Characterizations of Pseudolinear and Semistrictly Quasilinear Functions Vsevolod I. Ivanov Abstract In this paper we obtain several new complete characterizations of pseudolinear functions. Two of the results are of first-order and one is derivative free. All results are derived in terms of the Clarke-Rockafellar subdifferential. Additionally, we prove a characterization of the semistrictly quasilinear functions. It is similar to the derivative free characterization of the pseudolinear functions. We also find the conditions such that a semistrictly quasilinear function become pseudolinear. Keywords: Pseudolinear functions, semistrictly quasiliear functions, nonsmooth analysis, conditions for pseudoliearity and semistrictly quasiliearity Mathematical Subject Classification (2020): 26B25, 49J52 1 Introduction The concepts of pseudolinearity and semistrict quasilinearity provide in a very natural way generalizations of the linearity. These classes of functions include many useful functions. Among them are, for instance, the linear fractional functions. Several interesting properties of these functions appeared in [ 7 , 3 , 8 , 6 ] . In the paper by Ivanov [ 1 ] were obtained some new characterizations of pseudoconvex functions and semistrictly ones. Here we apply the obtained conditions for pseudoconvexity to derive three complete characterizations of pseudolinear functions in terms of the Clarke’s generalized directional derivative (Theorems 2 , 3 , 5 ). Theorem 2 is a generalization to non-differentiable functions of a theorem due to Chew and Choo [ 3 ] . In Theorem 4 , we obtain necessary conditions for pseudolinearity, but they are not sufficient for pseudolinearity. It is interesting which is the largest class of functions such that these conditions hold. We prove that they become necessary and sufficient when the function is semistrictly quasilinear, which implies that the mentioned class is the exact one. Every pseudolinear function, which is subdifferentiable with respect to the Clarke-Rockafellar subdifferential, is semistrictly quasilinear. We find the conditions that a semistrictly quasilinear function have to satisfy to become pseudolinear. 2 Preliminaries In this section, we present the necessary and sufficient conditions for a given function to be pseudoconvex, which were derived in [ 1 ] . In the next sections, we apply these results. In the sequel, we suppose that \mathbb{E} is a Banach space. We denote by \mathbb{E}^{*} its dual and the duality pairing between the vectors a\in\mathbb{E}^{*} and b\in\mathbb{E} by \langle a,b\rangle , by \mathbb{R} the set of reals. Definition 1 . Let f be a locally Lipschitz function, defined on some open set X in a Banach space \mathbb{E} . The Clarke’s generalized derivative at the point x\in X in direction v\in\mathbb{E} is defined by f^{0}(x,v)=\limsup_{y\to x;t\downarrow 0}\,[f(y+tv)-f(y)]/t. The Clarke’s subdifferential (or Clarke’s generalized gradient) of f at x is defined as follows: \partial f(x)=\{x^{*}\in\mathbb{E}^{*}\mid\langle x^{*},v\rangle\leq f^{0}(x,v),\quad\forall v\in\mathbb{E}\}. Definition 2 . A real-valued function f:X\to\mathbb{R} is called pseudoconvex (in terms of the Clarke directional derivative) iff the following implication is satisfied f(y)-\infty for every x\in\mathbb{E} . We want to apply these results to functions such that both f and -f are proper. Therefore f should be a finite function. The following result is a particular case of Lemma 3 in the paper [ 1 ] : Lemma 2 . Let f be a locally Lipschitz pseudoconvex real function, defined on some open convex set in a Banach space \mathbb{E} , which contains the convex set S . Then the following implication holds x\in S,\;y\in S,\;f(y)\leq f(x)\quad\Rightarrow\quad\langle x^{*},y-x\rangle\leq 0,\quad\forall x^{*}\in\partial f(x). Theorem 2 . Let f be a locally Lipsctitz f is pseudolinear on S with respect to the Clarke’s directional derivative if and only if there exists a positive function p:S\times S\times\mathbb{E}^{*}\to(0,+\infty) with f(y)-f(x)=p(x,y,x^{*})\,\langle x^{*},y-x\rangle,\quad\forall\;x\in S\,\;\forall y\in S,\;\forall x^{*}\in\partial^{\uparrow}f(x) (3) such that \partial^{\uparrow}f(x)\neq\emptyset . Proof. Let f be pseudolinear. We prove that there exists a function p satisfying ( 3 ). Consider the function, defined as follows: p(x,y,x^{*})=\left\{\begin{array}[]{ll}\frac{f(y)-f(x)}{\langle x^{*},y-x\rangle},&\quad\textrm{if}\quad\langle x^{*},y-x\rangle\neq 0\\ 1,&\quad\textrm{if}\quad\langle x^{*},y-x\rangle=0.\end{array}\right. (4) We prove that it is positive. Let \langle x^{*},y-x\rangle>0 . It follows from pseudoconvexity that f(y)\geq f(x) . Suppose that it is possible that f(y)=f(x) . Then by Lemma 2 we obtain that \langle x^{*},y-x\rangle\leq 0 , which is a contradiction. Let \langle x^{*},y-x\rangle<0 . Since \partial(sf)(x)=s\partial f(x) for every s\in\mathbb{R} , then -x^{*}\in\partial(-f)(x) . We conclude from here that f(y)\leq f(x) . By Lemma 2 we obtain that the case f(y)=f(x) is impossible. Therefore p>0 . We prove that the function p satisfies ( 3 ). It is enough to show that \langle x^{*},y-x\rangle=0 implies that f(y)=f(x) . Indeed, assume the contrary. If f(y)f(x) , then by pseudoconcavity we again get a contradiction. Let x\in S , y\in S , x^{*}\in\partial f(x) and equation ( 3 ) is satisfied. Obviously ( 4 ) implies that f is pseudolinear. ∎ Theorem 2 is a generalization to non-differentiable functions of a known claim due to Chew and Choo [ 3 , Proposition 2.1] . Theorem 3 . Let f be a locally Lipschitz real-valued function, defined on some open convex set in a Banach spase \mathbb{E} , which contains the convex set S . Then f is pseudolinear on S if and only if there exists a positive function p:S\times S\times\mathbb{E}^{*}\to(0,+\infty) with \begin{array}[]{c}p(x,y,x^{*})\;\langle x^{*},y-x\rangle+p(y,x,y^{*})\;\langle y^{*},x-y\rangle=0,\\ \forall(x,y)\in S\times S,\;\forall(x^{*},y^{*})\in\partial f(x)\times\partial f(y).\end{array} (5) Proof. Let f be pseudolinear. We prove that inequality ( 5 ) holds. Choose arbitrary x\in S , y\in S . It follows from Theorem 2 that there exists a function p:S\times S\times\mathbb{E}^{*}\to(0,+\infty) with f(y)-f(x)=p(x,y,x^{*})\;\langle x^{*},y-x\rangle,\quad\forall x^{*}\in\partial f(x) (6) and f(x)-f(y)=p(y,x,y^{*})\;\langle y^{*},x-y\rangle,\quad\forall y^{*}\in\partial f(y). (7) If we add ( 6 ) and ( 7 ), then we obtain ( 5 ). The converse claim follows from Theorem 1 . ∎ Theorem 4 . Let S be a convex set, included in some open convex set \Gamma in a Banach space \mathbb{E} . Suppose that f is a locally Lipschitz function, which is pseudolinear on S with respect to the Clarke’s derivative. Then for all x\in S , y\in S , and \lambda\in[0,1] there exists a number b>0 , which depend on x , y , \lambda such that the following conditions are satisfied: f[x+\lambda(y-x)]=\lambda\,b\,f(y)+(1-\lambda\,b)\,f(x), (8) 00 (14) we consider several cases: First, f(y)0 . It follows from q=1/p and ( 4 ) that q(z(\lambda),y)=(1-\lambda)\langle\nabla f(z(\lambda)),y-x\rangle/[f(y)-f(z(\lambda))]. (15) According to the continuous differentiability we obtain that \lim_{\lambda\downarrow 0}\,q(z(\lambda),y)=\frac{\nabla f(x)(y-x)}{f(y)-f(x)}=q(x,y). (16) Then we conclude from ( 12 ), ( 13 ), ( 16 ) that ( 14 ) holds. Second, f(y)>f(x) . We have f(y)>f(z(\lambda)) for all sufficiently small \lambda>0 . Thanks to ( 4 ) and ( 15 ), we obtain that ( 16 ) is satisfied again, and ( 14 ) holds. Third, f(y)=f(x) . We have that \nabla f(x)(y-x)=0 . Therefore q(x,y)=1 . It this case f[x+\lambda(y-x)]=f(x) for all \lambda\in[0,1] , and b(x,y,\lambda)=1 for every \lambda\in[0,1] . Therefore, the required equality ( 14 ) is satisfied. It is seen from ( 10 ) that for all x and y such that f(y)\neq f(x) we have q(x,y)=\langle\nabla f(x),y-x\rangle/[f(y)-f(x)]>0 . The converse claim (b) \Rightarrow (a) is trivial. ∎ Example 1 . Consider the function f:\mathbb{R}^{2}\to\mathbb{R} defined by f(x)=x_{2}/x_{1} , where S=\{x=(x_{1},x_{2})\mid x_{1}>0\}. The function f is pseudolinear over S . This function satisfies ( 8 ) and ( 9 ). We have [f(x+\lambda(y-x))-f(x)]/\lambda=(x_{1}y_{2}-x_{2}y_{1})/[x_{1}(x_{1}+\lambda(y_{1}-x_{1})] and b(x,y,\lambda)=(f(x+\lambda(y-x)))/(\lambda(f(y)-f(x)))=y_{1}/(x_{1}+\lambda(y_{1}-x_{1})). Therefore 0<\lambda\,b\leq 1 for all x\in S , y\in S , \lambda\in(0,1] . 4 On semistrictly quasilinear and pseudolinear functions It is interesting which is the class of functions such that the necessary conditions from Theorem 4 become both necessary and sufficient. In this section, we show that this property can be generalized and it becomes necessary and sufficient when the function is semistrictly quasilinear. Definition 3 . A function f , defined on a convex set S is called semistrictly quasiconvex iff for all x , y\in S , \lambda\in(0,1) the following implication holds: f(y)0 , which depend on x , y , \lambda such that 0<\lambda b(x,y,\lambda)<1 and Condition ( 8 ) is satisfied. Proof. Let f be both semistrictly quasiconvex and semistrictly quasiconcave. Consider the function b(x,y,\lambda) defined by b=\frac{f[x+\lambda(y-x)]-f(x)}{\lambda[f(y)-f(x)]}. Let f(y)0. It follows from Lemma 3 that f(y)f(x) is similar. It follows from semistrict quasiconvexity that f[x+\lambda(y-x)]\lambda\,b\,f(y)+(1-\lambda\,b)\,f(y)=f(y). It follows from ( 8 ) that f(y)f(x) , by pseudoconcavity we have \nabla f(x)(y-x)>0 , which is also a contradiction. Therefore ( 18 ) holds. ∎ Theorem 8 . Let S be a convex subset of an open set \Gamma in a Banach space \mathbb{E} . Suppose that f:\Gamma\to\mathbb{R} is a continuous function, which is both semistrictly quasiconvex and semistrictly quasiconcave on S and \partial^{\uparrow}f(x)\neq 0 , \partial^{\uparrow}(-f)(x)\neq 0 for all x\in S . Then f is pseudolinear with respect to the Clarke-Rockafellar subdifferential if and only if the both implications hold: x\in S,\;y\in S,\;\xi\in\partial f(x),\;\langle\xi,y-x\rangle=0\quad\Rightarrow\quad f(y)\geq f(x). (19) and x\in S,\;y\in S,\;\eta\in\partial^{\uparrow}(-f)(x),\;\langle\eta,y-x\rangle=0\quad\Rightarrow\quad f(y)\leq f(x). (20) Proof. Let f be pseudolinear. We prove implication ( 19 ). Take arbitrary x\in S , y\in S , \xi\in\partial f(x) such that \langle\xi,y-x\rangle=0 . If f(y)0,\;\xi\in\partial^{\uparrow}f(x)\quad{\rm implies}\quad f(y)\geq f(x). Indeed, it follows from \langle\xi,y-x\rangle>0 that f^{\uparrow}(x,y-x)>0 . By the definition of the Clarke-Rockafellar derivative, there exist \varepsilon>0 and sequences \{x_{i}\}_{i=1}^{\infty} , x_{i}\in\Gamma , \{t_{i}\}_{i=1}^{\infty} , t_{i}>0 such that x_{i}\to x , t_{i}\downarrow 0 and \inf_{u\in B(y-x,\varepsilon)}[f(x_{i}+t_{i}u)-f(x_{i})]/t_{i}>0,\quad\forall i. Taking the number i sufficiently large we ensure that x_{i}\in B(x,\varepsilon) . Therefore, we have y-x_{i}\in B(y-x,\varepsilon) and f[x_{i}+t_{i}(y-x_{i})]>f(x_{i}) . Using that f is lower semicontinuous and semistrictly quasiconvex we conclude from Proposition 1 that it is quasiconvex on S . Therefore, f(x_{i})0,\;\eta\in\partial^{\uparrow}(-f)(x)\quad{\rm imply}\quad f(y)\leq f(x). Indeed, it follows from \langle\eta,y-x\rangle>0 that (-f)^{\uparrow}(x,y-x)>0 . By the definition of the Clarke-Rockafellar derivative, there exist \varepsilon>0 and sequences \{x_{i}\}_{i=1}^{\infty} , x_{i}\in\Gamma , \{t_{i}\}_{i=1}^{\infty} , t_{i}>0 such that x_{i}\to x , t_{i}\downarrow 0 and \inf_{u\in B(y-x,\varepsilon)}[-f(x_{i}+t_{i}u)+f(x_{i})]/t_{i}>0,\quad\forall i. Taking the number i sufficiently large we ensure that x_{i}\in B(x,\varepsilon) . Therefore, we have y-x_{i}\in B(y-x,\varepsilon) and f[x_{i}+t_{i}(y-x_{i})]f[x_{i}+t_{i}(y-x_{i})]\geq f(y). Hence, f(x)\geq\limsup_{i\to\infty}f(x_{i})\geq f(y) . It follows from the converse implication that x\in S,\;y\in S,\;f(y)>f(x)\quad\textrm{imply}\quad\langle\eta,y-x\rangle\leq 0,\quad\forall\;\eta\in\partial^{\uparrow}(-f)(x). Therefore, by implication ( 20 ), we obtain that -f is pseudoconvex, which implies that the function f is pseudolinear. ∎ Corollary 1 . Let S be an open convex subset in a Banach space \mathbb{E} . Suppose that f:S\to\mathbb{R} is a locally Lipschitz semistrictly quasilinear on S function. Then f is pseudolinear with respect to the Clarke generalized directioanal derivative if and only if the following implication holds: x\in S,\;y\in S,\;\xi\in\partial f(x),\;\langle\xi,y-x\rangle=0\quad\Rightarrow\quad f(y)=f(x). and Remark 1 . Theorem 7 is not a consequence of Corollary 1 , because the Clarke subdifferential \partial f(x) does not coincides with the gradient \nabla f(x) when the function is Fréchet differentiable, but it is not necessarily continuously differentiable.