Avoid common mistakes on your manuscript.
1 Correction to: Journal of Optimization Theory and Applications (2021) 188:94–112 https://doi.org/10.1007/s10957-020-01779-7
2 Introduction
In our paper [6], there is a gap for the statement of Proposition 3.2 and Remark 3.2. In addition, on line 14 of page 110, the set \([x'\in C,\ \mathrm{supp}(x')=J,\ x'\rightarrow \overline{x}, \ x'\ne \overline{x}]\) may be empty. In this erratum, we provide the correct statements for Proposition 3.2 and Remark 3.2 and update the proof of Proposition 3.2.
3 Corrected Result
First, we give the correct statement of [6, Proposition 3.2 & Remark 3.2].
Proposition 2.1
([6, Proposition 3.2] corrected) (i) When \(h(\cdot )=\nu \Vert \cdot \Vert _0\) for a constant \(\nu >0\), if \(\psi \!:\mathbb {R}^p\rightarrow ]-\infty ,+\infty ]\) is a proper closed piecewise linear regular function, then for any \(\overline{x}\in \mathrm{dom}\,\psi \),
if \(\psi \) is an indicator function of some closed convex set \(C\subseteq \mathbb {R}^p\), then for any \(\overline{x}\in C\) such that \(\{x\in \mathbb {R}^p\,|\, x_i=0\ \mathrm{for}\ i\notin \mathrm{supp}(\overline{x})\}\cap \mathrm{ri}(C)\ne \emptyset \), it holds that
(ii) When \(h=\delta _{\varOmega }\), the indicator function of \(\varOmega :=\{x\in \mathbb {R}^p\!:\,\Vert x\Vert _0\le \kappa \}\) for an integer \(\kappa >0\), the results of part (i) hold at any \(\overline{x}\in \mathrm{dom}\psi \) with \(\Vert \overline{x}\Vert _0=\kappa \), and at any \(\overline{x}\in \mathrm{dom}\psi \) with \(\Vert \overline{x}\Vert _0<\kappa \) it holds that \(\partial (\psi +h)(\overline{x}) \subseteq \partial \psi (x)+\partial h(\overline{x})\).
Remark 2.1
([6, Remark 3.2] corrected) When \(\psi \) is a locally Lipschitz regular function, the first part of Proposition 2.1 still holds by [5, Theorem 9.13(b) & Corollary 10.9], and now equality (2) is also given in [2, Proposition 1.107(iii)] and [3, Prop. 1.29].
In what follows, we provide the proof of Proposition 2.1.
The proof of Proposition 2.1: First, we consider that \(\psi \) is a proper closed piecewise linear regular function. Fix any \(\overline{x}\in \mathrm{dom}\,\psi \). Notice that \(\mathrm{epi}\psi \) and \(\mathrm{epi}h\) are the union of finitely many polyhedral sets. By combining [4, Proposition 1] and [1, Section 3.2], it then follows that
From the first inclusion, \(\partial (\psi +h)(\overline{x})\supseteq \widehat{\partial }(\psi +h)(\overline{x})\supseteq \widehat{\partial }\psi (\overline{x})+\widehat{\partial } h(\overline{x})\), and the regularity of \(\psi \) and h, we obtain the equalities in (1). When \(\partial \psi (\overline{x})=\emptyset \), obviously, the equalities in (2) hold. So, it suffices to consider the case where \(\partial \psi (\overline{x})\ne \emptyset \). From the second inclusion in (3), it follows that
where \(K^{\circ }\) denotes the negative polar of a cone K. By combining this inclusion with [5, Exercise 8.23] and the second equality of (1), for any \(w\in \mathbb {R}^p\) we have
where \(\widehat{d}h(\overline{x})\) and \(dh(\overline{x})\), respectively, denote the regular subderivative and the subderivative of \(\psi +h\) at \(\overline{x}\), the equality is due to the regularity of \(\psi \) and h, and the second inequality is using [5, Corollary 10.9]. By [5, Corollary 8.19], this shows that \(\psi +h\) is regular, and hence \( \partial ^{\infty }(\psi +h)(\overline{x}) =[\widehat{\partial }(\psi +h)(\overline{x})]^{\infty } =[\partial \psi (\overline{x})+\partial h(\overline{x})]^{\infty }. \) Thus, we obtain the first part.
Next we consider the case \(\psi =\delta _C\). Fix any \(\overline{x}\in C\) with \(\mathrm{ri}(C)\cap L_{\overline{x}}\ne \emptyset \), where \(L_{\overline{x}}\!:=\{x\in \mathbb {R}^p\,|\, x_i=0\ \mathrm{for}\ i\notin \mathrm{supp}(\overline{x})\}\). Let \(J=\mathrm{supp}(\overline{x})\). We first argue
If there exists \(\varepsilon >0\) such that \([\mathbb {B}(\overline{x},\varepsilon )\backslash \{\overline{x}\}]\cap [C\cap L_{\overline{x}}]=\emptyset \), then \(\partial \delta _{C\cap L_{\overline{x}}}(\overline{x}) =\mathcal {N}_{C\cap L_{\overline{x}}}(\overline{x})=\mathbb {R}^p\), and the inclusion in (4) clearly holds. So, it suffices to consider that for any \(\varepsilon >0\), \([\mathbb {B}(\overline{x},\varepsilon )\backslash \{\overline{x}\}]\cap [C\cap L_{\overline{x}}]\ne \emptyset \). Pick any \(v\in \widehat{\partial }(\delta _C\!+\!h)(\overline{x})\). By the definition of regular subgradient, it follows that
which implies that \(v\in \widehat{\partial } \delta _{C\cap L_{\overline{x}}}(\overline{x}) =\partial \delta _{C\cap L_{\overline{x}}}(\overline{x})\). The inclusion in (4) holds. By combining (4) with [5, Corollary 10.9] and \(\partial h(\overline{x})=\mathcal {N}_{L_{\overline{x}}}(\overline{x})\), we have
where the second equality is due to \(\mathrm{ri}C \cap L_{\overline{x}}\ne \emptyset \). In fact, from the above arguments, we conclude that for all \(x\in C\ \mathrm{with}\ \mathrm{ri}(C)\cap L_{x}\ne \emptyset \),
Now we argue that \(\partial (\delta _C\!+h)(\overline{x})\subseteq \partial \delta _C(\overline{x})+\partial h(\overline{x})\). To this end, pick any \(v\in \partial (\delta _C+h)(\overline{x})\). Then, there exist sequences \(x^k\xrightarrow [\delta _{C}+h]{}\overline{x}\) and \(v^k\in \widehat{\partial }(\delta _{C}\!+\!h)(x^k)\) with \(v^k\rightarrow v\) as \(k\rightarrow \infty \). Since \(\delta _{C}(x^k)+h(x^k)\rightarrow \delta _{C}(\overline{x})+h(\overline{x})\), we must have \(x^k\in C\) and \(h(x^k)\rightarrow h(\overline{x})\) for all k large enough. The latter, along with \(\mathrm{supp}(x^k)\supseteq J\), implies that \(\mathrm{supp}(x^k)=J\) for all sufficiently large k. By invoking (6), for all sufficiently large k, \( v^k\in \partial \delta _{C}(x^k)+\partial h(x^k). \) By passing to the limit \(k\rightarrow \infty \) and using \(h(x^k)\rightarrow h(\overline{x})\), we obtain \(v\in \partial \delta _{C}(\overline{x})+\partial h(\overline{x})\). By the arbitrariness of v in \(\partial (\delta _C+h)(\overline{x})\), the stated inclusion follows. In particular, together with \(\partial (\delta _C\!+h)(\overline{x})\supseteq \widehat{\partial } (\delta _C\!+h)(\overline{x}) =\partial \delta _C(\overline{x})+\partial h(\overline{x})\) and (5),
Next we argue \(\partial ^{\infty }(\delta _C\!+h)(\overline{x})= \partial ^{\infty }\delta _{C\cap L_{\overline{x}}}(\overline{x})\). Pick any \(u\in \partial ^{\infty }(\delta _C\!+h)(\overline{x})\). Then, there exist sequences \(x^k\xrightarrow [\delta _{C}+h]{}\overline{x}\) and \(u^k\in \widehat{\partial }(\delta _{C}\!+\!h)(x^k)\) with \(\lambda _ku^k\rightarrow u\) for some \(\lambda _k\downarrow 0\) as \(k\rightarrow \infty \). By following the same arguments as above, \(\mathrm{supp}(x^k)=J\) for all k large enough. Together with (6) and \(u^k\in \widehat{\partial }(\delta _{C}\!+\!h)(x^k)\), we have \(u^k\in \widehat{\partial }\delta _{C\cap L_{\overline{x}}}(x^k)\) for all k large enough. Notice that \(x^k\xrightarrow [C\cap L_{\overline{x}}]{}\overline{x}\). So, \(u\in \partial ^{\infty }(\delta _C\!+\!\delta _{L_{\overline{x}}})(\overline{x})\) and \(\partial ^{\infty }(\delta _C\!+h)(\overline{x})\subseteq \partial ^{\infty }\delta _{C\cap L_{\overline{x}}}(\overline{x})\). Conversely, pick any \(u\in \partial ^{\infty }\delta _{C\cap L_{\overline{x}}}(\overline{x})\). There exist \(x^k\xrightarrow [C\cap L_{\overline{x}}]{}\overline{x}\) and \(u^k\in \widehat{\partial }\delta _{C\cap L_{\overline{x}}}(x^k)\) with \(\lambda _ku^k\rightarrow u\) for some \(\lambda _k\downarrow 0\) as \(k\rightarrow \infty \). Clearly, \((\delta _C+h)(x^k)\rightarrow (\delta _C+h)(\overline{x})\). Also, from (6) and \(u^k\in \widehat{\partial }\delta _{C\cap L_{\overline{x}}}(x^k)\), we have \(u^k\in \widehat{\partial }(\delta _{C}+h)(x^k)\). So, \(u\in \partial ^{\infty }(\delta _C+h)(\overline{x})\), and \(\partial ^{\infty }(\delta _C\!+h)(\overline{x})\supseteq \partial ^{\infty }\delta _{C\cap L_{\overline{x}}}(\overline{x})\). The stated equality follows. From [5, Exercise 8.14 & Proposition 8.12], \(\partial \delta _{C\cap L_{\overline{x}}}(\overline{x}) =\partial ^{\infty }\delta _{C\cap L_{\overline{x}}}(\overline{x}) =[\widehat{\partial }\delta _{C\cap L_{\overline{x}}}(\overline{x})]^{\infty }\). Thus,
Together with (7), we obtain the conclusion for \(h(\cdot )=\nu \Vert \cdot \Vert _0\). By following the same arguments as above, one may obtain the second part. \(\square \)
References
Ioffe, A.D., Outrata, J.V.: On metric and calmness qualification conditions in subdifferential calculus. Set Valued Var. Anal. 16, 199–227 (2008)
Mordukhovich, B.S.: Variational Analysis and Generalized Differentiation I: Basic Theory. Springer, Heidelberg (2006)
Mordukhovich, B.S.: Variational Analysis and Applications. Springer, Cham (2018)
Robinson, S.M.: Some continuity properties of polyhedral multifunctions. Math. Program. Stud. 14, 206–214 (1981)
Rockafellar, R.T., Wets, R.J.: Variational Analysis. Springer, New York (1998)
Wu, Y.Q., Pan, S.H., Bi, S.J.: Kurdyka–Łojasiewicz property of zero-norm composite functions. J. Optim. Theory App. 188, 94–112 (2021)
Acknowledgements
Funding was provided by National Natural Science Foundation of China (Grant Nos. 11971177 and 11701186).
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Boris S. Mordukhovich.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
About this article
Cite this article
Wu, Y., Pan, S. & Bi, S. Correction to: Kurdyka–Łojasiewicz Property of Zero-Norm Composite Functions. J Optim Theory Appl 190, 339–342 (2021). https://doi.org/10.1007/s10957-021-01855-6
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10957-021-01855-6