University College London · FACULTY OF MATHEMATICS

MATH0005 Chap.8 Gram-Schmidt and the Spectral Theorem

- one subject, every graph, every model, every mark
7 Chapters5-page Bible
Our own words - no uploaded lecturer files
Updated for this semester
Chapter 8 of 8 · MATH0005

Gram-Schmidt and the Spectral Theorem

Gram-Schmidt and the Spectral Theorem joins orthogonal geometry to spectral structure. Gram-Schmidt starts with an independent sequence and replaces each new vector by the residual left after subtracting every projection onto earlier orthogonal directions. The residual is perpendicular to those directions and preserves the accumulated span. Normalising the nonzero residuals produces an orthonormal basis.

A zero residual reveals that the original vector was already in the earlier span. Orthogonal matrices have orthonormal columns, satisfy Q transpose Q=I and preserve lengths and angles. For real symmetric matrices, the spectral theorem guarantees real eigenvalues and an orthonormal eigenbasis, so Q transpose A Q is diagonal. Eigenspaces for distinct eigenvalues are orthogonal.

Within a repeated eigenspace, Gram-Schmidt can create an orthonormal basis without losing the eigenvector property. The chapter distinguishes ordinary diagonalisation from orthogonal diagonalisation and uses three coordinated checks: Q transpose Q=I, AQ=QD and correct eigenvalue-column pairing. The closing synthesis shows why symmetry allows a transformation to be understood as independent scaling along perpendicular axes.

Quadratic forms reveal the same structure by assigning one signed coefficient to each orthonormal eigendirection after mixed coordinate effects have been removed. Eigen-coordinate reconstruction links those separated effects back to the original vector and provides a final basis check.

In this chapter

What this chapter covers

  • 01

    Gram-Schmidt

  • 02

    Projection subtraction

  • 03

    Orthonormal bases

  • 04

    Orthogonal matrices

  • 05

    Symmetric matrices

  • 06

    Spectral theorem

  • 07

    Orthogonal diagonalisation

Worked example · free

Apply Gram-Schmidt to a plane basis

Q [10 marks]. Apply Gram-Schmidt to b1=(1,1,0) and b2=(1,0,1), producing an orthonormal basis for their span. This original practice weighting is not an official university mark allocation.
The upper route removes all components in earlier directions and the lower step normalises the surviving residual, producing an orthonormal sequence with exactlThe upper route removes all components in earlier directions and the lower step normalises the surviving residual, producing an orthonormal sequence with exactly the original span.start b1,b2,b3independentsubtract earlierprojectionsnormalise eachnonzero resultq1,q2,q3orthonormalsame span, simpler coordinates
The upper route removes all components in earlier directions and the lower step normalises the surviving residual, producing an orthonormal sequence with exactly the original span.
  • 2Normalise b1 to q1=(1/sqrt(2),1/sqrt(2),0).
  • 3Subtract the q1 projection from b2 to obtain u2=(1/2,-1/2,1).
  • 3Normalise the residual to q2=(1,-1,2)/sqrt(6).
  • 2Verify q1 dot q2=0 and both lengths equal one, while the original vectors remain in the same span.
An orthonormal basis is ((1,1,0)/sqrt(2),(1,-1,2)/sqrt(6)), with sign changes also valid. The displayed points are original practice weighting.
Sia tip — Subtract every earlier projection, test the residual for orthogonality, and only then divide by its length.
Glossary

Key terms

Gram-Schmidt process
A procedure that converts an independent sequence into an orthogonal sequence with the same span.
Orthogonal matrix
A real square matrix whose columns form an orthonormal basis.
Symmetric matrix
A real square matrix equal to its transpose.
FAQ

Gram-Schmidt and the Spectral Theorem FAQ

What does a zero residual mean?

It means the current input vector already lies in the span of the earlier vectors, so the original sequence is dependent at that stage. There is no new direction to normalise, and dividing by the residual length would be invalid.

Why is an orthogonal matrix easy to invert?

Its columns are orthonormal, so Q transpose Q=I. Therefore Q inverse equals Q transpose. The same identity shows that multiplying by Q preserves dot products, lengths and angles in the real setting.

What does symmetry add to diagonalisation?

A real symmetric matrix has real eigenvalues and an orthonormal eigenbasis. This gives an orthogonal diagonalisation Q transpose A Q=D, which is stronger than merely having some invertible eigenvector matrix P.

Study strategy

Exam move

Run Gram-Schmidt as a sequence of local checks. Start with an independent list, normalise the first vector and, for each later vector, subtract projections onto every orthonormal direction already constructed. Dot the residual with each earlier vector before normalising. If the residual is zero, identify the dependence and stop that branch.

Accept equivalent nonzero scalings of residuals before normalisation and remember that final orthonormal vectors can differ by sign. For orthogonal matrices, verify Q transpose Q=I and use the identity to prove length preservation. Before invoking the spectral theorem, check the entry condition A transpose=A.

Find eigenspace bases, rely on orthogonality between distinct eigenspaces and apply Gram-Schmidt within any repeated eigenspace. Assemble Q from orthonormal eigenvectors and keep eigenvalues aligned in D. Verify both Q transpose Q=I and AQ=QD.

Keep error categories for projecting onto original rather than new directions, mixing normalised and unnormalised formulas, normalising a zero residual and assuming every diagonalizable matrix is symmetric. Finish with the structural benefit: orthogonal coordinates preserve geometry while diagonal form separates the map into independent scalar actions.

Practise a three-vector run with an audit after each subtraction: record all earlier dot products, the residual norm and the span relation that recovers the original vector. Include one dependent input so a zero residual is recognised before division. For spectral problems, separate three checkpoints: symmetry of A, orthonormality of Q and eigenpair alignment in AQ=QD.

Work one repeated-eigenvalue example where Gram-Schmidt is needed inside a single eigenspace. Finally, express a quadratic form in eigen-coordinates and explain how each diagonal coefficient controls one perpendicular direction rather than merely reporting the matrix product.

Working through Gram-Schmidt and the Spectral Theorem in MATH0005? Sia is AskSia’s AI Mathematics tutor — ask any MATH0005 Gram-Schmidt and the Spectral Theorem question and get a clear, step-by-step explanation grounded in how MATH0005 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.

A+Everything unlocked
Unlocks this Bible + all 32 of your subjects at University College London, plus 1,000+ more Bibles across the AskSia library.
Sia - your MATH0005 tutor, unlimited, worked the way the exam marks it
The full 5-page Bible + practice bank with worked solutions
Chrome extension - sync your LMS so Sia knows your deadlines
Bilingual EN / Chinese on every Bible and every Sia answer
$0.99 Trial
30-day money-back · cancel in one tap · how it works