The view that makes a line a point

A first-year descriptive geometry course spends several weeks on a drill. Given a line drawn in two views, construct a third view in which it appears at its true length; then construct a fourth in which it appears as a single point; then, from those, the true shape of a plane face containing it.

Drawn on paper the drill is a sequence of operations with folding lines, transfer distances and rules about which measurement comes from which view. It reads like a recipe, and it is one — but the recipe is not an approximation to anything, and that makes it unusual on this site. Almost every taught construction the wrong field measures turns out to be a rule of thumb standing in for a projection. This one is a projection, written as an instruction.

One identity, read three ways #

An orthographic projection drops the component of everything along the ray direction. So for a segment $\mathbf{v}$ and a unit ray direction $\hat{\mathbf{d}}$, what survives is the part of $\mathbf{v}$ perpendicular to $\hat{\mathbf{d}}$, and its length is

$$|\mathbf{v}| \sin\theta$$

where $\theta$ is the angle between the segment and the ray. That is the whole of it, and the drill’s two headline constructions are its two extreme values.

$\theta = 90°$: true length. The ray is perpendicular to the segment, nothing is dropped, and the drawn length is the real one.

$\theta = 0$: point view. The ray is along the segment, everything is dropped, and the segment draws as a single point.

True length is a circle of directions, not a construction #

The drill’s phrasing hides something worth saying out loud. It asks the student to construct the auxiliary view that shows the line at true length, as if there were one.

There is a whole circle of them. Any direction perpendicular to the segment gives $\sin\theta = 1$, and the directions perpendicular to a given line in space form a plane — so a whole circle of unit directions, a one-parameter family, each of which draws the line at exactly its own length.

That is why the drill is a drill rather than a search. The student is not finding the direction that works; every direction in a circle works, and the folding line’s placement chooses one of them for reasons of layout — which view it can take its transfer distances from, and where the new view will fit on the sheet.

The same is true of the point view: a segment has exactly one direction along it, up to sign, so the point view is unique. True length is a circle and point view is a point, and the asymmetry is a fact about the geometry rather than about the teaching.

Why a second auxiliary is needed for a face #

A plane face needs the same argument one dimension up, and the count changes.

The area of a face’s image is its true area times the cosine of the angle between the face’s normal and the ray direction — the same dropping of a component, applied to a two-dimensional patch. The face draws at true shape only when the ray is along its normal, and that is one direction rather than a circle.

So the drill’s first auxiliary cannot reach the true shape in one step. It brings an edge of the face to true length; the second, taken along that edge, puts the face edge-on; and the third, perpendicular to that edge-on line, is along the normal. The steps are three because the constraints are three, and a course that presented the sequence as a convention would be leaving out the only reason it has to be a sequence.

What the standard three views are #

The front, top and side views are auxiliary views with the direction chosen once and for all, and the choice is made for exactly the property this essay is about.

A rectilinear part has its edges along three perpendicular directions. Look along one of them, and the other two are at $\theta = 90°$ — true length, both of them — while the third is at $\theta = 0$ and draws as a point. So each standard view shows two of the part’s three families of edges at true length, which is what makes the drawing dimensionable with a ruler.

An axonometric view gives that up on purpose. Looking along a direction off all three axes puts every family at some intermediate $\theta$, so nothing is at true length and every edge is foreshortened by its own factor — which is what the axis scales are, and why they satisfy an identity rather than being free.

The face accounting, which is the same identity one dimension down #

The edge identity has a companion nobody writes beside it, and having both makes the drill’s structure obvious rather than conventional.

Write the unit ray direction as $\hat{\mathbf d} = (d_1, d_2, d_3)$. An edge along axis $i$ draws at $s_i = \sin\theta_i = \sqrt{1 - d_i^{2}}$, and a face whose normal is axis $i$ draws with area scale $a_i = |d_i|$. So

$$\sum_i s_i^{2} \;=\; 3 - 1 \;=\; 2, \qquad \sum_i a_i^{2} \;=\; 1.$$

Edge scales squared sum to two; face scales squared sum to one. One identity for the three directions a solid’s edges run in and one for the three planes its faces lie in, and they are the same statement of what an orthographic projection keeps — two of the three dimensions — counted once per edge family and once per face family.

The face version settles several things at a glance. A face at true shape needs $a_i = 1$, which forces the other two to zero: one face at true shape means the other two edge-on, which is exactly the elevation and is why a standard view shows one face square and two as lines. And an isometric direction has $|d_i| = 1/\sqrt3$ throughout, so every face draws at 0.5774 of its area while every edge draws at 0.8165 of its length — the two familiar isometric numbers, and now visibly the same number squared and subtracted from one.

True length is common and true shape is rare, by a factor of six #

The asymmetry the essay notes — a circle of true-length directions against a single point-view direction — has a dual, and putting the four together gives the pattern:

A line and a face’s normal are the same kind of object, so the two rows of each kind are one row. What is genuinely asymmetric is the pairing: the thing that is easy for a line is edge-on for a face, and the thing that is hard for a face is a point view for a line.

That converts into a number if the conditions are relaxed from exact to useful. Allowing five per cent:

A line draws within five per cent of true length over 31 per cent of all directions — the band $|\cos\theta| < 0.312$, whose area on the sphere is 0.312 of the whole.

A face draws within five per cent of true shape over 5 per cent of them — two caps of half-angle 18.2°, together a twentieth of the sphere.

A factor of six, and it is the practical content of the drill. A student asked for a true length has a wide target and can choose the auxiliary for layout; a student asked for a true shape has a narrow one and must construct it. The three-step sequence exists because the third step’s target is six times smaller than the first’s, which is a better reason than “the constraints are three” and is the same reason a curved plate has no true view at all: its normal is not one direction but a family, and no cap contains a family.

It also says what a drawing office is buying with its standard three views. Each of them puts two edge families at exactly true length and one face at exactly true shape — the corners of both tables at once — and a part whose faces are not axis-aligned gets none of it. The auxiliary view is the machinery for manufacturing, on demand, the coincidence that the three standard views provide for free on a rectilinear part and never for anything else.

One further reading of the two identities, because it prices what an axonometric gives up. Both sums are fixed, so improving one scale costs the others: raising an edge scale toward one drives the other two down, and the sum of squares is conserved. The direction that shares the loss equally is the isometric one, and it is the only direction that does — three equal $|d_i|$ forces $1/\sqrt3$ each. So isometric is not a convention chosen for convenience of drawing but the unique solution of an equal-shares constraint, which is what isometric actually means arrived at from the identity rather than from the name.

The measurement a foreshortened view supports, and the one it does not #

There is a practical rule buried here that the drill never quite states, and it is the reason a drawing office cares which view a length is taken from.

A length measured in a view is $|\mathbf{v}|\sin\theta$, and $\theta$ is not observable in that view. So a ruler laid on a drawing measures something true only when the reader knows the segment’s direction independently — which, in a standard view of a rectilinear part, they do.

That is exactly the failure the ruler essay measures on an isometric drawing: the same physical length draws at different lengths depending on which way it runs, so a single scale bar is wrong for everything except the directions it was drawn for.

The scale bar problem, stated exactly #

A drawing carries one scale bar and an isometric drawing has three foreshortenings, so the bar is right for at most one family of directions and wrong for the rest. Two consequences follow and they are usually confused with each other.

The first is that a length off the paper needs a direction before it can be converted, and the direction has to come from the reader’s understanding of the object rather than from the picture. The second is that a length not along any axis has a foreshortening of its own, lying somewhere between the extremes, and no scale bar can be drawn for it at all.

Which is why axonometric drawings are illustrations and orthographic views are documents. The illustration shows the object; the document is the thing a length is taken from.

The drill checked against the projection it claims to produce #

The site’s habit is that a taught construction is drawn beside the projection it is supposed to produce, so it can be checked rather than trusted. Run that here and the three steps come out exact:

• the edge’s imaged length in the first auxiliary equals its true length to 1e-16 of a metre;
• its imaged length in the second is zero to the same precision, which is what “a point” means numerically;
• the face’s imaged area in the third equals its true area to 1e-16 of a square metre.

Those are not tolerances. They are the arithmetic saying the construction is an identity, and it is worth contrasting with the numbers the same treatment produces elsewhere on this site: the four-centre ellipse falls 5.72% short at the ends of its major axis, the by-eye depth division misplaces a post by metres, and the taught cone of vision turns out to be a statement about the reader rather than the picture.

Where the drill’s transfer distances come from #

Worth one paragraph, because it is the part of the recipe that looks most like bookkeeping and is in fact the whole content.

Constructing a new view on paper means computing, for each point, its two coordinates in the new picture. One of them is shared with the view being projected from — the direction in the picture plane that both views have in common, which is why the new view is drawn “folded” about a line. The other is the coordinate along the old view’s ray direction, which the old view dropped and which some earlier view still has.

So the transfer distances are the recovery of the coordinate the previous projection destroyed, taken from a view that did not destroy it. The drill is a sequence of projections precisely because each one drops a different component, and the sheet as a whole holds all three.

That is also why two views are enough to start from and one is not. One orthographic view has dropped a coordinate outright, and no construction on the paper can supply it.

The perspective version of the same question is not the same operation #

There is a natural next question — what the equivalent of “look at it square on” is for a photograph — and the answer is instructive because the two operations are not analogues.

In a parallel view the true shape of a plane face is obtained by projecting again, along the normal. In a perspective picture it is obtained by applying a homography to the picture itself: the face is a plane, a photograph of a plane is a projective image of it, and the map back is determined by four correspondences or by the plane’s vanishing line and two more numbers.

The difference is where the information comes from. The auxiliary view needs a second view — the coordinate the first one dropped has to exist somewhere. The rectification needs no second view at all, because a perspective projection did not drop a coordinate cleanly; it folded the missing one into the divide, and the plane’s own vanishing line is enough to unfold it.

So the parallel case and the perspective case answer the same practical question by opposite routes: one adds a picture, the other reads a line off the picture it has. It is one more instance of the trade this site keeps meeting — a parallel projection preserves ratios along a line and destroys the evidence of depth outright, while a perspective one destroys the ratios and keeps a trace of the depth in the convergence.

The fold line is a rotation, and the paper hides it #

The drill’s folding line reads as an operation on the sheet, and it is worth one paragraph on what it is in space, because it is the same object the whole essay is about.

Two views are two picture planes, each perpendicular to its own ray direction. Two planes meet in a line. That line — projected into either picture — is the fold, and the reason a coordinate can be “transferred” across it is that the two planes share the direction along it, so a distance measured along that direction means the same thing in both pictures.

Which is why the transfer works in one direction and not the other. Along the fold, the two views agree exactly; perpendicular to it, one view carries a coordinate the other has dropped, and the drill is an accounting of which view still has which.

Seen that way, the sheet full of folding lines is a drawing of a chain of picture planes hinged along their intersections — the parallel-projection version of changing the picture plane while the centre stays put, with the centre at infinity throughout.

Choosing the direction is the whole method #

Stated compactly, everything above is one idea used four times.

A parallel view is chosen by naming a direction, and every property of the resulting picture follows from the angles between that direction and the things being drawn. A line at $90°$ is true; a line at $0°$ is a point; a face whose normal is at $0°$ is true; a face whose normal is at $\theta$ has its area multiplied by $\cos\theta$.

The reason this is worth separating from “descriptive geometry” as a subject is that the subject is usually presented as constructions on a sheet — folding lines, transfer distances, reference planes — and the constructions are the implementation. What is being decided each time is a direction in space, and a reader who holds that in mind can reconstruct any of the drills without remembering which measurement gets transferred from where.