Linear-Algebra-Done-Right ยท Unit 26 ยท Video 4 ยท Interactive Practice

The Trace Equals the Sum of the Eigenvalues

IKey Formulas

FormulaNameHypotheses
trโกA=A1,1+โ‹ฏ+An,n\operatorname{tr} A = A_{1,1} + \cdots + A_{n,n}Trace of a square matrixAA an nร—nn \times n matrix
trโก(AB)=โˆ‘j=1mโˆ‘k=1nAj,kBk,j=trโก(BA)\operatorname{tr}(AB) = \displaystyle\sum_{j=1}^{m}\sum_{k=1}^{n} A_{j,k}B_{k,j} = \operatorname{tr}(BA)Swap the order of the double sumAA is mร—nm \times n, BB is nร—mn \times m
trโกT=trโกM(T)\operatorname{tr} T = \operatorname{tr}\mathcal{M}(T), the same for every basisTrace of an operatorTโˆˆL(V)T \in \mathcal{L}(V); M(T,(u))=Cโˆ’1M(T,(v))โ€‰C\mathcal{M}(T, (u)) = C^{-1}\mathcal{M}(T, (v))\,C
trโกT=ฮป1+โ‹ฏ+ฮปn\operatorname{tr} T = \lambda_1 + \cdots + \lambda_n; ย q(z)=znโˆ’(trโกT)znโˆ’1+โ‹ฏ\ q(z) = z^n - (\operatorname{tr} T)z^{n-1} + \cdotsSum of eigenvalues; characteristic polynomialF=C\mathbf{F} = \mathbf{C}, n=dimโกVn = \dim V, each eigenvalue repeated as often as its multiplicity

Key Insight: A change of basis replaces M(T)\mathcal{M}(T) by Cโˆ’1M(T)CC^{-1}\mathcal{M}(T)C, and trโก((Cโˆ’1A)C)=trโก(C(Cโˆ’1A))=trโกA\operatorname{tr}\bigl((C^{-1}A)C\bigr) = \operatorname{tr}\bigl(C(C^{-1}A)\bigr) = \operatorname{tr} A, so the trace belongs to the operator.

IIThe Diagonal Changes, Its Sum Does Not

For T(x,y)=(4xโˆ’y,ย 2x+y)T(x, y) = (4x - y,\ 2x + y), a change of basis moves the diagonal entries of M(T)\mathcal{M}(T) but keeps their sum 5=2+35 = 2 + 3.

IIItrโก(AB)=trโก(BA)\operatorname{tr}(AB) = \operatorname{tr}(BA) by Swapping a Double Sum

Every product Aj,kBk,jA_{j,k}B_{k,j} is counted once in trโก(AB)\operatorname{tr}(AB), by rows, and once in trโก(BA)\operatorname{tr}(BA), by columns.

๐Ÿ’ก ABAB is 2ร—22 \times 2 and BABA is 3ร—33 \times 3, yet their traces agree; ABAB and BABA themselves need not be equal even when both are square.

IVQuiz Questions

Problem 1 ยท Count the Multiplicity

Given: TโˆˆL(C4)T \in \mathcal{L}(\mathbf{C}^4) whose eigenvalues are 33 with multiplicity 22, โˆ’1-1 with multiplicity 11, and 2i2i with multiplicity 11 โ€” find trโกT\operatorname{tr} T.

โœ… Correct! 3+3+(โˆ’1)+2i=5+2i3 + 3 + (-1) + 2i = 5 + 2i.
โŒ Each eigenvalue counts as often as its multiplicity. 33 appears twice: 3+3โˆ’1+2i3 + 3 - 1 + 2i.
โŒ The trace of a complex operator can be non-real. Keep the eigenvalue 2i2i in the sum.
โŒ Check the sign of โˆ’1-1. The sum uses the eigenvalues themselves, not their absolute values.
Show solution

On a complex vector space, trโกT\operatorname{tr} T is the sum of the eigenvalues, each repeated as often as its multiplicity (the multiplicities add up to dimโกC4=4\dim \mathbf{C}^4 = 4):

trโกT=3+3+(โˆ’1)+2i=5+2i.\operatorname{tr} T = 3 + 3 + (-1) + 2i = 5 + 2i.

Problem 2 ยท Trace from the Characteristic Polynomial

Given: TโˆˆL(C3)T \in \mathcal{L}(\mathbf{C}^3) with characteristic polynomial z3โˆ’4z2+z+6z^3 - 4z^2 + z + 6 โ€” find trโกT\operatorname{tr} T.

โœ… Correct! trโกT\operatorname{tr} T is the negative of the coefficient of z2z^2; indeed z3โˆ’4z2+z+6=(z+1)(zโˆ’2)(zโˆ’3)z^3 - 4z^2 + z + 6 = (z + 1)(z - 2)(z - 3) and โˆ’1+2+3=4-1 + 2 + 3 = 4.
โŒ Mind the sign. (zโˆ’ฮป1)(zโˆ’ฮป2)(zโˆ’ฮป3)=z3โˆ’(ฮป1+ฮป2+ฮป3)z2+โ‹ฏ(z - \lambda_1)(z - \lambda_2)(z - \lambda_3) = z^3 - (\lambda_1 + \lambda_2 + \lambda_3)z^2 + \cdots, so the trace is minus the coefficient.
โŒ The constant term is (โˆ’1)3ฮป1ฮป2ฮป3(-1)^3\lambda_1\lambda_2\lambda_3. It records the product of the eigenvalues, not their sum.
โŒ That is the coefficient of zz. The trace sits in the coefficient of znโˆ’1=z2z^{n-1} = z^2.
Show solution

With n=3n = 3, the characteristic polynomial is z3โˆ’(ฮป1+ฮป2+ฮป3)z2+โ‹ฏz^3 - (\lambda_1 + \lambda_2 + \lambda_3)z^2 + \cdots, so trโกT=โˆ’(โˆ’4)=4\operatorname{tr} T = -(-4) = 4.

Check: z3โˆ’4z2+z+6=(z+1)(zโˆ’2)(zโˆ’3)z^3 - 4z^2 + z + 6 = (z + 1)(z - 2)(z - 3), so the eigenvalues are โˆ’1-1, 22, 33, and โˆ’1+2+3=4-1 + 2 + 3 = 4.

Problem 3 ยท Traces of Products of Different Sizes

Given: a 2ร—32 \times 3 matrix AA and a 3ร—23 \times 2 matrix BB with AB=(1234)AB = \begin{pmatrix} 1 & 2 \\ 3 & 4 \end{pmatrix} โ€” find trโก(BA)\operatorname{tr}(BA).

โœ… Correct! trโก(BA)=trโก(AB)=1+4=5\operatorname{tr}(BA) = \operatorname{tr}(AB) = 1 + 4 = 5.
โŒ The identity holds for an mร—nm \times n AA and an nร—mn \times m BB. BABA is unknown, but its trace is not.
โŒ That adds every entry of ABAB. The trace adds only the diagonal entries.
โŒ That is only one diagonal entry of ABAB. The trace adds both: 1+41 + 4.
Show solution

For AA of size mร—nm \times n and BB of size nร—mn \times m,

trโก(AB)=โˆ‘j=1mโˆ‘k=1nAj,kBk,j=โˆ‘k=1nโˆ‘j=1mBk,jAj,k=trโก(BA).\operatorname{tr}(AB) = \sum_{j=1}^{m}\sum_{k=1}^{n} A_{j,k}B_{k,j} = \sum_{k=1}^{n}\sum_{j=1}^{m} B_{k,j}A_{j,k} = \operatorname{tr}(BA).

Here m=2m = 2, n=3n = 3, and trโก(AB)=1+4=5\operatorname{tr}(AB) = 1 + 4 = 5, so trโก(BA)=5\operatorname{tr}(BA) = 5 although BABA itself is not determined by ABAB.

Problem 4 ยท Trace of an Operator on Polynomials

Given: TโˆˆL(P2(R))T \in \mathcal{L}(\mathcal{P}_2(\mathbf{R})) defined by Tp=xpโ€ฒ+p(1)Tp = xp' + p(1), where p(1)p(1) is the constant polynomial โ€” find trโกT\operatorname{tr} T.

โœ… Correct! In the basis 1,x,x21, x, x^2: T1=1T1 = 1, Tx=1+xTx = 1 + x, Tx2=1+2x2Tx^2 = 1 + 2x^2, so the diagonal is 1,1,21, 1, 2.
โŒ The term p(1)p(1) also reaches the diagonal. T1=0+1=1T1 = 0 + 1 = 1, so the first diagonal entry is 11, not 00.
โŒ That adds every entry of M(T)\mathcal{M}(T). The constants 11 in TxTx and Tx2Tx^2 lie above the diagonal.
โŒ Not quite. Write T1T1, TxTx, Tx2Tx^2 in the basis 1,x,x21, x, x^2 and add the coefficient of 11 in T1T1, of xx in TxTx, of x2x^2 in Tx2Tx^2.
Show solution

Use the basis 1,x,x21, x, x^2:

T1=0+1=1,Tx=x+1,Tx2=2x2+1.T1 = 0 + 1 = 1, \qquad Tx = x + 1, \qquad Tx^2 = 2x^2 + 1. M(T)=(111010002),trโกT=1+1+2=4.\mathcal{M}(T) = \begin{pmatrix} 1 & 1 & 1 \\ 0 & 1 & 0 \\ 0 & 0 & 2 \end{pmatrix}, \qquad \operatorname{tr} T = 1 + 1 + 2 = 4.

The matrix is upper triangular, so the eigenvalues are 1,1,21, 1, 2, and their sum is 44 as well.

Solved: 0 / 4