108 X. Zhang / Information Sciences 330 (2016) 104–124
=
(μ
β
1
)
2
− (μ
β
2
)
2
+ (v
β
2
)
2
− (v
β
1
)
2
+ (μ
β
2
)
2
(v
β
1
)
2
− (μ
β
1
)
2
(v
β
2
)
2
2 − (μ
β
1
)
2
− (v
β
1
)
2
2 − (μ
β
2
)
2
− (v
β
2
)
2
=
(1 − (μ
β
2
)
2
)(1 − (v
β
1
)
2
) − (1 − (μ
β
1
)
2
)(1 − (v
β
2
)
2
)
2 − (μ
β
1
)
2
− (v
β
1
)
2
2 − (μ
β
2
)
2
− (v
β
2
)
2
If β
1
≤ β
2
, according to Definition 2.2 we clearly know that 0 ≤ u
β
1
≤ u
β
2
≤ 1and1≥ v
β
1
≥ v
β
2
≥ 0. Thus we have:
(1 − (μ
β
2
)
2
)(1 − (v
β
1
)
2
) ≤ (1 − (μ
β
1
)
2
)(1 − (v
β
2
)
2
),
2 − (μ
β
1
)
2
− (v
β
1
)
2
2 − (μ
β
2
)
2
− (v
β
2
)
2
> 0,
Obviously, P(β
1
) − P (β
2
) =
(1−(μ
β
2
)
2
)(1−(
v
β
1
)
2
)−(1−(μ
β
1
)
2
)(1−(
v
β
2
)
2
)
(2−(μ
β
1
)
2
−(
v
β
1
)
2
)(2−(μ
β
2
)
2
−(
v
β
2
)
2
)
≤ 0, i.e., P(β
1
) ≤ P(β
2
), which completes the proof of
the Proposition 2.2.
Based on the closeness index P
(β), we present a new ranking method for PFNs.
Definition 2.7. Let
β
j
= P(u
β
j
, v
β
j
)(j = 1, 2) be two PFNs, P (β
1
) and P(β
2
) be the closeness indices of β
1
and β
2
, respectively,
then
(1) If P
(β
1
) < P(β
2
),thenβ
1
≺
P
β
2
;
(2) If P
(β
1
) > P(β
2
),thenβ
1
P
β
2
;
(3) If P
(β
1
) = P(β
2
),thenβ
1
∼
P
β
2
.
Example 2.2 (Continued Example 2.1). Let
β
1
= P(
√
5/3, 2/3) and β
2
= P(2/3,
√
3/3) be two PFNs, according to Definition 2.6
the following results are obtained:
P
(β
1
) =
1 −
(2/3)
2
2 − (
√
5/3)
2
− (2/3)
2
= 5/9, P(β
2
) =
1 −
(
√
3/3)
2
2 − (2/3)
2
− (
√
3/3)
2
= 6/11
Thus, P (β
1
) > P(β
2
). According to the comparison law provided in Definition 2.7, it is easy to see that β
1
P
β
2
.Obviously,
the comparison results of Examples 2.1 and 2.2 show that the proposed closeness index-based ranking method for PFNs is more
reasonable than the score-based ranking method [48].
In many real-world evaluation processes of MCDM problems, the values of the membership degree and non-membership
degree in a PFS are not easy for the decision maker to assign exact values. Instead, it is convenient for the decision maker to
employ intervals for expressing his/her preference about the membership degree and the non-membership degree. For this
purpose, in what follows we extend the concept of PFS to propose the new concept of IVPFS.
3. Interval-valued Pythagorean fuzzy number
The objective of this section is threefold. First, we present the new concept of IVPFNs and develop the basic operations of
IVPFNs. Second, we introduce an interval-valued Pythagorean fuzzy distance measure. Third, we propose a closeness index for
IVPFN and present the closeness index-based ranking method for comparing the magnitude of IVPFNs.
Definition 3.1. Let X be an ordinary finite nonempty set. An IVPFS
˜
P in X is defined by:
˜
P =
{< x,
˜
P
(
˜μ
˜
P
(x),
˜
v
˜
P
(x)
)
> |x ∈ X} (3.1)
where ˜
μ
˜
P
(x) ⊆ [0, 1] and ˜ν
˜
P
(x) ⊆ [0, 1] are interval values, ˜μ
L
˜
P
(x) and ˜μ
U
˜
P
(x) are the lower and upper of interval value ˜μ
˜
P
(x),re-
spectively, ˜
ν
L
˜
P
(x) and ˜ν
U
˜
P
(x) are the lower and upper of interval value ˜ν
˜
P
(x), respectively, and they satisfy ( ˜μ
U
˜
P
(x))
2
+ ( ˜ν
U
˜
P
(x))
2
≤
1.
For every x ∈ X,˜
π
˜
P
(x) = [˜π
L
˜
P
(x), ˜π
U
˜
P
(x)] is called an interval-valued Pythagorean fuzzy index of x to
˜
P,where ˜π
L
˜
P
(x) =
1 − ( ˜μ
U
˜
P
(x))
2
− (
˜
v
U
˜
P
(x))
2
and ˜π
U
˜
P
(x) =
1 − ( ˜μ
L
˜
P
(x))
2
− (
˜
v
L
˜
P
(x))
2
.Clearly,theIVPFSreducestothePFSif ˜μ
L
˜
P
(x) = ˜μ
U
˜
P
(x) and
˜
ν
L
˜
P
(x) = ˜ν
U
˜
P
(x); and the IVPFS reduces to an IVIFS if ˜μ
U
˜
P
(x) + ˜ν
U
˜
P
(x) ≤ 1. That is to say, the PFS and the IVIFS are two special cases
of IVPFS. For simplicity, we called
˜
P
( ˜μ
˜
P
(x),
˜
v
˜
P
(x)) an IVPFN denoted by
˜
β =
˜
P([˜μ
L
˜
β
, ˜μ
U
˜
β
], [
˜
v
L
˜
β
,
˜
v
U
˜
β
]) where ( ˜μ
U
˜
β
)
2
+ (
˜
v
U
˜
β
)
2
≤ 1. It is
noted that the IVPFN
˜
β is reduced to an interval-valued intuitionistic fuzzy number (IVIFN) if ˜μ
U
˜
β
+
˜
v
U
˜
β
≤ 1. The main difference
between IVPFN and IVIFN is their different constraint conditions. Usually, the space of the IVPFN
˜
β is larger than the space of the
IVIFN ˜
α =
˜
I([˜μ
L
˜
α
, ˜μ
U
˜
α
], [
˜
v
L
˜
α
,
˜
v
U
˜
α
]), which is shown in Fig. 2.
Remark 3.1. IVPFN is a good tool to model the imprecise and ambiguous information in the real-life decision evaluation process,
which is usually employed by decision makers to express their assessment values for alternatives on criteria. For example, the
IVPF assessment value
˜
β
ij
=
˜
P([0.6, 0.8], [0.4, 0.5]) provided by the decision maker can express the meaning that the degree to