Rates of convergence for iterative solutions of

arXiv:1908.06734v1 [math.OC] 19 Aug 2019

equations involving set-valued accretive operators

Ulrich Kohlenbach and Thomas Powell
Department of Mathematics
Technische Universität Darmstadt
Schlossgartenstraße 7
64289 Darmstadt, Germany
Email: {kohlenbach,powell}@mathematik.tu-darmstadt.de

Abstract
This paper studies proofs of strong convergence of various iterative
algorithms for computing the unique zeros of set-valued accretive oper-
ators that also satisfy some weak form of uniform accretivity at zero.
More precisely, we extract explicit rates of convergence from these proofs
which depend on a modulus of uniform accretivity at zero, a concept
first introduced by A. Koutsoukou-Argyraki and the first author in 2015.
Our highly modular approach, which is inspired by the logic-based proof
mining paradigm, also establishes that a number of seemingly unrelated
convergence proofs in the literature are actually instances of a common
pattern.

Keywords: Accretive operators, uniform accretivity, uniformly smooth Banach

spaces, Ishikawa iterations, rates of convergence, proof mining.
Mathematics Subject Classification (2010): 47H05, 47J25, 03F10.

1 Introduction
The problem of approximating zeroes of accretive set-valued operators A : X →
2X has been widely studied since the 70’s. This is primarily due to the im-
portance of these operators in modelling abstract Cauchy problems such as
evolution equations (see e.g. [2, 3]), as well as - for Hilbert spaces H = X (and
under the name monotone operators) - their relevance in convex optimization
for the computation of minima of lower semi-continuous functions f , where then
A = ∂f is the subdifferential of f (see e.g. [4]) and the zero set of A coincides
with the set of minimizers of f.
In the context of Hilbert spaces, a standard tool for approximating a zero
of A is the famous Proximal Point Algorithm (PPA) (due to [20, 22]) which
iterates, for varying coefficients λn > 0 satisfying appropriate conditions, the

1
(single-valued and firmly nonexpansive) resolvent JλA = (I + λA)−1 of A :

xn+1 = Jλn A xn .

Only recently [12] has the PPA been studied quantitatively in the general setting
of uniformly convex Banach spaces.
For arbitrary Banach spaces, various different types of iterations from metric
fixed point theory have been used to compute zeros, such as the Krasnoselski-
Mann or Ishikawa iterations. Just as for the PPA, in general these iterations
converge only weakly, and even when strong convergence holds (e.g. in the finite
dimensional case or for ‘Halpern-type’ modifications), there is usually - already
for X = R - no computable (in the sense of Church-Turing) rate of convergence
(see e.g. [21]).
This situation changes when A satisfies some form of strong accretivity, which
ensures that A has at most one zero 0 ∈ Aq. General theorems from logic
guarantee, for a broad range of situations, that in the presence of uniqueness one
can use quantitative data from the uniqueness proofs (e.g. so-called moduli of
uniform uniqueness) to give rates of convergence for procedures which compute
approximate solutions to problems (such as finding zeroes or fixed points). For
all this see e.g. [10].
Most forms of strong (quasi-)accretivity are stronger and more restricted
instances of what is called uniform accretivity at zero in [13, Definition 10],
which is given a quantitative form via a modulus Θ of accretivity at zero.
The purpose of this paper is two-fold:
1. to show that in typical cases of known strongly convergent algorithms
computing the unique zero of a strongly accretive operator A, one can
extract from the convergence proof an explicit rate of convergence in terms
of a modulus Θ of accretivity at zero;
2. to provide, using the concept of uniform accretivity at zero together with
the logical analysis of the convergence proofs, a modular and unified ac-
count of strong convergence results in the literature which at first glance
appear unrelated.
This is exemplified by selecting as test cases the implicit iteration schema from
[1] together with the explicit Ishikawa-type schemes used in [19] and in [6] (the
latter paper being further generalized e.g. in [18]). In particular, we recover
as special cases the quantitative results in [1]. Whereas the main convergence
theorems in [1] (Theorems 2.1 and 4.1) hold in arbitrary Banach spaces and
without any continuity assumption on A, the convergence results in [6] (Theorem
4.1) and [19] (Theorem 2.2) use the uniform continuity of A (w.r.t. the Hausdorff
metric) while [6] (Theorem 4.2) and [18] (Theorem 2.1) instead use that X is
uniformly smooth. Although the assumptions on A being uniformly continuous
and, respectively, on X being uniformly smooth are very different, it turns
out they both follow as instances of the same technical lemma. The rates of
convergence we extract in these cases then also depend (in addition to Θ) on

2
moduli of uniform continuity for A and, respectively, for the duality mapping
of X, where in the latter case such a modulus can be computed in terms of a
modulus of uniform smoothness for X (see [14]).
The various forms of strong (quasi-)accretivity used in the aforementioned
results are all covered by mostly more restrictive versions of our concept of
uniform accretivity at zero (note that [1] uses uniform accretivity to denote a
concept which is much more restrictive than our notion of uniform accretivity
at zero even when we drop the restriction ‘at zero’ as it corresponds to ψ-
strong accretivity as defined in Definition 2.3.(a) with ψ additionally assumed
to be strictly increasing). Therefore our results strengthen various convergence
theorems not just quantitatively but also qualitatively.
Since the convergence proofs we study all apply to situations where A can
be shown to have a unique zero, in our quantitative results we always assume
both the existence of a zero and well-definedness of the approximating sequence
at hand, which typically allows us to omit certain extra assumptions made in
the original papers.
Although no concepts or methods from logic are mentioned explicitly in this
paper, our approach has been motivated by the tools of the proof mining pro-
gram which uses logic-based proof transformations for the extraction of effective
bounds from prima facie noneffective proofs (see [10]). In the case of the prox-
imal point algorithm, this approach - again based on the concept of uniform
accretivity (specialized to the monotone case in Hilbert spaces) - has been used
in [17]. For a recent survey on proof mining in general see [11].

2 Preliminaries
N := {0, 1, 2, 3, . . .} denotes the set of nonnegative integers.
Throughout this paper, X will be a real Banach space with dual space X ∗ . The
∗
normalized duality mapping J : X → 2X is defined by
2 2
J(x) := {j ∈ X ∗ : hx, ji = kxk = kjk }.

We will make frequent use of the following well known geometric inequality.
Lemma 2.1. For all x, y ∈ X and j ∈ J(x + y) we have
2 2
kx + yk ≤ kxk + 2hy, ji.

Proof. Let j ∈ J(x + y). Then

2 2 2
kx + yk = hx + y, ji ≤ kxk · kx + yk + hy, ji ≤ 21 (kxk + kx + yk ) + hy, ji

and the result follows.

A mapping A : X → 2X will be called an operator on X. The domain of A
is defined by D(A) := {x ∈ X : Ax 6= ∅}. We sometimes write (x, u) ∈ A for
u ∈ Ax. The range R(A) of A is defined as R(A) := {y ∈ X : ∃x ∈ X(y ∈ Ax)}.

3
2.1 Accretive operators
For a detailed survey of the various notions of accretivity, including quantitative
forms which come equipped with moduli, the reader is encouraged to consult
[13, Section 2.1]. Here, we simply outline the key definitions which play a role
in the present paper.
Definition 2.2. An operator A is said to be accretive if for all u ∈ Ax and v ∈ Ay
there exists some j ∈ J(x − y) such that hu − v, ji ≥ 0.
The notion of accretivity was independently introduced (in a slightly differ-
ent but equivalent form) by Browder [5], Kato [8] and Komura [16]. However,
convergence proofs of the kind we study here typically appeal to various stronger,
uniform forms of accretivity:
Definition 2.3. (a) Let ψ : [0, ∞) → [0, ∞) be a continuous function with
ψ(0) = 0 and ψ(x) > 0 for x > 0. Then an operator A : D(A) → 2X
is said to be ψ-strongly accretive if
∀(x, u), (y, v) ∈ A ∃j ∈ J(x − y) (hu − v, ji ≥ ψ(kx − yk)kx − yk).

(b) Let φ : [0, ∞) → [0, ∞) be a continuous function with φ(0) = 0 and φ(x) > 0
for x > 0. Then an operator A : D(A) → 2X is said to be uniformly φ-
accretive if
∀(x, u), (y, v) ∈ A ∃j ∈ J(x − y) (hu − v, ji ≥ φ(kx − yk)).
It turns out that for all of the results we study in this paper, the above no-
tions can be replaced by the following more general property of being uniformly
accretive at zero, introduced by Garcı́a-Falset in [7] and given a quantitative
form by the first author in [13].
Definition 2.4. An an accretive operator A : D(A) → 2X with 0 ∈ Aq is said to
be uniformly accretive at zero if
∀ε, K > 0 ∃δ > 0 ∀(x, u) ∈ A
(∗)
(kx − qk ∈ [ε, K] → ∃j ∈ J(x − q) (hu, ji ≥ δ)).
Moreover, any function Θ(·) (·) : (0, ∞) × (0, ∞) → (0, ∞) such that δ := ΘK (ε)
satisfies (∗) is called a modulus of uniform accretivity at zero for A.
In particular, we observe that if A is uniformly φ-accretive, a modulus of
uniform accretivity at zero for A is given by
ΘK (ε) := inf{φ(x) : x ∈ [ε, max{ε, K}]},
and in the case where φ is also strictly increasing, we can simply define ΦK (ε) :=
φ(ε).
Remark 2.5. Though technically speaking, moduli of uniform accretivity at
zero are defined relative to some given q ∈ D(A) with 0 ∈ Aq, one can actually
show that such a q, if it exists, is necessarily unique. Moreover, a modulus of
uniqueness for q can be constructed in terms of a modulus of uniform accretivity
at zero, as is made precise in [13, Remark 2].

4
 Accretivity of an operator A is typically associated with a corresponding
notion of pseudocontractivity for the operator (I − A). In the case of uniformly
accretive operators at zero, the correspondence is given as follows:
Lemma 2.6. Suppose that A : D(A) → 2X with 0 ∈ Aq is uniformly accretive
at zero with modulus Θ(·) (·). Then

∀ε, K > 0 ∀(x, u) ∈ (I − A)

2
(kx − qk ∈ [ε, K] → ∃j ∈ J(x − q) (hu − q, ji ≤ kx − qk − ΘK (ε))).

Proof. If u = (I − A)x then u = x − ū for ū ∈ Ax, and thus if kx − qk ∈ [ε, K]

there exists some j ∈ J(x − q) such that hū, ji ≥ ΘK (ε). Therefore
2 2
hu − q, ji = hx − q, ji − hū, ji = kx − qk − hū, ji ≤ kx − qk − ΘK (ε).

3 An abstract technical lemma

We begin by presenting an abstract quantitative lemma, which forms the main
unifying scheme of the paper. This technical lemma captures a key combina-
torial idea which is shared by numerous proofs of strong convergence theorems
involving accretive operators, and as we will see, quantitative versions of those
theorems can be obtained in an entirely modular fashion by instantiating the
parameters of our lemma in a suitable way. What is particularly interesting
is that in each case we study, those parameters are obtained by appealing to
quantitative versions of assumptions which are seemingly unrelated, which here
include properties imposed on the operator A (Sections 5 and 6) or alternatively
attributes of the underlying space X (Section 7). Moreover, our abstract result
applies to different approximating schemes, including implicit schemes (Sections
4 and 5) in addition to Ishikawa methods (Sections 6 and 7).

3.1 Rates of convergence and divergence

We begin by specifying quantitative versions of a couple of fundamental notions.
Definition 3.1. Let (αn ) be a sequence of nonnegative reals such that αn → 0.
A rate of convergence for (αn ) is a function φ : (0, ∞) → N such that

∀ε > 0 ∀n ≥ φ(ε) (αn ≤ ε).

P∞
sequence of nonnegative reals such that i=0 αi =
Definition 3.2. Let (αn ) be aP
∞
∞. A rate of divergence for i=0 αi is a function r : N × (0, ∞) → N such that

r(N,x)
X
∀N ∈ N ∀x > 0 ( αi ≥ x).
i=N

5
Remark 3.3. The quantitative formulation of divergence above is also used by
the first author in [9]. Note that a more traditional rate of divergence would be
a function f : (0, ∞) → N satisfying
f (x)
X
∀x > 0 ( αi ≥ x),
i=0

which can be converted into a rate of divergence in our sense by setting r(N, x) :=
Pn−1
f (x + S(N )) where S : N → (0, ∞) is any function satisfying i=0 αi ≤ S(n),
since then we have
f (x+S(N )) f (x+S(N )) n−1
X X X
αi = αi − αi ≥ (x + S(N )) − S(N ) = x.
i=N i=0 i=0

In particular, if the αi are bounded above by some K, we can simply set

r(N, x) := f (x + K · N ).

3.2 The technical lemma

We now present our unifying lemma, which generalises similar abstract results
in the literature, such as Lemma 2.2 of [1] and Lemma 2.1 of [19], the latter
having been given a quantitative form as Lemma 1 of [15].
Lemma 3.4. Let (θn ) and (αn ) be sequences of nonnegative reals such that
P ∞
i=0 αi diverges, and suppose that for any ε > 0 there exists some δ > 0 and
N ∈ N such that

(∗) ∀n ≥ N (ε < θn+1 → θn+1 ≤ θn − αn · δ).

Then θn → 0 as n → ∞. Moreover, if:

(i) K ∈ (0, ∞) satisfies θn < K for all n ∈ N,
P∞
(ii) r : N × (0, ∞) → N is a rate of divergence for i=0 αi ,
(iii) N : (0, ∞) → N and ϕ : (0, ∞) → (0, ∞) witness property (∗) in the sense
that for all ε > 0 we have

∀n ≥ N (ε)(ε < θn+1 → θn+1 ≤ θn − αn · ϕ(ε)),

then ΨK,r,N,ϕ(ε) := r(N (ε), K/ϕ(ε)) + 1 is a rate of convergence for (θn ).

Proof. We first observe that for any ε > 0 and n ≥ N (ε) we have

θn ≤ ε → θn+1 ≤ ε.

Otherwise, if there were some n ≥ N (ε) with θn ≤ ε and ε < θn+1 we would
have
ε < θn+1 ≤ θn − αn · ϕ(ε) ≤ θn ≤ ε.

6
Therefore to establish θn → 0 it suffices to find, for each ε > 0, a single n ∈ N
with θn ≤ ε. Fixing some ε > 0 and j ≥ N (ε), suppose for that θn+1 > ε for all
n ∈ N with N (ε) ≤ n ≤ j. Then in particular we would have

αn · ϕ(ε) ≤ θn − θn+1

for all n in this range, and thus

j
X j
X
ϕ(ε) αn ≤ (θn − θn+1 ) = θN (ε) − θj+1 ≤ θN (ε) < K.
n=N (ε) n=N (ε)

But this is a contradiction for j := r(N (ε), K/ϕ(ε)), and thus θn ≤ ε for some
n ≤ j + 1, which means that m ≥ j + 1 ≥ n we also have θm ≤ ε.
Remark 3.5. Condition (i) of Lemma 3.4 is not strictly necessary, as we do
not require boundedness of (θn ) to establish θn → 0. However, as a rate of
convergence we would then obtain e.g. Ψ(θn ),r,N,ϕ := r(N (ε), (θN (ε) + 1)/ϕ(ε))
which is dependent on the (θn ). In each subsequent applications of this result,
we are able to supply a uniform bound K for our sequence (θn ), in which case
our lemma results in a rate of convergence which is independent of the (θn ).
We conclude this section by observing that we can reformulate Lemma 3.4 so
that it no longer makes directPreference to a rate of divergence for ∞
P
i=0 i , but
α
∞
rather uses the divergence of i=0 αi implicitly. This will later allow to connect
our quantitative convergence theorems to the numerical results presented in [1].
Lemma 3.6. Let (θn ), (αn ), K, N and ϕ be as in Lemma 3.4, and assume in
addition that αn > 0 for all n ∈ N. Suppose that f : (0, ∞) → (0, ∞) is strictly
decreasing and continuous with f (ε) → ∞ as ε → 0 and
N (ε)−1
X K
f (ε) ≥ αi +
i=0
ϕ(ε)
P−1
for all ε > 0 (where we adopt the convention i=0 αi = 0). Then for sufficiently
large n ∈ N we have
n−1
!
X
−1
θn ≤ f αi .
i=0

Proof. First note that f must have an inverse f −1 : (a, ∞) → (0, ∞)

Pfor a :=
inf{f (x) : x ∈ (0, ∞)}. Define n0 ∈ N to be the least such that ni=00
αi ∈
(a, ∞), and for n ≥ n0 + 1 define
n−1
!
X
−1
εn := f αi .
i=0

7
Applying Lemma 3.4 for r(N, x) defined to be the least j ≥ N such that
Pj
i=N αi ≥ x, we have θm ≤ εn for all m ≥ j + 1, where j ≥ N (εn ) is the
least such that
j
X K
αi ≥ .
ϕ(εn )
i=N (εn )

Now observing that

n−1 N (εn )−1 n−1
X X K X K
αi = f (εn ) ≥ αi + and thus αi ≥
i=0 i=0
ϕ(εn ) ϕ(εn )
i=N (εn )

it follows that n − 1 ≥ j and therefore n ≥ j + 1, which means that θn ≤ εn .

Thus the lemma holds for all n ≥ n0 + 1.

3.3 Outline of the remainder of the paper

We now turn our attention towards concrete convergence theorems involving
strongly accretive operators A. We focus on a series of examples, where in each
case we utilise Lemma 3.4 together with a modulus of uniform accretivity at
zero for A to carry out a quantitative analysis of the proof in question, resulting
in a series of new, quantitative convergence results which each fall underneath
the same unifying scheme.

4 A simple implicit scheme

Our first result will be a quantitative analysis of the following result of Alber
et al. [1], which is based on a straightforward implicit approximation method
generated by a uniformly accretive operator.
Theorem 4.1 (Theorem 2.1 of [1]). Let D be a closed subset of X and A :
D → 2X a ψ-strongly accretive operator for some strictly increasing ψ, which
satisfies the range condition (RC)
D ⊂ (I + rA)(D), ∀r > 0.

Then the following assertions hold:

(a) There exists a unique q ∈ D such that 0 ∈ Aq.
(b) If (αi ) is a sequence of positive reals with ∞
P
i=0 αi = ∞, then if the sequence
(xn ) starting from some x0 ∈ D satisfies
xn+1 = xn − αn un , un ∈ Axn+1

we have kxn − qk → 0.
Our quantitative analysis in this case and in all those that follow will focus
on the extraction of an explicit rate of convergence for kxn − qk → 0 from the
corresponding proof of this fact. In doing so, we adopt the following pattern:

8
 We assume from the outset the existence of some q satisfying 0 ∈
Aq, and take some arbitrary sequence (xn ) satisfying the relevant
approximation scheme.
By focusing exclusively on the proof that kxn − qk → 0, we are typically able to
weaken certain conditions of the original theorem, which are often needed only
to establish the existence of a zero 0 ∈ Aq or to ensure that the sequence of
approximations (xn ) is well-defined. As such, we obtain a rate of convergence
for kxn − qk → 0 which is valid in a much more general setting. At the same
time, the original results guarantee us that there is always a natural context in
which a zero and a corresponding approximation sequence do indeed exist!
In the case of Theorem 4.1 above, for the purpose of our quantitative con-
vergence result, we are able to dispense with the range condition together with
the assumption that D is closed, and can take A to be an arbitrary operator
which is uniformly accretive at zero.
Theorem 4.2. Let A : D(A) → 2X with 0 ∈ Aq be uniformly accretive at
zero with modulus Θ. Let (αi ) be a sequence of nonnegative reals such that
P ∞
i=0 αi = ∞ with modulus of divergence r, and suppose that (xn ) and (un ) are
sequences satisfying xn ∈ D(A) and
xn+1 = xn − αn un , un ∈ Axn+1
for all n ∈ N. Finally, let K > 0 be such that kx0 − qk < K. Then kxn − qk → 0
with rate of convergence
ΦΘ,r,K (ε) := r(0, K 2 /ΘK (ε)) + 1.

Proof. We first observe that for any j ∈ J(xn+1 − q) we have

2
kxn+1 − qk =hxn+1 − q, ji
=hxn − q, ji − αn hun , ji (1)
≤kxn − qk · kxn+1 − qk − αn hun , ji.

We argue by induction that kxn − qk < K for all n ∈ N: For n = 0 this holds
by assumption, while the induction step follows directly from (1) together with
the accretivity of A, which ensures that hun , ji ≥ 0 for some j ∈ J(xn+1 − q).
Now, fixing some n ∈ N and ε > 0, suppose that ε < kxn+1 − qk. Then since
in particular we would have kxn+1 − qk ∈ [ε, K], by uniform accretivity of A at
zero it follows that there exists some j ∈ J(xn+1 − q) such that hun , ji ≥ ΘK (ε).
Substituting this into (1) and dividing by kxn+1 − qk we obtain
αn ΘK (ε) αn ΘK (ε)
kxn+1 − qk ≤ kxn − qk − < kxn − qk − . (2)
kxn+1 − qk K
We are now able to apply Lemma 3.4 for θn := kxn − qk. Conditions (i) and
(ii) of the lemma are clearly satisfied by K and r, while for condition (iii) we
set N (ε) := 0 and ϕ(ε) := ΘK (ε)/K, and our rate of convergence is obtained
directly.

9
Remark 4.3. In [1], the operator A is assumed to be ψ-strongly accretive for
some strictly increasing ψ. Under the additional assumption that 0 ∈ Aq, A
must then also be uniformly accretive at zero with modulus ΘK (ε) = ψ(ε) · ε,
since for u ∈ Ax with kx − qk ∈ [ε, K] there is, by ψ-strong accretivity, some
j ∈ J(x − q) such that
hu, ji ≥ ψ(kx − qk)kx − qk ≥ ψ(ε) · ε = ΘK (ε).
However, in the case of ψ-strong accretivity, we can reformulate (2) as
αn ψ(kxn+1 − qk)kxn+1 − qk
kxn+1 − qk ≤ kxn − qk − ≤ kxn − qk − αn ψ(ε)
kxn+1 − qk
and thus an improved rate of convergence for kxn − qk → 0 is given by
Φψ,r,K := r(0, K/ψ(ε)) + 1.
In this case, by appealing to Lemma 3.6, we obtain an implicit rate of conver-
gence closely related to Theorem 3.1 of [1].
Corollary 4.4. Let A : D(A) → 2X with 0 ∈ Aq be a ψ-strongly accretive
operator for some strictly increasing ψ, and otherwise let (αi ), r, (xn ), (un ) and
K be as in Theorem 4.2. Then kxn − qk → 0 with
!
K
kxn − qk < ψ −1 Pn−1
i=0 αi

sufficiently large n ∈ N.
Proof. We apply Lemma 3.6 with parameters instantiated as in Remark 4.3 i.e.
N (ε) = 0 and ϕ(ε) = ψ(ε). Then in particular, we can define our bounding
function f : (0, ∞) → (0, ∞) by f (ε) := K/ψ(ε), observing that as ε → 0 then
ψ(ε) → 0 and hence f (ε) → ∞. The result then follows by observing that
f −1 (x) = ψ −1 (K/x).
Remark 4.5. Note that Corollary 4.4 is broadly analogous but not identical to
the corresponding Theorem 3.1 of [1], which is to be expected, since
P the latter
uses an integral comparison rather than a rate of divergence for ∞i=0 αi .

5 An implicit scheme using approximating operators

Our second case study is also taken from [1]. Here, the implicit scheme studied
in the previous section is modified to one of the form
xn+1 = xn − αn un , un ∈ An xn+1
for some sequence of operators (An ), where in order to maintain convergence of
the (xn ) to some zero q when it exists, a convergence property for the (An ) is
required. In [1] this takes the form of approximation relative to the Hausdorff
distance.

10
Definition 5.1. Let A and An : D → 2X be operators defined on some subset D
of X for n ∈ N. We say that the sequence (An ) approximates the operator A if
there exists a sequence of positive reals (hn ) with hn → 0 as n → ∞ such that
∀x ∈ D ∀n ∈ N (H(An x, Ax) ≤ hn ξ(kxk))
where ξ : [0, ∞) → [0, ∞) is some given function and H denotes the Hausdorff
distance between sets, defined as usual by
H(P, Q) = max{ sup inf kx − yk, sup inf kx − yk}.
x∈X y∈Y y∈Y x∈X

We analyse the following generalisation of Theorem 2.1 of [1].

Theorem 5.2 (Theorem 4.1 of [1]). Let D be a closed subset of X and A : D →
2X a ψ-strongly accretive operator for some strictly increasing ψ which satisfies
the range condition (RC). Suppose that the sequence of operators An : D → 2X
approximates A and each An satisfies the strong range condition that for any
r > 0 and u ∈ D there exists a unique x ∈ D with
u ∈ (I + rAn )x.
If (αi ) is a sequence of positive reals with ∞
P
i=0 αi = ∞, (xn ) is the sequence
starting from some x0 ∈ D and defined by
xn+1 = xn − αn un , un ∈ An xn+1 ,
and (xn ) is bounded, then kxn − qk → 0 for the unique zero 0 ∈ Aq (which exists
by Theorem 4.1).
As in the previous section, we simply assume the existence of some 0 ∈
Aq, and as such, omit all range conditions from our quantitative version of
this result. However, that (An ) approximates the operator A is essential for
convergence of the (xn ). Just as various notions of strong accretivity are replaced
by uniform accretivity at zero, we define below a general, uniform variant of the
approximation property which reflect the more restricted way in which that
property is actually used in the proof of Theorem 5.2.
Definition 5.3. We define the predicate H ∗ ⊆ 2X × 2X × (0, ∞) by
H ∗ [P, Q, a] :≡ ∀u ∈ P ∃v ∈ Q (ku − vk ≤ a).
Definition 5.4. Let A and An : D → 2X be operators with D(A) = D(An ) = D.
We say that (An ) uniformly approximates A with the rate µ(·) (·) : (0, ∞) ×
(0, ∞) → N of uniform approximation if
∀K, ε > 0 ∀n ≥ µK (ε) ∀x ∈ D (kxk ≤ K → H ∗ [An x, Ax, ε]) .
Lemma 5.5. Suppose that (An ) approximates A (in the sense of Definition
5.1) with respect to (hn ) and some ξ : [0, ∞) → [0, ∞), and moreover there is a
function ξ ∗ : (0, ∞) → (0, ∞) satisfying
∀x, y ∈ [0, ∞)(x ≤ y ∧ y > 0 → ξ(x) ≤ ξ ∗ (y)).

11
Then (An ) uniformly approximates A. Moreover, if φ : (0, ∞) → N is a rate of
convergence for hn → 0 then

µL (ε) := φ(2ε/3ξ ∗ (L))

is a rate of uniform approximation for (An ) and A.

Proof. We first observe that for P, Q ∈ 2X , whenever H(P, Q) < a for some
a ∈ (0, ∞) it follows that H ∗ [P, Q, a]. To see this, note that H(P, Q) < a
implies in particular that for all u ∈ P we have

inf ku − vk < a
v∈V

and thus there must exists some v ∈ V with ku − vk < a. Now, fixing some
n, K (with K > 0) and x ∈ D with kxk ≤ K, we have

H(An x, Ax) ≤ hn ξ(kxk) ≤ hn ξ ∗ (K) < 32 hn ξ ∗ (K)

where for the last step we use ξ ∗ (K) > 0. Now let n ≥ φ(2ε/3ξ ∗ (K)), then
3
2 hn ξ (K) ≤ ε and so H (An x, Ax, ε).
∗ ∗

We are now ready to state and prove our quantitative formulation of Theo-
rem 5.2.
Theorem 5.6. Let A : D → 2X with 0 ∈ Aq be uniformly accretive at zero
with modulus Θ, and An : D → 2X be a sequence of operators which uniformly
approximates
P∞ A with rate µ. Let (αi ) be a sequence of nonnegative reals such
that i=0 αi = ∞ with modulus of divergence r, and suppose that (xn ) and (un )
are sequences satisfying xn ∈ D and

xn+1 = xn − αn un , un ∈ An xn+1

for all n ∈ N. Finally, suppose K, K ′ ∈ (0, ∞) satisfy kxn − qk < K for all
n ∈ N and kqk < K ′ . Then kxn − qk → 0 with rate of convergence

K2
   
ΘK (ε)
ΦΘ,µ,r,K,K ′ (ε) := r µK+K ′ , + 1.
2K ΘK (ε)

Proof. Using the assumption that (An ) uniformly approximates A, together

with the assumption that xn+1 ∈ D and kxn+1 k ≤√kxn+1 − qk + kqk < K + K ′
for each n ∈ √
N, we have H ∗ [An xn+1 , Axn+1 , ΘK ( ε)/2K] for all n ≥ N (ε) :=
µK+K ′ (ΘK ( ε)/2K) . In particular, this means that for all n ≥ N (ε) there
exists some vn ∈ Axn+1 such that
√
ΘK ( ε)
kun − vn k ≤ 2K . (3)

12
Now, for any j ∈ J(xn+1 − q) we have
2 2
kxn+1 − qk =kxn − αn un − qk
L.2.1
≤ kxn − qk2 − 2αn hun , ji
2
=kxn − qk + 2αn hvn − un , ji − 2αn hvn , ji (4)
2
≤kxn − qk + 2αn kvn − un k · kxn+1 − qk − 2αn hvn , ji
√
2 ΘK ( ε)
≤kxn − qk − 2αn (hvn , ji − 2 ).
where for the last step we use (3) by which
√
ΘK ( ε)
kxn+1 − qk · kvn − un k < K · 2K .
√
Now suppose that ε < kxn+1 − qk2 , and thus kxn+1 − qk ∈ [ ε, K]. Then by
uniform accretivity
√ of A at zero there exists some j ∈ J(xn+1 − q) such that
hvn , ji ≥ ΘK ( ε) and hence
√ √ √
hvn , ji − ΘK 2( ε) ≥ ΘK ( ε) − ΘK 2( ε)
√ (5)
= 12 ΘK ( ε)
2
Substituting (5) into (4), for n ≥ N (ε) and ε < kxn+1 − qk we have
2 2
kxn+1 − qk ≤ kxn − qk − αn · ϕ(ε)
√ 2
for ϕ(ε) := ΘK ( ε). Therefore applying Lemma 3.4 for θn := kxn − qk ≤ K 2 ,
2
where condition (i) is witnessed by K , (ii) by r and (iii) by N and ϕ as defined
2
above, we obtain a rate of convergence for kxn − qk → 0, which can be modified
to a rate of convergence for kxn − qk → 0 by substituting ε2 for ε.
We conclude our study of [1] with a final quantitative result which forms a
more direct counterpart of Theorem 4.1 in [1], which brings together Lemma 5.5
and Theorem 5.6 above, and in addition incorporates the discussion on pp.100-
101 of [1], in which boundedness of the
P∞ kxn k is replaced by the a priori condition
that the An are each accretive and n=0 αn hn < ∞.
Theorem 5.7. Let A : D → 2X with 0 ∈ Aq be uniformly accretive at zero
with modulus Θ, and An : D → 2X be a sequence of accretive operators each
satisfying the range condition (RC) which approximates A with respect to (hn )
and some ξ : [0, ∞) → [0, ∞). Let φ : (0, ∞) → N be a rate of convergence for
hn → 0 and ξ ∗ : (0, ∞) → (0, ∞) a function satisfying
∀x, y ∈ [0, ∞)(x ≤ y ∧ y > 0 → ξ(x) ≤ ξ ∗ (y)).
P∞
In addition, let (αi ) be a sequence of nonnegative reals such that i=0 αi = ∞
with modulus of divergence r, and suppose that (xn ) and (un ) are sequences
satisfying xn ∈ D and
xn+1 = xn − αn un , un ∈ An xn+1

13
for all n ∈ N. Finally,
Pn suppose that K0 , K1 , K2 ∈ (0, ∞) satisfy kx0 − qk < K0 ,
kqk < K1 and i=0 αi hi < K2 for all n ∈ N. Then kxn − qk → 0 with rate of
convergence

K2
   
ΘK (ε)
ΦΘ,φ,ξ∗ ,r,K0 ,K1 ,K2 (ε) := r φ , +1
3K · ξ ∗ (K + K1 ) ΘK (ε)

for K := K0 + K2 · ξ ∗ (K1 ).
Proof. Since An satisfies the range condition, we have q ∈ (I +αn An )(D), which
means there exist a pair of sequences (yn ) and (vn ) with

q = yn + αn vn , vn ∈ An yn

for all n ∈ N. We now observe that since kqk < K1 we have H(An q, Aq) <
hn ξ ∗ (K1 ), and thus since 0 ∈ Aq there exists some wn ∈ An q satisfying kwn k <
hn ξ ∗ (K1 ). Now for any j ∈ J(yn − q) we have

kyn − qk2 = hyn − q, ji = −αn hvn , ji = −αn hwn , ji − αn hvn − wn , ji. (6)

Since An is accretive there exists some j ∈ J(yn − q) such that hvn − wn , ji ≥ 0

and substituting this into (6) we get
2
kyn − qk ≤ −αn hwn , ji ≤ αn kwn kkyn − qk

and therefore
kyn − qk ≤ αn kwn k ≤ αn hn ξ ∗ (K1 ) (7)
By a similar calculation we see that for j ∈ J(xn+1 − yn ) we have
2
kxn+1 − yn k = hxn+1 − yn , ji = hxn − q, ji − αn hun − vn , ji

and again by accretivity of An on un ∈ An xn+1 and vn ∈ An yn we see that

kxn+1 − yn k ≤ kxn − qk (8)

Putting (7) and (8) together we obtain

kxn+1 − qk ≤ kxn+1 − yn k + kyn − qk ≤ kxn − qk + αn hn ξ ∗ (K1 )

and therefore
n−1
X
kxn − qk ≤ kx0 − qk + αi hi ξ ∗ (K1 ) < K0 + ξ ∗ (K1 ) · K2 .
i=0

This establishes boundedness of kxn −qk for n ∈ N. We can now apply Theorem
5.6 for K := K0 + ξ ∗ (K1 ) · K2 , K ′ := K1 and (by Lemma 5.5) µL (ε) :=
φ(2ε/3ξ ∗ (L)) to obtain the given rate of convergence.

14
6 An Ishikawa scheme for uniformly continuous operators
Our next result is a quantitative analysis of a theorem due to Moore and Nnoli,
which rather than the implicit schemes studied in the previous section deals
with an explicit Ishikawa method for approximating zeroes of accretive operators
A. Here, convergence is made possible by demanding that the operator A be
uniformly continuous in the following sense.
Definition 6.1. Let CB(X) denote the family of all nonempty subsets of X
which are closed and bounded. An operator A : D(A) → CB(X) ⊂ 2X is said
to be uniformly continuous if it satisfies

∀ε > 0 ∃δ > 0 ∀x, y ∈ X(kx − yk ≤ δ → H(Ax, Ay) ≤ ε).

Theorem 6.2 (Theorem 2.2 of [19]). Let A : X → 2X be a uniformly continuous

and uniformly quasi-accretive operator with nonempty closed values such that
the range of (I − A) is bounded and 0 ∈ Aq for somePq∞∈ X. Let (αn ), (βn ) be
sequences in [0, 12 ) such that αn → 0, βn → 0 and i=0 αi = ∞. Finally, let
(xn ) be the sequences generated from some x0 ∈ X via the Ishikawa scheme

xn+1 = (1 − αn )xn + αn un , un ∈ (I − A)yn

yn = (1 − βn )xn + βn vn , vn ∈ (I − A)xn

Then (xn ) converges strongly to q.

Remark 6.3. The notion of quasi-accretivity (cf. [19, Definition 1.3]) is essen-
tially a formulation of uniform φ-accretivity for zeroes, and will in any case be
replaced by our notion of uniform accretivity at zero.
We now present our computational interpretation of the above theorem,
which in particular replaces uniform continuity of A with the following quantita-
tive condition involving the Hausdorff-like predicate H ∗ introduced in Definition
5.3.
Definition 6.4. Let A : D(X) → 2X be an operator. A functional ̟ : (0, ∞) →
(0, ∞) is called a modulus of uniform continuity for A if it satisfies

∀ε > 0 ∀x, y ∈ D(A)(kx − yk ≤ ̟(ε) → H ∗ [Ax, Ay, ε]).

Remark 6.5. Note that given some A : X → CB(X), if ω : (0, ∞) → (0, ∞)

is a traditional modulus of uniform continuous with respect to the Hausdorff
metric, in that it satisfies

∀ε > 0 ∀x, y ∈ D(A)(kx − yk ≤ ω(ε) → H(Ax, Ay) ≤ ε), (9)

then ̟(ε) := ω(ε/2) is a modulus of uniform continuity for A in the sense of

Definition 6.4. To see this, note that if kx − yk ≤ ω(ε/2) then H(Ax, Ay) ≤
ε/2 < ε, which implies that for any u ∈ Ax there must exist some v ∈ Ay
with ku − vk < ε, which is just H ∗ [Ax, Ay, ε]. However, possessing a modulus
of uniform continuity ̟ is more general than possessing some ω satisfying (9),

15
since in particular the latter only makes sense when H(Ax, Ay) always exists,
whereas our formulation allows us to drop assumptions about the range of A,
and so in particular we do not require A to always return nonempty closed
values.
Theorem 6.6. Let A : D(X) → 2X with 0 ∈ Aq be uniformly accretive at
zero with modulus Θ, and in addition suppose that A has a modulus of uniform
continuity ̟ : (0, ∞) → (0, ∞). Let (αn ), (βn ) be sequences
P∞ in [0, 21 ) such that
αn , βn → 0 with joint rate of convergence φ and i=0 αi = ∞ with rate of
divergence r. Suppose that (xn ), (yn ), (un ) and (vn ) are sequences satisfying
xn , yn ∈ D(A) and
xn+1 = (1 − αn )xn + αn un , un ∈ (I − A)yn
yn = (1 − βn )xn + βn vn , vn ∈ (I − A)xn
Finally, suppose that K0 , K1 ∈ (0, ∞) satisfy kwk < K0 for all w ∈ R(I − A)
and kx0 − qk < K1 .
Then kxn − qk → 0 as n → ∞ with rate of convergence
ΦΘ,̟,φ,r,K0 ,K1 (ε)
K2
      
1 1 ΘK (ε) ΘK (ε)
:= r φ min , min ,̟ , +1
4 3K 16K 16K ΘK (ε)
for K := 2K0 + K1 .
Proof. We first show by induction that kxn −qk < K for K = 2K0 +K1 . For the
base case we have kx0 − qk < K1 < K, and for the induction step we calculate:
kxn+1 − qk = k(1 − αn )(xn − q) + αn (un − q)k
≤ (1 − αn )kxn − qk + αn kun − qk
< (1 − αn )K + αn K
=K
where to establish kun − qk < K we use that q = (I − A)q and hence q ∈
R(I − A), and - by assumption - un ∈ R(I − A), from which we see that
kun − qk ≤ kun k + kqk < 2K0 < K. By an analogous argument we then also
see that
kyn − qk ≤ (1 − βn )kxn − qk + βn kvn − qk < K
for all n ∈ N. We are now also able to show that
kyn − xn+1 k = kαn xn − βn xn + βn vn − αn un k
≤ αn kxn k + βn kxn k + βn kvn k + αn kun k (10)
≤ 3(αn + βn )K ≤ 3K,
where for the last step we use that kun k, kvn k < K0 < K and kxn k ≤ kxn −
qk + kqk < K + K0 < 2K. Appealing to the joint rate of convergence φ for
αn , βn → 0 we see that for δ̃ ≤ δ/3K
∀n ≥ φ(δ̃)(kyn − xn+1 k ≤ δ). (11)

16
For the remainder of the proof we fix some ε > 0 and suppose that ε < kxn+1 −
√
qk2 and thus kxn+1 − qk ∈ [ ε, K]. We now suppose that
 n √ √ o
(∗) n ≥ φ min 14 , Θ48KK ( ε) ̟(ΘK ( ε)/16K)
2 , 3K .
√
Then by (11) we have for δ0 := ΘK ( ε)/16K

kyn − xn+1 k ≤ δ0 and H ∗ [Ayn , Axn+1 , δ0 ],

where the second property follows from the fact that ̟ is a modulus of uniform
continuity for A and in addition kyn −xn+1 k ≤ ̟(δ0 ). Now, since un ∈ (I −A)yn
we have un = yn − λn for some λn ∈ Ayn , and similarly vn = xn − σn for some
σn ∈ Axn . But since H ∗ [Ayn , Axn+1 , δ0 ] there must also be some σ̄n+1 ∈ Axn+1
with kλn − σ̄n+1 k ≤ δ0 , and thus setting v̄n+1 := xn+1 − σ̄n+1 ∈ (I − A)xn+1
we have

kun −v̄n+1 k = kyn −λn −xn+1 + σ̄n+1 k ≤ kyn −xn+1 k+kλn − σ̄n+1 k ≤ 2δ0 . (12)

Now, using Lemma 2.1 on xn+1 − q = x + y for x := (1 − αn )(xn − q) and

y := αn (un − q) we see that for any j ∈ J(xn+1 − q) we have
2
kxn+1 − qk
2
≤ (1 − αn )2 kxn − qk + 2αn hun − q, ji
2 2
= (1 − 2αn )kxn − qk + α2n kxn − qk + 2αn hun − q, ji
2
(13)
< (1 − 2αn )kxn − qk + αn (αn K 2 + 2hun − q, ji)
2
= (1 − 2αn )kxn − qk + αn (αn K 2 + 2hv̄n+1 − q, ji + 2hun − v̄n+1 , ji)
≤ (1 − 2αn )kxn − qk2 + αn (αn K 2 + 2hv̄n+1 − q, ji + 4δ0 K)

where for the last step we use (12) to establish

hun − v̄n+1 , ji ≤ kun − v̄n+1 k · kxn+1 − qk ≤ 2δ0 · K.

√
Now, since kxn+1 − qk ∈ [ ε, K], by Lemma 2.6 there is some j ∈ J(xn+1 − q)
such that √
2
hv̄n+1 − q, ji ≤ kxn+1 − qk − ΘK ( ε)
and substituting this into (13) we obtain
2 2 √
(1 − 2αn )kxn+1 − qk ≤ (1 − 2αn )kxn − qk + αn (αn K 2 − 2ΘK ( ε) + 4δ0 K).

Dividing both sides by (1 − 2αn ) > 0 we get

√
   
2 2 2αn αn K
kxn+1 − qk ≤ kxn − qk − ΘK ( ε) + (αn K + 4δ0 )
1 − 2αn 1 − 2αn
√
 
2 αn K
≤ kxn − qk − 2αn ΘK ( ε) + (αn K + 4δ0 )
1 − 2αn

17
and therefore
2 2 √
kxn+1 − qk ≤ kxn − qk − αn (2ΘK ( ε) − δ1 ) (14)

for  
K
δ1 := (αn K + 4δ0 ).
1 − 2αn
(∗) also implies that (using that αn ≤ 1/4 implies 1/(1 − 2αn ) ≤ 2)
√  √ 
√
 
ΘK ( ε) ΘK ( ε
δ1 ≤ 2K 2
K + 4 < ΘK ( ε)
18K 16K

and thus by (14), under the assumption that ε ≤ kxn+1 − qk2 we have shown
2 2
kxn+1 − qk ≤ kxn − qk − αn · ϕ(ε)
√
for ϕ(ε) := ΘK ( ε) and for all n ≥ N (ε).
2
We can now apply Lemma 3.4 for θn := kxn − qk and (αn ). Conditions (i) and
2
(ii) are satisfied for K and r respectively, and we have established condition (iii)
for ϕ and N as defined above. The stated rate of convergence is then obtained
directly from the rate of convergence for kxn − qk2 given by the lemma, which
as before is converted to one for kxn − qk by substituting ε2 for ε.

7 An Ishikawa scheme for uniformly smooth spaces

Our final application concerns another Ishikawa scheme, but in contrast to the
previous section, uniform continuity of A is now exchanged for uniform smooth-
ness of the underlying space. This results in a somewhat different approach for
establishing strong convergence, but is nevertheless still subsumed under our
general framework. Convergence results pertaining to Ishikawa schemes based
on accretive operators can be found in several places in the literature. The
quantitative result presented here is based on an extension of [6, Theorem 4.2]
due to Lin [18, Theorem 2.1], the latter involving an Ishikawa scheme based on
two accretive operators. We first establish a quantitative version of the notion
of uniform smoothness.
Definition 7.1. A Banach space X is uniformly smooth if for all ε > 0 there
exists some δ > 0 such that

(∗) ∀x, y ∈ X(kxk = 1 ∧ kyk ≤ δ → kx + yk + kx − yk ≤ 2 + εkyk).

Any function τ : (0, ∞) → (0, ∞) such that δ = τ (ε) satisfies (∗) is called a
modulus of uniform smoothness for X.
It is well known that in uniformly smooth spaces, the normalized duality
mapping J is single-valued and uniformly continuous. A quantitative formula-
tion of this fact follows directly from Proposition 2.5 of [14].

18
Lemma 7.2. Let X be uniformly smooth with modulus τ . Define ωτ : (0, ∞) ×
(0, ∞) → (0, ∞) by

ε2  ε
ωτ (d, ε) := ·τ , ε ∈ [0, 2), d ≥ 1
12d 2d
with ωτ (d, ε) := ωτ (1, ε) for d < 1 and ωτ (d, ε) := ωτ (d, 2) for ε > 0. Then the
single-valued duality map J : X → X ∗ is norm-to-norm uniformly continuous
on bounded subsets with modulus ωτ , that is, for all d, ε > 0 and x, y ∈ X with
kxk, kyk ≤ d we have

kx − yk ≤ ωτ (d, ε) → kJx − Jyk ≤ ε.

Theorem 7.3 (Theorem 2.1 of [18] (cf. Remark 2.2)). Let X be uniformly
smooth, and A1 , A2 : D → 2D be two uniformly φ-accretive operators for D a
nonempty, closed and convex subset of X, such that the ranges of (I − A1 ) and
(I − P
A2 ) are bounded. Let (αn ), (βn ) be sequences in [0, 1) such that αn , βn → 0
∞
and i=0 αi = ∞. For any f, x0 ∈ D let (xn ) be generated via the Ishikawa
scheme
xn+1 = (1 − αn )xn + αn (f + un ), un ∈ (I − A1 )yn
yn = (1 − βn )xn + βn (f + vn ),vn ∈ (I − A2 )xn .
(
f ∈ A1 q
Then whenever the system of operator equations has some solution
f ∈ A2 q
q ∈ D, then (xn ) converges strongly to q.
We now present a quantitative analysis of the above result, where for sim-
plicity we set f = 0.
Theorem 7.4. Let X be uniformly smooth with modulus τ , and A1 , A2 : D →
2X with 0 ∈ Ai q for i = 1, 2 be uniformly accretive at zero, with Θ a modulus
of uniform accretivity for A1 . Let (αn ), (βn ) be sequences
P∞ in [0, 1) such that
αn , βn → 0 with joint rate of convergence φ and i=0 αi = ∞ with rate of
divergence r. Suppose that (xn ), (yn ), (un ) and (vn ) are sequences satisfying
xn , yn ∈ D(A) and

xn+1 = (1 − αn )xn + αn un , un ∈ (I − A1 )yn

yn = (1 − βn )xn + βn vn , vn ∈ (I − A2 )xn .

Finally, suppose that K0 , K1 ∈ (0, ∞) satisfy kwk < K0 for all w ∈ R(I − Ai )
for i = 1, 2 and kx0 − qk < K1 . Then kxn − qk → 0 as n → ∞ with rate of
convergence

ΦΘ,τ,φ,r,K0,K1 (ε)
K2
     
1 ε 3ΘK (ε/2) ΘK (ε/2)
:= r φ min , , ωτ K, , +1
6K 2 32K 16K ΘK (ε/2)
for K := 2K0 + K1 and ωτ as defined in Lemma 7.2.

19
Proof. To begin with, we claim that kxn − qk, kyn − qk, kun − qk, kvn − qk <
K := 2K0 + K1 for all n ∈ N and moreover kyn − xn+1 k ≤ 3(αn + βn )K < 6K,
and therefore kyn − xn+1 k ≤ δ for any n ≥ φ(δ̃) with δ̃ ≤ δ/6K. All of this is
established entirely analogously to the beginning of the proof of Theorem 6.6,
which uses just the Ishikawa equations together with basic properties of normed
spaces: Note our generalisation of the Ishikawa scheme to two maps is dealt
with by the assumption that K0 is a joint bound for the ranges of (I − A1 ) and
(I − A2 ).
Let us now define jn := J(xn − q) and jn′ := J(yn − q) for each n ∈ N. By
an application of Lemma 2.1 we have
2 2
kxn+1 − qk ≤ (1 − αn )2 kxn − qk + 2αn hun − q, jn+1 i
2
≤ (1 − αn )2 kxn − qk + 2αn hun − q, jn′ i + 2αn hun − q, jn+1 − jn′ i
2
≤ (1 − αn )2 kxn − qk + 2αn hun − q, jn′ i + 2αn Kcn
(15)
for cn := kjn+1 − jn′ k, where for the last step we use hun − q, jn+1 − jn′ i ≤
kun − qk · kjn+1 − jn′ k ≤ Kcn . An analogous calculation yields
2 2
kyn − qk ≤ (1 − βn )2 kxn − qk + 2βn hvn − q, jn′ i
2
(16)
≤ (1 − βn )2 kxn − qk + 2βn hvn − q, jn i + 2βn Kdn
for dn := kjn′ − jn k. Now, by accretivity of A2 at zero we have hvn − q, jn i ≤
2
kxn − qk and substituting this into (16) we get
2 2
kyn − qk ≤ (1 + βn2 )kxn − qk + 2βn Kdn (17)
For the remainder of the proof we fix some ε > 0 and suppose that ε < kxn+1 −
2
qk . We now suppose that
 n√ √  √  o
ε ΘK ( ε/2)
(∗) n ≥ N (ε) := φ min 12K , 64K 2 , ωτ K, ΘK16K
( ε/2)
/6K .
√
Then, in particular, kyn − xn+1 k ≤ ε/2. This then implies that
√ √
ε < kxn+1 − qk ≤ kxn+1 − yn k + kyn − qk ≤ ε/2 + kyn − qk
√
and so kyn − qk ∈ [ ε/2, K]. By Lemma 2.6 we then have hun − q, jn′ i ≤
2 √
kyn − qk − ΘK ( ε/2), and thus using (17):
2 √
hun − q, jn′ i ≤ (1 + βn2 )kxn − qk + 2βn Kdn − ΘK ( ε/2).
Finally, substituting this into (15) we obtain
2
kxn+1 − qk
2 2 √
≤ (1 + α2n )kxn − qk + 2αn βn2 kxn − qk + 4αn βn Kdn − 2αn ΘK ( ε/2) + 2αn Kcn
√
< kxn − qk2 + α2n K 2 + 2αn βn2 K 2 + 4αn βn Kdn − 2αn ΘK ( ε/2) + 2αn Kcn
2
= kxn − qk − αn · δ
(18)

20
for √
δ := 2ΘK ( ε/2) − (αn K 2 + 2βn2 K 2 + 4βn Kdn + 2Kcn ).
√
Define δ0 := ΘK ( ε/2)/8. (∗) implies αn , βn ≤ δ0 /8K 2 and so in turn αn K 2 <
δ0 , 2βn2 K 2 < δ0 (using βn < 1), and 4βn Kdn < δ0 . For the latter note that

dn = kjn′ − jn k ≤ kjn k + kjn′ k = kxn − qk + kyn − qk < 2K.

Finally, let ωτ be defined as in Lemma 7.2. Then again by (∗) we have k(xn+1 −
q) − (yn − q)k = kxn+1 − yn k ≤ ωτ (K, δ0 /2K) and thus by Lemma 7.2 it follows
that cn = kjn+1 − jn′ k ≤ δ0 /2K and thus 2Kcn ≤ δ0 .
2
√ together we conclude that if ε < kxn+1 − qk and n ≥ N (ε)
Putting all this
we have δ ≥ ΘK ( ε/2) and thus by (18):

kxn+1 − qk2 ≤ kxn − qk2 − αn · ϕ(ε)

√
for ϕ(ε) := ΘK ( ε/2).
We can now apply Lemma 3.4 as in the proof of Theorem 6.6 for θn := kxn −qk2
on parameters K 2 , r, ϕ and N to obtain the stated rate of convergence.
Remark 7.5. The precise statement of Theorem 7.4 is consistent with Remark
2.2 of [18], in that we only require a modulus of uniform accretivity for one of
the operators (though we require both to be accretive).

8 Acknowledgement
This work has been supported by the German Science Foundation DFG (Project
KO 1737/6-1).

References
[1] Y. Alber, S. Reich, and D. Shoikhet. Iterative approximations of null points
of uniformly accretive operators with estimates of the convergence rate.
Communications in Applied Analysis, 6(1):89–104, 2002.

[2] V. Barbu. Nonlinear semigroups and differential equations in Banach spaces,

Noordhoff International Publishing, Leyden, The Netherlands, 1976.
[3] V. Barbu. Nonlinear differential equations of monotone types in Banach
spaces. Springer Monographs in Mathematics, x+272pp., Springer-Verlag.
[4] H.H. Bauschke, P.L. Combettes, Convex Analysis and Monotone Opera-
tor Theory in Hilbert Spaces, Springer, New York-Dordrecht-Heidelberg-
London, 2010.
[5] F. E. Browder, Nonlinear accretive operators in Banach spaces. Bulletin of
the American Mathematical Society, 73:470–476, 1967.

21
[6] S.S. Chang. On Chidume’s open questions and approximate solution of
multi-valued strongly accretive mapping equations in Banach spaces. Jour-
nal of Mathematical Analysis and Applications, 216:94–111, 1994.
[7] J. Garcı́a-Falset. The asymptotic behavior of the solutions of the Cauchy
problem generated by φ-accretive operators. Journal of Mathematical Anal-
ysis and Applications, 310:594–608, 2005.
[8] T. Kato. Nonlinear semigroups and evolution equations. J. Math. Soc.
Japan, 19: 508–520, 1967.
[9] U. Kohlenbach. A quantitative version of a theorem due to Borwein-Reich-
Shafrir. Numer. Funct. Anal. and Optimiz., 22:641–656, 2001.
[10] U. Kohlenbach. Applied Proof Theory: Proof Interpretations and their Use
in Mathematics. Springer Monographs in Mathematics. xx+536pp., Springer
Heidelberg-Berlin, 2008.
[11] U. Kohlenbach. Proof-theoretic Methods in Nonlinear Analysis. In: Proc.
ICM 2018, B. Sirakov, P. Ney de Souza, M. Viana (eds.), Vol. 2, pp. 61-82.
World Scientific 2019.
[12] U. Kohlenbach. Quantitative results on the Proximal Point Algorithm in
uniformly convex Banach spaces. Preprint 2019, submitted.
[13] U. Kohlenbach and A. Koutsoukou-Argyraki. Rates of convergence and
metastability for abstract Cauchy problems generated by accretive opera-
tors. Journal of Mathematical Analysis and Applications, 423:1089–1112,
2015.
[14] U. Kohlenbach and L. Leuştean. On the computational content of con-
vergence proofs via Banach limits. Philosophical Transactions of the Royal
Society A, 370:3449–3463, 2012.
[15] D. Körnlein and U. Kohlenbach. Effective rates of convergence for Lips-
chitzian pseudocontractive mappings in general Banach spaces. Nonlinear
Analysis, 74:5253–5267, 2011.
[16] Y. Komura. Nonlinear semi-groups in Hilbert space. J. Math. Soc. Japan,
19: 493–507, 1967.
[17] L. Leuştean, A. Nicolae, A. Sipoş. An abstract proximal point algorithm.
Journal of Global Optimization, 72:553–577, 2018.
[18] W. Lin. The iterative approximation of solutions to system of multi-valued
nonlinear operator equations. Journal of Applied Analysis, 10(2):303–309,
2004.
[19] C. Moore and B. V. C. Nnoli. Iterative solution of nonlinear equations
involving set-valued uniformly accretive operators. Computers and Mathe-
matics with Applications, 42(1–2):131–140, 2001.

22
[20] B. Martinet. Régularisation d’inéquations variationnelles par approxima-
tions successives. Rev. Française Informat. Recherche Opérationnelle, 4:154–
158, 1970.
[21] E. Neumann. Computational problems in metric fixed point theory and
their Weihrauch degrees. Logical Methods in Computer Science, 11, 44pp.,
2015.
[22] R. T. Rockafellar. Monotone operators and the proximal point algorithm.
SIAM J. Control Optim., 14:877–898, 1976.

23