Illustrated · 1 min

Consider the two n-dimensional vectors: a⃗=<a1,a2,...,an> and b⃗=<b1,b2,...,bn>.\vec{a}= \left< a_1,a_2,...,a_n \right>\text{ and } \vec{b}=\left< b_1,b_2,...,b_n \right>.

A simple calculation shows ∣a⃗∣2∣b⃗∣2=(a12+a22+...+an2)(b12+b22+...+bn2).|\vec{a}|^2|\vec{b}|^2=\left( a_1^2+a_2^2+...+a_n^2\right)\left( b_1^2+b_2^2+...+b_n^2\right).

In addition: (a⃗⋅b⃗)2=(a1b1+a2b2+...+anbn)2\left(\vec{a}\cdot\vec{b}\right)^2=\left(a_1b_1+a_2b_2+...+a_nb_n\right)^2.

Let θ\theta be the angle between vectors a⃗\vec{a} and b⃗\vec{b} when the two are placed with their initial points at the same location.

We know that (a⃗⋅b⃗)=∣a⃗∣∣b⃗∣cos⁡θ\left( \vec{a}\cdot\vec{b} \right) = \lvert \vec{a}\rvert \lvert \vec{b} \rvert \cos\theta.

Since −1≤cos⁡θ≤1-1\le \cos\theta \le1, 0≤cos⁡2θ≤10\le \cos^2\theta \le1.

Therefore, (a⃗⋅b⃗)2=∣a⃗∣2∣b⃗∣2cos⁡2θ≤∣a⃗∣2∣b⃗∣2.\left( \vec{a}\cdot\vec{b} \right) ^2= \lvert \vec{a}\rvert ^2 \lvert \vec{b} \rvert^2 \cos^2\theta \le \lvert \vec{a}\rvert^2 \lvert \vec{b}\rvert^2.

Plugging in our earlier expressions yields the Cauchy-Schwarz Inequality: (a12+a22+...+an2)(b12+b22+...+bn2)≥(a1b1+a2b2+...+anbn)2.\left( a_1^2+a_2^2+...+a_n^2 \right)\left(b_1^2+b_2^2+...+b_n^2 \right) \ge \left(a_1b_1+a_2b_2+...+a_nb_n\right)^2.

The vectors also give us a very nice way of seeing when equality holds.

Clearly, equality holds if and only if cos⁡2θ=1\cos^2 \theta=1, which means that the vectors a⃗\vec{a} and b⃗\vec{b} are parallel.

This happens if and only if there is a real number kk such that ka⃗=b⃗k\vec{a}=\vec{b}, or if one or both of the vectors are the zero vector.

Thus, equality holds if and only if one or more of the following conditions hold (we must make a special case to avoid division by zero): b1=b2=...=bn=0b_1=b_2=...=b_n=0 a1b1=a2b2=...=anbn\frac{a_1}{b_1}=\frac{a_2}{b_2}=...=\frac{a_n}{b_n}

Here is another (and very simple) proof for the Cauchy-Schwartz Inequality.

Consider:

(a1x−b1)2=a12x2−2a1b1x+b12≥0(a_1x-b_1)^2=a_1^2x^2-2a_1b_1x+b_1^2\ge 0

(a2x−b2)2=a22x2−2a2b2x+b22≥0(a_2x-b_2)^2=a_2^2x^2-2a_2b_2x+b_2^2\ge0

(a3x−b3)2=a32x2−2a3b3x+b32≥0(a_3x-b_3)^2=a_3^2x^2-2a_3b_3x+b_3^2\ge0

......

(anx−bn)2=an2x2−2anbnx+bn2≥0(a_nx-b_n)^2=a_n^2x^2-2a_nb_nx+b_n^2\ge0

Add all of the above inequalities and we get:

(a12+a22+...+an2)x2−2(a1b1+a2b2+...+anbn)x+(b12+b22+...+bn2)?0(a_1^2+a_2^2+...+a_n^2)x^2-2(a_1b_1+a_2b_2+...+a_nb_n)x+(b_1^2+b_2^2+...+b_n^2)?0

Above, we have a quadratic inequality in the form ax2+bx+c≥0ax^2+bx+c\ge0

On a graph, this will look something like these:

This is because (a12+a22+...+an2)x2−2(a1b1+a2b2+...+anbn)x+(b12+b22+...+bn2)(a_1^2+a_2^2+...+a_n^2)x^2-2(a_1b_1+a_2b_2+...+a_nb_n)x+(b_1^2+b_2^2+...+b_n^2) will always be greater or equal to 0. In these cases, since the graph either intersects the x-axis at one point (has one solution) or does not intersect it at all (no solution), we can use the discriminant b2−4ac≤0b^2-4ac\le0 where

b=−2(a1b1+a2b2+...+anbn),a=(a12+a22+...+an2),c=(b12+b22+...+bn2)b=-2(a_1b_1+a_2b_2+...+a_nb_n), a=(a_1^2+a_2^2+...+a_n^2), c=(b_1^2+b_2^2+...+b_n^2)

Thus, 4(a1b1+a2b2+...+anbn)2−4(a12+a22+...+an2)(b12+b22+...+bn2)≤04(a_1b_1+a_2b_2+...+a_nb_n)^2-4(a_1^2+a_2^2+...+a_n^2)(b_1^2+b_2^2+...+b_n^2)\le0

(a1b1+a2b2+...+anbn)2≤(a12+a22+...+an2)(b12+b22+...+bn2)(a_1b_1+a_2b_2+...+a_nb_n)^2\le(a_1^2+a_2^2+...+a_n^2)(b_1^2+b_2^2+...+b_n^2)

Finally, we have the original Cauchy-Schwartz inequality!

Type a concept, like “slope” or “photosynthesis”.

↑ ↓ to choose · Enter to openSearch explanations and problems