Convexity is the geometric expression of “no dents.” It is a simple condition with a surprisingly rigid algebra around it. Once a set is convex, linear functionals see it cleanly, intersections behave predictably, and many existence questions reduce to finite combinatorics. Convex geometry is therefore less a catalog of shapes than a toolkit: a small collection of principles that can be combined to control problems about feasibility, approximation, and structure.
Throughout, work in $\mathbf R^d$ with its usual affine structure. A \subset $K\subset \mathbf R^d$ is convex if for every $x,y\in K$ and every $t\in[0,1]$, the point $(1-t)x+ty$ lies in $K$. The convex hull $\mathrm{conv}(S)$ of a set $S$ is the smallest convex set containing it, equivalently the set of all finite convex combinations of points of $S$.
Popular Streaming Pick4K Streaming Stick with Wi-Fi 6Amazon Fire TV Stick 4K Plus Streaming Device
Amazon Fire TV Stick 4K Plus Streaming Device
A mainstream streaming-stick pick for entertainment pages, TV guides, living-room roundups, and simple streaming setup recommendations.
- Advanced 4K streaming
- Wi-Fi 6 support
- Dolby Vision, HDR10+, and Dolby Atmos
- Alexa voice search
- Cloud gaming support with Xbox Game Pass
Why it stands out
- Broad consumer appeal
- Easy fit for streaming and TV pages
- Good entry point for smart-TV upgrades
Things to know
- Exact offer pricing can change often
- App and ecosystem preference varies by buyer
Supporting hyperplanes and the separation principle
The first major theorem is that convex sets can be “touched” by hyperplanes in a way that reflects their geometry.
A hyperplane is a level set $H=\{x\in \mathbf R^d: \langle u,x\rangle=\alpha\}$ for some nonzero vector $u$ and scalar $\alpha$. The corresponding closed half-spaces are $\langle u,x\rangle\le \alpha$ and $\langle u,x\rangle\ge \alpha$. A hyperplane supports a convex set $K$ if $K$ lies in one closed half-space and intersects $H$.
A finite-dimensional separation theorem can be phrased in several equivalent forms. One of the most useful is:
If __GCNKDDTOK_0__) and __GCNKDDTOK_1__) are disjoint convex sets and __GCNKDDTOK_2__) is compact while __GCNKDDTOK_3__) is closed, then there exists a hyperplane that strictly separates them.
A geometric proof uses the distance function. Since $K$ is compact and $L$ is closed, the function $x\mapsto \mathrm{dist}(x,L)$ achieves its minimum on $K$, and there exist points $x_0\in K$, $y_0\in L$ realizing the minimal distance between the sets. Consider the vector $u=x_0-y_0$. A short argument using convexity shows that
Thus the hyperplane through the midpoint $(x_0+y_0)/2$ orthogonal \to $u$ separates $K$ and $L$.
Even when hypotheses for strict separation fail, weak separation often still holds. This is the geometric avatar of the Hahn–Banach principle: linear functionals can extend in a way that respects convex constraints. In finite dimensions, the geometric picture is often the most efficient way to remember what the theorem is really saying.
Extreme points, faces, and the geometry of linear functionals
A linear functional $\ell(x)=\langle u,x\rangle$ on a compact convex set $K$ attains its maximum, and the maximizers form a face of $K$. A face $F\subset K$ is a convex \subset with the property that if a line segment in $K$ has an interior point in $F$, then the entire segment lies in $F$. Faces are exactly the sets cut out by supporting hyperplanes.
Extreme points are the zero-dimensional faces: a point $p\in K$ is extreme if it cannot be written as a nontrivial convex combination of two other points in $K$. The set of extreme points can be complicated, but in finite dimensions there is a basic reconstruction principle: every compact convex set is the convex hull of its extreme points. This is a finite-dimensional reflection of the Krein–Milman theorem.
For polytopes, faces and extreme points are explicitly combinatorial. If $P$ is the convex hull of finitely many points, then its extreme points are among those generators, and its faces correspond to subsets selected by linear inequalities. This is why linear optimization is naturally geometric: maximizing a linear functional over a polytope forces the solution to occur on a face, often at a vertex.
Carathéodory and the finite support phenomenon
Convex combinations are defined by allowing arbitrarily many points in principle, but in $\mathbf R^d$ there is a sharp bound on how many are needed.
Carathéodory’s theorem. If $x\in \mathrm{conv}(S)\subset \mathbf R^d$, then $x$ can be written as a convex combination of at most $d+1$ points of $S$.
A proof uses affine dependence. If $x$ is written as a convex combination of many points, then the corresponding vectors are affinely dependent once there are more than $d+1$ of them. That means there exist coefficients $\alpha_i$, not all zero, with $\sum \alpha_i=0$ and $\sum \alpha_i s_i=0$. Perturb the convex coefficients along $\alpha_i$ while keeping the sum and barycenter fixed. One can push in the positive direction until some coefficient hits zero, reducing the number of points without changing $x$. Repeating yields a representation with at most $d+1$ points.
This finite support principle shows up everywhere. It says that convexity in $\mathbf R^d$ is fundamentally finite-dimensional, even when the set $S$ is infinite.
Radon’s theorem and the first combinatorial collapse
Radon’s theorem is the first strong combinatorial statement in convexity.
Radon’s theorem. Any set of $d+2$ points in $\mathbf R^d$ can be partitioned into two disjoint subsets whose convex hulls intersect.
A standard proof again uses affine dependence. Given $d+2$ points $p_1,\dots,p_{d+2}$, there exist scalars $\alpha_i$ not all zero with $\sum \alpha_i=0$ and $\sum \alpha_i p_i=0$. Split indices into $I_+=\{i:\alpha_i>0\}$ and $I_-=\{i:\alpha_i<0\}$. Rewrite the dependence as
and normalize both sides so coefficients sum to one. This produces a point that lies in $\mathrm{conv}(\{p_i:i\in I_+\})$ and also in $\mathrm{conv}(\{p_j:j\in I_-\})$, giving the intersection.
Radon’s theorem is the engine behind Helly’s theorem and other intersection results. It is a statement about how convex hulls must overlap once there are “too many” points relative to the dimension.
Helly’s theorem: intersection from local data
Helly’s theorem is a cornerstone because it turns a global intersection problem into a finite test.
Helly’s theorem. Let $\mathcal F$ be a finite family of convex subsets of $\mathbf R^d$. If every subfamily of size at most $d+1$ has nonempty intersection, then the whole family has nonempty intersection.
A proof can be built from Radon’s theorem. One approach is a minimal counterexample argument: assume the theorem fails and choose a counterexample family of minimal size. In such a family, the intersection of all sets is empty, but the intersection of any proper subfamily is nonempty. Pick for each set $K_i$ a point $x_i$ that lies in the intersection of all the other sets. Consider the set of points $\{x_i\}$. By Radon’s theorem, partition them into two subsets with intersecting convex hulls. Convexity then forces the corresponding subfamilies to have a common point, contradicting minimality. The logic is delicate but the geometric mechanism is simple: affine dependence forces overlap, and overlap transfers to intersections of convex sets.
Helly’s theorem has immediate consequences in computational geometry and feasibility. If a system of convex constraints in $\mathbf R^d$ is infeasible, there is already an infeasible subsystem with at most $d+1$ constraints. In small dimension this means infeasibility has a succinct certificate.
Duality via polars and support functions
Convex geometry becomes more powerful when paired with duality. For a convex body $K$ containing the origin in its interior, its polar is
This exchanges containment with reverse containment: if $K\subset L$, then $L^\circ\subset K^\circ$. It also turns supporting hyperplanes of $K$ into boundary points of $K^\circ$, so geometric extremality becomes dual to functional constraints.
Another dual viewpoint is the support function
For compact convex $K$, the support function determines $K$ uniquely, because
Support functions convert Minkowski sums into addition:
Thus many geometric operations become linear after passing to support functions, and inequalities about convex bodies often become inequalities about functions on the sphere.
This duality is a reason separation theorems are so central: separation is exactly the statement that linear functionals detect disjointness or boundary behavior of convex sets.
A small catalog of reusable proof moves
Convex geometry is not learned by memorizing theorems alone. It is learned by internalizing a few proof moves that reappear in different disguises.
* Reduce to finite data using Carathéodory: if a statement is about membership in a convex hull, look for a $d+1$-point witness.
* Produce intersection points using Radon: affine dependence yields a point represented two ways.
* Convert global intersection questions to local ones via Helly: infeasibility has a small certificate.
* Turn geometry into inequalities by choosing an appropriate functional: supporting hyperplanes and support functions are the correct probes.
* Use compactness to extract extremal points: many separation arguments are hidden “minimize distance” arguments.
These moves interact well. For example, a typical feasibility argument uses Helly to reduce \to a small subfamily, then uses separation to produce a hyperplane certificate.
Worked picture: feasibility and certificates
Consider the problem of deciding whether a point lies in the intersection of a collection of convex sets $K_1,\dots,K_m\subset \mathbf R^d$. If the intersection is empty, Helly says there exist indices $i_1,\dots,i_{d+1}$ such that $K_{i_1}\cap\dots\cap K_{i_{d+1}}=\varnothing$. This is already a finite certificate of infeasibility. In many settings, separation strengthens this: one can find a hyperplane that separates one of the sets from the intersection of the others, giving an explicit inequality certificate.
Even when the sets are described implicitly, this viewpoint is stabilizing. Convexity turns feasibility into a problem where the dimension controls the complexity of minimal obstructions. That is a geometric statement, but it has algorithmic consequences.
Convexity as geometry with linear algebra inside
Convex geometry earns its role as a toolkit because it sits at the interface of affine structure and linear functionals. The shape data is encoded by intersection patterns of half-spaces; the extremal data is encoded by where functionals attain maxima; the combinatorial data is encoded by how many points are needed to witness membership.
The unifying theme is that convexity aligns geometry with linear algebra. Once a set has no dents, the linear probes are honest, and the dimension gives sharp bounds on how complicated witnesses must be. Separation, Helly, and duality are three faces of that alignment.
Books by Drew Higgins
Leave a Reply