The methodology named LIFE (Linear-in-Flux-Expressions) was developed with the purpose of simulating and analyzing large metabolic systems. With LIFE, the number of model parameters is reduced by accounting for correlations among the parameters of the system. Perturbation analysis on LIFE systems results in less overall variability of the system, leading to results that more closely resemble empirical data. These systems can be associated to graphs, and characteristics of the graph give insight into the dynamics of the system.
This work addresses two main problems: 1. for fixed metabolite levels, find all fluxes for which the metabolite levels are an equilibrium, and 2. for fixed fluxes, find all metabolite levels which are equilibria for the system. We characterize the set of solutions for both problems, and show general results relating stability of systems to the structure of the associated graph. We show that there is a structure of the graph necessary for stable dynamics. Along with these general results, we show how stability analysis from the fields of network flows, compartmental systems, control theory and Markov chains apply to LIFE systems.
Citation: Nathaniel J. Merrill, Zheming An, Sean T. McQuade, Federica Garin, Karim Azer, Ruth E. Abrams, Benedetto Piccoli. Stability of metabolic networks via Linear-in-Flux-Expressions[J]. Networks and Heterogeneous Media, 2019, 14(1): 101-130. doi: 10.3934/nhm.2019006
[1] | Miao Fu, Yuqin Zhang . Results on monochromatic vertex disconnection of graphs. AIMS Mathematics, 2023, 8(6): 13219-13240. doi: 10.3934/math.2023668 |
[2] | Fan Wu, Xinhui An, Baoyindureng Wu . Sombor indices of cacti. AIMS Mathematics, 2023, 8(1): 1550-1565. doi: 10.3934/math.2023078 |
[3] | Yinkui Li, Jiaqing Wu, Xiaoxiao Qin, Liqun Wei . Characterization of $ Q $ graph by the burning number. AIMS Mathematics, 2024, 9(2): 4281-4293. doi: 10.3934/math.2024211 |
[4] | A. El-Mesady, Y. S. Hamed, M. S. Mohamed, H. Shabana . Partially balanced network designs and graph codes generation. AIMS Mathematics, 2022, 7(2): 2393-2412. doi: 10.3934/math.2022135 |
[5] | Yixin Zhang, Yanbo Zhang, Hexuan Zhi . A proof of a conjecture on matching-path connected size Ramsey number. AIMS Mathematics, 2023, 8(4): 8027-8033. doi: 10.3934/math.2023406 |
[6] | Haicheng Ma, Xiaojie You, Shuli Li . The singularity of two kinds of tricyclic graphs. AIMS Mathematics, 2023, 8(4): 8949-8963. doi: 10.3934/math.2023448 |
[7] | Yuan Zhang, Haiying Wang . Some new results on sum index and difference index. AIMS Mathematics, 2023, 8(11): 26444-26458. doi: 10.3934/math.20231350 |
[8] | Syafrizal Sy, Rinovia Simanjuntak, Tamaro Nadeak, Kiki Ariyanti Sugeng, Tulus Tulus . Distance antimagic labeling of circulant graphs. AIMS Mathematics, 2024, 9(8): 21177-21188. doi: 10.3934/math.20241028 |
[9] | Muhammad Kamran Jamil, Muhammad Imran, Aisha Javed, Roslan Hasni . On the first general Zagreb eccentricity index. AIMS Mathematics, 2021, 6(1): 532-542. doi: 10.3934/math.2021032 |
[10] | Sakander Hayat, Bagus Imanda, Asad Khan, Mohammed J. F. Alenazi . Three infinite families of Hamilton-connected convex polytopes and their detour index. AIMS Mathematics, 2025, 10(5): 12343-12387. doi: 10.3934/math.2025559 |
The methodology named LIFE (Linear-in-Flux-Expressions) was developed with the purpose of simulating and analyzing large metabolic systems. With LIFE, the number of model parameters is reduced by accounting for correlations among the parameters of the system. Perturbation analysis on LIFE systems results in less overall variability of the system, leading to results that more closely resemble empirical data. These systems can be associated to graphs, and characteristics of the graph give insight into the dynamics of the system.
This work addresses two main problems: 1. for fixed metabolite levels, find all fluxes for which the metabolite levels are an equilibrium, and 2. for fixed fluxes, find all metabolite levels which are equilibria for the system. We characterize the set of solutions for both problems, and show general results relating stability of systems to the structure of the associated graph. We show that there is a structure of the graph necessary for stable dynamics. Along with these general results, we show how stability analysis from the fields of network flows, compartmental systems, control theory and Markov chains apply to LIFE systems.
A self-mapping $ \mathcal{F} $ on a convex, closed, and bounded subset $ K $ of a Banach space $ U $ is known as nonexpansive if $ \lVert{\mathcal{F}u -\mathcal{F}v}\rVert $ $ \leq $ $ \lVert{u-v}\rVert $, $ u, v\in U $ and need not essentially possess a fixed point. It is widely known that a point $ u \in U $ is a fixed point or an invariant point if $ \mathcal{F}u = u $. However, some researchers ensured the survival of a fixed point of nonexpansive mapping in Banach spaces utilizing suitable geometric postulates. Numerous mathematicians have extended and generalized these conclusions to consider several nonlinear mappings. One such special class of mapping is Suzuki generalized nonexpansive mapping (SGNM). Many extensions, improvements and generalizations of nonexpansive mappings are given by eminent researchers (see [8,9,10,13,15,17,19,21,22,25], and so on). On the other hand, Krasnosel'skii [16] investigated a novel iteration of approximating fixed points of nonexpansive mapping. A sequence $ \{u_i\} $ utilizing the Krasnosel'skii iteration is defined as: $ u_1 = u, u_{i+1} = (1-\alpha)u_i + \alpha \mathcal{F}u_i $, where $ \alpha \in(0, 1) $ is a real constant. This iteration is one of the iterative methods which is the extension of the celebrated Picard iteration [24], $ u_{i+1} = \mathcal{F}u_i $. The convergence rate of the Picard iteration [24] is better than the Krasnosel'skii iteration although the Picard iterative scheme is not essentially convergent for nonexpansive self-mappings. It is interesting to see that the fixed point of a self-mapping $ \mathcal{F} $ is also a fixed point of the iteration $ \mathcal{F}^n $ $ (n\in \mathbb{N}) $, of the self-mapping $ \mathcal{F} $ but the reverse implication is not feasible. Recently several authors presented extended and generalized results for better approximation of fixed points (see [1,3,11,23,26,27]).
We present convergence and common fixed point conclusions for the associated $ \alpha $-Krasnosel'skii mappings satisfying condition (E) in the current work. Also, we support these with nontrivial illustrative examples to demonstrate that our conclusions improve, generalize and extend comparable conclusions of the literature.
We symbolize $ F(\mathcal{F}) $, to be the collection of fixed points of a self-mapping $ \mathcal{F} $, that is, $ F(\mathcal{F}) $ = $ \{u\in U : \mathcal{F}u = u\} $. We begin with the discussion of convex Banach spaces, $ \alpha $-Krasnosel'skii mappings and the condition (E) (see [12,18,20,23]).
Definition 2.1. [14] A Banach space $ U $ is uniformly convex if, for $ \epsilon \in (0, 2]\ $ $ \exists $ $ \delta > 0 $ satisfying, $ \lVert\frac{u + v}{2}\rVert $ $ \leq 1-\delta $ so that $ \lVert u-v\rVert > \epsilon $ and $ \lVert u \rVert = \lVert v\rVert = 1 $, $ u, v \in U $.
Definition 2.2. [14] A Banach space $ U $ is strictly convex if, $ \lVert \frac{u+v}{2}\rVert < 1 $ so that $ u \neq v, \lVert u\rVert = \lVert v\rVert = 1 $, $ u, v \in U $.
Theorem 2.1. [5] Suppose $ U $ is a uniformly convex Banach space. Then $ \exists $ a $ \gamma > {0} $, satisfying $ \lVert{\frac{1}{2}(\, u+v)\, }\rVert $$ \leq[\, 1-\gamma\frac{\epsilon}{\delta}]\, \delta $ for every $ \epsilon, \; {\delta > 0} $ so that $ \lVert{u-v}\rVert\geq{\epsilon} $, $ \; \lVert{u}\rVert\leq{\delta} $ and $ \lVert{v}\rVert\leq\delta $, for $ u, v \in U $.
Theorem 2.2. [14] The subsequent postulates are equivalent in a Banach space $ U $:
(i) $ U $ is strictly convex.
(ii) $ u = 0 $ or $ v = 0 $ or $ v = cu $ for $ c > {0} $, whenever $ \lVert { u + v }\rVert $ = $ \lVert{u}\rVert + \lVert{v}\rVert, u, v \in U $.
Definition 2.3. Suppose $ \mathcal{F} $ is a self-mapping on a non-void subset $ V $ of a Banach space $ U $.
(i) Suppose for $ u \in U $, $ \exists $ $ v \in V $ so that for all $ w \in V $, $ \lVert{v-u}\rVert $ $ \leq $$ \lVert{w-u}\rVert $. Then $ v $ is a metric projection [6] of $ U $ onto $ V $, and is symbolized by $ P_V(.) $. The mapping $ P_V(u) : U \rightarrow V $ is the metric projection if $ P_V(x) $ exists and is determined uniquely for each $ x\in U $.
(ii) $ \mathcal{F} $ satisfies condition $ (E_{\mu}) $ [23] on $ V $ if $ \exists\; $ $ \; \mu\geq{1} $, satisfying $ \lVert{u-\mathcal{F}v}\rVert \leq \mu\lVert{u-\mathcal{F}u}\rVert+\lVert{u-v}\rVert, \; \; u, v\in V $. Moreover, $ \mathcal{F} $ satisfies condition $ (E) $ on $ V $, if $ \mathcal{F} $ satisfies $ (E_{\mu}) $.
(iii) $ \mathcal{F} $ satisfies condition (E) [23] and $ F(\mathcal{F}) \neq 0 $, then $ \mathcal{F} $ is quasi-nonexpansive.
(iv) $ \mathcal{F} $ is a generalized $ \alpha $-Reich-Suzuki nonexpansive [21] if for an $ \alpha \in [\, 0, 1)\, $, $ \frac{1}{2}\lVert{u-\mathcal{F}u}\rVert \leq \lVert{u-v}\rVert \implies \lVert{\mathcal{F}u-\mathcal{F}v}\rVert \leq $ max $ \{ \alpha\lVert{\mathcal{F}u-u}\rVert + \alpha \lVert{\mathcal{F}v-v}\rVert + (\, 1-2\alpha)\, \lVert{u-v}\rVert, \; \alpha\lVert{\mathcal{F}u-v}\rVert + \alpha\lVert{\mathcal{F}v-u}\rVert + (\, 1-2\alpha)\, \lVert{u-v}\rVert\} $, $ \forall \; u, v \in V $.
(v) A self-mapping $ \mathcal{F}_{\alpha}:V\to V $ is an $ \alpha $-Krasnosel'skii associated with $ \mathcal{F} $ [2] if, $ \mathcal{F}_{\alpha} u = (1-\alpha) u + \alpha \mathcal{F}u $, for $ \alpha \in(0, 1) $, $ u \in V $.
(vi) $ \mathcal{F} $ is asymptotically regular [4] if $ \lim\limits_{n\to\infty} \lVert{\mathcal{F}^nu-\mathcal{F}^{n+1}u}\rVert = 0 $.
(vii) $ \mathcal{F} $ is a generalized contraction of Suzuki type [2], if $ \exists $ $ \beta\in(0, 1) $ and $ \alpha_1, \alpha_2, \alpha_3 \in [\, 0, 1]\, $, where $ \alpha_1 + 2\alpha_2 + 2\alpha_3 = 1 $, satisfying $ \beta\lVert{u-\mathcal{F}u}\rVert \leq \lVert{u-v}\rVert $ implies
$ ‖Fu−Fv‖≤α1‖u−v‖+α2(‖u−Fu‖+‖v−Fv‖)+α3(‖u−Fv‖+‖v−Fu‖) ,u,v∈U. $ |
(viii) $ \mathcal{F} $ is $ \alpha $-nonexpansive [7] if $ \exists $ an $ \alpha < 1 $ satisfying
$ ‖Fu−Fv‖≤α‖Fu−v‖+α‖Fv−u‖+(1−2α)‖u−v‖,u,v∈U. $ |
Theorem 2.3. [5] A continuous mapping on a non-void, convex and compact subset $ V $ of a Banach space $ U $ has a fixed point in $ V $.
Pant et al.[23] derived a proposition that if $ \beta = \frac{1}{2} $, then a generalized contraction of Suzuki type is a generalized $ \alpha $-Reich-Suzuki nonexpansive. Moreover, the reverse implication may not necessarily hold.
Lemma 2.1. [2] Let $ \mathcal{F} $ be a generalized contraction of the Suzuki type on a non-void subset $ V $ of a Banach space $ U $. Let $ \beta\in[\, \frac{1}{2}, 1)\, $, then
$ ‖u−Fv‖≤(2+α1+α2+3α31−α2−α3)‖u−Fu‖+‖u−v‖. $ |
Proposition 2.1. [23] Let $ \mathcal{F} $ be a generalized contraction of the Suzuki type on a non-void subset $ V $ of a Banach space $ U $, then $ \mathcal{F} $ satisfies condition (E).
The converse of this proposition is not true, which can be verified by the following example.
Example 2.1. Suppose $ U = (\mathbb{R}^2, \left\|.\right\|) $ with the Euclidean norm and $ V = [-1, 1]\times[-1, 1] $ be a subset of $ U $. Let $ \mathcal{F}:V\to V $ be defined as
$ F(u1,u2)={(u12,u2),if|u1|≤12(−u1,u2),if|u1|>12. $ |
Case I. Let $ x = (u_1, u_2), y = (v_1, v_2) $ with $ |u_1|\leq\frac{1}{2} $, $ |v_1|\leq\frac{1}{2} $. Then,
$ ‖Fx−Fy‖=‖(u12,u2)−(v12,v2)‖=√(u1−v1)24+(u2−v2)2≤√(u1−v1)2+(u2−v2)2=‖x−y‖, $ |
which implies
$ ‖x−Fy‖≤‖x−Fx‖+‖Fx−Fy‖≤‖x−Fx‖+‖x−y‖. $ |
Case II. If $ |u_1|\leq\frac{1}{2} $, $ |v_1| > \frac{1}{2} $
$ ‖x−Fy‖=√(u1+v1)2+(u2−v2)2‖x−y‖=√(u1−v1)2+(u2−v2)2‖x−Fx‖=|u1|2. $ |
Consider
$ ‖x−Fy‖=√(u1−v1)2+(u2−v2)2+4u1v1≤√(u1−v1)2+(u2−v2)2+4|u1|≤√(u1−v1)2+(u2−v2)2+4|u1|. $ |
Hence,
$ \left\|x-\mathcal{F}y\right\|\leq 8\left\|x-\mathcal{F}x\right\|+\left\|x-y\right\|. $ |
Here $ \mu = 8 $ satisfies the inequality.
Case III. If $ |u_1| > \frac{1}{2} $, $ |v_1|\leq\frac{1}{2} $
$ ‖x−Fy‖=√(u1−v12)2+(u2−v2)2‖x−y‖=√(u1+v1)2+(u2−v2)2‖x−Fx‖=2|u1|. $ |
Consider
$ ‖x−Fy‖=√(u1−v12)2+(u2−v2)2≤√(u1−v1)2+(u2−v2)2≤√(u1−v1)2+(u2−v2)2+|u1|≤√(u1−v1)2+(u2−v2)2+2|u1|. $ |
So,
$ \left\|x-\mathcal{F}y\right\|\leq \left\|x-\mathcal{F}y\right\|+\left\|x-y\right\|. $ |
Case IV. If $ |u_1| > \frac{1}{2} $ and $ |v_1| > \frac{1}{2} $, then
$ ‖x−Fy‖=√(u1+v1)2+(u2−v2)2‖x−y‖=√(u1−v1)2+(u2−v2)2‖x−Fx‖=2|u1|. $ |
Since $ |u_1| > \frac{1}{2} $ and $ |v_1| > \frac{1}{2} $, by simple calculation as above, we attain
$ ‖x−Fy‖≤μ‖x−Fx‖+‖x−y‖. $ |
Thus, $ \mathcal{F} $ satisfies condition (E) for $ \mu = 4 $.
Now, suppose $ x = (\frac{1}{2}, 1) $ and $ y = (1, 1) $, so
$ β‖x−Fx‖=β(12−14)=β4≤‖x−y‖=12. $ |
Clearly, $ \left\|\mathcal{F}x-\mathcal{F}y\right\| = \sqrt{(\frac{5}{4})^2+(1-1)^2} = \frac{5}{4} $.
Consider
$ α1‖x−y‖+α2(‖x−Fx‖+‖y−Fy‖)+α3(‖x−Fy‖+‖y−Fx‖)=α1‖(12,1)−(1,1)‖+α2(‖(12,1)−(14,1)‖+‖(1,1)−(−1,1)‖)+α3(‖(12,1)−(−1,1)‖+‖(1,1)−(14,1)‖)=α12+α24+2α2+3α32+3α34=α12+94(α2+α3)=α12+94(1−α12)(by Definition 2.3 (vii))=α12+98−9α18=98−5α18. $ |
Since $ \alpha_1, \alpha_2, \alpha_3\geq 0 $, therefore
$ ‖Fx−Fy‖>α1‖x−y‖+α2(‖x−Fy‖+‖y−Fy‖)+α3(‖x−Fy‖+‖y−Fx‖), $ |
which is a contradiction.
Thus, $ \mathcal{F} $ is not a generalized contraction of the Suzuki type.
Now, we establish results for a pair of $ \alpha $-Krasnosel'skii mappings using condition (E).
Theorem 3.1. Let $ \mathcal{F}_i $, for $ i\in\{1, 2\} $, be self-mappings on a non-void convex subset $ V $ of a uniformly convex Banach space $ U $ and satisfy condition (E) so that $ F(\mathcal{F}_1\cap \mathcal{F}_2)\neq \phi $. Then the $ \alpha $-Krasnosel'skii mappings $ \mathcal{F}_{i_{\alpha}} $, $ \alpha \in (\, 0, 1)\, $ and $ i\in\{1, 2\} $ are asymptotically regular.
Proof. Let $ v_0 \in V $. Define $ v_{n+1} = \mathcal{F}_{i_{\alpha}} v_n $ for $ i\in\{1, 2\} $ and $ n \in N\cup\{0\} $. Thus,
$ Fiαvn=yn+1=(1−α)vn+αFivnfori∈{1,2}, $ |
and
$ Fiαvn−vn=Fiαvn−Fiαvn−1=α(Fivn−vn)fori∈{1,2}. $ |
It is sufficient to show that $ \lim\limits_{n\to\infty} \lVert{\mathcal{F}_i v_n-v_n}\rVert = 0 $ to prove $ \mathcal{F}_{i_{\alpha}} $ is asymptotically regular.
By definition, for $ u_0 \in F(\mathcal{F}_1\cap \mathcal{F}_2) $, we have
$ ‖u0−Fivn‖≤‖u0−vn‖fori∈{1,2} $ | (3.1) |
and for $ i\in\{1, 2\} $,
$ ‖u0−vn+1‖=‖u0−Fiαvn‖=‖u0−(1−α)vn−αFivn‖≤(1−α)‖u0−vn‖+α‖u0−Fivn‖=(1−α)‖u0−vn‖+α‖u0−vn‖=‖u0−vn‖. $ | (3.2) |
Thus, the sequence $ \{\lVert u_0-v_n\rVert\} $ is bounded by $ s_0 = \lVert u_0-v_0\rVert $. From inequality (3.2), $ v_n \to u_0 $ as $ n \to \infty $, if $ v_{n_0} = u_0 $, for some $ n_0\in \mathbb{N} $. So, assume $ v_n\neq u_0 $, for $ n \in \mathbb{N} $, and
$ wn=u0−vn‖u0−vn‖anden=u0−Fivn‖u0−vn‖,fori∈{1,2}. $ | (3.3) |
If $ \alpha\leq\frac{1}{2} $ and using Eq (3.3), we obtain
$ ‖u0−vn+1‖=‖u0−Fiαvn‖,fori∈{1,2}=‖u0−(1−α)vn−αFivn‖,fori∈{1,2}=‖u0−vn+αvn−αFivn−2αu0+2αu0+αvn−αvn‖,fori∈{1,2}=‖(1−2α)u0−(1−2α)vn+(2αu0−αvn−αFivn)‖,fori∈{1,2}≤(1−2α)‖u0−vn‖+α‖2u0−vn−Fivn‖=2α‖u0−vn‖‖wn+en2‖+(1−2α)‖u0−vn‖. $ | (3.4) |
As the space $ U $ is uniformly convex with $ \lVert w_n\rVert \leq 1 $, $ \lVert e_n \rVert \leq 1 $ and $ \lVert w_n-e_n\rVert = \frac{\lVert v_n-\mathcal{F}_iv_n\rVert}{ \lVert u_0-v_n\rVert} \geq \frac{\lVert v_n-\mathcal{F}_iv_n\rVert}{s_0} = \epsilon $ (say) for $ i\in\{1, 2\} $, we obtain
$ ‖wn+en‖2≤1−δ‖vn−Fivn‖sofori∈{1,2}. $ | (3.5) |
From inequalities (3.4) and (3.5),
$ ‖u0−vn+1‖≤(2α(1−δ‖vn−Fivn‖so)+(1−2α))‖u0−vn‖=(1−2αδ(‖vn−Fivn‖s0) )‖u0−vn‖. $ | (3.6) |
By induction, it follows that
$ ‖u0−vn+1‖≤n∏j=1(1−2αδ(‖vn−Fivn‖s0))s0. $ | (3.7) |
We shall prove that $ \lim\limits_{n\to\infty}\lVert \mathcal{F}_iv_n-v_n\rVert = 0 $ for $ i\in\{1, 2\} $. On the contrary, consider that $\{ \lVert \mathcal{F}_iv_n-v_n\rVert\} $ for $ i\in\{1, 2\} $ is not converging to zero, and we have a subsequence $ \{v_{n_k}\}, $ of $ \{v_n\}, $ satisfying $ \lVert \mathcal{F}_iv_{n_k}-v_{n_k} \rVert $ converges to $ \zeta > 1 $. As $ \delta\in [\, 0, 1]\, $ is increasing and $ \alpha\leq\frac{1}{2} $, $ 1-2\alpha\delta\frac{\lVert v_k-\mathcal{F}_iv_k\rVert}{s_0}\in [\, 0, 1]\, $, $ i\in\{1, 2\} $, for all $ k \in \mathbb{N} $. Since $ \lVert \mathcal{F}_iv_{n_k}-v_{n_k}\rVert \to \zeta $ so, for sufficiently large $ k, \; \; \lVert \mathcal{F}_iv_{n_k}-v_{n_k}\rVert\geq\frac{\zeta}{2} $, from inequality (3.7), we have
$ ‖u0−vnk+1‖≤s0(1−2αδ(ζ2−s0))(nk+1). $ | (3.8) |
Making $ k\to\infty $, it follows that $ v_{n_{k}}\to u_0 $. By inequality (3.1), we get $ \mathcal{F}_{i} v_{n_{k}} \to u_0 $ and $ \lVert v_{n_{k}} -\mathcal{F}_{i} v_{n_{k}} \rVert \to 0 $ as $ k\to \infty $, which is a contradiction. If $ \alpha > \frac{1}{2} $, then $ 1-\alpha < \frac{1}{2} $, because $ \alpha \in (\, 0, 1)\, $. Now, for $ i\in\{1, 2\} $
$ ‖u0−vn+1‖=‖u0−(1−α)vn−αFivn‖=‖u0−vn+αvn−αFivn+(2−2α)u0−(2−2α)u0+Fivn−Fivn+αFivn−αFivn‖=‖(2u0−vn−Fivn)−α(2u0−vn−Fivn)+2α(u0−Fivn)−(u0−Fivn)‖≤(1−α)‖2u0−vn−Fivn‖+(2α−1)‖u0−vn‖≤2(1−2α)‖u0−vn‖‖wn+en‖2+(2α−1)‖u0−vn‖. $ |
By the uniform convexity of $ U $, we attain, for $ i\in\{1, 2\} $,
$ ‖x0−yn+1‖≤(2(1−α)−2(1−α)δ‖yn−Fiyn‖so+(1−2α))‖x0−yn‖. $ | (3.9) |
By induction, we get
$ ‖u0−vn+1‖≤n∏j=1(1−2(1−α)δ(‖vj−Fivj‖s0))s0. $ |
Similarly, it can be easily proved that $ \lVert \mathcal{F}_{i}v_n-v_n\rVert \to 0 $ as $ n \to \infty $, which implies that $ \mathcal{F}_{i_{\alpha}} $ for $ i\in\{1, 2\} $, is asymptotically regular.
Next, we demonstrate by a numerical experiment that a pair of $ \alpha $-Krasnosel'skii mappings are asymptotically regular for fix $ \alpha \in (0, 1) $.
Example 3.1. Assume $ U = (R^2, ||.||) $ with Euclidean norm and $ V = \{ u\in R^2 : \lVert u\rVert \leq1 \} $, to be a convex subset of $ U $. $ \mathcal{F}_i $ for $ i\in \{1, 2\} $ be self-mappings on $ V $, satisfying
$ F1(u1,u2)=(u1,u2)F2(u1,u2)=(u12,0) $ |
Then, clearly both $ \mathcal{F}_1 $ and $ \mathcal{F}_2 $ satisfy the condition $ (E) $ and $ F(\mathcal{F}_1\cap \mathcal{F}_2) = (0, 0) $. Now, we will show that the $ \alpha $-Krasnosel'skii mappings $ \mathcal{F}_{i_{\alpha}} $ for $ \alpha \in (\, 0, 1)\, $ and $ i\in\{1, 2\} $ are asymptotically regular.
Since $ \mathcal{F}_1 $ is the identity map, $ \alpha $- Krasnosel'skii mapping $ \mathcal{F}_{1\alpha} $ is also identity and hence asymptotically regular.
Now, we show $ \mathcal{F}_{2\alpha} $ is asymptotically regular, let $ u = (u_1, u_2) \in V $
$ F2α(u1,u2)=(1−α)(u1,u2)+αF2(u1,u2)=((1−α)u1,(1−α)u2)+α(u12,0)=(u1−αu12,(1−α)u2), $ |
$ F22α(u1,u2)=(1−α)(u1−αu12,(1−α)u2)+αF2(u1−αu12,(1−α)u2)=(x1+α2u12−3αu12,(1−α)2u2)+(αu2−α2u14,0)=(u1−αu1+α2u14,(1−α)2x2). $ |
Continuing in this manner, we get
$ fn2α(u1,u2)=((u1−α2)n,(1−α)nu2). $ |
Since $ (u_1, u_2) \in V $ and $ \alpha \in(0, 1) $, we get that $ \lim\limits_{n\to \infty}(u_1-\frac{\alpha}{2})^n = 0 $ and $ \lim\limits_{n\to \infty}(1-\alpha)^n = 0 $. Now, consider
$ limn→∞‖Fn2α(u1,u2)−Fn+12α(u1,u2)‖=supu∈Mlimn→∞‖(u1−α2)n−(u1−α2)n+1,((1−α)n−(1−α)n+1)x2‖=0. $ |
Hence, $ \mathcal{F}_{2\alpha} $ is also asymptotically regular.
Theorem 3.2. Let $ \mathcal{F}_i $ be quasi-nonexpansive self-mappings on a non-void and closed subset $ V $ of a Banach space $ U $ for $ i\in \{1, 2\} $, and satisfy condition (E) so that $ F(\mathcal{F}_1\cap \mathcal{F}_2) \neq 0 $. Then, $ F(\mathcal{F}_1\cap \mathcal{F}_2) $ is closed in $ V $. Also, if $ U $ is strictly convex, then $ F(\mathcal{F}_1 \cap \mathcal{F}_2) $ is convex. Furthermore, if $ U $ is strictly convex, $ V $ is compact, and $ \mathcal{F} $ is continuous, then for any $ s_0 \in V, \alpha \in (0, 1) $, the $ \alpha $-Krasnosel'skii sequence $ \{\mathcal{F}^{n}_{i_{\alpha}} (s_0)\}, $ converges to $ s\in F(\ \mathcal{F}_1\cap \mathcal{F}_2)\ $.
Proof. (i) We assume $ \{s_n\} \in F(\ \mathcal{F}_1\cap \mathcal{F}_2)\ $ so that $ s_n \to s\in F(\mathcal{F}_1\cap \mathcal{F}_2) $ as $ n \to \infty $. Hence, $ \mathcal{F}_{i}s_n = s_n $ for $ i\in\{1, 2\} $. Next, we show that $ \mathcal{F}_is = s $ for $ i\in \{1, 2\} $. Since $ \mathcal{F}_i $ are quasi-nonexpansive, we get
$ ‖sn−Fis‖≤‖sn−s‖fori∈{1,2}, $ |
that is, $ \mathcal{F}_is = s $ for $ i = 1, 2 $, hence $ F(\mathcal{F}_2\cap \mathcal{F}_2) $ is closed.
(ii) $ V $ is convex since $ U $ is strictly convex. Also fix $ \gamma \in (\ 0, 1)\ $ and $ u, v \in F(\mathcal{F}_1\cap \mathcal{F}_2)\ $ so that $ u\neq v $. Take $ s = \gamma u + (1-\gamma)v \in V $. Since mapping $ \mathcal{F}_i $ satisfy condition (E),
$ ‖u−Fis‖≤‖u−Fiu‖+‖u−s‖=‖u−s‖fori∈{1,2}. $ |
Similarly,
$ ‖v−Fis‖≤‖v−s‖fori∈{1,2}. $ |
Using strict convexity of $ U $, there is a $ \theta \in [\ 0, 1]\ $ so that $ \mathcal{F}_is = \theta u + (1-\theta) v $ for $ i = 1, 2 $
$ (1−θ)‖u−v‖=‖Fiu−Fis‖≤‖u−s‖=(1−γ) ‖u−v‖,fori∈{1,2}, $ | (3.10) |
and
$ θ‖u−v‖=‖Fiv−Fis‖≤‖v−s‖=γ‖u−v‖,fori∈{1,2}. $ | (3.11) |
From inequalities (3.10) and (3.11), we obtain
$ 1−θ≤1−γandθ≤γimplies thatθ=γ. $ |
Hence, $ \mathcal{F}_{i}s = s $ for i = 1, 2, implies $ s\in F(\mathcal{F}_1\cap \mathcal{F}_2)\ $.
(iii) Let us define $ \{ s_n\} $ by $ s_n = \mathcal{F}^{n}_{i_{\alpha}}s_0, s_0 \in V $, where $ \mathcal{F}_{i_{\alpha}}s_0 = (1-\alpha) s_0 + \alpha \mathcal{F}_is_0, \alpha\in (\ 0, 1)\ $. We have a subsequence $ \{s_{n_k}\} $ of $ \{s_n\} $ converging to some $ s\in V $, since $ V $ is compact. Using the Schauder theorem and the continuity of $ \mathcal{F}_i $, we have $ F(\mathcal{F}_1\cap \mathcal{F}_2)\ \neq\phi $. We shall demonstrate that $ s \in F(\mathcal{F}_1\cap \mathcal{F}_2) $. Let $ w_0 \in F(\mathcal{F}_1\cap \mathcal{F}_2) $, consider
$ ‖sn−w0‖=‖Fniαs0−w0‖≤‖Fn−1iαs0−w0|=‖sn−1−w0‖. $ |
Therefore, $ \{ \lVert s_n-w_0\rVert \} $ converges as it is a decreasing sequence that is bounded below by $ 0 $. Moreover, since $ \mathcal{F}_{i_{\alpha}} $ for $ i = 1, 2 $ is continuous, we have
$ ‖w0−s0‖=limk→∞‖snk+1−so‖=limk→∞‖Fiαsnk−s0‖=‖Fiαs−s0‖=‖(1−α)s+αFis−s0‖≤(1−α)‖s−s0‖+α‖Fis−s0‖fori∈{1,2}. $ | (3.12) |
Since $ \alpha > 0 $, we get
$ ‖s−s0‖≤‖Fis−s0‖,fori∈{1,2}. $ | (3.13) |
Since $ \mathcal{F}_i $ are quasi-nonexpansive maps, we get
$ ‖Fis−s0‖≤‖s−s0‖,fori∈{1,2}, $ | (3.14) |
and from inequalities (3.13) and (3.14), we get
$ ‖Fis−s0‖=‖s−s0‖,fori∈{1,2}. $ | (3.15) |
Now, from inequality (3.12), we have
$ ‖s−s0‖≤‖(1−α)s+αFis−s0‖,fori∈{1,2}≤(1−α)‖s−s0‖+α‖Fis−s0‖,fori∈{1,2}=‖s−s0‖, $ |
which implies that
$ ‖(1−α)s+αFis−s0‖=(1−α)‖s−s0‖+α‖Fis−s0‖,fori∈{1,2}. $ |
Since $ U $ is strictly convex, either $ \mathcal{F}_is-s_0 = a(s-s_0) $ for some $ a \gneq 0 $ or $ s = s_0 $. From Eq (15), it follows that $ a = 1 $, then, $ \mathcal{F}_{i}s = s $ for $ i = 1, 2 $ and $ s \in F(\mathcal{F}_1 \cap \mathcal{F}_2) $. Since $ \lim\limits_{n \to \infty} \lVert s_n-s_0\rVert $ exists and $ \{s_{nk}\} $ converges strongly to $ s $. Hence, $ \{s_n\} $ converges strongly to $ s \in F(\mathcal{F}_1 \cap \mathcal{F}_2) $.
The next conclusion for metric projection is slightly more fascinating.
Theorem 3.3. Let $ \mathcal{F}_i $ be quasi-nonexpansive self-mappings on a non-void, closed, and convex subset $ V $ of a uniformly convex Banach space $ U $ for $ i\in\{1, 2\} $, and satisfies condition (E) so that $ F(\mathcal{F}_1 \cap \mathcal{F}_2)\neq \phi $. Let $ P: U \to F(\mathcal{F}_1 \cap \mathcal{F}_2) $ be the metric projection. Then, for every $ u \in U $, the sequence $ \{P\mathcal{F}^{n}_iu\} $ for $ i = \{1, 2\} $, converges to $ s \in F(\mathcal{F}_1 \cap \mathcal{F}_2) $.
Proof. Let $ u \in V $. For $ n, m \in N $
$ ‖PFniu−Fniu‖≤‖PFmiu−Fniu‖,forn≥m,i∈{1,2}. $ | (3.16) |
Since $ u\in F(\mathcal{F}_1\cap \mathcal{F}_2)\ $, $ n \in N $ and $ \mathcal{F}_i $ are quasi-nonexpansive maps, $ \mbox{for}\; i\in \{1, 2\} $ we have
$ ‖PFmiu−Fniu‖=‖PFmiu−FiFn−1iu‖≤‖PFmiu−Fn−1iu‖. $ |
Therefore, for $ n \geq m $, it follows that
$ ‖PFmiu−Fniu‖≤‖PFmiu−Fmiu‖,fori∈{1,2}. $ | (3.17) |
From inequalities (3.16) and (3.17), we have
$ ‖PFniu−Fniu‖≤‖PFmiu−Fmiu‖,fori∈{1,2}, $ |
which implies that $ \lim\limits_{n \to \infty} \lVert P\mathcal{F}^{n}_iu-\mathcal{F}^{n}_iu\rVert $ exists. Taking $ \lim\limits_{n \to \infty} \lVert P\mathcal{F}^{n}_iu-\mathcal{F}^{n}_iu\rVert = l $.
If $ l = 0 $, then we have an integer $ n_0(\ \epsilon)\ $ for $ \epsilon > 0 $, satisfying
$ ‖PFniu−Fniu‖>ϵ4,fori∈{1,2}, $ | (3.18) |
for $ n \geq n_0 $. Therefore, if $ n \geq m \geq n_0 $ and using inequalities (3.17) and (3.18), we have, for $ i \in \{1, 2\} $,
$ ‖PFniu−PFmiu‖≤‖PFniu−PFn0iu‖+‖PFn0iu−Fmiu‖≤‖PFniu−Fniu‖+‖Fniu−PFn0iu‖+‖PFmiu−Fmiu‖+‖Fmiu−PFn0iu‖≤‖PFniu−Fniu‖+‖Fn0iu−PFn0iu‖+‖PFmiu−Fmiu‖+‖Fn0iu−PFn0iu‖≤ϵ4+ϵ4+ϵ4+ϵ4=ϵ. $ |
That is, $ \{P\mathcal{F}^{n}_iu\} $ for $ i = \{1, 2\} $ is a Cauchy sequence in $ F(\mathcal{F}_1\cap \mathcal{F}_2) $. Using the completeness of $ U $ and the closedness of $ F(\mathcal{F}_1\cap \mathcal{F}_2) $ from the above theorem, $ \{P\mathcal{F}^{n}_ix\} $ for $ i = 1, 2, $ converges in $ F(\mathcal{F}_1 \cap \mathcal{F}_2) $. Taking $ l > 0 $, we claim that the sequence $ \{P\mathcal{F}^{n}_iu\} $ for $ i = 1, 2, $ is a Cauchy sequence in $ U $. Also we have, an $ \epsilon _{0} > 0 $ so that, for each $ n_0 \in N $, we have some $ r_0, s_0 \geq n_0 $ satisfying
$ ‖PFr0iu−PFs0iu‖≥ϵ0,fori∈{1,2}. $ |
Now, we choose a $ \theta > 0 $
$ (l+θ)(1−δϵ0l+θ)<θ. $ |
Let $ m_0 $ be as large as possible such that for $ q\geq m_0 $
$ l≤‖PFqiu−Fqiu‖≤l+θ. $ |
For this $ m_0 $, there exist $ q_1, q_2 $ such that $ q_1, q_2 > m_0 $ and
$ ‖PFq1iu−PFq2iu‖≥ϵ0fori∈{1,2}. $ |
Thus, for $ q_0 \geq max \{q_1, q_2\} $, we attain
$ ‖PFq1ix−Fq0ix‖≥‖PFq1ix−Fq1ix‖<l+θ, $ |
and
$ ‖PFq2ix−Fq0ix‖≥‖PFq1ix−Fq1ix‖<l+θfori∈{1,2}. $ |
Now, using the uniform convexity of $ U $, we attain
$ l≤‖PFq0ix−Fq0ix‖≤‖PFq1ix+PFq2ix2−Fq0ix‖,fori∈{1,2}≤( l+θ) (1−δϵ0l+θ)<θ, $ |
a contradiction. Hence for every $ u \in V $, the sequence $ \{P\mathcal{F}^{n}_iu\} $ for $ i = 1, 2, $ converges to some $ s \in F(\mathcal{F}_1 \cap \mathcal{F}_2) $.
We have proved some properties of common fixed points and also showed that if two mappings have common fixed points, then their $ \alpha $-Krasnosel'skii mappings are asymptotically regular. To show the superiority of our results, we have provided an example. Further, we have proved that the $ \alpha $-Krasnosel'skii sequence and its projection converge to a common fixed whose collection is closed.
Researchers would like to thank the Deanship of Scientific Research, Qassim University for funding publication of this project.
The authors declare no conflict of interest.
[1] |
Efficient generation and selection of virtual populations in quantitative systems pharmacology models. Systems Pharmacology (2016) 5: 140-146. ![]() |
[2] | (1993) Algebraic Graph Theory. Cambridge university press. |
[3] |
A. Bressan and B. Piccoli, Introduction to Mathematical Control Theory, AIMS series on applied mathematics, Philadelphia, 2007. |
[4] |
F. Bullo, Lectures on Network Systems, Edition 1, 2018, (revision 1.0 -May 1, 2018), 300 pages and 157 exercises, CreateSpace, ISBN 978-1-986425-64-3. |
[5] |
J. S. Caughman and J. J. P. Veerman, Kernels of directed graph laplacians, The Electronic Journal of Combinatorics, 13 (2006), Research Paper 39, 8 pp. |
[6] |
E. Çinlar, Introduction to stochastic processes, Prentice-Hall, Englewood Cliffs, N. J., 1975. |
[7] |
P. De Leenheer, The Zero Deficiency Theorem, Notes for the Biomath Seminar I - MAP6487, Fall 09, Oregon State University (2009), Available on-line: http://math.oregonstate.edu/ deleenhp/teaching/fall09/MAP6487/notes-zero-def.pdf |
[8] |
Dynamics of open chemical systems and the algebraic structure of the underlying reaction network. Chemical Engineering Science (1974) 29: 775-787. ![]() |
[9] |
Maximal flow through a network. Canadian Journal of Mathematics (1956) 8: 399-404. ![]() |
[10] |
D. Gale, H. Kuhn and A. W. Tucker, Linear Programming and the Theory of Games -Chapter XII, in Koopmans, Activity Analysis of Production and Allocation, 1951, 317-335 |
[11] |
J. Gunawardena, A linear framework for time-scale separation in nonlinear biochemical systems, PloS One, 7 (2012), e36321. |
[12] |
G. T. Heineman, G. Pollice and S. Selkow, Chapter 8: Network flow algorithms, in Algorithms in a Nutshell, Oreilly Media, 2008, 226-250. |
[13] |
Laplacian dynamics on general graphs. Bulletin of Mathematical Biology (2013) 75: 2118-2149. ![]() |
[14] |
Qualitative theory of compartmental systems. SIAM Review (1993) 35: 43-79. ![]() |
[15] |
Timescale analysis of rule based biochemical reaction networks. Biotechnology Progress (2012) 28: 33-44. ![]() |
[16] |
Asymptotic behavior of nonlinear compartmental systems: Nonoscillation and stability. IEEE Transactions on Circuits and Systems (1978) 25: 372-378. ![]() |
[17] |
An O(|V|3) algorithm for finding maximum flows in networks. Information Processing Letters (1978) 7: 277-278. ![]() |
[18] |
S. T. McQuade, Z. An, N. J. Merrill, R. E. Abrams, K. Azer and B. Piccoli, Equilibria for large metabolic systems and the LIFE approach, In 2018 Annual American Control Conference (ACC). IEEE, (2018) pp. 2005-2010. |
[19] |
S. T. McQuade, R. E. Abrams, J. S. Barrett, B. Piccoli and Karim Azer, Linear-in-flux-expressions methodology: Toward a robust mathematical framework for quantitative systems pharmacology simulators, Gene Regulation and Systems Biology, 11 (2017). |
[20] |
C. D. Meyer, Matrix Analysis and Applied Linear Albegra, Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA, 2000. |
[21] | (2006) Systems Biology. Cambridge University Press. |
[22] | Using quantitative systems pharmacology for novel drug discovery. Expert Opinion on Drug Discovery (2015) 10: 1315-1331. |
[23] |
Theory for the systemic definition of metabolic pathways and their use in interpreting metabolic function from a pathway-oriented perspective. Journal of Theoretical Biology (2000) 203: 229-248. ![]() |
[24] |
A network dynamics approach to chemical reaction networks. International Journal of Control (2016) 89: 731-745. ![]() |
A directed graph
A directed graph
A directed graph where vertices
A directed cycle graph
Reverse Cholesterol Transport Network from [19]. This network contains 6 vertices which represent metabolites, 10 edges which represent fluxes and 2 virtual vertices
The trajectories of the values of metabolites over 25 hours