Distinct Eigenvalues Give Independent Eigenvectors
proof, eigenvalues, eigenvectors, diagonalization
1 Claim
The key theorem behind diagonalization is:
different eigenvalues force genuinely different directions.
Theorem 1 (Theorem) Let \(A \in \mathbb{R}^{n \times n}\), and let \(v_1,\dots,v_k\) be eigenvectors of \(A\) with distinct eigenvalues \(\lambda_1,\dots,\lambda_k\).
Then \(v_1,\dots,v_k\) are linearly independent.
2 Why This Proof Matters
This is the theorem that starts turning eigenvalue computation into structural understanding.
It matters because:
- it explains why distinct eigenvalues automatically help diagonalization
- it lets you count independent modes in a dynamical system
- it keeps reappearing in graph and spectral arguments
3 Dependencies
- \(Av_i = \lambda_i v_i\) for each eigenpair
- a linearly independent list is the same as a list with only the trivial linear relation
4 Strategy Before Details
Use induction on the number of eigenvectors.
The main trick is to start from a hypothetical linear relation and then apply \((A-\lambda_k I)\) to kill the last vector while keeping the others alive with shifted eigenvalues.
5 Full Proof
Proof. We prove the statement by induction on \(k\).
For \(k=1\), the statement is immediate because an eigenvector is nonzero.
Now assume the theorem holds for \(k-1\) distinct eigenvalues, and suppose
\[ c_1 v_1 + \cdots + c_k v_k = 0 \]
for eigenvectors \(v_1,\dots,v_k\) with distinct eigenvalues \(\lambda_1,\dots,\lambda_k\).
Apply \(A-\lambda_k I\) to both sides:
\[ (A-\lambda_k I)(c_1 v_1 + \cdots + c_k v_k) = 0. \]
Because \((A-\lambda_k I)v_k = 0\), this becomes
\[ c_1(\lambda_1-\lambda_k)v_1 + \cdots + c_{k-1}(\lambda_{k-1}-\lambda_k)v_{k-1} = 0. \]
The eigenvalues are distinct, so each coefficient \((\lambda_j-\lambda_k)\) is nonzero for \(j<k\).
Therefore the vectors \(v_1,\dots,v_{k-1}\) satisfy a linear relation. By the induction hypothesis, they are linearly independent, so
\[ c_1(\lambda_1-\lambda_k)=\cdots=c_{k-1}(\lambda_{k-1}-\lambda_k)=0. \]
Hence
\[ c_1=\cdots=c_{k-1}=0. \]
Returning to the original relation, we get
\[ c_k v_k = 0. \]
Since \(v_k \neq 0\), it follows that \(c_k=0\).
So all coefficients are zero, which proves that \(v_1,\dots,v_k\) are linearly independent.
6 Step Annotations
- The operator \(A-\lambda_k I\) is chosen to kill one selected eigenvector exactly.
- The other eigenvectors survive, but with nonzero scaling factors because their eigenvalues are different.
- Distinctness of eigenvalues is what prevents those scaling factors from vanishing.
7 Why The Assumptions Matter
- The eigenvalues must be distinct. Repeated eigenvalues do not automatically give dependence, but they do not guarantee independence either.
- The theorem does not say every matrix with repeated eigenvalues fails to diagonalize. It says distinct eigenvalues are a sufficient source of independence.
8 Common Failure Modes
- forgetting to use the distinctness assumption
- applying \(A\) instead of \(A-\lambda_k I\), which does not isolate the last vector
- concluding “repeated eigenvalues imply dependence,” which is false
9 Reusable Pattern
The proof teaches a durable move:
apply an operator that annihilates one chosen mode and rescales the others.
This same pattern appears in spectral decomposition arguments and minimal-polynomial reasoning.
10 Where This Shows Up Again
- diagonalization tests
- spectral methods for repeated matrix powers
- graph operators where eigenvectors define independent modes
11 Exercises
- Adapt the proof to the case \(k=2\) without induction.
- Give an example of a matrix with repeated eigenvalue that still has two independent eigenvectors.
- Explain why the theorem implies that a matrix with \(n\) distinct eigenvalues is diagonalizable.
12 Sources and Further Reading
- MIT 18.06: Lecture 21, Eigenvalues and Eigenvectors -
First pass- official source for the theorem and its role in diagonalization. Checked2026-04-24. - Hefferon, Linear Algebra -
Second pass- stronger proof and exercise treatment of eigenvalue theory. Checked2026-04-24. - A Comprehensive Survey on Spectral Clustering with Graph Structure Learning -
Paper bridge- current reminder that independent eigen-directions still drive modern spectral methods. Checked2026-04-24.