−1 ← µj−1 + µ′j−1
µj+1 ← µj+1 + µ′j−1 .
6: Obtain new list {(µv , θv )}i−1
v=1 :

µv ← µv+1
θv ← θv+1
for v ∈ {j + 1, . . . i − 1}.
7: i ← i − 1.
8: end while
9: Output P Q ≡ {(µv , θv )}Q v=1 .

From the above theorem, we have the following corollary.

Corollary 2: Let Error(P) be defined by (59). Then we have
θC
Error(P + ) − ≤ Error(P Q+ ) < Error(P + )
2[Q − 1]2
1 1 1 θC
− Error(P + ) < − Error(P Q+ ) ≤ − Error(P + ) +
2 2 2 2[Q − 1]2
 
θC
Error(P + ) 1 − ≤ Error(P Q+ ) < Error(P + )
[1 − θC ][Q − 1]2
  
1 1 1 θC
− Error(P Q+ ) < − Error(P + ) ≤ − Error(P + ) 1 + .
2 2 2 M (P)[Q − 1]2

V. E XPERIMENTAL R ESULTS
In this section, we introduce our experimental results. In them, we calculate the bit error probability of the polar source code
without decoder side information, where we define PX (1) ≡ 0.11. The initial joint probability distribution is P ≡ {(1, θ)},
where θ ≡ PX (0) − PX (1) = 0.78.
Our results can be applied to the polar channel code for the binary symmetric channel, where the crossover probability
is PY |X (1|0) = PY |X (0|1) = 0.11. Although we can reduce the linear channel code to linear source code without decoder
side information by considering the syndrome decoding of additive noise, we introduce an alternative approach. Assume that
ε = PY |X (1|0) = PY |X (0|1) and the channel input distribution is uniform, where PX (0) = PX (1) = 1/2. Then we have
PY (0) = PY (1) = 1/2 and PX|Y (x|y) = PY |X (x|y) for all x, y ∈ {0, 1}. The initial joint probability distribution is given as
P ≡ {(1/2, θ0), (1/2, θ1 )}, where θy is defined as
θy ≡ PX|Y (0|y) − PX|Y (1|y)
= ∓y [1 − 2ε] (81)
C
using bipolar-binary conversion ∓y of y ∈ {0, 1}. Since |θ0 | = |θ1 | = |1 − 2ε|, we have canonical representation P of joint
probability as
P C = {(1, |1 − 2ε|)}.
By applying (38), we can compute the bit error probability of the polar channel code with initial distribution P C = {(1, |1−2ε|)},
which is the same as the initial joint probability distribution of the polar source code. By letting ε ≡ 0.11, we have P C =
{(1, 0.78)}.
 17

In the experimental results, we consider the polar codes with n ∈ {8, 10}, where the block length is 2n ∈ {256, 1024}.
Tables I and II show the bit error probabilities, where the set [0n : 1n ] of indexes is sorted in descending order of the bit error
probability. A negative value, −γ (γ > 0), means that the bit error probability is 0.5 − γ.
When n = 8, we exactly compute the bit error probabilities as (47), where we use only Algorithms 1 and 2. Since the initial
joint distribution consists of rational numbers, we can employ algorithms with infinite precision using the four fundamental
rules of arithmetic. After the exact computation of the bit error probability, the fifth digit is rounded up to obtain the upper
bound in Table I.
When n = 10, we compute the approximation of the bit error probability with the greedy degrading-/upgrading-merge
algorithms (Algorithms 3 and 4) before the ± operations. Although Q ≡ 128 is sufficient to determine the order of the
bit error probabilities, we set Q ≡ 512 to improve the approximation. In Algorithm 4, we reduce the objective function of
minimization (76) to
µv [θv − θv−1 ] µv [θv+1 − θv ]
µ′v−1 µ′v+1 [θv+1 − θv−1 ] = · · [θv+1 − θv−1 ]
θv+1 − θv−1 θv+1 − θv−1
µ2 [θv − θv−1 ][θv+1 − θv ]
= v
θv+1 − θv−1
µ2v [θv − θv−1 ][θv+1 − θv ]
=
[θv+1 − θv ] + [θv − θv−1 ]
µ2v
= 1 1 , (82)
θv −θv−1 + θv+1 −θv

where the first equality comes from (71) and (72). Using this relation, we can reduce the number of multiplications, where
the computation of the inverse of a rational number is easy. The fifth digit of the bit error probability is suitably truncated or
rounded up in Table II to make the margin of the upper and lower bounds.
Remark 6: Both in Tables I and II, the bit error probability corresponding to bn = 0k 10n−k−1 (±n = −k + −n−k−1 ) is
constant for all k ∈ {0, 1, . . . , n − 1}. In the following, we show that this fact is generally true. When P ≡ {(1, θ)}, we have
k k
P − = {(1, θ2 )}
from (12) and (13). Then we have
( k+1 k
! k+1
!)
−k + 1 + θ2 2θ2 1 − θ2
P = , , ,0
2 1 + θ2k+1 2
from (15), (16) and
k k
M (P − +
) = θ2
from (8). Then we have
k
+−n−k−1 k n−k−1
M (P − ) = [θ2 ]2
n−1
= θ2 , (83)
which does not depend on k ∈ {0, 1, . . . , n − 1}, where the first equality comes from (56).

VI. P ROOFS
A. Proof of Lemma 2
The first equality is shown directly from (8), (12), and (13). The second equality is shown as
   
X X µv µv [1 + θv θv ]  θv + θv  X X µv µv [1 − θv θv ]  θv − θv 
+ 0 1 0 1  1 0  0 1 0 1  1 0 
M (P ) =  1 + θv θv  +  1 − θv θv 
2 0 1 2 0 1
v0 ∈V v1 ∈V v0 ∈V v1 ∈V
X X µv µv |θv + θv | X X µv µv |θv − θv |
0 1 1 0 0 1 1 0
= +
2 2
v0 ∈V v1 ∈V v0 ∈V v1 ∈V
X X
= µv0 µv1 max{|θv0 |, |θv1 |}, (84)
v0 ∈V v1 ∈V

where the first equality comes from (8), (15), and (16), the second equality comes from the fact that θv0 , θv1 ∈ [−1, 1] and
|θv1 ± θv0 | = 0 when 1 ± θv1 θv0 = 0, and the last equality comes from Lemma 13 in Appendix. 
 18

B. Proof of Lemma 3
Although the lemma was previously given [1, Proposition 5.1], we show it directly with Lemma 2 and the fact that θv ∈
[−1, 1]. Inequalities (26) are shown as
X X
M (P − ) = µv0 µv1 |θv0 θv1 |
v0 ∈V v1 ∈V
X X
≤ µv0 µv1 |θv0 |
v0 ∈V v1 ∈V
X
= µv0 |θv0 |
v0 ∈V
= M (P) (85)
and
X X
M (P + ) = µv0 µv1 max{|θv0 |, |θv1 |}
v0 ∈V v1 ∈V
X X
≥ µv0 µv1 |θv0 |
v0 ∈V v1 ∈V
X
= µv0 |θv0 |
v0 ∈V
= M (P). (86)
Equality (27) is shown as
X X
M (P − ) = µv0 µv1 |θv0 θv1 |
v0 ∈V v1 ∈V
X X
= µv0 |θv0 | µv1 |θv1 |
v0 ∈V v1 ∈V
" #" #
X X
= µv0 |θv0 | µv1 |θv1 |
v0 ∈V v1 ∈V
= M (P)2 . (87)
Inequality (28) is shown as
X X X X
M (P − ) + M (P + ) = µv0 µv1 |θv0 θv1 | + µv0 µv1 max{|θv0 |, |θv1 |}
v0 ∈V v1 ∈V v0 ∈V v1 ∈V
X X X X
≤ µv0 µv1 min{|θv0 |, |θv1 |} + µv0 µv1 max{|θv0 |, |θv1 |}
v0 ∈V v1 ∈V v0 ∈V v1 ∈V
X X
= µv0 µv1 [|θv0 | + |θv1 |]
v0 ∈V v1 ∈V
X X
= µv0 |θv0 | + µv1 |θv1 |
v0 ∈V v1 ∈V
= 2M (P), (88)
where the inequality comes from relation |θv0 θv1 | = |θv0 ||θv1 | ≤ min{|θv0 |, |θv1 |}, which comes from the fact that θv0 , θv1 ∈
[−1, 1]. 

C. Proof of Lemma 4
Since Error(U |V ) = [1 − M (P(U |V ))]/2 is the MAP decision error probability of gessing U from V , inequality 1 −
M (P(U |V )) ≤ H(U |V ) comes from the fact that 2Error(U |V ) ≤ H(U |V ) shown in [6], and inequality H(U |V ) ≤ h([1 −
M (P(U |V ))]/2) comes from the Fano inequality. 

D. Proof of Lemma 5
From the definition (8) of M , value M (P) does not change when θv is replaced by |θv |. 
 19

E. Proof of Lemma 6
Relation P A−A = P −A is shown as
P A−A = {(µv0 µv1 , ||θv0 ||θv1 ||) : (v0 , v1 ) ∈ V × V}
= {(µv0 µv1 , |θv0 θv1 |) : (v0 , v1 ) ∈ V × V}
= P −A . (89)
A+A +A
Next we show relation P = P . We have
   
µv0 µv1 [1 ∓u |θv0 ||θv1 |]  |θv0 | ∓u |θv1 | 
P A+A = , : (u, v0 , v1 ) ∈ U × V × V
2 1 ∓u |θv0 ||θv1 | 
   
µv0 µv1 [1 ∓u θv0 θv1 ]  θv0 ∓u θv1 
P +A = , : (u, v , v ) ∈ U × V × V ,
1 ∓u θv0 θv1 
0 1
2
where ∓u is the bipolar-binary conversion of u. When θv0 θv1 ≥ 0, we have the fact that θv0 , θv1 ≥ 0 or θv0 , θv1 ≤ 0. Then
we have relation P A+A = P +A immediately from the fact that
1 ∓u |θv0 ||θv1 | = 1 ∓u θv0 θv1 ≥ 0
||θv0 | ∓u |θv1 || = |θv0 ∓u θv1 |.
When θv0 θv1 < 0, we have the fact that θv0 < 0 < θv1 or θv1 < 0 < θv0 . Then we have
1 ∓u |θv0 ||θv1 | = 1 ±u θv0 θv1 ≥ 0
||θv1 | ∓u |θv0 || = |θv1 ±u θv0 |,
which implies that
     
µv0 µv1 [1 ∓u |θv0 ||θv1 |]  |θv0 | ∓u |θv1 |  µv0 µv1 [1 ±u θv0 θv1 ]  θv0 ±u θv1 
, = ,
2 1 ∓u |θv0 ||θv1 |  2  1 ± u θv θv 
0 1

Then we have relation P A+A = P +A using a bijection:

(
(u, v0 , v1 ) if θv0 θv1 ≥ 0
π(u, v0 , v1 ) ≡
(u ⊕ 1, v0 , v1 ) if θv0 θv1 < 0.


F. Proof of Lemma 7
We have
X
M (P) = µv |θv |
v∈V:
µv >0
X X
= µv |θ|
θ∈Θ v:θv =θ
" #
X X
= µv |θ|
θ∈Θ v:θv =θ
Σ
= M (P ). (90)


G. Proof of Lemma 8
First, we show relation P Σ−Σ = P −Σ . Define µθ and Θ− as
X
µθ ≡ µv . (91)
v:θv =θ
Θ− ≡ {θv0 θv1 : (v0 , v1 ) ∈ V × V s.t. µv0 µv1 > 0}
= {θ0 θ1 : (θ0 , θ1 ) ∈ Θ × Θ s.t. µθ0 µθ1 > 0} , (92)
 20

where Θ is defined by (37). Then we have

  

 

 X  
P Σ−Σ =  µ µ ,
θ0 θ1 θ  : θ ∈ Θ −

 

 (θ0 ,θ1 ): 
θ θ =θ
 0 1  

 


   


 X X  

  −
=  µv0 µv1 , θ : θ ∈ Θ

   


 (θ0 ,θ1 ): (v0 ,v1 ):  


 θ0 θ1 =θ θv0 =θ0 

θv1 =θ1
  

 

 X  
=  
µv0 µv1 , θ : θ ∈ Θ −

 

 (v0 ,v1 ): 
θv0 θv1 =θ
−Σ
=P , (93)
where the second equality comes from (91) and the third equality comes from Lemma 14, which appears in Appendix, by
letting v ≡ (v0 , v1 ), θ ≡ (θ0 , θ1 ), and g(θ) ≡ θ0 θ1 , and the last equality comes from (92).
Next we show relation P Σ+Σ = P +Σ . Let
 
+ θv0 ∓u θv1 µθv0 µθv1 [1 ∓u θv0 θv1 ]
Θ ≡ : u ∈ {0, 1}, (v0 , v1 ) ∈ V × V s.t. >0
1 ∓u θv0 θv1 2
 
θ0 ∓ u θ1 µθ µθ [1 ∓u θ0 θ1 ]
= : u ∈ {0, 1}, θ0, θ1 ∈ Θ s.t. 0 1 >0 , (94)
1 ∓ u θ0 θ1 2
where ∓u is the bipolar-binary conversion of u. Then we have
  

 


 
 X µ µ [1 ∓ θ θ ]




Σ+Σ  θ 0 θ 1 u 0 1  +
P =  , θ : θ ∈ Θ

  (u,θ0 ,θ1 ): 2  


 

 θ1 ∓u θ0 =θ 
1∓u θ0 θ1
  

 


   


 X X  

 µ µ
θ v0 θ v1 [1 ∓ θ θ
u v0 v1 ]  +
=  , θ : θ ∈ Θ

 2  


  (u,θ0 ,θ1 ): (v0 ,v1 ):  


 1∓ θ1 ∓u θ0
=θ θ v 0 =θ 0


θ θ
u 0 1
θv1 =θ1
  

 


  

 X µ µ [1 ∓ θ θ ]  
 v0 v1 u v0 v1  +
=  , θ : θ ∈ Θ

 2  


 (u,v0 ,v1 ): 

 θv ∓u θv 1 0 =θ

1∓u θv0 θv1

+Σ
=P , (95)
where the third equality comes from Lemma 14, which appears in Appendix, by letting s ≡ (v0 , v1 ), θ ≡ (θ0 , θ1 ), and
θ1 ∓ u θ0
g(θ) ≡
1 ∓ u θ0 θ1
for each u ∈ {0, 1}, and the last equality comes from (94). 

H. Proof of Lemma 9
Let
ΘA ≡ {|θv | : v ∈ V s.t. µv > 0}
= {|θ| : θ ∈ Θ s.t. µθ > 0} , (96)
 21

where Θ is defined by (37) and µθ is defined by (91). Then we have

  
 X 
P ΣAΣ =  µθ ′ , θ  : θ ∈ Θ A
 ′ ′ 
θ :|θ |=θ
  
 X X 
=  µv , θ  : θ ∈ Θ A
 ′ ′ 
θ :|θ |=θ v:θv =θ ′
  
 X 
=  µv , θ  : θ ∈ Θ A
 
v:|θv |=θ
AΣ
=P , (97)
where the third equality comes from Lemma 14 in Appendix by letting v ≡ v, θ ≡ θ′ , and g(θ) ≡ |θ′ |, and the last equality
comes from (96). 

I. Proof of Theorem 2
Equation (38) is shown as
M (P C ) = M (P AΣ )
= M (P A )
= M (P), (98)
where the second equality comes from Lemmas 7 and the third equality comes from Lemmas 5. Eq. (39) is shown as
P C±C = P AΣ±AΣ
= P AΣ±ΣAΣ
= P A±ΣAΣ
= P A±AΣ
= P ±AΣ
= P ±C , (99)
where the second equality comes from Lemma 9, the third comes from Lemma 8, the fourth comes from Lemma 9, and the
fifth comes from Lemma 6. Eq. (40) is shown as
M (P C± ) = M (P C±A )
= M (P C±AΣ )
= M (P C±C )
= M (P ±C )
= M (P ±AΣ )
= M (P ±A )
= M (P ± ), (100)
where the first equality comes from Lemmas 5, the second comes from Lemmas 7 the fourth comes from (39), the sixth comes
from Lemmas 7, and the last one comes from Lemmas 5. 

J. Proof of Lemma 10
We have
M (P + ) = M (P C+ )
C X
X C
= µv0 µv1 max{θv0 , θv1 }
v0 =1 v1 =1
X 0 −1
C vX C
X C
X C
X
= µv0 µv1 θv0 + µv0 µv0 θv0 + µv0 µv1 θv1
v0 =1 v1 =1 v0 =1 v0 =1 v1 =v0 +1
 22

X 0 −1
C vX C
X X 1 −1
C vX
= µv0 µv1 θv0 + µv0 µv0 θv0 + µv0 µv1 θv1
v0 =1 v1 =1 v0 =1 v1 =1 v0 =1
C v−1
X X C
X
=2 µv µv ′ θ v + µv µv θ v
v=1 v ′ =1 v=1
C
" v−1
#
X X
= µv θ v µv + 2 µv ′ , (101)
v=1 v ′ =1

where the first equality comes from (40), the second equality comes from (25), and parameters are renamed in the fifth equality.


K. Proof of Lemma 11
Equation (53) is shown as
  P 
X X  µv θv 
b =   v∈q −1 (b
v )
M (P) µv   P 
 v∈q−1 (b v ) µv 
b
b∈V
v v∈q−1 (b
v)
X X
= µv |θv |
b v∈q−1 (b
b∈V
v v)
X
= µv |θv |
v∈V
= M (P), (102)
where the first equality comes from (51), the second comes from the assumption, and the third comes from the fact that
{q −1 (b
v )}vb∈Vb forms a partition of V. Eq. (54) is immediately shown from (53). Eq. (55) is shown as
b Q ) = M (P
M (P b EΣ )
bE)
= M (P
= M (P)
b
= M (P), (103)
where the second equality comes from Lemma 7 and the last equality comes from the first equality. 

L. Proof of Lemma 12
First, we show (56) as
k k−1 k−2 k
M (P − ) = M (P − )2 = [M (P − )2 ]2 = · · · = M (P)2
by applying (27).
Next we show the left hand side inequality of (57). Since g(ξ) ≡ max{αξ, β} = max{β, αξ} is a convex function of ξ, we
have
X
M (P −+ ) = µv00 µv01 µv10 µv11 max{|θv00 ||θv01 |, |θv10 ||θv11 |}
v00 ,v01 ,v10 ,v11
( )
X X X
≥ µv00 µv10 max |θv00 | µv01 |θv01 |, |θv10 | µv11 |θv11 |
v00 ,v10 v01 v11
X
= µv00 µv10 max {|θv00 |M (P), |θv10 |M (P)}
v00 ,v10
X
= M (P) µv00 µv10 max {|θv00 |, |θv10 |}
v00 ,v10
+
= M (P)M (P ), (104)
where the first and third equalities come from the assumption (45), and the Jensen inequality is applied twice in the inequality.
Next we show the right hand side inequality of (57). We have
X
M (P −+ ) = µv00 µv01 µv10 µv11 max{|θv00 θv01 |, |θv10 θv11 |}
v00 ,v01 ,v10 ,v11
 23

X
≤ µv00 µv01 µv10 µv11 max{|θv00 |, |θv01 |} max{|θv10 |, |θv11 |}
v00 ,v01 ,v10 ,v11
X X
= µv00 µv10 max{|θv00 |, |θv10 |} µv01 µv11 max{|θv01 |, |θv11 |}
v00 ,v10 v01 ,v11
" #" #
X X
= µv00 µv10 max{|θv00 |, |θv10 |} µv01 µv11 max{|θv01 |, |θv11 |}
v00 ,v10 v01 ,v11
+ 2
= M (P )
= M (P +− ), (105)
where the first equality comes from Lemma 2 and the last equality comes from (27).
Next we show the left hand side inequality of (58). By applying (57), we have
k k−1 k−1
M (P − +
) ≥ M (P − )M (P − +
)
k−1 k−2 k−2
≥ M (P − )M (P − )M (P − +
)
..
.
k−1
≥ M (P − ) · · · M (P)M (P + ). (106)
From (56), we have
"k−1 #
k Y i−1
M (P − +
)≥ M (P − ) M (P + )
i=0
"k−1 #
Y
2i
= M (P) M (P + )
i=0
Pk−1
2i
= M (P) i=0 M (P + )
k
= M (P)2 −1
M (P + ). (107)
Finally, we show the right hand side inequality of (58). By applying (57), we have
k k−1
M (P − +
) ≤ M (P − +−
)
k−2 2
≤ M (P − +−
)
..
.
k
≤ M (P +− )
k
= M (P + )2 , (108)
where the equality comes from (56). 

M. Proof of Theorem 3
Similar to the proof in [13], we have
+
M (Pi+ ) − M (Pi−1 )= min µv µv+1 [θv+1 − θv ] .
v∈{1,...,i−1}
" #3
r
= 3 min µv µv+1 [θv+1 − θv ]
v∈{1,...,i}
 3
p
3
= min µv µv+1 [θv+1 − θv ]
v∈{1,...,i−1}
" i−1
#3
1 Xp 3
≤ µv µv+1 [θv+1 − θv ]
i − 1 v=1
" i−1 # " i−1 # " i−1 #
1 X X X
≤ µv µv+1 [θv+1 − θv ]
[i − 1]3 v=1 v=1 v=1
[1 − µi ][1 − µ1 ] [θi − θ1 ]
= , (109)
[i − 1]3
 24

where the first equality comes from (63), the first inequality comes from the fact that the minimum is smaller than the average,
and the second inequality comes from the Hölder inequality. Since θi is an increasing function of i ∈ {1, . . . , C}, we have
θi
M (Pi+ ) − M (Pi−1
+
)≤
[i − 1]3
θC
≤ . (110)
[i − 1]3
The inequalities of (65) and (66) are shown from (64) and the fact that
C
X  
M (P + ) − M (P Q+ ) = M (Pi+ ) − M (Pi−1
+
)
i=Q+1
C
X θC
≤
[i − 1]3
i=Q+1
θC
≤ , (111)
Q2
where the equality comes from (63) and the first inequality comes from (110). The left hand side inequality of (67) is shown
as
θC
M (P Q+ ) ≥ M (P + ) − 2
Q
 
+ θC
= M (P ) 1 −
M (P + )Q2
 
+ θC
≥ M (P ) 1 − , (112)
M (P)Q2
where the first inequality comes from (111) and the second inequality comes from (26). The right hand side inequality of (68)
is shown as
 
Q+
 +
 θC
1 − M (P ) ≤ 1 − M (P ) 1 +
[1 − M (P + )]Q2
 
  θC
≤ 1 − M (P + ) 1 + , (113)
[1 − θC ]Q2
where the first inequality comes from (65) and the last inequality comes from the relation M (P + ) ≤ θC that comes from
(25). 

N. Proof of Theorem 4
Similar to the proof in [13], we have
+
M (Pi−1 ) − M (Pi+ ) = min µ′v µ′v+1 [θv+1 − θv−1 ]
v∈{2,...,i−1}
µ2v [θv − θv−1 ] [θv+1 − θv ]
= min
v∈{2,...,i−1} θv+1 − θv−1
≤ min µ2v [θv+1 − θv ]
v∈{2,...,i−1}
" #3
r
= 3 min µ2v [θv+1 − θv ]
v∈{2,...,i}
 3
p
= min 3
µ2v [θv+1 − θv ]
v∈{2,...,i−1}
" i−1
#3
1 Xp
≤ 3
µ2v [θv+1 − θv ]
i − 2 v=2
" i−1 # " i−1 # " i−1 #
1 X X X
≤ µv µv [θv+1 − θv ]
[i − 2]3 v=2 v=2 v=2
[1 − µi − µ1 ]2 [θi − θ2 ]
= (114)
[i − 2]3
 25

where the first equality comes from (74), the second equality comes from (71) and (72), the first inequality comes from the
fact that θv is an increasing function of v, the second inequality comes from the fact that the minimum is smaller than the
average, and the third inequality comes from the Hölder inequality. Since θv is an increasing function of v ∈ {1, . . . , C}, we
have
+ θi
M (Pi−1 ) − M (Pi+ ) ≤
[i − 2]3
θC
≤ . (115)
[i − 2]3
The inequalities of (77) and (78) are shown from (64) and the fact that
C
X  +

M (P Q+ ) − M (P + ) = M (Pi−1 ) − M (Pi+ )
i=Q+1
C
X θC
≤
[i − 2]3
i=Q+1
θC
≤ , (116)
[Q − 1]2
where the equality comes from (74) and the first inequality comes from (115). The right hand side inequalities of (79) are
shown as
 
Q+ + θC
M (P ) ≤ M (P ) 1 +
M (P + )[Q − 1]2
 
+ θC
≤ M (P ) 1 + , (117)
M (P)[Q − 1]2
where the first inequality comes from (116) and and the second comes from (26). The right hand side inequality of (78) is
shown as
 
Q+
 +
 θC
1 − M (P ) ≥ 1 − M (P ) 1 −
[1 − M (P + )][Q − 1]2
 
  θC
≥ 1 − M (P + ) 1 − , (118)
[1 − θC ][Q − 1]2
where the first inequality comes from (78) and the last inequality comes from (25). 

ACKNOWLEDGMENTS
The author thanks Dr. Y. Sakai and Prof. K. Iwata for the introduction of the polar codes and helpful discussions. The author
also thanks Dr. K. Arai for helpful discussions.

A PPENDIX
Lemma 13:
|θ + θ′ | + |θ′ − θ|
max{|θ|, |θ′ |} =
2
Proof: The equality is trivial when θ, θ′ ≥ 0 or θ, θ′ ≤ 0 using the relation
θ + θ′ + |θ′ − θ|
max{θ, θ′ } = . (119)
2
For the rest of the proof, we assume that θ ≤ 0 ≤ θ′ without loss of generality. In this case, we have
|θ| + |θ′ | + ||θ′ | − |θ||
max{|θ|, |θ′ |} =
2
−θ + θ′ + |θ′ − [−θ]|
=
2
| − θ + θ′ | + |θ′ + θ|
=
2
|θ + θ′ | + |θ′ − θ|
= , (120)
2
where the first equality comes from (119) and the third comes from the fact that −θ + θ′ > 0.
The following lemma seems to be trivial but we show it for completeness.
 26

Lemma 14: Let v ≡ (v1 , . . . , vk ), θv = (θv1 , . . . , θvk ), and θ = (θ1 , . . . , θk ). For given function g, we have
[
{v : θ v = θ} = {v : g(θv ) = θ}.
θ:g(θ)=θ

for all θ.
Proof: We prove the lemma by showing
[
{v : θ v = θ} ⊂ {v : g(θv ) = θ} (121)
θ:g(θ)=θ
[
{v : θv = θ} ⊃ {v : g(θv ) = θ}. (122)
θ:g(θ)=θ

First, assume that [

v∈ {v : θv = θ}.
θ:g(θ)=θ

Then there is vector θ such that g(θ) = θ and θv = θ. This implies that g(θv ) = θ and v ∈ {v : g(θv ) = θ}. Then we have
(121).
Next assume that
v ∈ {v : g(θv ) = θ}.
By letting θi ≡ θvi for each i ∈ {1, . . . , k}, we have
g(θ) = g(θv )
= θ. (123)
Then we have vector θ such that g(θ) = θ and θv = θ. This implies that v ∈ {v : g(θv ) = θ} and (122).

R EFERENCES
[1] M. Alsan, “Re-proving channel polarization theorems: a extremality and robustness analysis,” 2015, [online] Available:
https://infoscience.epfl.ch/record/203886/files/EPFL TH6403.pdf.
[2] E. Arıkan, “Channel polarization: a method for constructing capacity-achieving codes for symmetric binary-input memoryless channels,” IEEE Trans.
Inform. Theory, vol. IT-55, no. 7, pp. 3051–3073, Jul. 2009.
[3] E. Arıkan, “Source polarization,” Proc. 2010 IEEE Int. Symp. Inform. Theory, Austin, U.S.A., June 13–18, 2010, pp. 899–903.
[4] E. Arıkan, “Systematic polar coding,” IEEE Communications Letters, vol. 15, no. 8, pp. 860–862, Aug. 2010.
[5] E. Arıkan and E. Telatar, “On the rate of channel polarization,” Proc. 2009 IEEE Int. Symp. Inform. Theory, Seoul, Korea, Jun. 28–Jul. 3, 2009, pp.
1493-1495.
[6] J. T. Chu and J. C. Chueh, “Inequalities between information measures and error probability,” J. Franklin Inst., vol. 282, pp. 121–125, Aug. 1966.
[7] T. C. Gulcu, M. Ye, and A. Barg, “Construction of polar codes for arbitrary discrete memoryless channels,” IEEE Trans. Inform. Theory, vol. 64, no.
1, pp. 309–321, Jan. 2018.
[8] J. Honda and H. Yamamoto, “Polar coding without alphabet extension for asymmetric models,” IEEE Trans. Inform. Theory, vol. IT-59, no. 12, pp. 7829–
7838, Dec. 2013.
[9] S. B. Korada and R. L. Urbanke, “Polar codes are optimal for lossy source coding,” IEEE Trans. Inform. Theory, vol. IT-56, no. 4, pp. 1751–1768, Apr.
2010.
[10] R. Mori and T. Tanaka, “Source and channel polarization over finite fields and Reed-Solomon matrices,” IEEE Trans. Inform. Theory, vol. IT-60, no. 5,
pp. 2720–2736, May 2014.
[11] J. Muramatsu, “Successive-cancellation decoding of linear source code,” Proceedings of the 2019 IEEE Information Theory Workshop, Visby, Sweden,
Aug. 25–28, 2019. Extended version available arXiv:1903.11787[cs.IT], 2019.
[12] J. Muramatsu, “Binary polar codes based on bit error probability,” [online] Available: arXiv:2101.08433[cs.IT], 2021.
[13] R. Pedarsani, S. H. Hassani, I. Tal, and E. Telatar, “On the construction of polar codes,” Proc. 2011 IEEE Int. Symp. Inform. Theory, St. Petersburg,
Russia, Jul. 31–Aug. 5, 2011, pp. 11–15.
[14] E. Şaşoğlu, “Polarization and polar codes,” Fund. Trends Commun. Inf. Theory, vol. 8, no. 4, pp. 259–381, Oct. 2012.
[15] I. Tal and A. Vardy, “How to construct polar codes,” IEEE Trans. Inform. Theory, vol. IT-59, no. 10, pp. 6563–6582, Oct. 2013.
 27

TABLE I
B IT E RROR P ROBABILITY OF P OLAR C ODE WITH B LOCK L ENGTH 28 = 256

b8 Upper bound b8 Upper bound

00000000 −1.189 × 10−28 00110010 −9.816 × 10−3
00000001 −7.710 × 10−15 01010001 −1.041 × 10−2
00000010 −7.710 × 10−15 01001100 −1.265 × 10−2
00000100 −7.710 × 10−15 10010001 −1.362 × 10−2
00001000 −7.710 × 10−15 10001100 −1.602 × 10−2
00010000 −7.710 × 10−15 01010010 −1.742 × 10−2
00100000 −7.710 × 10−15 00110100 −1.759 × 10−2
01000000 −7.710 × 10−15 10010010 −2.333 × 10−2
10000000 −7.710 × 10−15 01100001 −2.721 × 10−2
00000011 −9.313 × 10−8 00001111 −2.946 × 10−2
00000101 −1.086 × 10−7 00111000 −3.142 × 10−2
00001001 −1.202 × 10−7 01010100 −3.251 × 10−2
00010001 −1.238 × 10−7 00010111 −3.954 × 10−2
00100001 −1.241 × 10−7 10100001 −4.300 × 10−2
01000001 −1.241 × 10−7 01100010 −4.302 × 10−2
10000001 −1.241 × 10−7 10010100 −4.541 × 10−2
00000110 −1.397 × 10−7 00011011 −5.379 × 10−2
00001010 −1.901 × 10−7 00100111 −5.455 × 10−2
00010010 −2.319 × 10−7 01011000 −5.931 × 10−2
00100010 −2.456 × 10−7 01000111 −6.394 × 10−2
01000010 −2.477 × 10−7 10100010 −6.727 × 10−2
10000010 −2.481 × 10−7 10000111 −6.765 × 10−2
00001100 −3.141 × 10−7 00011101 −6.942 × 10−2
00010100 −5.698 × 10−7 01100100 −7.258 × 10−2
00100100 −8.016 × 10−7 00101011 −7.759 × 10−2
01000100 −9.004 × 10−7 00011110 −8.498 × 10−2
10000100 −9.340 × 10−7 10011000 −8.633 × 10−2
00011000 −1.513 × 10−6 01001011 −9.793 × 10−2
00101000 −3.880 × 10−6 11000001 −9.994 × 10−2
01001000 −6.249 × 10−6 00101101 −1.024 × 10−1
10001000 −7.642 × 10−6 10001011 −1.086 × 10−1
00110000 −1.930 × 10−5 10100100 −1.104 × 10−1
01010000 −5.638 × 10−5 00110011 −1.150 × 10−1
10010000 −9.531 × 10−5 01101000 −1.189 × 10−1
00000111 −3.854 × 10−4 00101110 −1.259 × 10−1
01100000 −3.961 × 10−4 01001101 −1.360 × 10−1
00001011 −4.995 × 10−4 11000010 −1.420 × 10−1
00010011 −6.382 × 10−4 00110101 −1.505 × 10−1
00001101 −6.707 × 10−4 01010011 −1.569 × 10−1
00100011 −6.933 × 10−4 10001101 −1.570 × 10−1
01000011 −7.023 × 10−4 01001110 −1.716 × 10−1
10000011 −7.037 × 10−4 10101000 −1.726 × 10−1
00001110 −8.425 × 10−4 01110000 −1.796 × 10−1
00010101 −9.856 × 10−4 00110110 −1.841 × 10−1
10100000 −9.936 × 10−4 10010011 −1.852 × 10−1
00100101 −1.243 × 10−3 00111001 −1.986 × 10−1
01000101 −1.336 × 10−3 11000100 −2.016 × 10−1
10000101 −1.364 × 10−3 10001110 −2.035 × 10−1
00010110 −1.384 × 10−3 01010101 −2.057 × 10−1
00011001 −1.637 × 10−3 01100011 −2.334 × 10−1
00100110 −2.079 × 10−3 00111010 −2.383 × 10−1
00011010 −2.432 × 10−3 10110000 −2.442 × 10−1
01000110 −2.440 × 10−3 10010101 −2.451 × 10−1
10000110 −2.572 × 10−3 01010110 −2.483 × 10−1
00101001 −2.715 × 10−3 01011001 2.341 × 10−1
01001001 −3.503 × 10−3 11001000 2.305 × 10−1
00011100 −3.743 × 10−3 00011111 2.199 × 10−1
10001001 −3.891 × 10−3 00111100 2.143 × 10−1
00101010 −4.417 × 10−3 10100011 2.129 × 10−1
11000000 −5.796 × 10−3 01100101 2.118 × 10−1
00110001 −6.044 × 10−3 10010110 2.057 × 10−1
01001010 −6.159 × 10−3 01011010 1.910 × 10−1
10001010 −7.093 × 10−3 10011001 1.844 × 10−1
00101100 −7.742 × 10−3 01100110 1.690 × 10−1
 28

b8 Upper bound b8 Upper bound

11010000 1.657 × 10−1 11110000 1.077 × 10−3
00101111 1.643 × 10−1 01111001 9.443 × 10−4
10100101 1.547 × 10−1 11001011 8.848 × 10−4
01101001 1.521 × 10−1 10110101 5.891 × 10−4
01011100 1.456 × 10−1 01111010 5.128 × 10−4
10011010 1.409 × 10−1 11001101 3.489 × 10−4
11000011 1.218 × 10−1 10110110 3.133 × 10−4
01101010 1.151 × 10−1 01111100 2.698 × 10−4
10100110 1.147 × 10−1 11010011 2.279 × 10−4
01001111 1.137 × 10−1 10111001 2.177 × 10−4
11100000 1.107 × 10−1 01011111 1.939 × 10−4
00110111 1.097 × 10−1 11001110 1.832 × 10−4
10011100 9.992 × 10−2 10111010 1.145 × 10−4
10101001 9.902 × 10−2 11010101 8.492 × 10−5
01110001 9.796 × 10−2 10111100 5.894 × 10−5
10001111 8.192 × 10−2 11100011 4.712 × 10−5
01101100 8.007 × 10−2 11010110 4.390 × 10−5
11000101 7.856 × 10−2 10011111 3.408 × 10−5
00111011 7.106 × 10−2 11011001 2.831 × 10−5
01110010 7.039 × 10−2 01101111 1.994 × 10−5
10101010 6.970 × 10−2 11100101 1.690 × 10−5
01010111 6.330 × 10−2 11011010 1.453 × 10−5
10110001 5.634 × 10−2 11100110 8.643 × 10−6
11000110 5.348 × 10−2 11011100 7.370 × 10−6
01110100 4.672 × 10−2 11101001 5.523 × 10−6
00111101 4.597 × 10−2 11101010 2.814 × 10−6
10101100 4.517 × 10−2 01110111 2.365 × 10−6
11001001 4.491 × 10−2 10101111 1.802 × 10−6
10110010 3.785 × 10−2 11110001 1.676 × 10−6
10010111 3.758 × 10−2 11101100 1.422 × 10−6
01011011 3.526 × 10−2 11110010 8.506 × 10−7
00111110 3.070 × 10−2 11110100 4.291 × 10−7
11001010 2.928 × 10−2 01111011 4.092 × 10−7
01111000 2.878 × 10−2 11111000 2.156 × 10−7
01100111 2.783 × 10−2 10110111 1.450 × 10−7
10110100 2.363 × 10−2 01111101 1.196 × 10−7
11010001 2.296 × 10−2 01111110 5.998 × 10−8
01011101 2.040 × 10−2 11001111 5.543 × 10−8
11001100 1.771 × 10−2 10111011 2.058 × 10−8
10011011 1.761 × 10−2 10111101 5.666 × 10−9
11010010 1.441 × 10−2 11010111 3.351 × 10−9
10111000 1.381 × 10−2 10111110 2.837 × 10−9
01101011 1.313 × 10−2 11011011 3.892 × 10−10
01011110 1.264 × 10−2 11100111 1.312 × 10−10
10100111 1.186 × 10−2 11011101 1.026 × 10−10
11100001 1.054 × 10−2 11011110 5.132 × 10−11
10011101 9.104 × 10−3 11101011 1.468 × 10−11
11010100 8.460 × 10−3 11101101 3.826 × 10−12
01101101 6.578 × 10−3 11101110 1.913 × 10−12
11100010 6.407 × 10−3 11110011 1.423 × 10−12
10011110 5.317 × 10−3 11110101 3.693 × 10−13
01110011 4.951 × 10−3 11110110 1.847 × 10−13
11011000 4.707 × 10−3 11111001 9.440 × 10−14
10101011 4.612 × 10−3 11111010 4.720 × 10−14
01101110 3.767 × 10−3 11111100 2.361 × 10−14
11100100 3.682 × 10−3 01111111 8.090 × 10−15
11000111 2.719 × 10−3 10111111 1.809 × 10−17
01110101 2.306 × 10−3 11011111 7.304 × 10−21
10101101 2.053 × 10−3 11101111 9.793 × 10−24
11101000 2.024 × 10−3 11110111 9.457 × 10−26
10110011 1.410 × 10−3 11111011 6.283 × 10−27
01110110 1.273 × 10−3 11111101 1.572 × 10−27
10101110 1.121 × 10−3 11111110 7.856 × 10−28
00111111 1.110 × 10−3 11111111 2.439 × 10−54
 29

TABLE II
B IT E RROR P ROBABILITY OF P OLAR C ODE WITH B LOCK L ENGTH 210 = 1024

b10 Lower bound Upper bound b10 Lower bound Upper bound
0000000000 −1.599×10−111 −1.598×10−111 0000100101 −5.749 × 10−14 −5.748 × 10−14
0000000001 −2.828 × 10−56 −2.827 × 10−56 0001000101 −6.141 × 10−14 −6.140 × 10−14
0000000010 −2.828 × 10−56 −2.827 × 10−56 0010000101 −6.168 × 10−14 −6.167 × 10−14
0000000100 −2.828 × 10−56 −2.827 × 10−56 0100000101 −6.169 × 10−14 −6.168 × 10−14
0000001000 −2.828 × 10−56 −2.827 × 10−56 1000000101 −6.169 × 10−14 −6.168 × 10−14
0000010000 −2.828 × 10−56 −2.827 × 10−56 0000010110 −6.202 × 10−14 −6.201 × 10−14
0000100000 −2.828 × 10−56 −2.827 × 10−56 0000011001 −7.400 × 10−14 −7.399 × 10−14
0001000000 −2.828 × 10−56 −2.827 × 10−56 0000100110 −1.026 × 10−13 −1.025 × 10−13
0010000000 −2.828 × 10−56 −2.827 × 10−56 0000011010 −1.119 × 10−13 −1.118 × 10−13
0100000000 −2.828 × 10−56 −2.827 × 10−56 0001000110 −1.218 × 10−13 −1.217 × 10−13
1000000000 −2.828 × 10−56 −2.827 × 10−56 0010000110 −1.234 × 10−13 −1.233 × 10−13
0000000011 −1.784 × 10−28 −1.783 × 10−28 0100000110 −1.234 × 10−13 −1.233 × 10−13
0000000101 −2.081 × 10−28 −2.080 × 10−28 1000000110 −1.234 × 10−13 −1.233 × 10−13
0000001001 −2.304 × 10−28 −2.303 × 10−28 0000101001 −1.430 × 10−13 −1.429 × 10−13
0000010001 −2.374 × 10−28 −2.373 × 10−28 0000011100 −1.766 × 10−13 −1.765 × 10−13
0000100001 −2.379 × 10−28 −2.378 × 10−28 0110000000 −1.971 × 10−13 −1.970 × 10−13
0001000001 −2.379 × 10−28 −2.378 × 10−28 0001001001 −2.152 × 10−13 −2.151 × 10−13
0010000001 −2.379 × 10−28 −2.378 × 10−28 0010001001 −2.414 × 10−13 −2.413 × 10−13
0100000001 −2.379 × 10−28 −2.378 × 10−28 0100001001 −2.456 × 10−13 −2.455 × 10−13
1000000001 −2.379 × 10−28 −2.378 × 10−28 1000001001 −2.463 × 10−13 −2.462 × 10−13
0000000110 −2.676 × 10−28 −2.675 × 10−28 0000101010 −2.486 × 10−13 −2.485 × 10−13
0000001010 −3.642 × 10−28 −3.641 × 10−28 0000110001 −3.923 × 10−13 −3.922 × 10−13
0000010010 −4.464 × 10−28 −4.463 × 10−28 0001001010 −4.236 × 10−13 −4.235 × 10−13
0000100010 −4.738 × 10−28 −4.737 × 10−28 0010001010 −4.826 × 10−13 −4.825 × 10−13
0001000010 −4.756 × 10−28 −4.755 × 10−28 0100001010 −4.912 × 10−13 −4.911 × 10−13
0010000010 −4.757 × 10−28 −4.756 × 10−28 1000001010 −4.926 × 10−13 −4.925 × 10−13
0100000010 −4.757 × 10−28 −4.756 × 10−28 0000101100 −4.984 × 10−13 −4.983 × 10−13
1000000010 −4.757 × 10−28 −4.756 × 10−28 0000110010 −7.094 × 10−13 −7.093 × 10−13
0000001100 −6.020 × 10−28 −6.019 × 10−28 0001010001 −1.299 × 10−12 −1.298 × 10−12
0000010100 −1.116 × 10−27 −1.115 × 10−27 0001001100 −1.328 × 10−12 −1.327 × 10−12
0000100100 −1.676 × 10−27 −1.675 × 10−27 0000110100 −1.620 × 10−12 −1.619 × 10−12
0001000100 −1.886 × 10−27 −1.885 × 10−27 0010001100 −1.850 × 10−12 −1.849 × 10−12
0010000100 −1.902 × 10−27 −1.901 × 10−27 0100001100 −1.947 × 10−12 −1.946 × 10−12
0100000100 −1.903 × 10−27 −1.902 × 10−27 1000001100 −1.963 × 10−12 −1.962 × 10−12
1000000100 −1.903 × 10−27 −1.902 × 10−27 0001010010 −2.535 × 10−12 −2.534 × 10−12
0000011000 −3.048 × 10−27 −3.047 × 10−27 0010010001 −2.571 × 10−12 −2.570 × 10−12
0000101000 −1.046 × 10−26 −1.045 × 10−26 0100010001 −3.244 × 10−12 −3.243 × 10−12
0001001000 −2.315 × 10−26 −2.314 × 10−26 1000010001 −3.490 × 10−12 −3.489 × 10−12
0010001000 −2.914 × 10−26 −2.913 × 10−26 0000111000 −4.032 × 10−12 −4.031 × 10−12
0100001000 −3.016 × 10−26 −3.015 × 10−26 0010010010 −5.135 × 10−12 −5.134 × 10−12
1000001000 −3.033 × 10−26 −3.032 × 10−26 0100010010 −6.487 × 10−12 −6.486 × 10−12
0000110000 −7.796 × 10−26 −7.795 × 10−26 1000010010 −6.979 × 10−12 −6.978 × 10−12
0001010000 −8.435 × 10−25 −8.434 × 10−25 0001010100 −7.551 × 10−12 −7.550 × 10−12
0010010000 −3.304 × 10−24 −3.303 × 10−24 1010000000 −7.800 × 10−12 −7.799 × 10−12
0100010000 −5.261 × 10−24 −5.260 × 10−24 0001100001 −9.160 × 10−12 −9.159 × 10−12
1000010000 −6.089 × 10−24 −6.088 × 10−24 0001100010 −1.803 × 10−11 −1.802 × 10−11
0001100000 −4.197 × 10−23 −4.196 × 10−23 0010010100 −1.915 × 10−11 −1.914 × 10−11
0010100000 −1.814 × 10−21 −1.813 × 10−21 0100010100 −2.551 × 10−11 −2.550 × 10−11
0100100000 −1.221 × 10−20 −1.220 × 10−20 1000010100 −2.772 × 10−11 −2.771 × 10−11
1000100000 −2.729 × 10−20 −2.728 × 10−20 0001011000 −2.938 × 10−11 −2.937 × 10−11
0011000000 −1.112 × 10−18 −1.111 × 10−18 0001100100 −5.747 × 10−11 −5.746 × 10−11
0101000000 −8.085 × 10−17 −8.084 × 10−17 0010100001 −6.023 × 10−11 −6.022 × 10−11
1001000000 −6.602 × 10−16 −6.601 × 10−16 0010100010 −1.202 × 10−10 −1.201 × 10−10
0000000111 −1.687 × 10−14 −1.686 × 10−14 0010011000 −1.496 × 10−10 −1.495 × 10−10
0000001011 −2.187 × 10−14 −2.186 × 10−14 0100100001 −1.563 × 10−10 −1.562 × 10−10
0000010011 −2.813 × 10−14 −2.812 × 10−14 1000100001 −2.337 × 10−10 −2.336 × 10−10
0000001101 −2.937 × 10−14 −2.936 × 10−14 0001101000 −2.801 × 10−10 −2.800 × 10−10
0000100011 −3.067 × 10−14 −3.066 × 10−14 0100011000 −2.841 × 10−10 −2.840 × 10−10
0001000011 −3.085 × 10−14 −3.084 × 10−14 0100100010 −3.124 × 10−10 −3.123 × 10−10
0010000011 −3.085 × 10−14 −3.084 × 10−14 1000011000 −3.506 × 10−10 −3.505 × 10−10
0100000011 −3.085 × 10−14 −3.084 × 10−14 0010100100 −4.353 × 10−10 −4.352 × 10−10
1000000011 −3.085 × 10−14 −3.084 × 10−14 1000100010 −4.672 × 10−10 −4.671 × 10−10
0000001110 −3.690 × 10−14 −3.689 × 10−14 0100100100 −1.206 × 10−9 −1.205 × 10−9
0000010101 −4.380 × 10−14 −4.379 × 10−14 0011000001 −1.491 × 10−9 −1.490 × 10−9
 30

b10 Lower bound Upper bound b10 Lower bound Upper bound
0001110000 −1.571 × 10−9 −1.570 × 10−9 0100010101 −7.142 × 10−6 −7.141 × 10−6
1000100100 −1.836 × 10−9 −1.835 × 10−9 0001011001 −7.373 × 10−6 −7.372 × 10−6
0011000010 −2.972 × 10−9 −2.971 × 10−9 1000010101 −7.446 × 10−6 −7.445 × 10−6
0010101000 −3.046 × 10−9 −3.045 × 10−9 0011100000 −7.798 × 10−6 −7.797 × 10−6
1100000000 −9.032 × 10−9 −9.031 × 10−9 1010000010 −7.859 × 10−6 −7.858 × 10−6
0011000100 −1.068 × 10−8 −1.067 × 10−8 0101010000 −8.940 × 10−6 −8.939 × 10−6
0100101000 −1.152 × 10−8 −1.151 × 10−8 0001100101 −1.045 × 10−5 −1.044 × 10−5
0101000001 −1.272 × 10−8 −1.271 × 10−8 0010010110 −1.155 × 10−5 −1.154 × 10−5
1000101000 −2.025 × 10−8 −2.024 × 10−8 0001011010 −1.160 × 10−5 −1.159 × 10−5
0101000010 −2.540 × 10−8 −2.539 × 10−8 0100010110 −1.405 × 10−5 −1.404 × 10−5
0010110000 −2.876 × 10−8 −2.875 × 10−8 1000010110 −1.479 × 10−5 −1.478 × 10−5
1001000001 −3.634 × 10−8 −3.633 × 10−8 0010100011 −1.546 × 10−5 −1.545 × 10−5
1001000010 −7.262 × 10−8 −7.261 × 10−8 0001100110 −1.721 × 10−5 −1.720 × 10−5
0011001000 −7.428 × 10−8 −7.427 × 10−8 0010011001 −1.722 × 10−5 −1.721 × 10−5
0101000100 −9.430 × 10−8 −9.429 × 10−8 0001011100 −1.957 × 10−5 −1.956 × 10−5
0000001111 −1.951 × 10−7 −1.950 × 10−7 0001101001 −2.285 × 10−5 −2.284 × 10−5
0100110000 −2.055 × 10−7 −2.054 × 10−7 0100011001 −2.383 × 10−5 −2.382 × 10−5
0000010111 −2.660 × 10−7 −2.659 × 10−7 0100100011 −2.499 × 10−5 −2.498 × 10−5
1001000100 −2.755 × 10−7 −2.754 × 10−7 1000011001 −2.648 × 10−5 −2.647 × 10−5
0000011011 −3.693 × 10−7 −3.692 × 10−7 1010000100 −2.736 × 10−5 −2.735 × 10−5
0000100111 −4.059 × 10−7 −4.058 × 10−7 0110001000 −2.742 × 10−5 −2.741 × 10−5
0001000111 −4.890 × 10−7 −4.889 × 10−7 0010100101 −2.936 × 10−5 −2.935 × 10−5
0000011101 −4.890 × 10−7 −4.889 × 10−7 1000100011 −3.057 × 10−5 −3.056 × 10−5
0010000111 −4.966 × 10−7 −4.965 × 10−7 0010011010 −3.135 × 10−5 −3.134 × 10−5
0100000111 −4.968 × 10−7 −4.967 × 10−7 1001010000 −3.405 × 10−5 −3.404 × 10−5
1000000111 −4.968 × 10−7 −4.967 × 10−7 0001101010 −3.674 × 10−5 −3.673 × 10−5
1000110000 −5.270 × 10−7 −5.269 × 10−7 0100011010 −4.630 × 10−5 −4.629 × 10−5
0000011110 −6.131 × 10−7 −6.130 × 10−7 0100100101 −4.907 × 10−5 −4.906 × 10−5
0000101011 −6.170 × 10−7 −6.169 × 10−7 1000011010 −5.228 × 10−5 −5.227 × 10−5
0110000001 −6.278 × 10−7 −6.277 × 10−7 0010100110 −5.352 × 10−5 −5.351 × 10−5
0101001000 −7.373 × 10−7 −7.372 × 10−7 0001110001 −5.484 × 10−5 −5.483 × 10−5
0011010000 −7.675 × 10−7 −7.674 × 10−7 1000100101 −6.058 × 10−5 −6.057 × 10−5
0000101101 −8.724 × 10−7 −8.723 × 10−7 0001101100 −6.701 × 10−5 −6.700 × 10−5
0001001011 −9.061 × 10−7 −9.060 × 10−7 0010011100 −7.086 × 10−5 −7.085 × 10−5
0010001011 −9.820 × 10−7 −9.819 × 10−7 0011000011 −7.683 × 10−5 −7.682 × 10−5
0100001011 −9.911 × 10−7 −9.910 × 10−7 0010101001 −7.730 × 10−5 −7.729 × 10−5
1000001011 −9.926 × 10−7 −9.925 × 10−7 0001110010 −9.243 × 10−5 −9.242 × 10−5
0000110011 −1.080 × 10−6 −1.079 × 10−6 0100100110 −9.474 × 10−5 −9.473 × 10−5
0000101110 −1.139 × 10−6 −1.138 × 10−6 0101100000 −9.901 × 10−5 −9.900 × 10−5
0110000010 −1.250 × 10−6 −1.249 × 10−6 1000100110 −1.190 × 10−4 −1.189 × 10−4
0001001101 −1.590 × 10−6 −1.589 × 10−6 0100011100 −1.338 × 10−4 −1.337 × 10−4
0000110101 −1.603 × 10−6 −1.602 × 10−6 1100000001 −1.344 × 10−4 −1.343 × 10−4
0010001101 −1.922 × 10−6 −1.921 × 10−6 0010101010 −1.365 × 10−4 −1.364 × 10−4
0100001101 −1.973 × 10−6 −1.972 × 10−6 0011000101 −1.453 × 10−4 −1.452 × 10−4
1000001101 −1.982 × 10−6 −1.981 × 10−6 0100101001 −1.515 × 10−4 −1.514 × 10−4
0000110110 −2.194 × 10−6 −2.193 × 10−6 1010001000 −1.640 × 10−4 −1.639 × 10−4
0001010011 −2.199 × 10−6 −2.198 × 10−6 1000011100 −1.677 × 10−4 −1.676 × 10−4
1001001000 −2.372 × 10−6 −2.371 × 10−6 0001110100 −1.859 × 10−4 −1.858 × 10−4
0001001110 −2.581 × 10−6 −2.580 × 10−6 1000101001 −2.012 × 10−4 −2.011 × 10−4
0000111001 −2.625 × 10−6 −2.624 × 10−6 0110010000 −2.221 × 10−4 −2.220 × 10−4
0010010011 −3.201 × 10−6 −3.200 × 10−6 0101000011 −2.251 × 10−4 −2.250 × 10−4
0100010011 −3.602 × 10−6 −3.601 × 10−6 0010110001 −2.371 × 10−4 −2.370 × 10−4
0010001110 −3.686 × 10−6 −3.685 × 10−6 0011000110 −2.628 × 10−4 −2.627 × 10−4
1000010011 −3.736 × 10−6 −3.735 × 10−6 1100000010 −2.632 × 10−4 −2.631 × 10−4
0001010101 −3.747 × 10−6 −3.746 × 10−6 0100101010 −2.865 × 10−4 −2.864 × 10−4
0000111010 −3.842 × 10−6 −3.841 × 10−6 0010101100 −2.901 × 10−4 −2.900 × 10−4
0100001110 −3.910 × 10−6 −3.909 × 10−6 0011001001 −3.807 × 10−4 −3.806 × 10−4
1000001110 −3.948 × 10−6 −3.947 × 10−6 1001000011 −3.809 × 10−4 −3.808 × 10−4
1010000001 −3.950 × 10−6 −3.949 × 10−6 1000101010 −3.900 × 10−4 −3.899 × 10−4
0110000100 −4.387 × 10−6 −4.386 × 10−6 0010110010 −4.149 × 10−4 −4.148 × 10−4
0001010110 −5.874 × 10−6 −5.873 × 10−6 0001111000 −4.173 × 10−4 −4.172 × 10−4
0001100011 −5.908 × 10−6 −5.907 × 10−6 0101000101 −4.331 × 10−4 −4.330 × 10−4
0000111100 −6.033 × 10−6 −6.032 × 10−6 1001100000 −4.446 × 10−4 −4.445 × 10−4
0010010101 −6.174 × 10−6 −6.173 × 10−6 0100110001 −6.383 × 10−4 −6.382 × 10−4
 31

b10 Lower bound Upper bound b10 Lower bound Upper bound
0011001010 −6.635 × 10−4 −6.634 × 10−4 0001101101 −9.555 × 10−3 −9.554 × 10−3
0100101100 −7.359 × 10−4 −7.358 × 10−4 1000011011 −1.010 × 10−2 −1.009 × 10−2
1001000101 −7.412 × 10−4 −7.411 × 10−4 0010011101 −1.071 × 10−2 −1.070 × 10−2
0000011111 −7.891 × 10−4 −7.890 × 10−4 0001110011 −1.156 × 10−2 −1.155 × 10−2
1100000100 −7.983 × 10−4 −7.982 × 10−4 0110001010 −1.229 × 10−2 −1.228 × 10−2
0101000110 −8.068 × 10−4 −8.067 × 10−4 0001101110 −1.231 × 10−2 −1.230 × 10−2
0010110100 −8.798 × 10−4 −8.797 × 10−4 0011100100 −1.245 × 10−2 −1.244 × 10−2
1000110001 −1.025 × 10−3 −1.024 × 10−3 1010000110 −1.288 × 10−2 −1.287 × 10−2
0000101111 −1.098 × 10−3 −1.097 × 10−3 1100010000 −1.323 × 10−2 −1.322 × 10−2
1000101100 −1.114 × 10−3 −1.113 × 10−3 0100100111 −1.330 × 10−2 −1.329 × 10−2
0100110010 −1.170 × 10−3 −1.169 × 10−3 0101100001 −1.369 × 10−2 −1.368 × 10−2
1010010000 −1.193 × 10−3 −1.192 × 10−3 1000111000 −1.374 × 10−2 −1.373 × 10−2
0101001001 −1.207 × 10−3 −1.206 × 10−3 1001010010 −1.419 × 10−2 −1.418 × 10−2
0011010001 −1.214 × 10−3 −1.213 × 10−3 0101010100 −1.435 × 10−2 −1.434 × 10−2
0011001100 −1.401 × 10−3 −1.400 × 10−3 0010011110 −1.450 × 10−2 −1.449 × 10−2
1001000110 −1.408 × 10−3 −1.407 × 10−3 0010101011 −1.464 × 10−2 −1.463 × 10−2
0000110111 −1.542 × 10−3 −1.541 × 10−3 1000100111 −1.516 × 10−2 −1.515 × 10−2
0110000011 −1.574 × 10−3 −1.573 × 10−3 0100011101 −1.567 × 10−2 −1.566 × 10−2
0110100000 −1.600 × 10−3 −1.599 × 10−3 0001110101 −1.622 × 10−2 −1.621 × 10−2
0001001111 −1.879 × 10−3 −1.878 × 10−3 1010001001 −1.774 × 10−2 −1.773 × 10−2
1000110010 −1.928 × 10−3 −1.927 × 10−3 1000011101 −1.792 × 10−2 −1.791 × 10−2
0010111000 −2.012 × 10−3 −2.011 × 10−3 0110010001 −2.031 × 10−2 −2.030 × 10−2
0011010010 −2.051 × 10−3 −2.050 × 10−3 0011000111 −2.079 × 10−2 −2.078 × 10−2
0000111011 −2.116 × 10−3 −2.115 × 10−3 0010101101 −2.101 × 10−2 −2.100 × 10−2
1001001001 −2.170 × 10−3 −2.169 × 10−3 0001110110 −2.138 × 10−2 −2.137 × 10−2
0101001010 −2.171 × 10−3 −2.170 × 10−3 0101100010 −2.225 × 10−2 −2.224 × 10−2
0010001111 −2.606 × 10−3 −2.605 × 10−3 1100000011 −2.247 × 10−2 −2.246 × 10−2
0100110100 −2.738 × 10−3 −2.737 × 10−3 0100101011 −2.265 × 10−2 −2.264 × 10−2
0001010111 −2.755 × 10−3 −2.754 × 10−3 0110001100 −2.375 × 10−2 −2.374 × 10−2
0100001111 −2.771 × 10−3 −2.770 × 10−3 0100011110 −2.397 × 10−2 −2.396 × 10−2
1000001111 −2.800 × 10−3 −2.799 × 10−3 0001111001 −2.512 × 10−2 −2.511 × 10−2
0000111101 −2.801 × 10−3 −2.800 × 10−3 0010110011 −2.532 × 10−2 −2.531 × 10−2
0110000101 −2.937 × 10−3 −2.936 × 10−3 0011101000 −2.585 × 10−2 −2.584 × 10−2
1100001000 −3.261 × 10−3 −3.260 × 10−3 1000101011 −2.708 × 10−2 −2.707 × 10−2
0000111110 −3.522 × 10−3 −3.521 × 10−3 0010101110 −2.789 × 10−2 −2.788 × 10−2
0011100001 −3.862 × 10−3 −3.861 × 10−3 1011000000 −2.850 × 10−2 −2.849 × 10−2
0001011011 −3.873 × 10−3 −3.872 × 10−3 1001010100 −2.889 × 10−2 −2.888 × 10−2
1010000011 −3.944 × 10−3 −3.943 × 10−3 1001100001 −2.892 × 10−2 −2.891 × 10−2
1001001010 −3.996 × 10−3 −3.995 × 10−3 1000011110 −2.944 × 10−2 −2.943 × 10−2
0011010100 −4.115 × 10−3 −4.114 × 10−3 1010001010 −2.948 × 10−2 −2.947 × 10−2
0101010001 −4.165 × 10−3 −4.164 × 10−3 0101011000 −3.043 × 10−2 −3.042 × 10−2
0010010111 −4.501 × 10−3 −4.500 × 10−3 0011001011 −3.188 × 10−2 −3.187 × 10−2
0101001100 −4.850 × 10−3 −4.849 × 10−3 0110010010 −3.209 × 10−2 −3.208 × 10−2
1000110100 −4.866 × 10−3 −4.865 × 10−3 0001111010 −3.410 × 10−2 −3.409 × 10−2
0001100111 −4.891 × 10−3 −4.890 × 10−3 0100101101 −3.552 × 10−2 −3.551 × 10−2
0001011101 −5.192 × 10−3 −5.191 × 10−3 0010110101 −3.612 × 10−2 −3.611 × 10−2
0110000110 −5.205 × 10−3 −5.204 × 10−3 0101000111 −3.745 × 10−2 −3.744 × 10−2
0100010111 −5.211 × 10−3 −5.210 × 10−3 1100000101 −3.859 × 10−2 −3.858 × 10−2
1000010111 −5.400 × 10−3 −5.399 × 10−3 0101100100 −4.004 × 10−2 −4.003 × 10−2
0011100010 −6.444 × 10−3 −6.443 × 10−3 1100100000 −4.222 × 10−2 −4.221 × 10−2
0001011110 −6.567 × 10−3 −6.566 × 10−3 0100110011 −4.440 × 10−2 −4.439 × 10−2
0100111000 −6.949 × 10−3 −6.948 × 10−3 1000101101 −4.494 × 10−2 −4.493 × 10−2
0001101011 −7.005 × 10−3 −7.004 × 10−3 0011001101 −4.525 × 10−2 −4.524 × 10−2
1010100000 −7.107 × 10−3 −7.106 × 10−3 1001100010 −4.621 × 10−2 −4.620 × 10−2
0101010010 −7.136 × 10−3 −7.135 × 10−3 1010010001 −4.662 × 10−2 −4.661 × 10−2
0010011011 −7.230 × 10−3 −7.229 × 10−3 0010110110 −4.785 × 10−2 −4.784 × 10−2
0110001001 −7.275 × 10−3 −7.274 × 10−3 0001111100 −4.928 × 10−2 −4.927 × 10−2
1010000101 −7.328 × 10−3 −7.327 × 10−3 1001000111 −5.040 × 10−2 −5.039 × 10−2
1001010001 −8.158 × 10−3 −8.157 × 10−3 0100101110 −5.062 × 10−2 −5.061 × 10−2
0111000000 −8.326 × 10−3 −8.325 × 10−3 0110100001 −5.279 × 10−2 −5.278 × 10−2
0011011000 −9.203 × 10−3 −9.202 × 10−3 0011110000 −5.333 × 10−2 −5.332 × 10−2
0100011011 −9.353 × 10−3 −9.352 × 10−3 0011010011 −5.441 × 10−2 −5.440 × 10−2
1001001100 −9.414 × 10−3 −9.413 × 10−3 1010001100 −5.442 × 10−2 −5.441 × 10−2
0010100111 −9.450 × 10−3 −9.449 × 10−3 0010111001 −5.509 × 10−2 −5.508 × 10−2
 32

b10 Lower bound Upper bound b10 Lower bound Upper bound
0110010100 −5.523 × 10−2 −5.522 × 10−2 0100111100 −1.783 × 10−1 −1.782 × 10−1
1000110011 −5.845 × 10−2 −5.844 × 10−2 0011101001 −1.797 × 10−1 −1.796 × 10−1
0101001011 −5.940 × 10−2 −5.939 × 10−2 1110000000 −1.840 × 10−1 −1.839 × 10−1
0000111111 −5.947 × 10−2 −5.946 × 10−2 0011011100 −1.857 × 10−1 −1.856 × 10−1
0011001110 −5.962 × 10−2 −5.961 × 10−2 0001111011 −1.874 × 10−1 −1.873 × 10−1
1001011000 −6.010 × 10−2 −6.009 × 10−2 1000111010 −1.893 × 10−1 −1.892 × 10−1
1100000110 −6.063 × 10−2 −6.062 × 10−2 1100010010 −1.919 × 10−1 −1.918 × 10−1
0100110101 −6.603 × 10−2 −6.602 × 10−2 0101011001 −1.964 × 10−1 −1.963 × 10−1
1000101110 −6.776 × 10−2 −6.775 × 10−2 1001010101 −1.996 × 10−1 −1.995 × 10−1
1010010010 −7.118 × 10−2 −7.117 × 10−2 1000011111 −2.005 × 10−1 −2.004 × 10−1
0101101000 −7.302 × 10−2 −7.301 × 10−2 0110010011 −2.015 × 10−1 −2.014 × 10−1
0010111010 −7.351 × 10−2 −7.350 × 10−2 1010001011 −2.022 × 10−1 −2.021 × 10−1
0011010101 −7.516 × 10−2 −7.515 × 10−2 1011000001 −2.025 × 10−1 −2.024 × 10−1
1100001001 −7.570 × 10−2 −7.569 × 10−2 1001110000 −2.050 × 10−1 −2.049 × 10−1
0110100010 −7.718 × 10−2 −7.717 × 10−2 1010100100 −2.069 × 10−1 −2.068 × 10−1
1001100100 −7.944 × 10−2 −7.943 × 10−2 0111000100 −2.091 × 10−1 −2.090 × 10−1
0001011111 −8.172 × 10−2 −8.171 × 10−2 0110001110 −2.186 × 10−1 −2.185 × 10−1
1001001011 −8.238 × 10−2 −8.237 × 10−2 0010110111 −2.189 × 10−1 −2.188 × 10−1
0101001101 −8.604 × 10−2 −8.603 × 10−2 0011101010 −2.192 × 10−1 −2.191 × 10−1
0100110110 −8.957 × 10−2 −8.956 × 10−2 0101100101 −2.218 × 10−1 −2.217 × 10−1
1000110101 −9.021 × 10−2 −9.020 × 10−2 0001111101 −2.280 × 10−1 −2.279 × 10−1
0110000111 −9.063 × 10−2 −9.062 × 10−2 0100101111 −2.353 × 10−1 −2.352 × 10−1
0011100011 −9.529 × 10−2 −9.528 × 10−2 0101011010 −2.386 × 10−1 −2.385 × 10−1
0110011000 −9.609 × 10−2 −9.608 × 10−2 1100100001 −2.388 × 10−1 −2.387 × 10−1
0011010110 −9.691 × 10−2 −9.690 × 10−2 0011001111 −2.419 × 10−1 −2.418 × 10−1
1101000000 −1.001 × 10−1 −1.000 × 10−1 1001100011 −2.437 × 10−1 −2.436 × 10−1
0010111100 −1.017 × 10−1 −1.016 × 10−1 1000111100 −2.445 × 10−1 −2.444 × 10−1
0100111001 −1.026 × 10−1 −1.025 × 10−1 1001010110 −2.471 × 10−1 −2.470 × 10−1
0101010011 −1.027 × 10−1 −1.026 × 10−1 0011110001 −2.487 × 10−1 −2.486 × 10−1
1010100001 −1.085 × 10−1 −1.084 × 10−1 0110110000 −2.488 × 10−1 −2.487 × 10−1
1100001010 −1.104 × 10−1 −1.103 × 10−1 1100010100 2.476 × 10−1 2.477 × 10−1
0011011001 −1.109 × 10−1 −1.108 × 10−1 0110010101 2.474 × 10−1 2.475 × 10−1
0001101111 −1.116 × 10−1 −1.115 × 10−1 1011000010 2.467 × 10−1 2.468 × 10−1
1010010100 −1.139 × 10−1 −1.138 × 10−1 1010001101 2.397 × 10−1 2.398 × 10−1
0101001110 −1.140 × 10−1 −1.139 × 10−1 0001111110 2.354 × 10−1 2.355 × 10−1
0111000001 −1.154 × 10−1 −1.153 × 10−1 0101100110 2.351 × 10−1 2.352 × 10−1
0110100100 −1.174 × 10−1 −1.173 × 10−1 0010111011 2.312 × 10−1 2.313 × 10−1
1001001101 −1.221 × 10−1 −1.220 × 10−1 1001011001 2.297 × 10−1 2.298 × 10−1
0010011111 −1.249 × 10−1 −1.248 × 10−1 0011101100 2.291 × 10−1 2.292 × 10−1
1000110110 −1.256 × 10−1 −1.255 × 10−1 0111001000 2.274 × 10−1 2.275 × 10−1
0101110000 −1.264 × 10−1 −1.263 × 10−1 1100000111 2.258 × 10−1 2.259 × 10−1
0011100101 −1.284 × 10−1 −1.283 × 10−1 1010101000 2.257 × 10−1 2.258 × 10−1
0110001011 −1.329 × 10−1 −1.328 × 10−1 1000101111 2.194 × 10−1 2.195 × 10−1
1001101000 −1.332 × 10−1 −1.331 × 10−1 0101101001 2.149 × 10−1 2.150 × 10−1
0100111010 −1.346 × 10−1 −1.345 × 10−1 1010010011 2.111 × 10−1 2.112 × 10−1
0101010101 −1.403 × 10−1 −1.402 × 10−1 1100100010 2.105 × 10−1 2.106 × 10−1
0011011010 −1.410 × 10−1 −1.409 × 10−1 0110100011 2.100 × 10−1 2.101 × 10−1
1010000111 −1.413 × 10−1 −1.412 × 10−1 0101011100 2.090 × 10−1 2.091 × 10−1
1100010001 −1.445 × 10−1 −1.444 × 10−1 0011110010 2.088 × 10−1 2.089 × 10−1
1000111001 −1.450 × 10−1 −1.449 × 10−1 0110010110 2.041 × 10−1 2.042 × 10−1
1001010011 −1.467 × 10−1 −1.466 × 10−1 0100110111 2.017 × 10−1 2.018 × 10−1
0001110111 −1.474 × 10−1 −1.473 × 10−1 0011010111 2.003 × 10−1 2.004 × 10−1
1010100010 −1.497 × 10−1 −1.496 × 10−1 1001100101 1.982 × 10−1 1.983 × 10−1
0111000010 −1.552 × 10−1 −1.551 × 10−1 1010001110 1.926 × 10−1 1.927 × 10−1
0011100110 −1.611 × 10−1 −1.610 × 10−1 1011000100 1.900 × 10−1 1.901 × 10−1
1001001110 −1.631 × 10−1 −1.630 × 10−1 0010111101 1.861 × 10−1 1.862 × 10−1
1100001100 −1.638 × 10−1 −1.637 × 10−1 0110011001 1.843 × 10−1 1.844 × 10−1
0010101111 −1.696 × 10−1 −1.695 × 10−1 1001011010 1.833 × 10−1 1.834 × 10−1
0100011111 −1.717 × 10−1 −1.716 × 10−1 1100011000 1.819 × 10−1 1.820 × 10−1
0101100011 −1.725 × 10−1 −1.724 × 10−1 0101101010 1.725 × 10−1 1.726 × 10−1
0110101000 −1.756 × 10−1 −1.755 × 10−1 0101001111 1.719 × 10−1 1.720 × 10−1
1010011000 −1.765 × 10−1 −1.764 × 10−1 1101000001 1.649 × 10−1 1.650 × 10−1
0101010110 −1.768 × 10−1 −1.767 × 10−1 0111010000 1.622 × 10−1 1.623 × 10−1
0110001101 −1.775 × 10−1 −1.774 × 10−1 0011110100 1.600 × 10−1 1.601 × 10−1
 33

b10 Lower bound Upper bound b10 Lower bound Upper bound
0110100101 1.596 × 10−1 1.597 × 10−1 1100010101 5.508 × 10−2 5.509 × 10−2
1010110000 1.576 × 10−1 1.577 × 10−1 0011101101 5.434 × 10−2 5.435 × 10−2
1100001011 1.569 × 10−1 1.570 × 10−1 1011000011 5.406 × 10−2 5.407 × 10−2
1100100100 1.568 × 10−1 1.569 × 10−1 1011100000 5.296 × 10−2 5.297 × 10−2
1010010101 1.560 × 10−1 1.561 × 10−1 0110101100 5.047 × 10−2 5.048 × 10−2
1001100110 1.540 × 10−1 1.541 × 10−1 0101111000 4.863 × 10−2 4.864 × 10−2
0010111110 1.497 × 10−1 1.498 × 10−1 0111001001 4.848 × 10−2 4.849 × 10−2
0011011011 1.483 × 10−1 1.484 × 10−1 1010101001 4.633 × 10−2 4.634 × 10−2
0100111011 1.477 × 10−1 1.478 × 10−1 1010011100 4.576 × 10−2 4.577 × 10−2
1000110111 1.467 × 10−1 1.468 × 10−1 0011110011 4.462 × 10−2 4.463 × 10−2
0101110001 1.466 × 10−1 1.467 × 10−1 0101011101 4.416 × 10−2 4.417 × 10−2
0110011010 1.440 × 10−1 1.441 × 10−1 0110010111 4.272 × 10−2 4.273 × 10−2
1001101001 1.353 × 10−1 1.354 × 10−1 0110110010 4.148 × 10−2 4.149 × 10−2
1011001000 1.350 × 10−1 1.351 × 10−1 1000111110 4.096 × 10−2 4.097 × 10−2
1001011100 1.341 × 10−1 1.342 × 10−1 1110000100 4.083 × 10−2 4.084 × 10−2
0011100111 1.289 × 10−1 1.290 × 10−1 1100100011 3.932 × 10−2 3.933 × 10−2
0101101100 1.267 × 10−1 1.268 × 10−1 1100010110 3.625 × 10−2 3.626 × 10−2
1101000010 1.232 × 10−1 1.233 × 10−1 0011101110 3.615 × 10−2 3.616 × 10−2
0110100110 1.219 × 10−1 1.220 × 10−1 1001110100 3.542 × 10−2 3.543 × 10−2
1010100011 1.213 × 10−1 1.214 × 10−1 1010001111 3.458 × 10−2 3.459 × 10−2
0111000011 1.205 × 10−1 1.206 × 10−1 1101010000 3.298 × 10−2 3.299 × 10−2
1010010110 1.167 × 10−1 1.168 × 10−1 0111001010 3.208 × 10−2 3.209 × 10−2
1001001111 1.136 × 10−1 1.137 × 10−1 1011000101 3.176 × 10−2 3.177 × 10−2
0101010111 1.132 × 10−1 1.133 × 10−1 1001011011 3.166 × 10−2 3.167 × 10−2
0011111000 1.119 × 10−1 1.120 × 10−1 1010101010 3.027 × 10−2 3.028 × 10−2
0101110010 1.108 × 10−1 1.109 × 10−1 0101101011 2.998 × 10−2 2.999 × 10−2
1100001101 1.077 × 10−1 1.078 × 10−1 1100011001 2.958 × 10−2 2.959 × 10−2
1100101000 1.072 × 10−1 1.073 × 10−1 0101011110 2.873 × 10−2 2.874 × 10−2
0011011101 1.071 × 10−1 1.072 × 10−1 0011110101 2.609 × 10−2 2.610 × 10−2
0100111101 1.069 × 10−1 1.070 × 10−1 0110110100 2.579 × 10−2 2.580 × 10−2
0110101001 1.068 × 10−1 1.069 × 10−1 1110001000 2.513 × 10−2 2.514 × 10−2
0111100000 1.055 × 10−1 1.056 × 10−1 0111010001 2.478 × 10−2 2.479 × 10−2
0110011100 1.023 × 10−1 1.024 × 10−1 0010111111 2.376 × 10−2 2.377 × 10−2
1010011001 1.005 × 10−1 1.006 × 10−1 1010110001 2.271 × 10−2 2.272 × 10−2
1001101010 9.949 × 10−2 9.950 × 10−2 1001100111 2.229 × 10−2 2.230 × 10−2
1000111011 9.366 × 10−2 9.367 × 10−2 1100100101 2.213 × 10−2 2.214 × 10−2
1110000001 8.875 × 10−2 8.876 × 10−2 0110011011 2.141 × 10−2 2.142 × 10−2
1100010011 8.829 × 10−2 8.830 × 10−2 1001111000 2.092 × 10−2 2.093 × 10−2
1011010000 8.801 × 10−2 8.802 × 10−2 1011000110 2.003 × 10−2 2.004 × 10−2
1101000100 8.568 × 10−2 8.569 × 10−2 0111001100 1.951 × 10−2 1.952 × 10−2
0011101011 8.431 × 10−2 8.432 × 10−2 1101100000 1.856 × 10−2 1.857 × 10−2
0110001111 8.215 × 10−2 8.216 × 10−2 1100011010 1.855 × 10−2 1.856 × 10−2
0111000101 8.172 × 10−2 8.173 × 10−2 1010101100 1.819 × 10−2 1.820 × 10−2
1010100101 8.053 × 10−2 8.054 × 10−2 1001011101 1.711 × 10−2 1.712 × 10−2
1001110001 8.004 × 10−2 8.005 × 10−2 0101101101 1.635 × 10−2 1.636 × 10−2
0100111110 7.851 × 10−2 7.852 × 10−2 1011001001 1.630 × 10−2 1.631 × 10−2
0011011110 7.785 × 10−2 7.786 × 10−2 0011110110 1.626 × 10−2 1.627 × 10−2
0110101010 7.704 × 10−2 7.705 × 10−2 0111010010 1.562 × 10−2 1.563 × 10−2
0101110100 7.670 × 10−2 7.671 × 10−2 0110100111 1.549 × 10−2 1.550 × 10−2
1100001110 7.628 × 10−2 7.629 × 10−2 0110111000 1.489 × 10−2 1.490 × 10−2
1010011010 7.109 × 10−2 7.110 × 10−2 1110010000 1.456 × 10−2 1.457 × 10−2
0101011011 7.093 × 10−2 7.094 × 10−2 1010110010 1.410 × 10−2 1.411 × 10−2
1100110000 6.714 × 10−2 6.715 × 10−2 1100100110 1.356 × 10−2 1.357 × 10−2
1001101100 6.677 × 10−2 6.678 × 10−2 1101000011 1.350 × 10−2 1.351 × 10−2
1110000010 6.208 × 10−2 6.209 × 10−2 1010010111 1.325 × 10−2 1.326 × 10−2
0110110001 6.114 × 10−2 6.115 × 10−2 0011111001 1.310 × 10−2 1.311 × 10−2
1001010111 6.095 × 10−2 6.096 × 10−2 0101110011 1.232 × 10−2 1.233 × 10−2
1000111101 6.062 × 10−2 6.063 × 10−2 0110011101 1.111 × 10−2 1.112 × 10−2
0001111111 6.004 × 10−2 6.005 × 10−2 0111100001 1.095 × 10−2 1.096 × 10−2
0111000110 5.688 × 10−2 5.689 × 10−2 1100011100 1.078 × 10−2 1.079 × 10−2
0101100111 5.637 × 10−2 5.638 × 10−2 1100101001 1.076 × 10−2 1.077 × 10−2
1001110010 5.555 × 10−2 5.556 × 10−2 1001011110 1.024 × 10−2 1.025 × 10−2
1010100110 5.539 × 10−2 5.540 × 10−2 1011001010 9.892 × 10−3 9.893 × 10−3
1101001000 5.518 × 10−2 5.519 × 10−2 0101101110 9.795 × 10−3 9.796 × 10−3
 34

b10 Lower bound Upper bound b10 Lower bound Upper bound
1001101011 9.556 × 10−3 9.557 × 10−3 1100111000 7.321 × 10−4 7.322 × 10−4
0111010100 9.159 × 10−3 9.160 × 10−3 1001110110 7.147 × 10−4 7.148 × 10−4
1010110100 8.158 × 10−3 8.159 × 10−3 1110001001 6.668 × 10−4 6.669 × 10−4
1110100000 8.045 × 10−3 8.046 × 10−3 1101010010 6.418 × 10−4 6.419 × 10−4
0011111010 7.860 × 10−3 7.861 × 10−3 1001111001 4.968 × 10−4 4.969 × 10−4
1011010001 7.256 × 10−3 7.257 × 10−3 0111001101 4.671 × 10−4 4.672 × 10−4
1101000101 6.846 × 10−3 6.847 × 10−3 1011101000 4.576 × 10−4 4.577 × 10−4
0011011111 6.767 × 10−3 6.768 × 10−3 1011000111 4.529 × 10−4 4.530 × 10−4
0111100010 6.618 × 10−3 6.619 × 10−3 0110110110 4.381 × 10−4 4.382 × 10−4
0100111111 6.522 × 10−3 6.523 × 10−3 1100011011 4.046 × 10−4 4.048 × 10−4
0110011110 6.489 × 10−3 6.490 × 10−3 1101100001 4.037 × 10−4 4.038 × 10−4
0110101011 6.439 × 10−3 6.440 × 10−3 1010101101 4.008 × 10−4 4.009 × 10−4
1100101010 6.351 × 10−3 6.352 × 10−3 1110001010 3.597 × 10−4 3.598 × 10−4
0101110101 6.120 × 10−3 6.121 × 10−3 1101010100 3.395 × 10−4 3.396 × 10−4
1100001111 5.632 × 10−3 5.633 × 10−3 0011110111 3.363 × 10−4 3.364 × 10−4
1011001100 5.615 × 10−3 5.616 × 10−3 0111010011 2.972 × 10−4 2.973 × 10−4
1010011011 5.170 × 10−3 5.171 × 10−3 0110111001 2.958 × 10−4 2.959 × 10−4
0111011000 5.071 × 10−3 5.072 × 10−3 1001111010 2.634 × 10−4 2.635 × 10−4
1100110001 4.504 × 10−3 4.505 × 10−3 0111001110 2.459 × 10−4 2.460 × 10−4
1001101101 4.487 × 10−3 4.488 × 10−3 1110010001 2.443 × 10−4 2.444 × 10−4
1010111000 4.466 × 10−3 4.467 × 10−3 1010110011 2.421 × 10−4 2.422 × 10−4
0011111100 4.425 × 10−3 4.426 × 10−3 1011110000 2.356 × 10−4 2.357 × 10−4
1111000000 4.291 × 10−3 4.292 × 10−3 1100100111 2.252 × 10−4 2.253 × 10−4
1011010010 4.245 × 10−3 4.246 × 10−3 1101100010 2.166 × 10−4 2.167 × 10−4
1101000110 3.937 × 10−3 3.938 × 10−3 1010101110 2.099 × 10−4 2.100 × 10−4
0111100100 3.766 × 10−3 3.767 × 10−3 1110001100 1.884 × 10−4 1.885 × 10−4
1100101100 3.529 × 10−3 3.530 × 10−3 1101011000 1.755 × 10−4 1.756 × 10−4
1110000011 3.525 × 10−3 3.526 × 10−3 0110111010 1.552 × 10−4 1.553 × 10−4
0111000111 3.493 × 10−3 3.495 × 10−3 1100011101 1.470 × 10−4 1.471 × 10−4
0101110110 3.488 × 10−3 3.489 × 10−3 1001111100 1.362 × 10−4 1.363 × 10−4
1010100111 3.230 × 10−3 3.231 × 10−3 1001011111 1.306 × 10−4 1.307 × 10−4
1001110011 3.070 × 10−3 3.071 × 10−3 1110010010 1.300 × 10−4 1.301 × 10−4
1101001001 3.028 × 10−3 3.029 × 10−3 0101101111 1.267 × 10−4 1.268 × 10−4
0110101101 2.924 × 10−3 2.925 × 10−3 1011001011 1.179 × 10−4 1.180 × 10−4
1011100001 2.827 × 10−3 2.828 × 10−3 1101100100 1.131 × 10−4 1.132 × 10−4
0101111001 2.616 × 10−3 2.617 × 10−3 0111010101 1.102 × 10−4 1.103 × 10−4
1100110010 2.566 × 10−3 2.567 × 10−3 1010110101 8.711 × 10−5 8.712 × 10−5
1001101110 2.505 × 10−3 2.506 × 10−3 0011111011 8.393 × 10−5 8.394 × 10−5
1011010100 2.351 × 10−3 2.352 × 10−3 1110100001 8.140 × 10−5 8.141 × 10−5
1010011101 2.267 × 10−3 2.268 × 10−3 0110111100 7.974 × 10−5 7.975 × 10−5
0111101000 2.047 × 10−3 2.048 × 10−3 1100011110 7.595 × 10−5 7.596 × 10−5
0110110011 1.982 × 10−3 1.983 × 10−3 1110010100 6.759 × 10−5 6.760 × 10−5
1000111111 1.749 × 10−3 1.750 × 10−3 0110011111 5.883 × 10−5 5.884 × 10−5
1101001010 1.697 × 10−3 1.698 × 10−3 1101101000 5.809 × 10−5 5.810 × 10−5
1110000101 1.627 × 10−3 1.628 × 10−3 0111100011 5.715 × 10−5 5.716 × 10−5
0110101110 1.608 × 10−3 1.609 × 10−3 0111010110 5.705 × 10−5 5.706 × 10−5
1011100010 1.602 × 10−3 1.603 × 10−3 1100101011 5.377 × 10−5 5.378 × 10−5
0011101111 1.561 × 10−3 1.562 × 10−3 1010110110 4.485 × 10−5 4.486 × 10−5
0101111010 1.453 × 10−3 1.454 × 10−3 1110100010 4.281 × 10−5 4.282 × 10−5
1100010111 1.422 × 10−3 1.423 × 10−3 1011001101 4.079 × 10−5 4.080 × 10−5
1100110100 1.393 × 10−3 1.394 × 10−3 0111011001 3.635 × 10−5 3.636 × 10−5
1001110101 1.325 × 10−3 1.326 × 10−3 1110011000 3.456 × 10−5 3.457 × 10−5
1011011000 1.251 × 10−3 1.252 × 10−3 1101110000 2.947 × 10−5 2.948 × 10−5
1010011110 1.233 × 10−3 1.234 × 10−3 0011111101 2.861 × 10−5 2.862 × 10−5
0111001011 1.176 × 10−3 1.177 × 10−3 1010111001 2.788 × 10−5 2.789 × 10−5
1101010001 1.172 × 10−3 1.173 × 10−3 1111000001 2.510 × 10−5 2.511 × 10−5
0111110000 1.077 × 10−3 1.078 × 10−3 1011010011 2.315 × 10−5 2.316 × 10−5
1010101011 1.034 × 10−3 1.035 × 10−3 1110100100 2.209 × 10−5 2.210 × 10−5
0101011111 9.917 × 10−4 9.918 × 10−4 1101000111 2.092 × 10−5 2.093 × 10−5
1101001100 9.109 × 10−4 9.110 × 10−4 1011001110 2.086 × 10−5 2.087 × 10−5
1110000110 8.914 × 10−4 8.915 × 10−4 0111100101 2.006 × 10−5 2.007 × 10−5
1011100100 8.700 × 10−4 8.701 × 10−4 0111011010 1.864 × 10−5 1.865 × 10−5
0110110101 8.223 × 10−4 8.224 × 10−4 1100101101 1.783 × 10−5 1.784 × 10−5
0101111100 7.749 × 10−4 7.750 × 10−4 0101110111 1.698 × 10−5 1.699 × 10−5
 35

b10 Lower bound Upper bound b10 Lower bound Upper bound
0011111110 1.460 × 10−5 1.461 × 10−5 1110010011 2.723 × 10−8 2.724 × 10−8
1010111010 1.424 × 10−5 1.425 × 10−5 1101011010 2.676 × 10−8 2.677 × 10−8
1111000010 1.306 × 10−5 1.307 × 10−5 1101100101 2.269 × 10−8 2.270 × 10−8
1110101000 1.125 × 10−5 1.126 × 10−5 1011110100 2.266 × 10−8 2.267 × 10−8
0111100110 1.024 × 10−5 1.025 × 10−5 1001111110 1.501 × 10−8 1.502 × 10−8
1100110011 9.559 × 10−6 9.561 × 10−6 1101011100 1.340 × 10−8 1.341 × 10−8
0111011100 9.456 × 10−6 9.457 × 10−6 0110111101 1.230 × 10−8 1.231 × 10−8
1100101110 9.068 × 10−6 9.069 × 10−6 1101100110 1.137 × 10−8 1.138 × 10−8
1001101111 8.762 × 10−6 8.763 × 10−6 1011111000 1.134 × 10−8 1.135 × 10−8
1011010101 7.615 × 10−6 7.616 × 10−6 1100011111 1.062 × 10−8 1.063 × 10−8
1010111100 7.204 × 10−6 7.205 × 10−6 1110010101 7.727 × 10−9 7.728 × 10−9
1111000100 6.701 × 10−6 6.702 × 10−6 1101101001 6.207 × 10−9 6.208 × 10−9
0111101001 6.376 × 10−6 6.377 × 10−6 0110111110 6.161 × 10−9 6.162 × 10−9
1110110000 5.686 × 10−6 5.687 × 10−6 0111010111 6.074 × 10−9 6.076 × 10−9
1101001011 4.189 × 10−6 4.190 × 10−6 1110010110 3.871 × 10−9 3.872 × 10−9
0110101111 4.066 × 10−6 4.067 × 10−6 1010110111 3.796 × 10−9 3.797 × 10−9
1011010110 3.862 × 10−6 3.863 × 10−6 1110100011 3.134 × 10−9 3.135 × 10−9
1011100011 3.530 × 10−6 3.531 × 10−6 1101101010 3.109 × 10−9 3.110 × 10−9
1111001000 3.402 × 10−6 3.403 × 10−6 1110011001 2.093 × 10−9 2.094 × 10−9
0111101010 3.242 × 10−6 3.243 × 10−6 1101110001 1.638 × 10−9 1.639 × 10−9
0101111011 3.154 × 10−6 3.155 × 10−6 1101101100 1.556 × 10−9 1.557 × 10−9
1100110101 3.017 × 10−6 3.018 × 10−6 1110011010 1.048 × 10−9 1.049 × 10−9
1010011111 2.339 × 10−6 2.340 × 10−6 1110100101 8.741 × 10−10 8.742 × 10−10
1011011001 2.283 × 10−6 2.284 × 10−6 1101110010 8.202 × 10−10 8.203 × 10−10
0111110001 1.874 × 10−6 1.875 × 10−6 1011001111 8.196 × 10−10 8.197 × 10−10
1111010000 1.716 × 10−6 1.717 × 10−6 0111011011 6.924 × 10−10 6.925 × 10−10
0111101100 1.636 × 10−6 1.637 × 10−6 1110011100 5.244 × 10−10 5.245 × 10−10
1100110110 1.524 × 10−6 1.525 × 10−6 1110100110 4.376 × 10−10 4.377 × 10−10
1101001101 1.284 × 10−6 1.285 × 10−6 0011111111 4.232 × 10−10 4.233 × 10−10
1011011010 1.153 × 10−6 1.154 × 10−6 1101110100 4.103 × 10−10 4.104 × 10−10
1110000111 1.138 × 10−6 1.139 × 10−6 1010111011 4.073 × 10−10 4.074 × 10−10
1011100101 1.114 × 10−6 1.115 × 10−6 1111000011 3.111 × 10−10 3.112 × 10−10
0111110010 9.493 × 10−7 9.494 × 10−7 1110101001 2.344 × 10−10 2.345 × 10−10
0101111101 9.489 × 10−7 9.490 × 10−7 1101111000 2.052 × 10−10 2.053 × 10−10
1100111001 8.771 × 10−7 8.772 × 10−7 0111100111 2.051 × 10−10 2.052 × 10−10
1111100000 8.621 × 10−7 8.622 × 10−7 0111011101 1.822 × 10−10 1.823 × 10−10
1001110111 7.449 × 10−7 7.450 × 10−7 1100101111 1.717 × 10−10 1.718 × 10−10
1101001110 6.471 × 10−7 6.473 × 10−7 1110101010 1.173 × 10−10 1.174 × 10−10
1101010011 6.448 × 10−7 6.449 × 10−7 1010111101 1.062 × 10−10 1.063 × 10−10
1011011100 5.799 × 10−7 5.800 × 10−7 0111011110 9.115 × 10−11 9.116 × 10−11
1011100110 5.622 × 10−7 5.623 × 10−7 1111000101 8.548 × 10−11 8.549 × 10−11
0111110100 4.780 × 10−7 4.781 × 10−7 1110110001 6.114 × 10−11 6.115 × 10−11
0101111110 4.774 × 10−7 4.775 × 10−7 1110101100 5.868 × 10−11 5.869 × 10−11
1100111010 4.417 × 10−7 4.418 × 10−7 1010111110 5.311 × 10−11 5.312 × 10−11
0110110111 3.317 × 10−7 3.318 × 10−7 1111000110 4.277 × 10−11 4.278 × 10−11
1011101001 3.259 × 10−7 3.260 × 10−7 1110110010 3.059 × 10−11 3.060 × 10−11
0111111000 2.398 × 10−7 2.399 × 10−7 1011010111 2.936 × 10−11 2.937 × 10−11
1100111100 2.217 × 10−7 2.218 × 10−7 1111001001 2.273 × 10−11 2.274 × 10−11
1110001011 1.964 × 10−7 1.965 × 10−7 0111101011 2.180 × 10−11 2.181 × 10−11
1101010101 1.913 × 10−7 1.914 × 10−7 1110110100 1.530 × 10−11 1.531 × 10−11
1011101010 1.640 × 10−7 1.641 × 10−7 1111001010 1.137 × 10−11 1.138 × 10−11
1001111011 1.078 × 10−7 1.079 × 10−7 1110111000 7.651 × 10−12 7.652 × 10−12
0111001111 1.056 × 10−7 1.057 × 10−7 1111010001 5.904 × 10−12 5.905 × 10−12
1101010110 9.611 × 10−8 9.612 × 10−8 1111001100 5.688 × 10−12 5.689 × 10−12
1011110001 9.002 × 10−8 9.003 × 10−8 0111101101 5.655 × 10−12 5.656 × 10−12
1011101100 8.230 × 10−8 8.231 × 10−8 1100110111 5.282 × 10−12 5.283 × 10−12
1101100011 7.883 × 10−8 7.884 × 10−8 1111010010 2.953 × 10−12 2.954 × 10−12
1010101111 7.685 × 10−8 7.686 × 10−8 0111101110 2.828 × 10−12 2.829 × 10−12
1110001101 5.679 × 10−8 5.680 × 10−8 1011011011 2.754 × 10−12 2.755 × 10−12
1101011001 5.334 × 10−8 5.335 × 10−8 0111110011 1.975 × 10−12 1.976 × 10−12
1011110010 4.521 × 10−8 4.522 × 10−8 1111100001 1.509 × 10−12 1.510 × 10−12
0110111011 4.501 × 10−8 4.502 × 10−8 1111010100 1.476 × 10−12 1.477 × 10−12
1001111101 2.997 × 10−8 2.998 × 10−8 1101001111 9.573 × 10−13 9.579 × 10−13
1110001110 2.848 × 10−8 2.849 × 10−8 1111100010 7.550 × 10−13 7.551 × 10−13
 36

b10 Lower bound Upper bound b10 Lower bound Upper bound
1111011000 7.385 × 10−13 7.386 × 10−13 1111010101 6.046 × 10−24 6.047 × 10−24
1011011101 7.041 × 10−13 7.042 × 10−13 1111010110 3.023 × 10−24 3.024 × 10−24
1011100111 6.420 × 10−13 6.421 × 10−13 1111100011 1.605 × 10−24 1.606 × 10−24
0111110101 5.090 × 10−13 5.091 × 10−13 1111011001 1.513 × 10−24 1.514 × 10−24
0101111111 4.916 × 10−13 4.917 × 10−13 1111011010 7.565 × 10−25 7.566 × 10−25
1100111011 4.683 × 10−13 4.684 × 10−13 1111100101 4.018 × 10−25 4.019 × 10−25
1111100100 3.775 × 10−13 3.776 × 10−13 1011011111 3.796 × 10−25 3.797 × 10−25
1011011110 3.520 × 10−13 3.521 × 10−13 1111011100 3.782 × 10−25 3.783 × 10−25
0111110110 2.545 × 10−13 2.546 × 10−13 1111100110 2.009 × 10−25 2.010 × 10−25
1111101000 1.887 × 10−13 1.888 × 10−13 0111110111 1.993 × 10−25 1.994 × 10−25
0111111001 1.294 × 10−13 1.295 × 10−13 1111101001 1.005 × 10−25 1.006 × 10−25
1100111101 1.190 × 10−13 1.191 × 10−13 1111101010 5.026 × 10−26 5.027 × 10−26
1111110000 9.440 × 10−14 9.441 × 10−14 1111110001 2.513 × 10−26 2.514 × 10−26
0111111010 6.471 × 10−14 6.472 × 10−14 1111101100 2.513 × 10−26 2.514 × 10−26
1100111110 5.953 × 10−14 5.954 × 10−14 0111111011 1.305 × 10−26 1.306 × 10−26
1011101011 5.739 × 10−14 5.740 × 10−14 1111110010 1.256 × 10−26 1.257 × 10−26
0111111100 3.235 × 10−14 3.236 × 10−14 1100111111 1.235 × 10−26 1.237 × 10−26
1101010111 2.254 × 10−14 2.256 × 10−14 1111110100 6.284 × 10−27 6.285 × 10−27
1011101101 1.457 × 10−14 1.458 × 10−14 0111111101 3.263 × 10−27 3.264 × 10−27
1011101110 7.287 × 10−15 7.288 × 10−15 1111111000 3.142 × 10−27 3.143 × 10−27
1011110011 4.546 × 10−15 4.547 × 10−15 0111111110 1.631 × 10−27 1.632 × 10−27
1110001111 1.920 × 10−15 1.921 × 10−15 1011101111 1.597 × 10−28 1.598 × 10−28
1101011011 1.826 × 10−15 1.827 × 10−15 1011110111 1.030 × 10−30 1.031 × 10−30
1011110101 1.150 × 10−15 1.151 × 10−15 1101011111 2.015 × 10−31 2.017 × 10−31
1011110110 5.751 × 10−16 5.752 × 10−16 1011111011 6.560 × 10−32 6.561 × 10−32
1001111111 5.001 × 10−16 5.002 × 10−16 1011111101 1.640 × 10−32 1.641 × 10−32
1101011101 4.605 × 10−16 4.607 × 10−16 1011111110 8.200 × 10−33 8.201 × 10−33
1101100111 3.345 × 10−16 3.346 × 10−16 1101101111 4.282 × 10−35 4.284 × 10−35
1011111001 2.894 × 10−16 2.895 × 10−16 1110011111 4.686 × 10−37 4.689 × 10−37
1101011110 2.302 × 10−16 2.304 × 10−16 1101110111 2.246 × 10−37 2.247 × 10−37
1011111010 1.447 × 10−16 1.448 × 10−16 1101111011 1.414 × 10−38 1.415 × 10−38
0110111111 1.022 × 10−16 1.023 × 10−16 1101111101 3.535 × 10−39 3.536 × 10−39
1011111100 7.235 × 10−17 7.236 × 10−17 1101111110 1.767 × 10−39 1.768 × 10−39
1110010111 3.672 × 10−17 3.673 × 10−17 1110101111 7.744 × 10−41 7.745 × 10−41
1101101011 2.597 × 10−17 2.598 × 10−17 1110110111 3.826 × 10−43 3.827 × 10−43
1101101101 6.530 × 10−18 6.531 × 10−18 1110111011 2.401 × 10−44 2.402 × 10−44
1101101110 3.265 × 10−18 3.266 × 10−18 1111001111 7.252 × 10−45 7.254 × 10−45
1110011011 2.788 × 10−18 2.789 × 10−18 1110111101 6.003 × 10−45 6.004 × 10−45
1101110011 1.857 × 10−18 1.858 × 10−18 1110111110 3.001 × 10−45 3.002 × 10−45
1110011101 7.001 × 10−19 7.003 × 10−19 1111010111 3.540 × 10−47 3.541 × 10−47
1110100111 4.832 × 10−19 4.833 × 10−19 1111011011 2.220 × 10−48 2.221 × 10−48
1101110101 4.663 × 10−19 4.664 × 10−19 1111011101 5.550 × 10−49 5.551 × 10−49
1110011110 3.500 × 10−19 3.502 × 10−19 1111011110 2.775 × 10−49 2.776 × 10−49
1101110110 2.331 × 10−19 2.332 × 10−19 1111100111 1.591 × 10−49 1.592 × 10−49
1101111001 1.168 × 10−19 1.169 × 10−19 1111101011 9.973 × 10−51 9.974 × 10−51
1101111010 5.842 × 10−20 5.843 × 10−20 1111101101 2.493 × 10−51 2.494 × 10−51
1110101011 3.587 × 10−20 3.588 × 10−20 1111101110 1.246 × 10−51 1.247 × 10−51
1101111100 2.921 × 10−20 2.922 × 10−20 1111110011 6.241 × 10−52 6.242 × 10−52
0111011111 2.433 × 10−20 2.434 × 10−20 1111110101 1.560 × 10−52 1.561 × 10−52
1110101101 8.997 × 10−21 8.998 × 10−21 1111110110 7.802 × 10−53 7.803 × 10−53
1010111111 8.306 × 10−21 8.307 × 10−21 1111111001 3.901 × 10−53 3.902 × 10−53
1111000111 4.782 × 10−21 4.783 × 10−21 1111111010 1.950 × 10−53 1.951 × 10−53
1110101110 4.498 × 10−21 4.499 × 10−21 0111111111 1.161 × 10−53 1.162 × 10−53
1110110011 2.497 × 10−21 2.498 × 10−21 1111111100 9.752 × 10−54 9.753 × 10−54
1110110101 6.259 × 10−22 6.260 × 10−22 1011111111 2.955 × 10−64 2.956 × 10−64
1111001011 3.492 × 10−22 3.493 × 10−22 1101111111 1.897 × 10−77 1.898 × 10−77
1110110110 3.129 × 10−22 3.130 × 10−22 1110111111 5.295 × 10−89 5.297 × 10−89
1110111001 1.566 × 10−22 1.567 × 10−22 1111011111 4.211 × 10−97 4.212 × 10−97
1111001101 8.753 × 10−23 8.754 × 10−23 1111101111 8.632×10−102 8.633×10−102
1110111010 7.833 × 10−23 7.834 × 10−23 1111110111 3.383×10−104 3.384×10−104
1111001110 4.376 × 10−23 4.377 × 10−23 1111111011 2.114×10−105 2.115×10−105
1110111100 3.916 × 10−23 3.917 × 10−23 1111111101 5.287×10−106 5.288×10−106
1111010011 2.414 × 10−23 2.415 × 10−23 1111111110 2.643×10−106 2.644×10−106
0111101111 2.356 × 10−23 2.357 × 10−23 1111111111 5.479×10−211 5.480×10−211