The Gold Standard of Vectors | Parallel Transport Around a Sphere Without a Globally Fixed Direction

In a flat introductory diagram, we often slide an arrow around without changing the vector it represents. This is useful for free displacement vectors in a Euclidean plane. A tangent vector on a surface carries another piece of information: the point where it is attached. Its permitted directions belong to the tangent plane at that point.

Use the unit sphere centred at the origin in three-dimensional Euclidean space. A point p on it satisfies p·p = 1. A vector v tangent there satisfies p·v = 0. The dot is the ordinary Cartesian dot product. These two small equations identify the surface and the plane in which the arrow is allowed to lie.

At N = (0,0,1), the tangent plane contains vectors (a,b,0). At A = (1,0,0), it contains vectors (0,b,c). An arrow with components (1,0,0) is tangent at N but is normal to the sphere at A. Keeping its three numbers unchanged during that move would therefore fail the tangent condition.

Before calculating, write the pair consisting of the point and its arrow. This keeps a location statement from quietly becoming a direction statement. It also explains why the root guide’s free-arrow intuition needs an additional comparison rule in this setting.

Contents · Next chapter

All our arrows can be written in one surrounding three-dimensional coordinate system. That lets us subtract their component lists as ordinary vectors in that surrounding space. It does not make that subtraction the intrinsic comparison between tangent directions at different surface points.

Consider carrying the north-pole arrow (1,0,0) towards A. Holding it fixed relative to the surrounding room is a perfectly intelligible instruction. Requiring it to remain tangent to the sphere is another instruction. The two demands cannot both hold throughout this route. A clear explanation must choose which geometric job the arrow is doing.

Parallel transport supplies a particular surface comparison. We use the Levi-Civita connection of the sphere’s usual metric, rather than an arbitrary rule invented for this exercise. More advanced geometries can use other connections. Naming the rule matters because the word parallel alone cannot determine every possible way of comparing separated tangent spaces.

The existing Differential Geometry explanation introduces that wider framework. Here we make one concrete case inspectable. The aim is not to make the surrounding space disappear, but to use its coordinates while distinguishing the surface question from the room-fixed question.

Contents · Previous chapter · Next chapter

Let p(s) move smoothly on the unit sphere, and let v(s) remain tangent at p(s). For our chosen connection, parallel transport means that the ordinary derivative v′(s) has no component in the local tangent plane. Any change of the arrow in surrounding coordinates points along the surface normal.

This condition is often written as a zero covariant derivative along the path. We can work without that notation initially: project v′ onto the tangent plane and require the result to be zero. The normal to the unit sphere at p is p itself, so the rule says v′ = λp for some scalar λ.

Differentiate the tangent condition p·v = 0. The product rule gives p′·v + p·v′ = 0. Since p·p = 1, substituting v′ = λp yields λ = −p′·v. Consequently the transport equation is v′ = −(p′·v)p. This is a local equation, evaluated along the specified path.

The University of Mannheim’s Metrics and Connections notes provide the formal connection and projection setting. Our calculations below use the displayed equation directly, so the reader can check what the chosen rule actually does.

Contents · Previous chapter · Next chapter

A useful rule should come with checks that do not depend on a persuasive sketch. Differentiate the squared length: (v·v)′ = 2v·v′. Substituting the transport equation gives −2(p′·v)(v·p), which is zero because v is tangent. A transported arrow therefore keeps its length.

Now transport two tangent vectors v and w along the same route. The derivative of v·w is v′·w + v·w′. Each term contains the dot product of a tangent vector with p, so both vanish. Their inner product remains constant. For nonzero arrows, preserving both lengths and their dot product preserves the angle between them.

These statements compare arrows transported together along one route. They do not say that an arrow must retain its angle with an independently chosen compass direction, coordinate basis or later segment of the path. Those reference directions may change according to different rules.

The zero vector also transports consistently, remaining zero, although it has no direction angle. Preserve that exception when discussing angle conservation. Length preservation and an eventual change of direction at the starting point can coexist; the forthcoming loop will show exactly how.

Contents · Previous chapter · Next chapter

Take N = (0,0,1), A = (1,0,0), and B = (0,1,0). Our route goes from N to A, then A to B, then B back to N. Each leg follows the shorter quarter of a great circle. The three points bound one octant of the sphere.

Attach v₀ = (1,0,0) at N. It is tangent because N·v₀ = 0, and its length is one. Its direction initially points along the first leg towards A. We will keep the same fixed Cartesian axes throughout, avoiding a coordinate change at each corner.

The route has corners. Transport is defined along each smooth leg and composed across them, with the vector carried continuously through the corner. Turning onto a new path does not authorise an extra rotation of the arrow. This is a mathematical piecewise-smooth path, not a description of how a person’s body automatically turns while walking.

Write an empty table with columns for point, vector and tangent check. Predict the final result before reading on, but leave room to revise the prediction. The exercise becomes useful when the calculation resolves a specific expectation, rather than when a surprising answer is merely announced.

Contents · Previous chapter · Next chapter

Parameterise the first leg by p₁(s) = (sin s,0,cos s), where s runs from zero to π/2. It starts at N and ends at A. Its unit tangent is p₁′(s) = (cos s,0,−sin s).

Choose v₁(s) = (cos s,0,−sin s). At the start it is our assigned arrow (1,0,0). Its derivative is (−sin s,0,−cos s) = −p₁(s). Also p₁′·v₁ = 1. Therefore v₁′ = −(p₁′·v₁)p₁, exactly the parallel-transport equation.

At A the arrow is (0,0,−1). Its length remains one, and it is tangent to the sphere because its x-component is zero. Its surrounding-space direction has changed substantially. That is required for an arrow that follows this surface while remaining locally parallel under the chosen rule.

Notice what was proved rather than assumed. Following the path tangent works on this great-circle leg because its derivative is purely normal to the sphere. Following the tangent of an arbitrary curved path on a surface is not automatically parallel transport. The reason matters more than the visual resemblance to going straight.

Contents · Previous chapter · Next chapter

For the second leg, use p₂(s) = (cos s,sin s,0), again with zero ≤ s ≤ π/2. It moves from A to B along the equator. The incoming arrow is (0,0,−1), so try keeping that vector constant on this leg.

Its derivative is zero, which certainly has no tangent component. More explicitly, p₂′ = (−sin s,cos s,0), and p₂′·v = 0. The right side of the transport equation is therefore zero as well. The constant arrow satisfies the rule and remains tangent because every point on this leg has zero z-coordinate.

At B we still have (0,0,−1). The traveller’s path tangent has changed from the incoming leg’s direction to the equatorial direction. The transported arrow did not receive an additional turn to follow it. Confusing those two directions is a common source of an incorrect cancellation at the corners.

A simple check is to compare the arrows immediately before and after the corner at A in the same Cartesian coordinates. They should agree. If a proposed method abruptly replaces the arrow with the new route tangent, it is solving a different problem and must explain that extra operation.

Contents · Previous chapter · Next chapter

Parameterise the final leg by p₃(s) = (0,cos s,sin s), from B back to N. Its tangent is p₃′ = (0,−sin s,cos s). The incoming arrow is opposite that tangent at the start, so take v₃(s) = (0,sin s,−cos s).

Differentiate: v₃′ = (0,cos s,sin s) = p₃. Also p₃′·v₃ = −1. Thus −(p₃′·v₃)p₃ = p₃, verifying transport again. At the end, p₃ = N and the arrow is v_final = (0,1,0).

The initial and final arrows now live in the same tangent plane at the same point. Their dot product is zero, and each has unit length. They differ by a quarter-turn. With positive rotation defined from the positive x-direction towards the positive y-direction in this plane, the final rotation is +90°.

The table records one fully specified mathematical journey. The rotation belongs to this route and transport rule; it is not an arbitrary adjustment made at the endpoint.

Contents · Previous chapter · Next chapter

Nothing in the local equation required the final arrow to equal the initial arrow after a loop. It required a particular derivative at each point. The changing tangent planes make those local conditions accumulate into a nontrivial overall comparison.

The transport map around a loop is called its holonomy map. In our two-dimensional tangent plane, this example acts as a rotation. The word names the completed effect; the three leg calculations explain it. A label should not replace the chain of reasoning that makes the result credible.

There is also no violation of angle preservation. If we had transported a second arrow alongside the first, their mutual angle would have remained unchanged. At the end both would be rotated relative to the fixed starting basis. Compare like with like: two travelling arrows are different reference objects from one travelling arrow and one arrow held at the starting point.

The Riemannian Geometry guide connects this phenomenon with curvature. This article supplies a concrete bridge from elementary components to that statement. The distinction to carry forward is between preserving a relationship locally along a route and returning to a globally unchanged direction.

Contents · Previous chapter · Next chapter

The first calculation used a conveniently aligned unit arrow. To test whether the result was a special accident, start with v₀ = (a,b,0) at N. The transport equation is linear in v along a fixed route, so we can transport the x- and y-basis arrows separately and combine the results.

We already know that (1,0,0) finishes as (0,1,0). Start instead with (0,1,0). It remains constant on the first leg because it is perpendicular to that leg’s plane. On the equator it follows the tangent, becoming (−1,0,0) at B. That vector remains constant on the last leg.

Consequently (a,b,0) finishes as (−b,a,0). This is the familiar quarter-turn map in the north-pole tangent plane. It preserves a² + b². For the original changed example (3,4,0), the result is (−4,3,0), and both arrows have length five.

This extension checks a structural claim. It shows how one loop acts on every tangent vector at the chosen point, including the zero vector, rather than only on the arrow drawn in the first diagram. The calculation also prepares the reader to treat transport as a linear map between tangent spaces.

Contents · Previous chapter · Next chapter

Start again at N with (1,0,0), but compare two journeys ending at B. Route one goes directly from N to B along their shorter great-circle arc. Its plane is the yz-plane, so the x-directed arrow remains constant and arrives as (1,0,0).

Route two goes from N to A and then A to B. The worked calculations show that it arrives as (0,0,−1). Both resulting arrows are tangent at B, and both have unit length. They are perpendicular to each other. The starting point, destination and initial vector are the same; the routes differ.

That is path dependence in a form a learner can inspect without advanced notation. Naming only the endpoints is insufficient to specify the transport result. A calculation that produces one answer from endpoints alone has either silently selected a route or used another comparison convention.

This distinction is useful whenever local coordinate frames are compared across a curved surface. It does not mean every applied system must use this exact sphere model. It means the comparison rule and path belong in the explanation. Ask for them before deciding that two component lists ought to agree.

Contents · Previous chapter · Next chapter

Transport along a route and then exactly back along the same route returns the original vector. Reversing the path reverses the comparison map. It does not create a second independent copy of the same forward rotation.

For our loop, the forward map is H(a,b) = (−b,a). Its inverse is H⁻¹(a,b) = (b,−a). Composing them gives H⁻¹(H(a,b)) = (a,b). Starting with (3,4), the forward loop gives (−4,3); the reverse loop applied to that result restores (3,4).

If instead we start anew with (3,4) and take only the reverse loop, we obtain (4,−3). That is a negative quarter-turn under the original orientation convention. State which initial vector is used before comparing results. Reversing a route after completing it and starting a fresh reversed experiment are different sequences of operations.

This is a good place to practise map order. Write the first operation closest to the starting vector and the next operation outside it. A correct matrix or component rule can still be used incorrectly if the chronological order is lost. The verification is the actual reconstructed arrow, not a remembered slogan about going backwards.

Contents · Previous chapter · Next chapter

Let C = (cos α,sin α,0), with 0 < α < π. Follow N to A, then the equator from A to C through angle α, then the shorter meridian from C to N. Keep v₀ = (1,0,0).

The first leg still produces (0,0,−1). That vector remains constant along the equator. Parameterise the final meridian as p(s) = (cos s cos α,cos s sin α,sin s). The vector v(s) = (sin s cos α,sin s sin α,−cos s) starts with the correct incoming arrow and satisfies the same transport equation.

At N it becomes (cos α,sin α,0). The completed rotation is therefore α under the established orientation convention. Choosing α = 60° gives (1/2,√3/2,0). Choosing α = 120° gives (−1/2,√3/2,0). The original right-angle triangle is one member of this family.

This changed parameter is more informative than multiplying every coordinate by a new number. It changes the route’s geometry while preserving the reasoning structure. The learner must identify which two legs keep their earlier argument and which formula needs the new angle. That is a useful test of transfer.

Contents · Previous chapter · Next chapter

For this specific family, the spherical region between two meridians and the equator has area α on the unit sphere, when α is measured in radians. One way to see it is that the corresponding full lune has area 2α, and the equator divides it into equal halves.

Our directly calculated holonomy angle is also α. This is consistent with the curvature-area relationship for a round sphere. We have not derived the full Gauss–Bonnet theorem, and the elementary calculation should not be presented as a proof for every surface, loop or connection.

On a sphere of radius R, the analogous region has area R²α while Gaussian curvature is 1/R². Their product remains α. The same angular route therefore gives the same rotation even though its physical lengths and area scale. Holding a small physical area fixed while changing R is a different comparison.

Use the area connection to organise further questions, not to skip orientation. Signed rotation depends on the traversal convention, and angles are understood modulo a full turn when comparing final directions. For the restricted triangles used here, the component calculation keeps those choices visible and avoids ambiguity about which region or direction was intended.

Contents · Previous chapter · Next chapter

At the north pole, longitude is not a unique coordinate label. Many meridians meet there. This coordinate difficulty does not mean the tangent plane disappears or that every tangent vector loses its meaning.

Our fixed Cartesian description gives p = (0,0,1) and tangent plane z = 0. The starting and final arrows are ordinary vectors in that plane. Their angle is well defined even though a longitude-based chart may need a different local description at the pole.

A calculation using latitude and longitude can still be valid, but its basis vectors and coordinate domain must be handled properly. A formula that divides by a coordinate factor vanishing at a pole may be showing a chart limitation rather than a geometric catastrophe. The first repair is to identify the failing representation, then choose one that covers the point.

The Tensor Analysis explanation develops the distinction between geometric objects and coordinate components. In this exercise, three-component vectors offer a simple way to keep the geometry visible. They do not turn a two-dimensional tangent plane into a three-dimensional space of permitted tangent directions; the dot-product constraint still removes one degree of freedom.

Contents · Previous chapter · Next chapter

Two distinct, non-antipodal points on a round sphere determine a unique shorter great-circle arc. Antipodal points do not. Infinitely many great circles connect them with semicircles of the same length. A transport instruction that merely says take the shortest route is incomplete in that case.

For example, transport from N to the south pole can follow different meridians. The chosen meridian determines how the initial tangent arrow evolves. The differential equation is not defective; the path input has not been specified sufficiently.

Our worked legs avoid antipodal endpoints. Each is a specified quarter-circle, and the changed-angle exercise explicitly names its equatorial and meridional arcs. Retaining that information prevents an endpoint formula from silently selecting an arbitrary axis in a degenerate case.

This is a transferable mathematical habit. When a rule appears to give several answers, check whether the question supplied a unique object. A missing path, basis, orientation or boundary condition often explains the apparent conflict. Do not repair ambiguity by reporting one numerical result without stating the choice that produced it.

Contents · Previous chapter · Next chapter

A tempting numerical shortcut is to move the old arrow to the next point and project it into the new tangent plane. If q is a unit surface point, that projection is v − (v·q)q. It satisfies q·v_new = 0, so it passes the local tangent check.

It need not preserve length. In the extreme one-step move from N to A, projecting the initial arrow (1,0,0) onto A’s tangent plane gives the zero vector. Exact transport along our chosen quarter-circle instead gives (0,0,−1). The shortcut has destroyed the direction altogether.

With small steps, projection methods can approximate transport under suitable numerical treatment, but an approximation needs its own convergence and error analysis. Normalising after every projection may hide a length error while leaving direction error. It also fails if the projected vector is zero. A procedure should not be called exact because one constraint was restored.

For this lesson, the analytic quarter-circle formulae provide a reference against which a numerical experiment could be checked. The wider lesson is to test several invariants and a known solution. Passing perpendicularity alone does not establish that the intended comparison rule has been implemented.

Contents · Previous chapter · Next chapter

The sphere in this article stays fixed. We move a tangent vector’s base point along it. Stretching or shearing a whole surface is a different operation, with different questions about tangent vectors and normals.

The existing surface-normal transformation guide explains inverse-transpose behaviour and orientation under such transformations. Those rules should not be substituted for parallel transport merely because both discussions contain matrices and perpendicularity checks.

Similarly, perspective interpolation concerns how represented values are combined through a projection. It does not by itself define a connection on a sphere. A representation can be correct for its task while answering a different geometric question.

A useful comparison exercise asks what is held fixed in each problem. Here it is the surface, metric and transport rule, while the point follows a route. In a surface transformation problem, the map changes the geometry being represented. In interpolation, the task is a value between specified samples. Naming those jobs prevents similar-looking algebra from concealing a change of subject.

Contents · Previous chapter · Next chapter

If the loop feels impossible, begin with one tangent check. Ask whether (1,0,0) is tangent at N and at A. A learner who can compute the dot products but cannot interpret them needs the meaning of the tangent plane clarified before studying transport equations.

If the tangent condition is secure but the corner seems to rotate the arrow, separate the path tangent from the transported vector. Draw both at A. One describes the next motion; the other describes the arrow arriving from the previous leg. Their different jobs explain why they need not point the same way.

If the derivative rule is the obstacle, start with the equatorial leg. A constant vertical arrow is visibly tangent there, its derivative is zero, and the equation can be verified with minimal arithmetic. Then return to the first leg, where the derivative is nonzero but purely normal.

The teaching sequence follows the clarity of the Clementi mathematics reference: make the earliest missing relationship visible, stabilise it with a nearby case, and remove the prompt gradually. Advanced vocabulary is useful after the underlying comparison becomes explainable.

Contents · Previous chapter · Next chapter

Try this before using the answer: start at N with v = (2,−1,0), and follow the original N–A–B–N loop. Find the arrow after each leg, its final length, and the dot product between initial and final arrows.

Linearity gives the first-leg arrow (0,−1,−2). On the equator, the y-directed contribution follows the equatorial tangent while the vertical contribution stays fixed. At B the arrow is (1,0,−2). On the final meridian, the x-component remains fixed while the vertical contribution rotates into positive y. The final arrow is (1,2,0).

The initial and final squared lengths are both five. Their dot product is 2 × 1 + (−1) × 2 = 0. Thus this non-unit, initially oblique arrow also undergoes a quarter-turn. Checking intermediate tangency gives x = 0 at A, y = 0 at B and z = 0 at N.

If your final vector has the right length but a different direction, inspect the equatorial contribution and corner continuity. Length alone cannot distinguish the correct rotation from many incorrect rotations. A complete check uses the route-specific map as well as the preserved quantities.

Contents · Previous chapter · Next chapter

For a delayed transfer task, replace the unit sphere with a sphere of radius three, keeping the same three angular directions for its vertices. The points become (0,0,3), (3,0,0), and (0,3,0). Keep the arrow’s length five rather than multiplying it merely because the sphere is larger.

The tangent plane at a point is still perpendicular to its radius. The same spatial rotations along the corresponding great-circle arcs therefore give the same final arrow transformation. For starting arrow (3,4,0), the final arrow is (−4,3,0). Arc lengths have tripled, but the angular comparison is unchanged.

Now make a second change: keep the sphere radius fixed and choose an equatorial angle of 60°. The earlier general-angle calculation gives a 60° transport rotation. These two changes test different dependencies. One changes physical scale; the other changes the route’s angular shape.

Ask the learner to explain the prediction before calculating. If they automatically scale every vector with the sphere or keep the holonomy at 90° for every triangle, return to the specified quantities. Transfer means recognising what the changed condition actually affects, not reusing every number from the first example.

Contents · Previous chapter · Next chapter

Does locally parallel mean fixed in the surrounding room? No. On our sphere it means the derivative has zero tangent component under the specified connection. A vector can change its Cartesian components while satisfying that rule.

Does every curved-looking path cause a final rotation? No. The route and geometry matter. Going out and exactly back gives the identity map. A full great-circle circuit also returns a tangent vector to its initial direction on the round sphere, although it passes through changing tangent planes.

Must the arrow point along the path? No. Our second leg carries an arrow perpendicular to the route tangent. Any tangent starting vector can be transported; the path supplies where to compare it, not a requirement that it become the direction of travel.

Is the quarter-turn evidence of numerical error? Not in the exact worked case. Each leg satisfies the differential equation and joins continuously to the next. A numerical method should reproduce that known result within its declared approximation error.

Can another comparison rule give another answer? Yes. That is why the connection is named. This lesson concerns the usual Levi-Civita transport on the round sphere, not every conceivable connection on every surface.

Contents · Previous chapter · Next chapter

A reproducible solution records the surface and radius, the route with its traversal direction, the starting point and tangent vector, the fixed coordinate convention, and the transport rule. It then supplies enough intermediate calculations for another reader to check both the local equation and the final comparison.

For our original loop, three endpoint vectors alone would not establish parallel transport. Many arbitrary motions could connect them. The formulae along each leg show that the proposed fields satisfy the rule throughout, rather than only at their endpoints.

A useful short report can therefore include the three path formulae, the corresponding vector fields, one derivative check per leg, and the final dot product. Add the tangent and length checks as independent diagnostics. If the route has been approximated numerically, identify that approximation instead of presenting the numerical path as the analytic one.

This evidence habit scales beyond geometry. Separate the model definition, a candidate result, the conditions it must satisfy and the tests that were actually performed. A convincing picture can support understanding, but an explicit account lets a new reader reproduce the reasoning without inheriting the maker’s intuition.

Contents · Previous chapter · Next chapter

For the foundations, return to Vectors and practise components, dot products, bases and the distinction between locations and displacements. Those skills make every tangent check in this article more accessible.

For the geometry of separated tangent spaces, continue with Differential Geometry and Riemannian Geometry. For the coordinate language, use Tensor Analysis. These broader explanations introduce the machinery whose concrete effect we have calculated.

For different vector tasks, keep the surface-normal and perspective-interpolation guides distinct. Each develops its own transformation or representation question. Choose the one matching the object you actually need to compare.

The central learning check is now specific: can you explain why preserving length along a route does not force the endpoint arrow to be unchanged? Use How X Works to explore that habit in other subjects, and return to Learning Transfers into Life when planning the next independent task.

Contents · Previous chapter · Next chapter

The University of Mannheim’s Metrics and Connections supplies the formal background for metrics, connections and tangent projection. Michael Fowler’s University of Virginia lecture on Thomas Precession includes a sphere-based introduction to parallel transport; its later physics concerns a different application and is not claimed as an application model here. These source routes were checked on 23 October 2026.

The path parameterisations, general starting-arrow calculation, changed-angle cases, projection counterexample and practice values in this article are original worked teaching material. They are verified against the explicitly stated transport equation. No empirical observation or instrument performance is inferred from them.

For further study, derive the transport equation on a sphere of radius R directly by differentiating p·v = 0 with p·p = R². Then compare your result with the scale-change explanation. This gives a purposeful next extension: rebuild the rule from its constraints instead of relying on a formula copied from the unit-radius case.

Contents · Previous chapter · Return to Vectors