Cross product

The formula

The clean geometric statement — magnitude and direction, separately:

|a × b| = |a| · |b| · sin(θ)

Reads as: the length of the result is the two input lengths multiplied together, scaled by the sine of the angle between them. When the vectors are parallel, sin(θ) = 0 and the result vanishes. When they are perpendicular, sin(θ) = 1 and you get the biggest answer possible.

The direction is set by the right-hand rule: point your fingers along a, curl them toward b, and your thumb points along a × b. That direction is perpendicular to both inputs.

The component form — what a computer actually runs:

a × b = ( a_y·b_z − a_z·b_y ,
          a_z·b_x − a_x·b_z ,
          a_x·b_y − a_y·b_x )

Reads as: each output component is a difference of two cross-multiplied terms from the other two axes. Miss a sign and the whole vector flips.

The same thing written as a 3×3 determinant, which is how most textbooks show it:

        | i   j   k  |
a × b = | a_x a_y a_z |
        | b_x b_y b_z |

Reads as: expand the determinant along the top row of unit vectors and you recover the component form above. It is a memory aid, not new math.

Torque — the reason most engineers meet this at all:

τ = r × F

Reads as: torque equals the position vector from the pivot to the load, crossed with the force. Same force, longer arm, more torque — the sine term captures the fact that only the perpendicular part of the force does any turning.

Where you meet it

Torque wrench on a bolt at a review board. When someone asks why a fastener 6 inches out on a bracket sees more moment than one 2 inches out under the same load, τ = r × F is the answer. The magnitude is r·F·sin(θ), so a force pulling straight along the arm (θ = 0) produces zero torque no matter how hard you pull — a fact that surprises people until they picture pushing a door toward its hinge.
Surface normals in a CAD or mesh review. Take two edge vectors of a triangle, cross them, and the result points straight out of the face. Every FEA mesh, every CFD surface, every 3-D-printed STL relies on this to know which way is "out." Flip the vertex order and the normal points inward — the classic reason a printed part comes out inside-out or a shaded model looks black.
Magnetic force on a moving charge, on the RF or motor bench. The Lorentz force F = q·(v × B) is why an electron beam bends in a CRT and why the wire in a motor pushes sideways. The force is perpendicular to both the velocity and the field, which is exactly the geometry a cross product produces and nothing else does as cleanly.
Angular momentum on a spin-balance stand. L = r × p describes the momentum of a spinning rotor or a satellite reaction wheel. When a gyro precesses in a way that feels backwards, it is the cross product's perpendicularity, not intuition, that is right.

How it works

The cross product only lives in three dimensions (and, as a mathematical curiosity, seven — nowhere else does a clean binary product of two vectors return a vector). In 2-D there is no room for a perpendicular direction, so a "2-D cross product" is really just the scalar a_x·b_y − a_y·b_x — a signed area, not a vector. Engineers use that scalar all the time for "is this point left or right of that line," but it is not the full operation.

The single most-repeated mistake is treating it like ordinary multiplication. It is anticommutative: swapping the inputs flips the sign.

a × b = −( b × a )

Numerically, (3,0,0) × (1,2,0) = (0,0,6) but (1,2,0) × (3,0,0) = (0,0,−6). Order matters, and the sign is the direction of your torque, your normal, your force. Get the order wrong on a torque calc and you have the bolt loosening when your model says tightening.

It is also not associative. (a × b) × c is generally not a × (b × c). For a=(1,2,3), b=(0,1,4), c=(2,0,1) the first gives (−4,−3,8) and the second gives (−28,5,6). You cannot drop parentheses the way you can with scalar multiplication. When you do need to unwind a nested cross product, the triple-product ("BAC minus CAB") identity is the tool:

a × ( b × c ) = b·(a·c) − c·(a·b)

Both sides evaluate to (−28,5,6) for those vectors — verified, and it turns a messy nested cross into two dot products.

The gotcha that bites test engineers: parallel inputs give zero. (2,4,6) × (1,2,3) = (0,0,0) because the vectors point the same way. If your surface-normal routine returns a zero-length vector, two of your triangle's edges are collinear — a degenerate, zero-area facet. If your torque comes out zero with a real force applied, the force line passes through your pivot. Zero is information, not a bug.

One more limit: the cross product is a pseudovector. Reflect your whole coordinate system in a mirror and an ordinary (polar) vector like a position or a force flips the way you'd expect, but a cross-product result does not — it picks up an extra flip and points the "wrong" way. Anything built from a cross product inherits this: surface normals, torque, and angular momentum are all pseudovectors. This never matters if you keep a consistent right-handed frame, which is why every serious analysis nails down handedness before anything else. Mix a left-handed CAD import into a right-handed solver and your normals and torques silently invert.

History

The cross product came out of a decades-long argument about how to multiply directed quantities in space. On October 16, 1843, William Rowan Hamilton — walking along the Royal Canal in Dublin — worked out the quaternions, a four-part number system whose multiplication rule buried both the modern dot and cross products inside it, though not yet by those names ^[1]^[2]. His product of two "pure" quaternions spat out a scalar part and a vector part; the vector part was, in effect, the cross product waiting to be extracted. The very next year, 1844, Hermann Grassmann in Germany published his own general theory of extended quantities, arriving at a related product independently and from a completely different, more geometric direction ^[1]. In 1877 William Kingdon Clifford gave the two operations the names "scalar product" and "vector product" that engineers still use for the dot and cross ^[1].

The version an engineer recognizes today came from stripping quaternions down to what was useful. Around 1881, Josiah Willard Gibbs at Yale printed lecture notes for his students that split Hamilton's machinery into separate dot and cross products with the modern × notation, and Oliver Heaviside in England did essentially the same thing independently at about the same time ^[1]^[3]. Gibbs's notes circulated privately through the 1880s; a formally published Vector Analysis, assembled by his student Edwin Wilson, did not appear until 1901 ^[3]. Heaviside used the cleaned-up vector calculus to compress Maxwell's electromagnetism into the compact form still taught. The quaternion loyalists fought it hard, but the engineers won: by the early twentieth century the standalone cross product had displaced quaternions across physics and engineering.

Related tools

/tools/bolt-torque
/tools/shaft-power-torque
/tools/moment-of-inertia-shapes
/tools/convert-torque
/tools/convert-angle

Sources

Written by HE in our own words from the cited sources — engineering judgment included, your stamp still required. All entries →