Bounds, counting and linear structure

Read eigenvectors in Book Isle: scalar-multiple output still requires nonzero input

Compute a matrix action on two directions, checking eigenvalues 3 and 1, whole lines of nonzero multiples and the zero-vector exclusion.

English translation of the Chinese original, translated on . Original publication dates, attribution, code and version notes are retained.

A vector mapped to a scalar multiple of itself has a special role, but zero cannot establish every eigenvalue. Use an original two-by-two card in Book Isle to check multiplication, proportionality and nonzero input, then verify repeated action by decomposition. No automatic linear-algebra analysis is assumed.

Copy the matrix and column-vector convention

V1 sets A rows(2, 1), (1, 2), acting on real columns. Seek nonzero v and scalarλ with Av=λv. The scalar and two-component vector are different objects. Keep the equation and nonzero condition in one visible box, import a clear PDF/EPUB and bookmark it. For image matrices inspect rows and columns without assuming searchable arrays. This specified real example does not establish identical eigenvector behavior for every real matrix.

Check both output components for each candidate

For u=(1, 1), row-column multiplication gives(3, 3)=3u. For v=(1, −1), it gives(1, −1)=1v. Eigenvalues are 3 and 1, not sums of output components 6 and 0. One scalar must account for both components simultaneously. Retain each product and minus sign. Symmetry does not permit entrywise multiplication; u/v are local labels, not universal textbook ordering.

A=[[2,1],[1,2]]
u=(1,1); A*u=(3,3)=3*u
v=(1,-1); A*v=(1,-1)=1*v
w=(1,0); A*w=(2,1), not a scalar multiple of w

Use a noncandidate and zero to test the definition

For w=(1, 0), Aw=(2, 1), whereas everyλw has second component 0, so w is not an eigenvector. A0=0 andλ0=0 for everyλ; the definition excludes zero precisely because it cannot distinguishλ. Partial satisfaction is insufficient. Highlight nonzero and use w’s second component as a contradiction instead of appearance alone. Component ratios can help when defined, but zero components never justify division by zero.

Find eigenvalues while retaining whole lines

det(A−λI)=(2−λ)²−1=λ²−4λ+3=(λ−1)(λ−3), giving eigenvalues 1 and 3. For 3 the components are equal: every nonzero c(1, 1) qualifies. For 1 they are opposite: every nonzero c(1, −1) qualifies. Two directions are not only two eigenvectors or only unit vectors. At c=2 the outputs are(6, 6) and(2, −2), with the same eigenvalues rather than doubledλ.

Trace 2+2=4 equals 3+1, and determinant 4−1=3 equals 3×1. These catch some eigenvalue errors but do not establish arbitrary candidate vectors. Retain scalar checks for eigenvalues and vector multiplication for eigenpairs separately.

Use negative and zero eigenvalues as separate controls

For B=diag(−1, 2), input(1, 0) maps to(−1, 0), so eigenvalue−1 is valid and reverses direction. For C=diag(1, 0), nonzero input(0, 1) maps to zero, so eigenvalue 0 is valid. Zero eigenvalue is allowed; zero eigenvector remains excluded. These controls do not replace V1: A still has only 3 and 1. Record matrix, input, output, λ together so an excerpt retains its object.

Decompose an ordinary input to verify repeated action

x=(3, 1)=2u+v is not itself an eigenvector. Linearity gives Ax=6u+v=(7, 5), matching direct multiplication. A²x=18u+v=(19, 17), also directly A(7, 5). Each directional coefficient gains its correspondingλ; do not multiply the whole ordinary input by 3. Separate decomposition and direct products to catch lost coefficients or assigning 3 to v. This does not guarantee two real eigenvector directions for every matrix.

w=(u+v)/2 gives Aw=(3u+v)/2=(2, 1). Mixing directions with different eigenvalues generally does not yield one scalar multiple. A nonzero combination within one eigenspace retains its eigenvalue; the zero combination remains excluded.

x=(3,1)=2*u+v
A*x=6*u+v=(7,5)
A^2*x=18*u+v=(19,17)
B=[[-1,0],[0,2]]; B*(1,0)=(-1,0)
C=[[1,0],[0,0]]; C*(0,1)=(0,0)

Save definition and verification together in Book Isle

Highlight or record locations for V1’s definition and products. Write a candidateλ before checking both components instead of copying 3 and 1 alone. Return by V1 if searchingλ fails. Six may add output components; zero input proving everyλ misses nonzero; scaling confusion calls for c=2. After one correction repeat direct and decomposed products. A saved excerpt is reading evidence, not an automatically verified eigenpair.

A horizontally written coordinate pair still denotes a column input here. Display commas do not authorize row-vector right multiplication; changed orientation needs changed product evidence.

Reopen and reject overstrong conclusions

Reopen and classify u, v, w, zero before stating A has two eigenvalues but infinitely many nonzero scalar-multiple eigenvectors. Revisit B/C to retain negative and zeroλ, then repeat both actions on(3, 1). Acceptance preserves matrix identity and nonzero input, not a universal parallel-arrow label. Save derivation, sources and actual operations; reopening stays pending if unperformed. A real rotation with no real eigenvector directions needs fresh analysis rather than this symmetric example generalized to every real matrix.

Matrix inverses and product checks; Dot products and zero conditions.

Method references: MIT18.06 lecture21 eigenvalues and eigenvectors. Diagrams, report data, counterexamples and ledgers are original exercises.

Version checked on 17 November 2026: the Apple public lookup lists Book Isle 1.1.4, requiring iOS / iPadOS 15.6 or later; see the official App Store for published capabilities. Use self-made PDFs, EPUBs or clear text materials; bookmarks, highlights and notes depend on format and visible controls, without automatic solving or literary analysis. External preparation and paper predictions do not constitute device observations. Leave unperformed checks blank.