Read circle lifting and winding homotopy invariance. We work with in Euclidean three-space and its antipodal map . Only the stated low-dimensional results are proved; a higher-dimensional degree, homology or general Borsuk–Ulam theorem is not assumed.
Motivation
A sphere can be projected to a plane, but no continuous projection can distinguish every pair of opposite points. The obstruction is subtler than the statement that the sphere has more coordinates. If opposite points never have the same image, subtract their images and normalize: this would construct an odd map from the sphere to a circle.
The equator of that sphere gives a contradiction. An odd circle map must wind an odd number of times. Yet the equator contracts through a hemisphere, so any map defined on the entire sphere sends it to a circle loop of winding zero. The proof connects the local operation of normalization with a global obstruction already measured by the fundamental group.
Definitions
Definition 14.2.1
(Odd maps and antipodal-free sets)
An odd or antipodal map between spaces with specified antipodal operations satisfies . A set is antipodal-free if it contains no pair ; equivalently is empty. Since no sphere point equals its negative, these are always two distinct points.
Parametrize the equator by , . For a circle-valued loop use a real lift with ; its integer winding is . Winding is invariant under homotopies of loops even when their common endpoint value moves, as previously proved by lifting the homotopy and noting that endpoint differences are continuous integers.
Distances on below are restrictions of Euclidean distance. For a nonempty closed , is continuous and vanishes exactly on . Indeed follows from the triangle inequality; zero distance gives a convergent approximating sequence whose limit lies in the closed set. For the empty set only, define the auxiliary function to be the constant 1 instead, retaining exactly its empty zero set.
Theorem/proposition
Theorem 14.2.2
(Odd circle maps have odd winding)
If is continuous and odd, its winding is an odd integer, in particular nonzero. Consequently there is no continuous odd map .
Theorem 14.2.3
(Borsuk–Ulam in dimension two)
For every continuous there is with . Therefore admits no continuous injective map into , and in particular is not homeomorphic to any planar subspace.
Theorem 14.2.4
(Three-set antipodal covering obstruction)
If is covered by three closed sets, at least one contains an antipodal pair. The same conclusion holds for a cover by three open sets. Four closed antipodal-free sets suffice, as do four open antipodal-free sets, so the number three cannot be increased in this statement.
Proof sketch or proof idea
Proof
(Odd lifts and the hemisphere contraction)
Lift the loop to . Oddness gives
The left side is continuous and the right side is a discrete subset of . It is therefore one fixed number throughout the connected half-interval. Evaluate at and , then add: . Notice that can be negative; odd winding does not require positive winding.
Suppose now that were odd. Its equatorial restriction is an odd circle map and has nonzero winding. But the explicit hemisphere homotopy
lies in , begins at and ends at the constant north pole. For each it is a loop since the endpoints agree. Composing with gives a circle-loop homotopy to a constant, forcing winding zero. This contradiction proves the first theorem without asserting that the whole sphere is contractible.
Proof X-ray
The obstruction lives on the equator
Oddness forces an odd winding on a one-dimensional loop. Extension across one hemisphere forces zero winding of the same loop. The contradiction concerns these two requirements; it does not come from counting coordinates, differentiating a map or treating the sphere as contractible.
The one non-obvious move is in the first half: the half-turn difference is continuous and takes values in the discrete set , so it is constant on the connected interval . Continuity of is used to have a continuous lift, oddness to place the difference in , and connectedness to make it constant. In the second half, continuity of on the whole sphere is what allows the hemisphere homotopy to be pushed forward; a map defined only on the equator could not use it.
Proof
(Normalize the antipodal difference)
For Borsuk–Ulam, suppose at every point. Then
is well defined and continuous into the unit circle in . Its numerator changes sign under and its denominator does not, so is odd. The first theorem rules it out. Thus a coincident antipodal pair exists. An injective map would distinguish every two distinct points, including this pair, so the nonembedding conclusion follows immediately. No differentiability or local geometric regularity of was used.
Proof
(Distance functions, a positive margin and a tetrahedral cover)
For the closed-cover result let cover and suppose none contains an antipodal pair. Use the continuous zero-set functions defined above for . Borsuk–Ulam applied to gives with , . If belongs to either , so does , contradiction. Hence belongs to neither of the first two sets; their distances at are positive, so belongs to neither either. The cover then puts both and in , also a contradiction. Empty cover members cause no difficulty because their chosen function never vanishes.
For an open cover , a member equal to the entire sphere already contains antipodal points. Otherwise every complement is a nonempty closed set, and define
At each some open member contains a ball around , so . Compactness gives a positive minimum of . The sets defined by are closed, contained in , and cover . The closed-cover result puts a pair in one and hence in the corresponding . This proves the open version rather than assuming every open cover is already closed.
To prove sharpness, take the four unit tetrahedron vertices
Let consist of sphere points where the scalar product with is maximal among the four. These four sets are closed and cover, allowing ties. If and both lie in , maximality for and makes all four scalar products equal. The vertices sum to 0, so their common value is 0; they span , so , impossible on the sphere. Thus each is antipodal-free.
The compact sets and have positive distance : the distance on their compact product has a minimum, and zero would mean intersection. The open neighborhood of is still antipodal-free. Indeed if and were both in it, choose within of respectively. Then contradicts the defining distance . These four open neighborhoods cover because they contain the . This completes every part of the covering theorem.
Worked examples
Worked example 14.2.5
(Projection can have just one antipodal coincidence pair)
For , equality forces . On the only possibilities are the north and south poles, one unordered antipodal pair. The theorem guarantees at least one pair, not an entire equator of pairs or a positive-dimensional coincidence set.
Worked example 14.2.6
(The tetrahedral cover is a sharp geometric model)
The in the proof are spherical regions cut out by the inequalities . Along a boundary two or more indices may maximize, which is why these are a closed cover rather than a disjoint partition. Every point still belongs somewhere, and the scalar-product argument excludes opposite points even on shared boundaries. Enlarging them slightly gives the open four-set cover, so changing closed to open does not permit a three-set antipodal-free cover.
Worked example 14.2.7
(Target dimension and domain both matter)
The identity separates all antipodal pairs; the two-dimensional target hypothesis cannot be dropped. Conversely given by its usual inclusion is injective and separates antipodes, so one cannot replace the domain by an arbitrary space possessing an antipodal operation. The proof uses precisely an equator that is odd and bounds a hemisphere inside the specified domain.
Counterexample mode
Dropping continuity gives a direct counterexample
Claim: even without continuity, every map identifies some antipodal pair. For look at the first nonzero coordinate in the order . Give sign if that coordinate is positive, sign otherwise. Opposite points have opposite signs. Sending these signs to and therefore distinguishes every antipodal pair.
It is discontinuous: points with a small negative first coordinate and second coordinate tending to 1 approach , but their signs are negative while the limit sign is positive. The two sign classes also form an antipodal-free cover by arbitrary subsets. Thus neither continuity in Borsuk–Ulam nor the closed/open regularity in the covering theorem is decorative.
Worked example 14.2.8
(Coincidence need not mean a zero)
For , every antipodal pair has equal images because squaring forgets signs. The first coordinate is at least 2 and the second at least 3, so never vanishes. Shared values vary as the pair varies, and the theorem does not specify one distinguished image point. This example separates existence of a coincident pair, size of the coincidence set and existence of a zero: they are different questions. The oddness hypothesis in the exercise below, which shows that a continuous odd map has a zero, supplies an equation absent from this even map.
Common mistakes
Common mistake
(The common value need not be zero)
The theorem produces equal values at opposite domain points. It does not assert that the common value is zero, nor that every chosen pair of coordinates is independent. Translating by a constant changes the shared value without affecting the conclusion.
Common mistake
(Odd planar maps can vanish)
An odd map to can vanish; the impossibility concerns an odd map into , or equivalently a nowhere-zero odd planar map. Normalization is exactly the step that requires no zeros.
Common mistake
(The equator contracts in the sphere, not in itself)
The equator is not null-homotopic inside itself. It contracts inside the sphere, and a hypothetical map defined on the whole sphere would transport that contraction. Confusing these ambient spaces erases the core argument.
Summary
Odd circle symmetry forces odd winding; a hemisphere extension forces zero winding. Their incompatibility proves Borsuk–Ulam for and the planar nonembedding result. Distance functions turn this map theorem into a three-set cover obstruction, while an explicit tetrahedral four-set cover proves sharpness for both closed and open sets.
Exercises
Quick check
(Which integers occur as winding numbers of continuous odd maps ? Give constructions for all possibilities.)
Solution · Answer
The theorem excludes all even integers. For any odd integer , positive or negative, is continuous on the circle, satisfies , and has winding . These are exactly the possibilities; no claim that different maps of the same winding are identical is made.
Quick check
(Prove the one-dimensional statement: every continuous takes equal values at some antipodal pair.)
Solution · Answer
Put . It is continuous and . If we are done; otherwise and have opposite signs, so the intermediate value theorem gives a zero between them. This proof uses a scalar difference, not the unproved higher-dimensional theorem.
Quick check
(Show any cover of by two closed sets has one containing an antipodal pair. Handle an empty member explicitly.)
Solution · Answer
If one set is empty, the other is the whole circle and contains every pair. Otherwise let be the two closed sets and use . The previous solution supplies with equal distances. If one lies in , both do. If neither does, the cover puts both in . The zero-set property is why closedness matters.
Quick check
(If is continuous and odd, prove it has a zero. Why can the zero value be asserted here but not for arbitrary ?)
Solution · Answer
Borsuk–Ulam gives , while oddness gives . Hence . Without oddness a constant nonzero already has coincident antipodal values everywhere and no zero, showing exactly which additional equation was needed.
Quick check
(Show every continuous odd is onto. Explain why merely seeing an odd winding at a few sampled points is not a proof.)
Solution · Answer
If a circle point is omitted, the remaining punctured circle has a continuous single-valued argument in an interval of length 1. Composing with that argument gives a lift whose values at the two parameter endpoints agree, so winding 0, contradicting oddness. A finite sample does not determine a continuous lift or exclude hidden additional winding between samples; the proof requires the entire continuous map and its exact symmetry.