We use the notation of the statement. Throughout we use Concatenation Identifies a Product of Euclidean Spaces with a Euclidean Space , by which ι \iota ι is a linear bijection, and claim 3 of that lemma, by which
ι ( ξ , η ) ⋅ ι ( ξ ′ , η ′ ) = ξ ⋅ ξ ′ + η η ′ and ∥ ι ( ξ , η ) ∥ 2 = ∥ ξ ∥ 2 + η 2 \iota(\xi,\eta)\cdot\iota(\xi',\eta')=\xi\cdot\xi'+\eta\,\eta'\qquad\text{and}\qquad\lVert\iota(\xi,\eta)\rVert^{2}=\lVert\xi\rVert^{2}+\eta^{2} ι ( ξ , η ) ⋅ ι ( ξ ′ , η ′ ) = ξ ⋅ ξ ′ + η η ′ and ∥ ι ( ξ , η ) ∥ 2 = ∥ ξ ∥ 2 + η 2
for ξ , ξ ′ ∈ R n \xi,\xi'\in\mathbb{R}^{n} ξ , ξ ′ ∈ R n and η , η ′ ∈ R \eta,\eta'\in\mathbb{R} η , η ′ ∈ R . Algebraic manipulations of dot products use The Dot Product on R n \mathbb{R}^n R n Is Symmetric, Bilinear and Positive Definite , and ∥ v ∥ 2 = v ⋅ v \lVert v\rVert^{2}=v\cdot v ∥ v ∥ 2 = v ⋅ v is claim 1 of Properties of the Euclidean Norm on R n \mathbb{R}^n R n : Square, Distance, Vanishing, Coordinate Bounds, Homogeneity, Triangle Inequalities and Expansion of a Square .
Claim 1. Let z ∈ U z\in U z ∈ U . If z = y z=y z = y the asserted inequality reads f ( y ) ≥ f ( y ) f(y)\ge f(y) f ( y ) ≥ f ( y ) , which holds. So assume z ≠ y z\neq y z = y , and let θ \theta θ be the smaller of 1 1 1 and r / ∥ z − y ∥ r/\lVert z-y\rVert r / ∥ z − y ∥ , a real number with 0 < θ ≤ 1 0<\theta\le1 0 < θ ≤ 1 . Put w = θ z + ( 1 − θ ) y = y + θ ( z − y ) w=\theta z+(1-\theta)y=y+\theta(z-y) w = θ z + ( 1 − θ ) y = y + θ ( z − y ) , which lies in U U U because U U U is convex. By claim 5 of Properties of the Euclidean Norm on R n \mathbb{R}^n R n : Square, Distance, Vanishing, Coordinate Bounds, Homogeneity, Triangle Inequalities and Expansion of a Square , ∥ w − y ∥ = θ ∥ z − y ∥ ≤ r \lVert w-y\rVert=\theta\lVert z-y\rVert\le r ∥ w − y ∥ = θ ∥ z − y ∥ ≤ r , so w ∈ B ˉ ( y , r ) w\in\bar{B}(y,r) w ∈ B ˉ ( y , r ) and the hypothesis gives
f ( w ) ≥ f ( y ) + p ⋅ ( w − y ) = f ( y ) + θ p ⋅ ( z − y ) . f(w)\ge f(y)+p\cdot(w-y)=f(y)+\theta\,p\cdot(z-y). f ( w ) ≥ f ( y ) + p ⋅ ( w − y ) = f ( y ) + θ p ⋅ ( z − y ) .
On the other hand f f f is convex on U U U , so f ( w ) ≤ θ f ( z ) + ( 1 − θ ) f ( y ) f(w)\le\theta f(z)+(1-\theta)f(y) f ( w ) ≤ θ f ( z ) + ( 1 − θ ) f ( y ) . Combining and subtracting f ( y ) f(y) f ( y ) ,
θ p ⋅ ( z − y ) ≤ θ ( f ( z ) − f ( y ) ) , \theta\,p\cdot(z-y)\le\theta\bigl(f(z)-f(y)\bigr), θ p ⋅ ( z − y ) ≤ θ ( f ( z ) − f ( y ) ) ,
and dividing by the positive number θ \theta θ gives f ( z ) ≥ f ( y ) + p ⋅ ( z − y ) f(z)\ge f(y)+p\cdot(z-y) f ( z ) ≥ f ( y ) + p ⋅ ( z − y ) . As z ∈ U z\in U z ∈ U was arbitrary, p ∈ ∂ U f ( y ) p\in\partial_{U}f(y) p ∈ ∂ U f ( y ) by Subdifferential of a Real-Valued Function on a Convex Subset of R n \mathbb{R}^n R n (i) .
Claim 2. Since U U U is open, y y y is an interior point of U U U , so A Convex Function is Lipschitz on a Ball around an Interior Point provides r , M ∈ R r,M\in\mathbb{R} r , M ∈ R with 0 < r 0<r 0 < r , 0 ≤ M 0\le M 0 ≤ M , B ˉ ( y , r ) ⊆ U \bar{B}(y,r)\subseteq U B ˉ ( y , r ) ⊆ U and
∣ f ( z ) − f ( w ) ∣ ≤ M ∥ z − w ∥ for all z , w ∈ B ˉ ( y , r ) . (L) |f(z)-f(w)|\le M\,\lVert z-w\rVert\qquad\text{for all }z,w\in\bar{B}(y,r). \tag{L} ∣ f ( z ) − f ( w ) ∣ ≤ M ∥ z − w ∥ for all z , w ∈ B ˉ ( y , r ) . ( L )
Put c = f ( y ) + M r + 1 c=f(y)+Mr+1 c = f ( y ) + M r + 1 and
K = { ι ( z , t ) : z ∈ B ˉ ( y , r ) , t ∈ R , f ( z ) ≤ t and t ≤ c } ⊆ R n + 1 . K=\bigl\{\,\iota(z,t)\;:\;z\in\bar{B}(y,r),\ t\in\mathbb{R},\ f(z)\le t\ \text{ and }\ t\le c\,\bigr\}\subseteq\mathbb{R}^{n+1}. K = { ι ( z , t ) : z ∈ B ˉ ( y , r ) , t ∈ R , f ( z ) ≤ t and t ≤ c } ⊆ R n + 1 .
K K K is nonempty: f ( y ) ≤ f ( y ) f(y)\le f(y) f ( y ) ≤ f ( y ) and f ( y ) ≤ c f(y)\le c f ( y ) ≤ c , so ι ( y , f ( y ) ) ∈ K \iota(y,f(y))\in K ι ( y , f ( y )) ∈ K .
K K K is bounded: let ι ( z , t ) ∈ K \iota(z,t)\in K ι ( z , t ) ∈ K . Then ∥ z ∥ ≤ ∥ y ∥ + r \lVert z\rVert\le\lVert y\rVert+r ∥ z ∥ ≤ ∥ y ∥ + r by claim 6 of Properties of the Euclidean Norm on R n \mathbb{R}^n R n : Square, Distance, Vanishing, Coordinate Bounds, Homogeneity, Triangle Inequalities and Expansion of a Square , and by (L) f ( y ) − M r ≤ f ( z ) ≤ t ≤ c f(y)-Mr\le f(z)\le t\le c f ( y ) − M r ≤ f ( z ) ≤ t ≤ c , so ∣ t ∣ ≤ ∣ f ( y ) − M r ∣ + ∣ c ∣ |t|\le|f(y)-Mr|+|c| ∣ t ∣ ≤ ∣ f ( y ) − M r ∣ + ∣ c ∣ . Hence ∥ ι ( z , t ) ∥ 2 = ∥ z ∥ 2 + t 2 \lVert\iota(z,t)\rVert^{2}=\lVert z\rVert^{2}+t^{2} ∥ ι ( z , t ) ∥ 2 = ∥ z ∥ 2 + t 2 is bounded above by a number independent of the point, and K K K is bounded.
K K K is closed: we use Sequential Characterization of Closed Subsets of a Metric Space . Let ( ξ m ) m ∈ N (\xi_{m})_{m\in\mathbb{N}} ( ξ m ) m ∈ N be a sequence in K K K converging to ξ ∈ R n + 1 \xi\in\mathbb{R}^{n+1} ξ ∈ R n + 1 . Since ι \iota ι is a bijection we may write ξ m = ι ( z m , t m ) \xi_{m}=\iota(z_{m},t_{m}) ξ m = ι ( z m , t m ) and ξ = ι ( z , t ) \xi=\iota(z,t) ξ = ι ( z , t ) , and by linearity of ι \iota ι and the norm identity,
∥ ξ m − ξ ∥ 2 = ∥ ι ( z m − z , t m − t ) ∥ 2 = ∥ z m − z ∥ 2 + ( t m − t ) 2 , \lVert\xi_{m}-\xi\rVert^{2}=\lVert\iota(z_{m}-z,t_{m}-t)\rVert^{2}=\lVert z_{m}-z\rVert^{2}+(t_{m}-t)^{2}, ∥ ξ m − ξ ∥ 2 = ∥ ι ( z m − z , t m − t ) ∥ 2 = ∥ z m − z ∥ 2 + ( t m − t ) 2 ,
so ∥ z m − z ∥ ≤ ∥ ξ m − ξ ∥ \lVert z_{m}-z\rVert\le\lVert\xi_{m}-\xi\rVert ∥ z m − z ∥ ≤ ∥ ξ m − ξ ∥ and ∣ t m − t ∣ ≤ ∥ ξ m − ξ ∥ |t_{m}-t|\le\lVert\xi_{m}-\xi\rVert ∣ t m − t ∣ ≤ ∥ ξ m − ξ ∥ ; hence ( z m ) (z_{m}) ( z m ) converges to z z z and ( t m ) (t_{m}) ( t m ) to t t t . The closed ball B ˉ ( y , r ) \bar{B}(y,r) B ˉ ( y , r ) is closed by claim 3 of Elementary Properties of the Closed Ball in a Metric Space , so z ∈ B ˉ ( y , r ) z\in\bar{B}(y,r) z ∈ B ˉ ( y , r ) by Sequential Characterization of Closed Subsets of a Metric Space again. From t m ≤ c t_{m}\le c t m ≤ c and Order Properties of Limits of Real Sequences we get t ≤ c t\le c t ≤ c . Finally (L) gives ∣ f ( z m ) − f ( z ) ∣ ≤ M ∥ z m − z ∥ |f(z_{m})-f(z)|\le M\lVert z_{m}-z\rVert ∣ f ( z m ) − f ( z ) ∣ ≤ M ∥ z m − z ∥ , so ( f ( z m ) ) (f(z_{m})) ( f ( z m )) converges to f ( z ) f(z) f ( z ) , and from f ( z m ) ≤ t m f(z_{m})\le t_{m} f ( z m ) ≤ t m and Order Properties of Limits of Real Sequences we get f ( z ) ≤ t f(z)\le t f ( z ) ≤ t . Hence ξ ∈ K \xi\in K ξ ∈ K , and K K K is closed.
K K K is convex: let ι ( z 1 , t 1 ) , ι ( z 2 , t 2 ) ∈ K \iota(z_{1},t_{1}),\iota(z_{2},t_{2})\in K ι ( z 1 , t 1 ) , ι ( z 2 , t 2 ) ∈ K and let θ ∈ R \theta\in\mathbb{R} θ ∈ R with 0 ≤ θ ≤ 1 0\le\theta\le1 0 ≤ θ ≤ 1 . By linearity of ι \iota ι ,
θ ι ( z 1 , t 1 ) + ( 1 − θ ) ι ( z 2 , t 2 ) = ι ( θ z 1 + ( 1 − θ ) z 2 , θ t 1 + ( 1 − θ ) t 2 ) . \theta\,\iota(z_{1},t_{1})+(1-\theta)\,\iota(z_{2},t_{2})=\iota\bigl(\theta z_{1}+(1-\theta)z_{2},\ \theta t_{1}+(1-\theta)t_{2}\bigr). θ ι ( z 1 , t 1 ) + ( 1 − θ ) ι ( z 2 , t 2 ) = ι ( θ z 1 + ( 1 − θ ) z 2 , θ t 1 + ( 1 − θ ) t 2 ) .
The point θ z 1 + ( 1 − θ ) z 2 \theta z_{1}+(1-\theta)z_{2} θ z 1 + ( 1 − θ ) z 2 lies in B ˉ ( y , r ) \bar{B}(y,r) B ˉ ( y , r ) by claim 2 of Euclidean Balls are Convex ; the number θ t 1 + ( 1 − θ ) t 2 \theta t_{1}+(1-\theta)t_{2} θ t 1 + ( 1 − θ ) t 2 is at most c c c ; and since f f f is convex on U ⊇ B ˉ ( y , r ) U\supseteq\bar{B}(y,r) U ⊇ B ˉ ( y , r ) ,
f ( θ z 1 + ( 1 − θ ) z 2 ) ≤ θ f ( z 1 ) + ( 1 − θ ) f ( z 2 ) ≤ θ t 1 + ( 1 − θ ) t 2 . f\bigl(\theta z_{1}+(1-\theta)z_{2}\bigr)\le\theta f(z_{1})+(1-\theta)f(z_{2})\le\theta t_{1}+(1-\theta)t_{2}. f ( θ z 1 + ( 1 − θ ) z 2 ) ≤ θ f ( z 1 ) + ( 1 − θ ) f ( z 2 ) ≤ θ t 1 + ( 1 − θ ) t 2 .
Hence the combination lies in K K K .
Projections from below the graph. For k ∈ N k\in\mathbb{N} k ∈ N put ξ k = ι ( y , f ( y ) − 1 / k ) \xi_{k}=\iota(y,f(y)-1/k) ξ k = ι ( y , f ( y ) − 1/ k ) . Since ι \iota ι is injective, ξ k ∈ K \xi_{k}\in K ξ k ∈ K would force f ( y ) ≤ f ( y ) − 1 / k f(y)\le f(y)-1/k f ( y ) ≤ f ( y ) − 1/ k , which is false; so ξ k ∉ K \xi_{k}\notin K ξ k ∈ / K . By claim 1 of Nearest-Point Projection onto a Nonempty Closed Convex Subset of Euclidean Space , applied to the nonempty closed convex set K K K , the nearest point π K ( ξ k ) ∈ K \pi_{K}(\xi_{k})\in K π K ( ξ k ) ∈ K is defined; write π K ( ξ k ) = ι ( z k , t k ) \pi_{K}(\xi_{k})=\iota(z_{k},t_{k}) π K ( ξ k ) = ι ( z k , t k ) and put v k = ξ k − π K ( ξ k ) v_{k}=\xi_{k}-\pi_{K}(\xi_{k}) v k = ξ k − π K ( ξ k ) , which is nonzero because ξ k ∉ K \xi_{k}\notin K ξ k ∈ / K while π K ( ξ k ) ∈ K \pi_{K}(\xi_{k})\in K π K ( ξ k ) ∈ K . By claim 2 of that lemma, v k ⋅ ( w − π K ( ξ k ) ) ≤ 0 v_{k}\cdot(w-\pi_{K}(\xi_{k}))\le0 v k ⋅ ( w − π K ( ξ k )) ≤ 0 for every w ∈ K w\in K w ∈ K ; multiplying by the positive number 1 / ∥ v k ∥ 1/\lVert v_{k}\rVert 1/ ∥ v k ∥ and putting u k = v k / ∥ v k ∥ u_{k}=v_{k}/\lVert v_{k}\rVert u k = v k / ∥ v k ∥ , a point with ∥ u k ∥ = 1 \lVert u_{k}\rVert=1 ∥ u k ∥ = 1 by claim 5 of Properties of the Euclidean Norm on R n \mathbb{R}^n R n : Square, Distance, Vanishing, Coordinate Bounds, Homogeneity, Triangle Inequalities and Expansion of a Square , we get
u k ⋅ ( w − π K ( ξ k ) ) ≤ 0 for every w ∈ K . (N) u_{k}\cdot\bigl(w-\pi_{K}(\xi_{k})\bigr)\le0\qquad\text{for every }w\in K. \tag{N} u k ⋅ ( w − π K ( ξ k ) ) ≤ 0 for every w ∈ K . ( N )
Since ι ( y , f ( y ) ) ∈ K \iota(y,f(y))\in K ι ( y , f ( y )) ∈ K , claims 3 and 4 of Nearest-Point Projection onto a Nonempty Closed Convex Subset of Euclidean Space give
∥ π K ( ξ k ) − ι ( y , f ( y ) ) ∥ = ∥ π K ( ξ k ) − π K ( ι ( y , f ( y ) ) ) ∥ ≤ ∥ ξ k − ι ( y , f ( y ) ) ∥ = ∥ ι ( 0 , − 1 / k ) ∥ = 1 / k , \bigl\lVert\pi_{K}(\xi_{k})-\iota(y,f(y))\bigr\rVert=\bigl\lVert\pi_{K}(\xi_{k})-\pi_{K}(\iota(y,f(y)))\bigr\rVert\le\bigl\lVert\xi_{k}-\iota(y,f(y))\bigr\rVert=\lVert\iota(0,-1/k)\rVert=1/k, π K ( ξ k ) − ι ( y , f ( y )) = π K ( ξ k ) − π K ( ι ( y , f ( y ))) ≤ ξ k − ι ( y , f ( y )) = ∥ ι ( 0 , − 1/ k )∥ = 1/ k ,
and since by The Archimedean Property of the Real Numbers the numbers 1 / k 1/k 1/ k eventually fall below any prescribed positive real, the sequence ( π K ( ξ k ) ) k ∈ N (\pi_{K}(\xi_{k}))_{k\in\mathbb{N}} ( π K ( ξ k ) ) k ∈ N converges to ι ( y , f ( y ) ) \iota(y,f(y)) ι ( y , f ( y )) .
The points u k u_{k} u k all lie in B ˉ ( 0 , 1 ) \bar{B}(0,1) B ˉ ( 0 , 1 ) , which is bounded, so The Bolzano-Weierstrass Theorem in R n \mathbb{R}^n R n provides u ∈ R n + 1 u\in\mathbb{R}^{n+1} u ∈ R n + 1 and a strictly increasing sequence ( p l ) l ∈ N (p_{l})_{l\in\mathbb{N}} ( p l ) l ∈ N in N \mathbb{N} N such that ( u p l ) l ∈ N (u_{p_{l}})_{l\in\mathbb{N}} ( u p l ) l ∈ N converges to u u u . Choose l 0 l_{0} l 0 with ∥ u p l 0 − u ∥ < 1 / 2 \lVert u_{p_{l_{0}}}-u\rVert<1/2 ∥ u p l 0 − u ∥ < 1/2 ; then by claim 6 of Properties of the Euclidean Norm on R n \mathbb{R}^n R n : Square, Distance, Vanishing, Coordinate Bounds, Homogeneity, Triangle Inequalities and Expansion of a Square , 1 = ∥ u p l 0 ∥ ≤ ∥ u ∥ + ∥ u p l 0 − u ∥ 1=\lVert u_{p_{l_{0}}}\rVert\le\lVert u\rVert+\lVert u_{p_{l_{0}}}-u\rVert 1 = ∥ u p l 0 ∥ ≤ ∥ u ∥ + ∥ u p l 0 − u ∥ , so ∥ u ∥ ≥ 1 / 2 \lVert u\rVert\ge1/2 ∥ u ∥ ≥ 1/2 and in particular u ≠ 0 u\neq0 u = 0 . Write u = ι ( a , b ) u=\iota(a,b) u = ι ( a , b ) with a ∈ R n a\in\mathbb{R}^{n} a ∈ R n and b ∈ R b\in\mathbb{R} b ∈ R .
Fix w ∈ K w\in K w ∈ K . For every l l l , by (N) and the dot product identity,
u p l ⋅ ( w − π K ( ξ p l ) ) ≤ 0. u_{p_{l}}\cdot\bigl(w-\pi_{K}(\xi_{p_{l}})\bigr)\le0 . u p l ⋅ ( w − π K ( ξ p l ) ) ≤ 0.
Both factors converge, the first to u u u and the second to w − ι ( y , f ( y ) ) w-\iota(y,f(y)) w − ι ( y , f ( y )) ; since for any points α , α ′ , β , β ′ \alpha,\alpha',\beta,\beta' α , α ′ , β , β ′ one has, by The Dot Product on R n \mathbb{R}^n R n Is Symmetric, Bilinear and Positive Definite and The Cauchy-Schwarz Inequality on R n \mathbb{R}^n R n ,
∣ α ′ ⋅ β ′ − α ⋅ β ∣ ≤ ∣ ( α ′ − α ) ⋅ β ′ ∣ + ∣ α ⋅ ( β ′ − β ) ∣ ≤ ∥ α ′ − α ∥ ∥ β ′ ∥ + ∥ α ∥ ∥ β ′ − β ∥ , |\alpha'\cdot\beta'-\alpha\cdot\beta|\le|(\alpha'-\alpha)\cdot\beta'|+|\alpha\cdot(\beta'-\beta)|\le\lVert\alpha'-\alpha\rVert\,\lVert\beta'\rVert+\lVert\alpha\rVert\,\lVert\beta'-\beta\rVert , ∣ α ′ ⋅ β ′ − α ⋅ β ∣ ≤ ∣ ( α ′ − α ) ⋅ β ′ ∣ + ∣ α ⋅ ( β ′ − β ) ∣ ≤ ∥ α ′ − α ∥ ∥ β ′ ∥ + ∥ α ∥ ∥ β ′ − β ∥ ,
and the norms ∥ w − π K ( ξ p l ) ∥ \lVert w-\pi_{K}(\xi_{p_{l}})\rVert ∥ w − π K ( ξ p l )∥ are bounded, the left-hand sides converge to u ⋅ ( w − ι ( y , f ( y ) ) ) u\cdot(w-\iota(y,f(y))) u ⋅ ( w − ι ( y , f ( y ))) . By Order Properties of Limits of Real Sequences , writing w = ι ( z , t ) w=\iota(z,t) w = ι ( z , t ) ,
a ⋅ ( z − y ) + b ( t − f ( y ) ) ≤ 0 for every ι ( z , t ) ∈ K . (H) a\cdot(z-y)+b\,\bigl(t-f(y)\bigr)\le0\qquad\text{for every }\iota(z,t)\in K. \tag{H} a ⋅ ( z − y ) + b ( t − f ( y ) ) ≤ 0 for every ι ( z , t ) ∈ K . ( H )
The last coordinate is negative. Taking z = y z=y z = y and t = f ( y ) + M r + 1 = c t=f(y)+Mr+1=c t = f ( y ) + M r + 1 = c in (H), which is legitimate since f ( y ) ≤ c f(y)\le c f ( y ) ≤ c , gives b ( M r + 1 ) ≤ 0 b\,(Mr+1)\le0 b ( M r + 1 ) ≤ 0 , so b ≤ 0 b\le0 b ≤ 0 because 0 < M r + 1 0<Mr+1 0 < M r + 1 . Suppose b = 0 b=0 b = 0 . Then ∥ u ∥ 2 = ∥ a ∥ 2 \lVert u\rVert^{2}=\lVert a\rVert^{2} ∥ u ∥ 2 = ∥ a ∥ 2 , so a ≠ 0 a\neq0 a = 0 . For z ∈ B ˉ ( y , r ) z\in\bar{B}(y,r) z ∈ B ˉ ( y , r ) we have, by (L), f ( z ) ≤ f ( y ) + M ∥ z − y ∥ ≤ f ( y ) + M r < c f(z)\le f(y)+M\lVert z-y\rVert\le f(y)+Mr<c f ( z ) ≤ f ( y ) + M ∥ z − y ∥ ≤ f ( y ) + M r < c , so ι ( z , f ( z ) ) ∈ K \iota(z,f(z))\in K ι ( z , f ( z )) ∈ K and (H) gives a ⋅ ( z − y ) ≤ 0 a\cdot(z-y)\le0 a ⋅ ( z − y ) ≤ 0 . Taking z = y + r a / ∥ a ∥ z=y+r\,a/\lVert a\rVert z = y + r a / ∥ a ∥ , which lies in B ˉ ( y , r ) \bar{B}(y,r) B ˉ ( y , r ) because ∥ z − y ∥ = r \lVert z-y\rVert=r ∥ z − y ∥ = r , we get a ⋅ ( z − y ) = r ∥ a ∥ 2 / ∥ a ∥ = r ∥ a ∥ > 0 a\cdot(z-y)=r\,\lVert a\rVert^{2}/\lVert a\rVert=r\lVert a\rVert>0 a ⋅ ( z − y ) = r ∥ a ∥ 2 / ∥ a ∥ = r ∥ a ∥ > 0 , a contradiction. Hence b < 0 b<0 b < 0 .
Conclusion. Put p = ( 1 / ( − b ) ) a p=(1/(-b))\,a p = ( 1/ ( − b )) a . For z ∈ B ˉ ( y , r ) z\in\bar{B}(y,r) z ∈ B ˉ ( y , r ) we have ι ( z , f ( z ) ) ∈ K \iota(z,f(z))\in K ι ( z , f ( z )) ∈ K as just shown, so (H) gives a ⋅ ( z − y ) + b ( f ( z ) − f ( y ) ) ≤ 0 a\cdot(z-y)+b(f(z)-f(y))\le0 a ⋅ ( z − y ) + b ( f ( z ) − f ( y )) ≤ 0 , that is a ⋅ ( z − y ) ≤ ( − b ) ( f ( z ) − f ( y ) ) a\cdot(z-y)\le(-b)\bigl(f(z)-f(y)\bigr) a ⋅ ( z − y ) ≤ ( − b ) ( f ( z ) − f ( y ) ) . Dividing by the positive number − b -b − b and using The Dot Product on R n \mathbb{R}^n R n Is Symmetric, Bilinear and Positive Definite ,
f ( z ) ≥ f ( y ) + p ⋅ ( z − y ) for every z ∈ B ˉ ( y , r ) . f(z)\ge f(y)+p\cdot(z-y)\qquad\text{for every }z\in\bar{B}(y,r). f ( z ) ≥ f ( y ) + p ⋅ ( z − y ) for every z ∈ B ˉ ( y , r ) .
By claim 1, p ∈ ∂ U f ( y ) p\in\partial_{U}f(y) p ∈ ∂ U f ( y ) , so ∂ U f ( y ) \partial_{U}f(y) ∂ U f ( y ) is nonempty.