An orthonormal basis is a set of vectors that are all perpendicular to each other and each exactly one unit long. Together they span the whole space, so every vector in it can be written as a combination of them — and each combination coefficient is just a dot product.
This idea quietly powers much of modern mathematics and engineering: Fourier series, the QR decomposition inside nearly every numerical library, and the state vectors of quantum mechanics all rest on orthonormal bases.
This guide explains what "orthonormal" means, how it differs from merely "orthogonal," and how the Gram–Schmidt process builds an orthonormal basis from independent vectors. We work through the example [3, 1] and [1, 2], and the calculator above repeats the same steps — orthogonalize, normalize, verify — for any 2D or 3D input.
Understanding Orthonormal Bases
The word "orthonormal" blends two requirements: "ortho" (the vectors are perpendicular to one another) and "normal" (each vector is normalized to unit length). Mutually perpendicular vectors are automatically linearly independent, so n such vectors in n-dimensional space form a basis. Intuitively, an orthonormal basis is the nicest possible coordinate system: the familiar x-, y-, and z-axes are the prototype, generalized to any set of tilted directions.
The Two Conditions Every Orthonormal Set Must Satisfy
A set e₁, e₂, …, eₙ is orthonormal when the dot product of any two of its vectors equals 1 for the same vector and 0 for different ones — written with the Kronecker delta as eᵢ · e₎ = δᵢ₎. That is two demands:
- Mutual orthogonality: for distinct vectors, eᵢ · e₎ = 0. Every pair meets at a right angle.
- Unit length: for every vector, eᵢ · eᵢ = 1. Each basis vector is a pure direction with no stretching.
Both are easy to test, which is why the calculator reports them separately: pairwise dots that should come out near zero, and self dots that should come out near one.
A Concrete Picture in Two and Three Dimensions
In the plane, the standard orthonormal basis is e₁ = [1, 0] and e₂ = [0, 1]. But orthonormal bases are far from unique: rotating both vectors by any angle gives another valid one. The pair e₁ = [0.9487, 0.3162], e₂ = [−0.3162, 0.9487] — our orthonormalized [3, 1] and [1, 2] — is a basis tilted about 18.4° from the standard axes. In 3D the standard basis is [1, 0, 0], [0, 1, 0], [0, 0, 1], with infinitely many rotated versions alongside it.
Orthonormal vs Orthogonal: The Key Difference
An orthogonal set requires only perpendicularity — dot products between distinct vectors are zero — while saying nothing about lengths. An orthonormal set adds the demand that every vector has length exactly one. Every orthonormal set is automatically orthogonal, but the converse holds only if all vectors happen to be unit vectors. Classical Gram–Schmidt produces an orthogonal set; dividing each vector by its own length upgrades it to orthonormal.
Side-by-Side Comparison
- Orthogonal basis: pairwise perpendicular, lengths unrestricted. Coordinates require dividing each dot product by the basis vector's squared length.
- Orthonormal basis: pairwise perpendicular, every length one. Coordinates are pure dot products with no division.
- Conversion: divide each vector of an orthogonal basis by its norm; scaling never changes direction.
- Checking: orthogonality needs the pairwise dots; orthonormality needs those plus a length check per vector.
Why the Extra Normalization Step Pays Off
Normalization buys a large simplification for little arithmetic. In an orthonormal basis the coordinates of v are simply cᵢ = v · eᵢ, with no denominators; in a merely orthogonal basis each coordinate is (v · uᵢ) / (uᵢ · uᵢ), and those divisions propagate through every later calculation. Lengths simplify too: the squared length of v is just the sum of the squares of its orthonormal coordinates (Parseval's identity).
The Gram–Schmidt Process With Normalization
The Gram–Schmidt process converts any linearly independent set of vectors into an orthogonal set spanning the same space, one vector at a time; appending normalization turns the output into an orthonormal basis. The key idea is subtraction of projections: to make the second vector orthogonal to the first, subtract whatever part of it points along the first; to make the third orthogonal to the first two, subtract its parts along both, and so on. The calculator implements exactly this: Gram–Schmidt first, then normalization, then verification.
The Classical Steps
- Start with the first vector: u₁ = v₁ — nothing earlier exists to be perpendicular to.
- Orthogonalize the second: u₂ = v₂ − ((v₂ · u₁) / (u₁ · u₁)) u₁.
- Orthogonalize the third: subtract its projections onto u₁ and u₂; continue the pattern for further vectors.
- Normalize each vector: eᵢ = uᵢ / ||uᵢ||. Directions unchanged, lengths now one.
- Verify: every pairwise dot eᵢ · e₎ (i ≠ j) is zero and every self dot eᵢ · eᵢ equals one.
Normalizing at the End Versus as You Go
The classical route orthogonalizes everything first and normalizes at the end. The alternative normalizes each vector immediately after orthogonalizing it, reducing every later projection to a plain dot product — dividing by a unit vector's squared length means dividing by one. This variant, close to the modified Gram–Schmidt algorithm, is preferred in numerical software because it accumulates less rounding error. The calculator shows the intermediate norms so you can see what each division did.
What Can Go Wrong
- Linear dependence: if an input vector combines the earlier ones, its remainder is the zero vector, which cannot be normalized — the process must stop with an error.
- Near-dependence: almost-parallel vectors leave a tiny remainder whose normalization amplifies rounding error.
- Wrong input shape: every vector needs the same component count, and no space yields more independent vectors than its dimension.
Worked Example: Orthonormalizing the Vectors [3, 1] and [1, 2]
We now run the full procedure by hand on v₁ = [3, 1] and v₂ = [1, 2], the pair preloaded in the calculator. They are independent — neither is a multiple of the other — so an orthonormal basis of the plane exists.
Step 1 — Take the First Vector and Normalize It
The first orthogonal vector is u₁ = [3, 1], with squared length 3² + 1² = 10 and norm √10 ≈ 3.1623. Dividing gives e₁ = [3/√10, 1/√10] ≈ [0.9487, 0.3162]. Check: 0.9487² + 0.3162² ≈ 0.9000 + 0.1000 = 1.0000 — unit length, same direction as [3, 1].
Step 2 — Orthogonalize the Second Vector
Remove from v₂ = [1, 2] the part along u₁. The projection coefficient is (v₂ · u₁) / (u₁ · u₁) = 5/10 = 0.5, so the projection is [1.5, 0.5]. Subtracting: u₂ = [1, 2] − [1.5, 0.5] = [−0.5, 1.5], and indeed u₁ · u₂ = −1.5 + 1.5 = 0. The set is now orthogonal — but not yet orthonormal, since the lengths are √10 and √2.5.
Step 3 — Normalize and Verify the Result
The norm of u₂ is √(0.25 + 2.25) = √2.5 ≈ 1.5811, giving e₂ ≈ [−0.3162, 0.9487]. Verification: e₁ · e₂ ≈ −0.3000 + 0.3000 = 0.0000 (orthogonal), and e₂ · e₂ ≈ 0.1000 + 0.9000 = 1.0000 (unit length). The orthonormal basis is e₁ ≈ [0.9487, 0.3162], e₂ ≈ [−0.3162, 0.9487] — exactly what the calculator returns.
How to Use the Orthonormal Basis Calculator
The calculator performs the procedure above for two or three vectors in two or three dimensions, preloaded with the worked example so you can press Calculate immediately.
- Choose the dimension — 2D or 3D. The input hints update to match.
- Choose the number of vectors — 2 or 3. The third input row appears only for 3 vectors.
- Enter each vector as comma-separated numbers, e.g. 3, 1 or −0.5, 1.5, 2, with exactly as many numbers as the dimension.
- Press Calculate (or hit Enter in any input) to run Gram–Schmidt plus normalization.
- Read the three output sections: the orthonormal basis, the norms before normalization, and the verification table (pairwise dots ≈ 0, self dots ≈ 1).
Invalid input produces a precise message: a wrong component count, a non-numeric entry, or a dependent set — three vectors in the plane, for instance, can never be independent. Reset restores the worked example.
Key Properties of Orthonormal Bases
- Coordinates are dot products: v = Σ (v · eᵢ) eᵢ, each coefficient found independently with no system to solve.
- Parseval's identity: ||v||² = Σ (v · eᵢ)² — lengths are preserved exactly in the new coordinates.
- Clean projections: projecting v onto a subset's span means summing (v · eᵢ) eᵢ over that subset only.
- Unique coordinates: the expansion is one of a kind, and orthonormality makes it trivially computable.
- Rotations preserve orthonormality: a rotated orthonormal basis is still orthonormal — tilted bases are just as valid as axis-aligned ones.
- Orthogonal matrices: the basis vectors as columns give Q with QTQ = I, so QT = Q−1 and solving Qx = b is one multiplication.
- Best approximation: truncating after k terms gives the closest possible k-term approximation — the basis of compression from JPEG to PCA.
Where Orthonormal Bases Show Up in Practice
Orthonormal bases are load-bearing machinery across science and engineering: wherever directions must be separated cleanly or approximations must be optimal, one is usually involved.
Fourier Series and Signal Processing
A Fourier series expands a periodic signal as a sum of sines and cosines, which form an orthonormal basis (under the right inner product) for the space of signals. Each coefficient is an inner product of the signal with one basis function — the dot-product rule again. Filtering and compression operate in this frequency basis because they become coefficient-wise operations there.
QR Decomposition in Numerical Linear Algebra
QR factors a matrix A into an orthogonal Q and an upper-triangular R, computed by applying Gram–Schmidt to the columns of A — the columns of Q are the orthonormal basis the process produces. Least-squares fitting, eigenvalue computation, and most numerical libraries run through QR under the hood. This calculator performs the Q half of that decomposition for small matrices.
Quantum States and Measurement
In quantum mechanics the states of definite measurement outcomes form an orthonormal basis of the state space. Normalization encodes that probabilities sum to one; orthogonality encodes that distinct outcomes are mutually exclusive. Each outcome's probability is the squared magnitude of a dot product with its basis vector.
More Places You Already Use Them
Principal component analysis finds the orthonormal directions of greatest data variance. Computer graphics stores every camera orientation as an orthonormal basis (right, up, forward). Least-squares fitting projects measurements onto a model space, most stably through an orthonormal basis. Even everyday x-, y-, z-coordinates are the default orthonormal basis of 3D space.
Common Mistakes to Avoid
- Skipping normalization. Classical Gram–Schmidt alone gives an orthogonal basis — the division by each norm earns the second half of the name.
- Normalizing before orthogonalizing. Rescaling the raw inputs first orthogonalizes nothing; perpendicularity comes only from the projection subtractions, in order.
- Feeding in dependent vectors. A redundant vector leaves a zero remainder that cannot be normalized.
- Forgetting the denominator. The projection coefficient is (v · u) / (u · u); dropping it is valid only when u is already a unit vector.
- Assuming uniqueness. Sign flips, reorderings, and rotations all give different but equally valid orthonormal bases.
- Rounding too early. Round only the final vectors; rounding intermediate projections destroys the orthogonality just built.
Frequently Asked Questions
1. What is an orthonormal basis in simple terms?
A set of arrows at right angles to each other, each exactly one unit long, numerous enough to reach any point in the space by combining them. Describing any vector then takes only dot products — one per arrow.
2. How is an orthonormal basis different from an orthogonal basis?
An orthogonal basis requires only perpendicularity; its vectors may have any lengths. An orthonormal basis additionally requires unit length for every vector. Dividing each vector of an orthogonal basis by its own length converts it to orthonormal.
3. What does the Gram–Schmidt process do?
It converts any independent set of vectors into an orthogonal set spanning the same space: keep the first vector, then subtract from each later vector its projections onto all previously processed ones. A final normalization step makes the result orthonormal.
4. Why do we normalize vectors in Gram–Schmidt?
It rescales each orthogonalized vector to length one without changing its direction, upgrading the basis to orthonormal. Coordinates become plain dot products, and length and projection formulas simplify.
5. Can any set of vectors be turned into an orthonormal basis?
Only an independent set can. If a vector combines the previous ones, Gram–Schmidt reduces it to the zero vector, which cannot be normalized. Independent vectors in an n-dimensional space always yield an orthonormal basis of that space.
6. What happens if my vectors are linearly dependent?
The redundant vector's remainder is zero (or near-zero), and normalizing it would divide by zero. The calculator detects this and reports linear dependence instead of a meaningless result.
7. Does the order of the input vectors change the result?
Yes, generally. Gram–Schmidt fixes the first vector's direction and orthogonalizes everything else against it, so reordering the inputs yields a different but equally valid orthonormal basis. List the most important direction first if orientation matters.
8. How do I verify that a basis is orthonormal?
Compute all pairwise dot products: different vectors must dot to zero, and each vector dotted with itself must give one. The calculator automates both checks in its verification section.
9. What are the standard orthonormal bases of R² and R³?
In R² it is [1, 0] and [0, 1]; in R³ it is [1, 0, 0], [0, 1, 0], [0, 0, 1]. Any rotation of these axes is an equally valid orthonormal basis — which is why Gram–Schmidt output often looks tilted.
10. Are orthonormal bases unique?
No. Any space of dimension two or more has infinitely many: rotate, reflect, reorder, or flip a vector's sign and the set stays orthonormal. Gram–Schmidt selects one particular basis determined by your input order.
11. How are coordinates computed in an orthonormal basis?
Each coordinate is the dot product of the vector with the corresponding basis vector: the i-th coordinate of v is v · eᵢ. Without orthonormality you would also divide by each squared length or solve a linear system.
12. What is the connection between orthonormal bases and orthogonal matrices?
Orthonormal basis vectors as matrix columns form an orthogonal matrix Q with QTQ = I. Such matrices represent rotations and reflections, and since their transpose equals their inverse, systems involving them solve especially cheaply.
13. Where is the Gram–Schmidt process used in real life?
It drives the QR decomposition behind least-squares fitting, eigenvalue computation, and most numerical libraries. It also builds orthonormal signal sets, orthogonalizes regression predictors, and constructs camera frames in graphics.
14. What is the difference between classical and modified Gram–Schmidt?
Classical Gram–Schmidt subtracts all projections from each vector in one batch, usually normalizing at the end. Modified Gram–Schmidt subtracts them one at a time, updating the vector after each — mathematically equivalent, but with far less floating-point rounding error.
15. Can this calculator handle 3D vectors and three vectors at a time?
Yes. Select 3D and 3 vectors, then enter each vector as three comma-separated numbers, for example 1, 0, 0. The calculator runs the same orthogonalize-normalize-verify procedure and checks all pairwise dots and unit lengths.