3.3 Common solution of the variational inequality and the fixed point problems 59
3.3.2 Convergence analysis
Lemma 3.3.5. Let {xn} be a sequence generated by Algorithm 3.3.2 under Assumption 3.3.1. Then
∥zn−p∥2 ≤ ∥wn−p∥2−
1−λ2nµ2 λ2n+1
∥yn−wn∥2.
Proof. Letp∈Γ. Sinceyn=PCn(wn−λnwn),we obtain by the characteristic property of PCn that ⟨yn−wn+λnAwn, yn−p⟩ ≤0 and this implies that
⟨yn−wn, yn−p⟩ ≤ −λn⟨Awn, yn−p⟩. (3.3.4) Also, from the definition of zn inStep 4, and Lemma2.1.1, we have
∥zn−p∥2 ≤ ∥yn−λn(Ayn−Awn)−p∥2
=∥yn−p∥2+λ2n∥Ayn−Awn∥2−2λn⟨Ayn−Awn, yn−p⟩
=∥wn−p∥2+∥yn−wn∥2+ 2⟨yn−wn, wn−p⟩+λ2n∥Ayn−Awn∥2
−2λn⟨Ayn−Awn, yn−p⟩
=∥wn−p∥2+∥yn−wn∥2+λ2n∥Ayn−Awn∥2−2⟨yn−wn, yn−wn⟩ + 2⟨yn−wn, yn−p⟩
−2λn⟨Ayn−Awn, yn−p⟩
=∥wn−p∥2− ∥yn−wn∥2+λ2n∥Ayn−Awn∥2+ 2⟨yn−wn, yn−p⟩
−2λn⟨Ayn−Awn, yn−p⟩ (3.3.5)
From (3.3.4) and (3.3.5), we obtain
∥zn−p∥2 ≤ ∥wn−p∥2− ∥yn−wn∥2+λ2n∥Ayn−Awn∥2−2λn⟨Awn, yn−p⟩
−2λn⟨Ayn−Awn, yn−p⟩
=∥wn−p∥2− ∥yn−wn∥2+λ2n∥Ayn−Awn∥2−2λn⟨Ayn, yn−p⟩
≤ ∥wn−p∥2− ∥yn−wn∥2+λ2n µ2
λ2n+1∥yn−wn∥2−2λn⟨Ayn, yn−p⟩
≤ ∥wn−p∥2−
1−λ2nµ2 λ2n+1
∥yn−wn∥2. (3.3.6)
Consider the limit
n→∞lim
1− λ2nµ2 λ2n+1
= 1−µ2 >0.
Hence there exists n0 ≥ 0 such that for all n ≥ n0, we have that
1− λλ2n2µ2 n+1
> 0. Thus, from (3.3.6), we have that
∥zn−p∥ ≤ ∥wn−p∥.
Lemma 3.3.6. Let {xn} be a sequence generated by Algorithm 3.3.2 under Assumption 3.3.1. Then, {xn} is bounded.
Proof. First, we show that PΓ(I −D+γ1f) is a contraction of H. For all x, y ∈ H, we have
∥PΓ(I−D+γ1f)x−PΓ(I −D+γ1f)y∥ ≤ ∥(I−D+γ1f)x−(I−D+γ1f)y∥
≤ ∥(I−D)x−(I −D)y∥+γ1∥f x−f y∥
≤(1−γ2)∥x−y∥+γ1ρ∥x−y∥
= (1−(γ2−γ1ρ))∥x−y∥.
Hence, PΓ(I −D+γ1f) is a contraction. Let p ∈ Γ. Then from the definition of wn in Step 2, we have
∥wn−p∥=∥xn+αn(xn−xn−1)−p∥
=∥xn−p∥+αn∥xn−xn−1∥.
Also, from Step 2, we observe that αn∥xn−xn−1∥ ≤τn,∀n≥1, which implies that αn
θn∥xn−xn−1∥ ≤ τn
θn →0, as n→ ∞. (3.3.7)
Hence, there exists M1 >0 such that αn
θn∥xn−xn−1∥ ≤M1, ∀n ≥1. (3.3.8)
This implies that ∥wn−p∥ ≤ ∥xn−p∥+θnM1,∀n≥1 and hence
∥zn−p∥ ≤ ∥wn−p∥ ≤ ∥xn−p∥+θnM1. (3.3.9) From Step 5, we have
∥Tnzn−p∥ ≤ ∥(1−βn)zn+βnWnzn−p∥
≤(1−βn)∥zn−p∥+βn∥Wnzn−p∥
≤(1−βn)∥zn−p∥+βn∥zn−p∥
=∥zn−p∥ (3.3.10)
From (3.3.9) and (3.3.10), we obtain for all n ≥n0,
∥xn+1−p∥ ≤ ∥θnγ1f xn+ (1−θnD)Tnzn−p∥
=∥θn(γ1f xn−Dp) + (1−θnD)(Tnzn−p)∥
≤θn∥γ1f xn−Dp∥+ (1−θnγ2)∥Tnzn−p∥
≤θn∥γ1(f xn−f p) + (γ1f p−Dp)∥+ (1−θnγ2)∥zn−p∥
≤θnγ1ρ∥xn−p∥+θn∥γ1f p−Dp∥+ (1−θnγ2)(∥xn−p∥+θnM1)
≤(1−θn(γ2−γ1ρ))∥xn−p∥+θn∥γ1f p−Dp∥+θnM1
= (1−θn(γ2−γ1ρ))∥xn−p∥+θn(γ2−γ1ρ)∥γ1f p−Dp∥+M1 γ2−γ1ρ
≤maxn
∥xn−p∥,∥γ1f p−Dp∥+M1 γ2−γ1ρ
o
≤maxn
∥xn0 −p∥,∥γ1f p−Dp∥+M1
γ2 −γ1ρ
o
Hence, the sequence{xn}is bounded. Consequently,{wn},{yn}and{zn}are also bounded.
Lemma 3.3.7. Assume that {wn} and {yn} are sequences generated by Algorithm 3.3.2 such that
n→∞lim ||wn −yn|| = 0. If {wnj} converges weakly to some xˆ ∈ H as j → ∞, then xˆ ∈ V I(C, A).
Proof. Since wnj ⇀ x,ˆ then by the hypothesis of the lemma it follows that ynj ⇀ xˆ as j → ∞. Also, since ynj ∈ Cnj, then by the definition of Cn we get
h(wnj) +⟨ξnj, ynj −wnj⟩ ≤0. (3.3.11) By the boundedness of {wn} and by condition (A3), there exists a constant M > 0 such that||ξnj|| ≤M for all j ≥0.Then, from (3.3.11) we obtainh(wnj)≤M||wnj−ynj|| → 0 as j → ∞,and this in turn implies that lim inf
j→∞ h(wnj)≤0.Applying condition (A2), we haveh(ˆx)≤lim inf
j→∞ h(wnj)≤0.This implies that ˆx∈ C.From Lemma2.4.1, we obtain
⟨ynj −wnj +λnjAwnj, z−ynj⟩ ≥0, ∀ z ∈ C ⊆ Cnj.
Since A is monotone, we have
0≤ ⟨ynj −wnj, z−ynj⟩+λnj⟨Awnj, z−ynj⟩
=⟨ynj−wnj, z−ynj⟩+λnj⟨Awnj, z−wnj⟩+λnj⟨Awnj, wnj −ynj⟩
≤ ⟨ynj −wnj, z−ynj⟩+λnj⟨Az, z−wnj⟩+λnj⟨Awnj, wnj−ynj⟩.
Letting j → ∞, and since lim
j→∞||ynj −wnj||= 0, we have
⟨Az, z−x⟩ ≥ˆ 0, ∀ z ∈ C.
Applying Lemma 3.2.2, we have that ˆx∈V I(C, A).
Lemma 3.3.8. Let {xn} be a sequence generated by Algorithm 3.3.2 under Assumption 3.3.1. Then,
∥xn+1−p∥2 ≤(1−ηn)∥xn−p∥2 +ηnh θnγ22
2(γ2−γ1ρ)M3+ 3M2(1−θnγ2)2 2(γ2−γ1ρ)
αn
θn||xn−xn−1||
+ 1
(γ2−γ1ρ)⟨γ1f p−Dp, xn+1−p⟩i
− (1−θnγ2)2 (1−θnγ1ρ)
h
1− λ2nµ2 λ2n+1
∥yn−wn∥2 +βn(1−βn)∥Wnzn−zn∥2i
, where ηn = 2θ(1−θn(γ2−γ1ρ)
nγ1ρ) .
Proof. Let p ∈ Γ. Then from Step 2, by applying the Cauchy-Schwartz inequality and Lemma 2.1.1, we obtain
||wn−p||2 =||xn+αn(xn−xn−1)−p||2
=||xn−p||2+α2n||xn−xn−1||2+ 2αn⟨xn−p, xn−xn−1⟩
≤ ||xn−p||2+α2n||xn−xn−1||2+ 2αn||xn−xn−1||||xn−p||
=||xn−p||2+αn||xn−xn−1||(αn||xn−xn−1||+ 2||xn−p||)
≤ ||xn−p||2+ 3M2αn||xn−xn−1||
=||xn−p||2+ 3M2θnαn
θn||xn−xn−1||, (3.3.12) where M2 := supn∈
N{||xn−p||, αn||xn−xn−1||}>0.
Now, applying the last inequality Lemma 2.1.1 and (3.3.6), we have
∥Tnzn−p∥2 =∥(1−βn)(zn−p) +βn(Wnzn−p)∥2
≤(1−βn)∥zn−p∥2+βn∥Wnzn−p∥2 −βn(1−βn)∥Wnzn−zn∥2
≤(1−βn)∥zn−p∥2+βn∥zn−p∥2 −βn(1−βn)∥Wnzn−zn∥2
=∥zn−p∥2−βn(1−βn)∥Wnzn−zn∥2
≤ ∥wn−p∥2−
1−λ2nµ2 λ2n+1
∥yn−wn∥2−βn(1−βn)∥Wnzn−zn∥2
≤ ||xn−p||2+ 3M2θnαn
θn||xn−xn−1|| −
1− λ2nµ2 λ2n+1
∥yn−wn∥2
−βn(1−βn)∥Wnzn−zn∥2. (3.3.13)
Also, by applying Lemma 2.1.1 and (3.3.13), we have
∥xn+1−p∥2 =∥θnγ1f xn+ (1−θnD)Tnzn−p∥2
=∥θn(γ1f xn−Dp) + (1−θnD)(Tnzn−p)∥2
≤(1−θnγ2)2∥Tnzn−p∥2+ 2θn⟨γ1f xn−Dp, xn+1−p⟩
≤(1−θnγ2)2∥Tnzn−p∥2+ 2θnγ1⟨f xn−f p, xn+1−p⟩
+ 2θn⟨γ1f p−Dp, xn+1−p⟩
≤(1−θnγ2)2h
∥xn−p∥2+ 3M2θnαn
θn||xn−xn−1||
−
1− λ2nµ2 λ2n+1
∥yn−wn∥2−βn(1−βn)∥Wnzn−zn∥2i + 2θn⟨γ1f xn−f p, xn+1−p⟩+ 2θn⟨γ1f p−Dp, xn+1−p⟩
≤(1−θnγ2)2∥xn−p∥2+ 3M2(1−θnγ2)2θnαn
θn||xn−xn−1||
−(1−θnγ2)2
1−λ2nµ2 λ2n+1
∥yn−wn∥2−(1−θnγ2)2βn(1−βn)∥Wnzn−zn∥2 + 2θnγ1⟨f xn−f p, xn+1−p⟩+ 2θn⟨γ1f p−Dp, xn+1−p⟩
≤(1−θnγ2)2∥xn−p∥2+ 3M2(1−θnγ2)2θnαn θn
||xn−xn−1||
−(1−θnγ2)2
1−λ2nµ2 λ2n+1
∥yn−wn∥2
−(1−θnγ2)2βn(1−βn)∥Wnzn−zn∥2+ 2θnγ1ρ∥xn−p∥∥xn+1−p∥
+ 2θn⟨γ1f p−Dp, xn+1−p⟩
≤(1−θnγ2)2∥xn−p∥2+ 3M2(1−θnγ2)2θnαn
θn||xn−xn−1||
−(1−θnγ2)2h
1− λ2nµ2 λ2n+1
∥yn−wn∥2+βn(1−βn)∥Wnzn−zn∥2i +θnγ1ρ
∥xn−p∥2+∥xn+1−p∥2
+ 2θn⟨γ1f p−Dp, xn+1−p⟩
= ((1−θnγ2)2+θnγ1ρ)∥xn−p∥2+θnγ1ρ∥xn+1−p∥2 + 3M2(1−θnγ2)2θn
αn
θn||xn−xn−1||+ 2θn⟨γ1f p−Dp, xn+1−p⟩
−(1−θnγ2)2h
1− λ2nµ2 λ2n+1
∥yn−wn∥2+βn(1−βn)∥Wnzn−zn∥2i .
From this, we obtain
∥xn+1−p∥2 ≤ (1−2θnγ2+ (θnγ2)2+θnγ1ρ)
(1−θnγ1ρ) ∥xn−p∥2
+ θn
(1−θnγ1ρ) h
3M2(1−θnγ2)2αn
θn||xn−xn−1||+ 2⟨γ1f p−Dp, xn+1−p⟩i
− (1−θnγ2)2 (1−θnγ1ρ)
h
1−λ2nµ2 λ2n+1
∥yn−wn∥2+βn(1−βn)∥Wnzn−zn∥2i
= (1−2θnγ2+θnγ1ρ)
(1−θnγ1ρ) ∥xn−p∥2+ (θnγ2)2
(1−θnγ1ρ)∥xn−p∥2
+ θn
(1−θnγ1ρ) h
3M2(1−θnγ2)2αn
θn||xn−xn−1||+ 2⟨γ1f p−Dp, xn+1−p⟩i
− (1−θnγ2)2 (1−θnγ1ρ)
h
1−λ2nµ2 λ2n+1
∥yn−wn∥2+βn(1−βn)∥Wnzn−zn∥2i
≤
1− 2θn(γ2−γ1ρ) (1−θnγ1ρ)
∥xn−p∥2 + 2θn(γ2−γ1ρ)
(1−θnγ1ρ)
h θnγ22
2(γ2−γ1ρ)M3+3M2(1−θnγ2)2 2(γ2−γ1ρ)
αn
θn||xn−xn−1||
+ 1
(γ2−γ1ρ)⟨γ1f p−Dp, xn+1−p⟩i
− (1−θnγ2)2 (1−θnγ1ρ)
h
1−λ2nµ2 λ2n+1
∥yn−wn∥2+βn(1−βn)∥Wnzn−zn∥2i , where M3 = sup{∥xn−p∥2 :n ∈N} and thus we obtain the desired conclusion.
We are now in the position to give the main theorem for Algorithm 3.3.2.
Theorem 3.3.9. Let {xn} be a sequence generated by Algorithm 3.3.2 under Assumption 3.3.1. Suppose that {Wn} is the sequence defined by (3.2.1). Then, the sequence {xn} converges strongly to a point x∗ ∈ Γ, where x∗ = PΓ(I−D+γ1f)x∗ is a solution of the variational inequality
⟨(D−γ1f)x∗, x∗−x⟩ ≤0, ∀x∈Γ.
Proof. Since x∗ =PΓ(I−D+γ1f)x∗, we obtain from Lemma3.3.8 that
∥xn+1−x∗∥2 ≤(1−ηn)∥xn−x∗∥2 +ηnh θnγ22
2(γ2−γ1ρ)M3+3M2(1−θnγ2)2 2(γ2−γ1ρ)
αn
θn||xn−xn−1|| (3.3.14)
+ 1
(γ2−γ1ρ)⟨γ1f x∗−Dx∗, xn+1−x∗⟩i .
Now, we claim that the sequence {∥xn−x∗∥} converges to zero. To show this, by Lemma 2.5.36 it suffices to show that lim sup
k→∞
⟨γ1f x∗−Dx∗, xnk+1−x∗⟩ ≤0 for every subsequence
{∥xnk−x∗∥} of {∥xn−x∗∥}satisfying lim inf
k→∞ (∥xnk+1−x∗∥ − ∥xnk −x∗∥)≥0. (3.3.15) Suppose that {∥xnk−x∗∥}is a subsequence of {∥xn−x∗∥}such that (3.3.15) holds. From Lemma 3.3.8, we obtain
(1−θnkγ2)2 (1−θnkγ1ρ)
1−λ2n
kµ2 λ2nk+1
∥ynk−wnk∥2 ≤(1−ηnk)∥xnk −x∗∥2− ∥xnk+1−x∗∥2 +ηnkh θnkγ22
2(γ2−γ1ρ)M3 + 3M2(1−θnkγ2)2
2(γ2 −γ1ρ) αn
θnk||xnk −xnk−1||
+ 1
(γ2−γ1ρ)⟨γ1f x∗−Dx∗, xnk+1−x∗⟩i .
By (3.3.15) and the fact that lim
k→∞ηnk = 0 (since lim
k→∞θnk = 0), we obtain (1−θnkγ2)2
(1−θnkγ1ρ)
1− λ2n
kµ2 λ2nk+1
∥ynk −wnk∥2 →0, ask → ∞.
Consequently, we have
k→∞lim ∥ynk −wnk∥= 0. (3.3.16) Following similar argument, from Lemma 3.3.8 we have
k→∞lim ∥Wnkznk −znk∥= 0. (3.3.17) From the definition of znk inStep 4 and (3.3.16), we have
∥znk −wnk∥=∥ynk −λnk(Aynk−Awnk)−wnk∥
≤ ∥ynk −wnk∥+λnk∥Aynk −Awnk∥
≤ ∥ynk −wnk∥+λnk µ
λnk+1∥wnk−ynk∥
=
1 + λnkµ λnk+1
∥ynk −wnk∥ →0, ask → ∞, which implies that
k→∞lim ∥znk−wnk∥= 0. (3.3.18) From (3.3.16) and (3.3.18) we have
k→∞lim ∥ynk −znk∥= 0. (3.3.19)
Also, from (3.3.17), (3.3.18) and (3.3.19), we get
k→∞lim ∥wnk−ynk∥= 0 lim
k→∞∥Wnkznk−wnk∥= 0, lim
k→∞∥Wnkznk −ynk∥= 0.
(3.3.20) Now, from Step 2 and by Remark3.3.3, we get
k→∞lim ∥wnk −xnk∥= lim
k→∞αnk∥xnk −xnk−1∥= 0. (3.3.21) From (3.3.18), (3.3.20) and (3.3.21), we obtain
k→∞lim ∥znk −xnk∥= 0, lim
k→∞∥Wnkznk−xnk∥= 0. (3.3.22) By applying (3.3.17), we have
∥Tnkznk−znk∥=∥(1−βnk)znk+βnkWnkznk−znk∥ (3.3.23)
≤(1−βnk)∥znk−znk∥+βnk∥Wnkznk −znk∥ →0, k → ∞.
Now, by using (3.3.18), (3.3.21) and (3.3.23) we obtain
k→∞lim ||Tnkznk −wnk||= 0, lim
k→∞||Tnkznk−xnk||= 0. (3.3.24) Consequently, by applying the fact that lim
k→∞θnk = 0 we get
∥xnk+1−xnk∥=∥θnkγ1f xnk + (1−θnkD)Tnkznk−xnk∥
=∥(θnkγ1f xnk−θnkDxnk) + (1−θnkD)(Tnkznk−xnk)∥
≤θnk∥γ1f xnk −Dxnk∥+ (1−θnkγ2)∥Tnkznk −xnk∥ →0, k → ∞.
Hence
k→∞lim ∥xnk+1−xnk∥= 0. (3.3.25) Now, we show that wω(xn) ⊂ ∩∞i=1F(Si) = F(W). Let z ∈ wω(xn) and suppose that z /∈F(W),that is, W z ̸=z.From (3.3.22), we have thatwω(xn) =wω(zn) and by Lemma 2.1.15 we have
lim inf
k→∞ ||znk −z||<lim inf
k→∞ |znk −W z|||
≤lim inf
k→∞ {||znk−W znk||+||W znk −W z||}
≤lim inf
k→∞ {||znk−W znk||+||znk −z||}. (3.3.26) Since xnk ∈K for all k≥1 and wω(xn) =wω(zn), we obtain
||W znk −znk|| ≤ ||W znk −Wnkznk||+||Wnkznk −znk||
≤ sup
x∈K
||W x−Wnkx||+||Wnkznk−znk||.
By applying Lemma 3.2.5 and (3.3.17), we have lim
k→∞||W znk −znk||= 0. Combining this with (3.3.26) yields
lim inf
k→∞ ||znk−z||<lim inf
k→∞ ||znk −z||,
which is a contradiction. Hence, we have z ∈F(W) =∩∞i=1F(Si), i.e., wω(xn)⊂F(W) = T∞
i=1F(Si).
Next, we show thatwω(xn)⊂V I(C, A).By invoking Lemma3.3.7, it follows from (3.3.20) that z ∈V I(C, A).Thus, wω(xn)⊂Γ.
From the fact that lim
k→∞∥xnk−znk∥= 0,we have thatwω{xn}=wω{zn}.By Lemma3.3.6, we have that {xnk} is bounded which implies that there exists a subsequence {xnk j} of {xnk}such that xnk j ⇀x¯and
j→∞lim⟨γ1f x∗−Dx∗, xnk j −x∗⟩= lim sup
k→∞
⟨γ1f x∗−Dx∗, xnk−x∗⟩
= lim sup
k→∞
⟨γ1f x∗−Dx∗, znk−x∗⟩.
Since x∗ =PΓ(I−D+γ1f)x∗, we have that lim sup
k→∞
⟨γ1f x∗−Dx∗, xnk−x∗⟩= lim
j→∞⟨γ1f x∗−Dx∗, xnk j −x∗⟩
=⟨γ1f x∗ −Dx∗,x¯−x∗⟩
≤0. (3.3.27)
From (3.3.25) and (3.3.27), we have lim sup
k→∞
⟨γ1f x∗−Dx∗, xnk+1−x∗⟩= lim sup
k→∞
⟨γ1f x∗−Dx∗, xnk−x∗⟩
=⟨γ1f x∗−Dx∗,x¯−x∗⟩
≤0. (3.3.28)
By applying Lemma 2.5.36 to (3.3.14) and using (3.3.28) together with the fact that
n→∞lim θn= 0 and lim
n→∞
αn
θn∥xn−xn−1∥= 0,we have that lim
n→∞∥xn−x∗∥= 0.Therefore,{xn} converges strongly to x∗.
Next, we propose our second algorithm which is a slight modification of Algorithm (3.3.2).
Algorithm 3.3.3. Inertial method with adaptive step size strategy.
Step 0: Choose sequences {θn}∞n=1 and {τn}∞n=1 such that condition (B5) holds and let λ1 >0, µ∈(0,1), α≥3 and x0, x1 ∈ H be given arbitrarily. Set n:= 1.
Step 1: Given the iteratesxn−1 and xn for eachn≥1,chooseαnsuch that 0 ≤αn≤α¯n, where
¯ αn:=
(minn
n−1
n+α−1,∥x τn
n−xn−1∥
o
, if xn ̸=xn−1 n−1
n+α−1, otherwise.
(3.3.29) Step 2: Compute
wn =xn+αn(xn−xn−1).
Step 3: Construct the halfspace
Cn ={w∈ H |c(wn) +⟨ξn, w−wn⟩ ≤0}
and compute
yn =PCn(wn−λnAwn).
If yn =wn(Awn = 0), then setwn=zn and go to Step 5. Else go to Step 4.
Step 4: compute
zn =yn−λn(Ayn−Awn).
Step 5: Compute
xn+1 =θnγ1f wn+ (I−θnD)Tnzn, where
Tn = (1−βn)I+βnWn and Wn is the mapping defined by (3.2.1).
Step 6: Compute
λn+1 = (min
n µ||wn−yn||
||Awn−Ayn||, λn
o
, if Awn̸=Ayn
λn, otherwise. (3.3.30)
Set n :=n+ 1 and go back to Step 1.
Theorem 3.3.10. Let{xn}be a sequence generated by Algorithm3.3.3under Assumption 3.3.1. Suppose that {Wn} is the sequence defined by (3.2.1). Then, the sequence {xn} converges strongly to a point x∗ ∈ Γ, where x∗ = PΓ(I−D+γ1f)x∗ is a solution of the variational inequality
⟨(D−γ1f)x∗, x∗−x⟩ ≤0, ∀x∈Γ.
Proof. The proof of this result follows similar argument with the proof of the result of Theorem (3.3.9).
Taking γ1 = 1 and D =I in Theorem 3.3.9 (where I is the identity mapping), we obtain the following:
Algorithm 3.3.4. Inertial method with adaptive step size strategy.
Step 0: Choose sequences {θn}∞n=1 and {τn}∞n=1 such that condition (B5) holds and let λ1 >0, µ∈(0,1), α≥3 and x0, x1 ∈ H be given arbitrarily. Set n:= 1.
Step 1: Given the iteratesxn−1 and xn for eachn≥1,chooseαnsuch that 0 ≤αn≤α¯n, where
¯ αn:=
(minn
n−1
n+α−1,∥x τn
n−xn−1∥
o
, if xn ̸=xn−1 n−1
n+α−1, otherwise.
Step 2: Compute
wn =xn+αn(xn−xn−1).
Step 3: Construct the halfspace
Cn ={w∈ H |c(wn) +⟨ξn, w−wn⟩ ≤0}
and compute
yn =PCn(wn−λnAwn).
If yn =wn(Awn = 0), then setwn=zn and go to Step 5. Else go to Step 4.
Step 4: compute
zn =yn−λn(Ayn−Awn).
Step 5: Compute
xn+1 =θnf xn+ (I−θn)Tnzn, where
Tn = (1−βn)I+βnWn and Wn is the mapping defined by (3.2.1).
Step 6: Compute
λn+1 =
(minn µ||w
n−yn||
||Awn−Ayn||, λno
, if Awn̸=Ayn
λn, otherwise.
Set n :=n+ 1 and go back to Step 1.
Corollary 3.3.5. Let {xn} be a sequence generated by Algorithm 3.3.4under Assumption 3.3.1. Suppose that {Wn} is the sequence defined by (3.2.1). Then, the sequence {xn} converges strongly to a point x∗ ∈ Γ, where x∗ =PΓ(f)x∗ is a solution of the variational inequality
⟨(I −f)x∗, x∗−x⟩ ≤0, ∀x∈Γ.
Taking γ1 = 1, D = I (where I is the identity mapping) and Sn = S for all n ≥ 1 in Theorem 3.3.9, we obtain the following:
Algorithm 3.3.6. Inertial method with adaptive step size strategy.
Step 0: Choose sequences {θn}∞n=1 and {τn}∞n=1 such that condition (B5) holds and let λ1 >0, µ∈(0,1), α≥3 and x0, x1 ∈ H be given arbitrarily. Set n:= 1.
Step 1: Given the iteratesxn−1 and xn for eachn≥1,chooseαnsuch that 0 ≤αn≤α¯n, where
¯ αn:=
(minn
n−1
n+α−1,∥x τn
n−xn−1∥
o
, if xn ̸=xn−1 n−1
n+α−1, otherwise.
Step 2: Compute
wn =xn+αn(xn−xn−1).
Step 3: Construct the halfspace
Cn ={w∈ H |c(wn) +⟨ξn, w−wn⟩ ≤0}
and compute
yn =PCn(wn−λnAwn).
If yn =wn(Awn = 0), then setwn=zn and go to Step 5. Else go to Step 4.
Step 4: compute
zn =yn−λn(Ayn−Awn).
Step 5: Compute
xn+1 =θnf xn+ (I−θn)Tnzn, where
Tn= (1−βn)I+βnS.
Step 6: Compute
λn+1 =
(minn µ||w
n−yn||
||Awn−Ayn||, λno
, if Awn̸=Ayn
λn, otherwise.
Set n :=n+ 1 and go back to Step 1.
Corollary 3.3.7. Let {xn} be a sequence generated by Algorithm 3.3.6under Assumption 3.3.1. Then the sequence {xn} converges strongly to a point x∗ ∈Γ, where x∗ = PΓ(f)x∗ is a solution of the variational inequality
⟨(I −f)x∗, x∗−x⟩ ≤0, ∀x∈Γ.