The Gold Standard of Vectors | Why Affine Point Weights Must Sum to One

A point names a location. A free displacement vector describes a directed change between locations. Both can be written as two numbers after a coordinate system is chosen, but their meanings remain different.

Suppose A has coordinates (2, 4) and B has coordinates (10, 8), measured in metres on perpendicular axes with equal unit scales. The displacement from A to B is (8, 4) m. Its components describe how to move from the first point to the second.

The midpoint has coordinates (6, 6) m. Those numbers identify a location relative to the chosen origin. They do not describe the same object as the displacement (8, 4), even though both answers use ordered pairs.

Label the intended output before calculating. Are you constructing a point, combining displacement vectors or finding a scalar length? A familiar notation cannot make that decision for you. Many errors begin when a student treats every pair of numbers as the same mathematical type.

The existing How to be Good at Vectors guide provides a foundation route. Here we investigate a narrower question: which weighted combinations retain their intended point or displacement meaning when the coordinate description changes?

Contents · Next chapter

Keep A and B physically fixed. Put a new origin at the old coordinate location c = (5, −3), without rotating or rescaling the axes. The new coordinates of any point p are p′ = p − c.

Consequently A′ = (−3, 7) and B′ = (5, 11). The midpoint’s new coordinates should be (6, 6) − (5, −3) = (1, 9). Calculating the average of A′ and B′ gives exactly that result.

The displacement B′ − A′ is (8, 4), unchanged. Both endpoint coordinates lost the same c, so subtraction cancels it. The fixed physical segment did not become shorter or turn merely because the coordinate origin moved.

This is a passive change of coordinates. It differs from physically translating the points while leaving the coordinate system fixed. In the passive description coordinates lose c. In an active translation by t, point coordinates gain t. Keeping those two stories separate prevents sign errors later.

For a paper check, draw a small coordinate grid and label both origins. The original and new coordinate pairs should lead to the same marked points. The purpose of the drawing is to verify the interpretation, not to estimate a result that the algebra can establish exactly.

Contents · Previous chapter · Next chapter

Consider the coordinate expression R = 0.4A + 0.4B. In the original system it gives R = (4.8, 4.8). It looks like a weighted location between two points, but its weights add to 0.8 rather than one.

Repeating the same expression in the shifted coordinates gives 0.4A′ + 0.4B′ = (0.8, 7.2). If the original R were one fixed physical point, its new coordinates would instead be R − c = (−0.2, 7.8).

Those are different locations. The expression did not define an origin-independent construction from A and B alone. It implicitly used the chosen origin as an additional reference.

The discrepancy is (1, −0.6), which equals 0.2c. This is not a rounding error. The coefficient shortfall of 0.2 determines exactly how the unwanted origin dependence enters.

A useful diagnostic question is whether the expression might be meaningful for a different stated task. In coordinates it equals 0.4A + 0.4B + 0.2O, where O is the chosen origin point. Those three point weights sum to one. That is a legitimate construction involving O, but it is not a construction depending only on A and B. The repair must therefore identify the intended geometry before changing a coefficient.

Contents · Previous chapter · Next chapter

Let a coordinate expression be q = Σ wi pi, and let s = Σ wi be the sum of its weights. After translating the origin by c, every input point has coordinates pi′ = pi − c.

Applying the same expression gives Σ wi pi′ = Σ wi pi − (Σ wi)c = q − sc. If q represents the same physical point, its correct new coordinates are q′ = q − c.

For the construction to agree for every origin shift c, we therefore need s = 1. The condition follows from the transformation behaviour expected of a point. A zero shift alone cannot test it, because every expression agrees when c = 0.

The University of Maryland geometry notes distinguish point operations, free vectors and affine combinations. The proof and numerical cases in this article are developed independently to show the meaning of the coefficient condition.

Notice the scope of the statement. We require independence from the coordinate origin when using only the stated input points and fixed scalar weights. A coordinate-dependent construction may be perfectly intentional in another problem. The mistake is claiming geometric independence that its equation does not possess.

Contents · Previous chapter · Next chapter

For a weighted expression intended to represent a free vector, the components should remain unchanged under a pure origin translation with the same axes. The transformed expression q − sc must therefore equal q for every c. This requires s = 0.

The familiar displacement B − A has coefficients +1 and −1, whose sum is zero. So does 2B − 2A. The latter represents twice the displacement from A to B, with the same origin independence.

An expression such as A + B has coefficient sum two. Without an additional point or an explicit chosen origin, it does not have the transformation behaviour of either a free vector or a single point. Arithmetic on the coordinate lists is possible, but the intended geometric type remains unresolved.

For a constructive check, choose weights −0.25 and +0.25 on A and B. Their combination is 0.25(B − A) = (2, 1) m. Repeating it in the shifted coordinate system produces the same components.

The phrase “weights sum to zero” does not mean the resulting vector must be the zero vector. It means the translation terms cancel. Distinguish the sum of the scalar coefficients from the sum of the weighted point coordinates. Confusing those two sums loses the entire purpose of the test.

Contents · Previous chapter · Next chapter

The expression P = A + t(B − A) begins at a point and adds a displacement. Its geometric meaning is clear before any coordinates are substituted. Expanding it gives P = (1 − t)A + tB, and its point coefficients sum to one.

For A = (2, 4), B = (10, 8) and t = 0.25, the displacement B − A is (8, 4). Taking one quarter gives (2, 1), so P = (4, 5).

In the shifted frame, A′ = (−3, 7) and the displacement remains (8, 4). Adding the same quarter-displacement gives P′ = (−1, 8), which equals (4, 5) − (5, −3). The two descriptions agree.

This route is often easier to understand than starting with a weighted-point formula. It separates the starting location from the movement used to construct another location. Expanding afterward reveals why the weights take the form they do.

Ask a learner to identify the type at every step: B − A is a vector; t(B − A) is a vector; adding it to A gives a point. This small type check can repair an error that a page of unlabelled coordinate arithmetic would merely repeat.

Contents · Previous chapter · Next chapter

An affine combination has point weights summing to one. A convex combination additionally has nonnegative weights. For two distinct endpoints, convex combinations lie on the segment joining them.

The midpoint uses weights 0.5 and 0.5. The earlier quarter-point uses 0.75 and 0.25. Both are convex combinations and therefore also affine combinations.

Now take t = 1.5 in A + t(B − A). The result is (14, 10), using weights −0.5 on A and 1.5 on B. The weights still sum to one, so the point construction remains origin independent. The point lies beyond B, outside the segment.

Negative weights therefore do not automatically make a construction invalid. They change which geometric region it can occupy. Whether they are appropriate depends on the intended task. A point intended to interpolate between two endpoints should have t between zero and one; an extrapolated point may deliberately use another value.

For a useful comparison, draw three cases: t = 0.25, t = 0.5 and t = 1.5. Then ask which property all share and which property only the first two share. All have the correct affine transformation behaviour. Only the first two are guaranteed to remain within the segment under these conditions.

Contents · Previous chapter · Next chapter

Let A = (1, 2), B = (9, 2) and C = (1, 8). Choose weights 0.25, 0.5 and 0.25. Their sum is one, and the resulting point is P = (5, 3.5).

One way to check the calculation is to use A as the starting point. The vectors B − A and C − A are (8, 0) and (0, 6). Thus P = A + 0.5(B − A) + 0.25(C − A) = (1, 2) + (4, 0) + (0, 1.5).

This representation makes the remaining coefficient on A visible: 1 − 0.5 − 0.25 = 0.25. The construction starts at A and uses two independent displacements to locate P.

The table displays coordinate arithmetic, while the starting-point form explains the geometry. Keep both available when teaching. A correct table is useful; being able to reconstruct why those coefficients define a point is the stronger learning outcome.

Contents · Previous chapter · Next chapter

Using the same triangle, suppose P = (3, 5). Seek weights a, b and c satisfying P = aA + bB + cC and a + b + c = 1.

Start from A: P − A = (2, 3) = b(B − A) + c(C − A) = b(8, 0) + c(0, 6). Therefore b = 0.25 and c = 0.5. The remaining weight is a = 0.25.

These are barycentric coordinates relative to the specified ordered triangle vertices. They are coefficients tied to that frame of points, not ordinary horizontal and vertical coordinates. Changing the vertex order changes which coefficient belongs to which point.

Because A, B and C are noncollinear, the two difference vectors are independent and the representation is unique in their plane. If the three points were collinear, an arbitrary point off their line could not be represented, and points on the line could have nonunique three-weight descriptions.

Do not confuse the sum-to-one condition with a guarantee of uniqueness. The condition gives the correct affine behaviour. Independence of the defining point differences supplies the uniqueness in this example. Each assumption performs a different job and should remain visible when the diagram changes.

Contents · Previous chapter · Next chapter

Keep A = (1, 2), B = (9, 2) and C = (1, 8). Choose weights −0.25, 0.75 and 0.5. Their sum is one, and their coordinate combination is P = (7, 5).

Relative to A, this is A + 0.75(B − A) + 0.5(C − A). The two displacement fractions sum to 1.25, placing P beyond the triangle edge opposite A. The negative coefficient on A records that extrapolation.

The result is still a well-defined affine combination. It is not a convex combination of the three vertices. If a software routine is intended to keep a point inside this triangle, the negative coefficient is relevant evidence that the constraint is not satisfied.

This interpretation belongs to the stated nondegenerate triangle. For arbitrary redundant sets of points, a negative coefficient in one representation does not always establish that the point lies outside the entire convex hull, because another representation might exist. Preserve the uniqueness conditions.

For a transfer exercise, ask the learner to construct a point outside a different edge by making another vertex weight negative while keeping the sum one. Then ask them to explain which edge is crossed. That turns a coefficient test into an interpretable geometric relationship rather than a rule about disallowed minus signs.

Contents · Previous chapter · Next chapter

The arithmetic mean of n point coordinates uses n weights, each equal to 1/n. Their sum is one, so the resulting point changes correctly under an origin translation.

For points A = (1, 2), B = (9, 2) and C = (1, 8), the equal-weight mean is (11/3, 4). Moving the origin by (4, 1) should change this to (−1/3, 3), and averaging the shifted input coordinates gives that same result.

An unequal weighted average can also be affine. If raw positive weights are 2, 3 and 5, dividing them by their total ten gives coefficients 0.2, 0.3 and 0.5. These coefficients define a point without privileging the coordinate origin.

The normalisation step has meaning. Using the unnormalised expression 2A + 3B + 5C and calling it the same average would leave the coefficients summing to ten. Its coordinate transformation would include ten copies of the origin shift rather than one.

Check what the weights represent in an application. Mass weights, confidence weights and interpolation weights may serve different purposes. The sum-to-one condition ensures a particular geometric behaviour; it does not establish that the chosen weighting scheme is scientifically or practically justified for every task.

Contents · Previous chapter · Next chapter

Suppose a correct displacement is v = B − A. Its coefficients sum to zero. Trying to divide those coefficients by their sum to force a point average would require division by zero and would destroy the intended type.

Similarly, u = 2C − A − B is an origin-independent free vector because its coefficients sum to zero. It can be interpreted as twice the displacement from the midpoint of A and B to C. No point-normalisation step is needed.

Using the triangle coordinates, u = 2(1, 8) − (1, 2) − (9, 2) = (−8, 12). The midpoint of A and B is (5, 2), so twice the vector from that midpoint to C is 2(−4, 6) = (−8, 12), agreeing with the expression.

This is a useful repair task because the same coefficients can look unfamiliar while the geometry is simple. Ask what point or movement the expression actually describes before manipulating its weights.

A blanket instruction to “make every set of weights add to one” is therefore incorrect. Point constructions and free-vector constructions transform differently. The right constraint follows from the object you intend to represent. Keeping that object label visible is more reliable than applying one normalisation habit to every coordinate list.

Contents · Previous chapter · Next chapter

Now physically translate every input point by t while keeping the axes fixed. The transformed coordinates are F(p) = p + t. Applying weighted coefficients gives Σ wiF(pi) = Σ wi pi + st, where s is their sum.

For a point construction with s = 1, the output gains exactly t. The constructed point moves with the input geometry. For a free-vector construction with s = 0, the output remains unchanged because a uniform physical translation does not change endpoint displacements.

For the triangle point P = (5, 3.5), translate every vertex by t = (−2, 4). Reusing weights 0.25, 0.5 and 0.25 gives Pnew = (3, 7.5), exactly P + t.

For the vector B − A = (8, 0), translating both vertices by that same t leaves (8, 0). The physical points moved; the connecting displacement did not change.

Compare this with the earlier passive coordinate shift, in which fixed points lost c from their coordinate descriptions. The signs differ because the stories differ. A good solution states which change is being made, then derives the appropriate coordinate update. The coefficient conditions agree in both stories because they express the same distinction between point and vector behaviour.

Contents · Previous chapter · Next chapter

An affine map has coordinate form F(p) = Lp + t, where L is a linear map and t a translation vector. Applying it to an affine combination gives F(Σ wi pi) = LΣ wi pi + t.

Combining the transformed inputs instead gives Σ wiF(pi) = LΣ wi pi + (Σ wi)t. These agree when the point weights sum to one. Thus the same coefficients locate the corresponding point after an affine transformation.

This is a statement about combinations, not a claim that every affine map preserves distances or angles. A shear or nonuniform scaling can change those metric properties while still preserving the affine-combination relationship.

For an original example, use F(x, y) = (2x + y + 3, y − 1). With A = (1, 2), B = (9, 2), C = (1, 8) and weights 0.25, 0.5 and 0.25, P = (5, 3.5) maps to (16.5, 2.5).

The mapped vertices are (7, 1), (23, 1) and (13, 7). Combining them with the same weights gives (16.5, 2.5). The agreement verifies this example. The general algebra explains why it works for every input point satisfying the stated combination rule.

Contents · Previous chapter · Next chapter

The coefficient sum tells us how a construction behaves under affine coordinate relationships. It does not by itself provide a distance formula, an angle or a physical unit conversion.

For example, the midpoint of two points remains their midpoint after an affine map. But the distance between the mapped endpoints may differ from the original distance. If one axis is scaled by two and the other by one, lengths can change by different factors depending on direction.

A learner may correctly preserve the weights while incorrectly claiming that a triangle’s side lengths stayed fixed. Those are separate statements requiring separate checks. The vector root’s treatment of bases and transformations provides the broader context.

Keep the coordinate system’s geometry attached to length calculations. The familiar square-root-of-squares formula assumes an orthonormal Euclidean basis. Affine coefficients can be valid even when that convenient metric formula is not valid for the displayed coordinate components.

This distinction helps when moving from school diagrams into graphics, maps or transformed illustrations. The same point relationship may survive while an apparent angle or measured screen distance changes. Ask which property the equation establishes, then avoid adding a stronger metric conclusion merely because the calculation looked familiar.

Contents · Previous chapter · Next chapter

As an extension, a common affine-computing representation stores a two-dimensional point as (x, y, 1) and a displacement as (u, v, 0). The final coordinate distinguishes their transformation roles in this restricted setting.

A weighted combination of stored points has final coordinate equal to the sum of its weights. If that sum is one, it retains the point form. If it is zero, it has the displacement form. The bookkeeping mirrors the origin-shift derivation.

For the midpoint of A = (2, 4) and B = (10, 8), averaging (2, 4, 1) and (10, 8, 1) gives (6, 6, 1). Subtracting the first stored point from the second gives (8, 4, 0).

Do not treat this short introduction as a complete account of projective geometry or homogeneous-coordinate equivalence. Other conventions and transformations require additional interpretation, including handling a nonunit final coordinate. Here the extra component is being used to make the affine type distinction visible.

For practical code review, ask whether the system distinguishes point and displacement inputs or merely stores both in one generic numeric array. A shared representation does not guarantee shared meaning. The mathematical type should remain available through naming, documentation or an appropriate data structure.

Contents · Previous chapter · Next chapter

Use A = (−2, 1), B = (6, 1) and C = (2, 9). Let P have weights 0.5, 0.25 and 0.25. Compute P, then move the origin to c = (3, 2) and repeat the weighted calculation.

The original result is P = (1, 3). The shifted vertex coordinates are (−5, −1), (3, −1) and (−1, 7). Their weighted combination is (−2, 1), which equals P − c.

Now compute v = C − 0.5A − 0.5B. The coefficient sum is zero, so it should be origin independent. In the original frame, the midpoint of A and B is (2, 1), giving v = (0, 8). The shifted-frame calculation gives the same vector.

For a final variation, use weights 0.5, 0.25 and 0.20 on the three points. Their sum is 0.95. The weighted coordinate result no longer has the transformation behaviour of a point construction from those three points alone. Predict the discrepancy before calculating: it is 0.05c between the recomputed expression and the correctly shifted original coordinate result.

Explain each conclusion with its intended output type. The arithmetic is short; identifying why the transformation check should succeed or fail is the transferable part.

Contents · Previous chapter · Next chapter

“Any weighted sum of point coordinates is automatically a point.” In a chosen coordinate system the arithmetic gives a list of numbers, but an origin-independent affine construction needs the point weights to sum to one. Otherwise an additional reference may be implicit.

“Negative weights are always a mistake.” Negative weights can define valid affine extrapolations. Nonnegative weights are an additional requirement for a convex combination, not the definition of every valid point construction.

“Weights summing to zero give the zero vector.” They give origin-independent free-vector behaviour under the stated combination rule. The resulting vector can be nonzero, as B − A demonstrates.

“Keeping barycentric weights preserves all geometry.” It preserves the affine-combination relationship under affine maps. Distance, angle and other metric properties require their own conditions.

For each repair, ask the learner to supply one numerical counterexample and one general equation. The counterexample makes the misunderstanding visible. The equation explains the condition that replaces it. This is stronger than memorising a collection of exceptions without seeing how they follow from point and vector transformation behaviour.

Contents · Previous chapter · Next chapter

Suppose an intended point average uses three coefficients equal to one third. A report rounds each to 0.333, giving a displayed sum of 0.999. If those rounded numbers are used as the actual computational coefficients, the expression no longer has exact affine point behaviour.

For a coordinate-origin shift c = (1000, 2000), the discrepancy between the recomputed expression and the correctly shifted original result is (1 − 0.999)c = (1, 2). The large shift amplifies a small coefficient-sum error.

This is a numerical representation issue layered on top of the geometry. Displaying rounded weights does not necessarily mean a program used rounded weights internally. Keep the stored precision separate from the printed precision when reviewing a result.

If the intended weights are known exactly, preserve an exact or sufficiently accurate computational representation and check the sum under the actual numerical method. Do not normalise arbitrary data automatically without confirming that they were intended to define a point combination.

The lesson is practical: the origin-invariance equation can diagnose a small numerical error as well as a large conceptual one. The expected transformation behaviour gives a testable property, rather than relying on whether a plotted point appears close enough at one convenient scale.

Contents · Previous chapter · Next chapter

Return to R = 0.4A + 0.4B. One repair is to change the weights to match an intended two-point average. Another is to acknowledge a third point: the selected origin O, with coefficient 0.2.

Those repairs describe different constructions. Replacing the expression with 0.5A + 0.5B gives the midpoint. Keeping 0.4A + 0.4B + 0.2O gives a point pulled toward the specified O. The first depends only on A and B; the second deliberately depends on three points.

If O is an actual fixed landmark rather than whichever coordinate origin happens to be selected, it too must receive new coordinates when the description changes. Replacing its coordinates with zero again after moving the axes would silently change the landmark and therefore the problem.

Ask the learner to list every geometric input to the construction. A formula that seems mysterious often becomes clear when a hidden reference point is made explicit. The goal is not to ban coordinate-dependent definitions; it is to ensure that the claimed dependencies match the actual equation.

Contents · Previous chapter · Next chapter

Must the weights always be fractions between zero and one?

No. They must sum to one for an affine point combination. Requiring every weight to be nonnegative gives a convex combination. Negative weights can describe extrapolation.

Can a point have more than one weighted representation?

Yes, when the defining point set is redundant or degenerate. Three noncollinear vertices in a plane give unique barycentric coordinates there. Collinear or additional redundant points change that uniqueness question.

Why not add points as ordinary vectors?

You can calculate with their position coordinates after selecting an origin. The question is whether the resulting expression has the geometric meaning claimed independently of that origin. The coefficient-sum test reveals the dependence.

Does this belong to every introductory school course?

No. It is an extension connecting familiar endpoint subtraction and ratio constructions with affine geometry. Use the foundation guides first if the distinction between a point and a displacement is still unclear.

Contents · Previous chapter · Next chapter

For foundations, return to Vectors, How to be Good at Vectors and Coordinate Geometry. For a graphics application, read Triangle Rasterisation, Barycentric Coordinates and Pixel Coverage. That guide’s rendering problem is distinct from the origin-independence question developed here.

For formal context, use the University of Maryland geometry and geometric-programming notes and the University of Waterloo graphics notes. The original examples in this article let readers reconstruct the coefficient conditions rather than simply adopt a definition.

The wider connection is Learning That Transfers into Life within How X Works. Choose a new origin, repeat one point construction and one displacement construction, and explain why their coordinates respond differently. When that explanation remains clear after the numbers change, the vector idea has become usable.

Contents · Previous chapter · Return to Vectors