Learn faster with bite‑sized practice that actually sticks.

Gram–Schmidt: A Worked-Example Reflection Sheet (Checklist Style)

Gram–Schmidt is like turning a messy group of “kinda pointing different ways” vectors into a neat set of perpendicular, unit-length arrows. Same span, cleaner coordinates, happier math.

Below is a step-labeled reflection sheet you can follow every time—no drama, just structure.

Big Picture Goal

Given linearly independent vectors $v_1, v_2, \dots, v_k$ in an inner-product space, we want an orthonormal set $q_1, q_2, \dots, q_k$ that spans the same subspace.

• Orthogonal means: $\langle q_i, q_j\rangle = 0 \text{ for } i\neq j$
• Normal (normalized) means: $\|q_i\| = 1$
• Orthonormal means both.

The Checklist (with a tiny worked example structure)

We’ll write the output vectors in two stages:

• Orthogonal (not yet unit): $u_1, u_2, \dots, u_k$
• Orthonormal (unit): $q_i = \dfrac{u_i}{\|u_i\|}$

Step 0 — Start with independent vectors

✅ Input: $v_1, v_2, \dots, v_k$ must be linearly independent.

Why? If one vector is “already made from the others,” Gram–Schmidt eventually tries to normalize a zero vector, and that’s a hard stop.

Step 1 — First vector: keep it (then normalize later)

Set:
$u_1 = v_1$

Then later:
$q_1 = \dfrac{u_1}{\|u_1\|}$

Idea: the first direction is free—you’re just choosing your first axis.

Step 2 — Make the next vector perpendicular by subtracting its “shadow”

For the second vector:

1. Project $v_2$ onto $u_1$:
   $\operatorname{proj}_{u_1}(v_2) = \frac{\langle v_2, u_1\rangle}{\langle u_1, u_1\rangle}u_1$

2. Subtract that projection:
   $u_2 = v_2 - \operatorname{proj}_{u_1}(v_2)$

3. Normalize:
   $q_2 = \dfrac{u_2}{\|u_2\|}$

Mental picture: you remove the part of $v_2$ that points along $u_1$, leaving only the “new direction.”

Step 3 — Repeat: subtract all earlier shadows

For the third vector, you subtract the shadows onto each previous orthogonal vector:

$u_3 = v_3 - \operatorname{proj}_{u_1}(v_3) - \operatorname{proj}_{u_2}(v_3)$

Then normalize:
$q_3 = \dfrac{u_3}{\|u_3\|}$

In general, for $i \ge 2$:

$u_i = v_i - \sum_{j=1}^{i-1} \operatorname{proj}_{u_j}(v_i)$

and

$q_i = \dfrac{u_i}{\|u_i\|}$

✅ Checklist wording:

• Subtract projections onto all earlier $u_j$
• Confirm $u_i \neq 0$
• Normalize to get $q_i$

Step 4 — Verify orthonormality (quick inner product audit)

Once you have $q_1,\dots,q_k$, verify:

1. Unit length:
   $\langle q_i, q_i\rangle = 1$

2. Perpendicular for different indices:
   $\langle q_i, q_j\rangle = 0 \quad (i\neq j)$

If these hold, you’ve built an orthonormal set.

Self-Audit: Catch Problems Before They Bite

How to spot near-dependence (a.k.a. “why is my vector almost zero?”)

Sometimes the input vectors are technically independent, but almost dependent. Then Gram–Schmidt can produce a tiny $\|u_i\|$ and things get numerically wobbly.

Watch for:

• Very small norm after subtraction: $\|u_i\| \approx 0$
• Huge coefficients in projections (because you’re dividing by $\langle u_j,u_j\rangle$)
• A feeling that $v_i$ is “mostly explained” by earlier vectors

What it means conceptually:

• You’re trying to create a “new direction,” but there’s barely any new direction left.

(Practical note: in computation-heavy settings, people often use Modified Gram–Schmidt for better numerical stability—but the idea is the same.)

Where conjugation appears (complex inner products)

If your vectors have complex entries, the inner product uses complex conjugates.

Common convention:
$\langle x, y\rangle = \sum_i \overline{x_i}\,y_i$

That conjugation shows up in the projection formula because projections use inner products:

$\operatorname{proj}_{u}(v) = \frac{\langle v, u\rangle}{\langle u, u\rangle}u$

✅ Self-check:

• In complex spaces, make sure your inner product is conjugate-linear in one argument (depending on convention).
• If you forget conjugation, your “orthogonality check” can fail in confusing ways.

Plain-Language Reflection Prompt (Explain the why)

In your own words, briefly explain the purpose of:

1. starting with independent vectors,
2. subtracting projections,
3. normalizing,
4. checking inner products at the end.

Takeaway

Gram–Schmidt is a tidy routine: keep the parts that are new, remove the parts that are repeats, then scale everything to length 1. It’s like organizing a closet—same clothes, suddenly everything fits nicely.