Finishing the Matrix Exponential with Jordan Form – Finding ChatGPT’s Typos in Maple Software

Author Jonathan David · Math Monthly · Watch on YouTube

Written Lesson

Book Source:

Gustafson, G. B. (n.d.). Differential equations and linear algebra: A course for science and engineering (Vols. I–II). University of Utah.

https://math.utah.edu/~gustafso/debook/deBookGG.pdf


QUESTION 11.4.15

$$e^{\left(\begin{smallmatrix}1&1\\0&2\end{smallmatrix}\right)t}=\begin{pmatrix}e^t&e^{2t}-e^t\\0&e^{2t}\end{pmatrix}$$

ANSWER [There does not appear to be any answers in this text.]

Matrix Exponential Identities. Verify from exponential identities.

$\mathcal{F}_1$ [Formulae] First, we find a necessary formula. After reviewing the book, we find the following formula. We will start with this but there are many more.

$$e^{\textcolor{red}{A}t}=\sum_{n=0}^{\infty}A^n\frac{t^n}{n!}$$

$\mathcal{F}_2$ [Formulate] Whenever verifying or showing, you often start with the left side and more until it equals the right side.

$$e^{\textcolor{red}{\left(\begin{smallmatrix}1&1\\0&2\end{smallmatrix}\right)}t}=\sum_{n=0}^{\infty}\textcolor{red}{\begin{pmatrix}1&1\\0&2\end{pmatrix}}^{\!n}\frac{t^n}{n!}$$

At this point, what do we do? There could be a simple linear algebra trick but likely, we need to diagonalize. However: Again: We must reference what we are allowed to use in the book. Let us check it again.

In this text, the “Jordan Form,” appears match our situation.

$$e^{\textcolor{red}{A}t}=P^{-1}e^{Jt}P\qquad\leftrightarrow\qquad J=P\textcolor{red}{A}P^{-1}$$
The Math Monthly Journal

NOTE To the untrained student, you may look at this and think it is diagonalization. However, on page 721, you will find that it is referring to a triangular matrix, which we have. Recognize it is upper triangular.

Pg. 721 “Because the Jordan form is a triangular matrix, then $T=J$ gives an algorithm for generation of columns in matrix $P$.”

From this goose chase, we see that the previous formula, $J=P\textcolor{red}{A}P^{-1}=T$. Then,

$$e^{\textcolor{red}{A}t}=P^{-1}e^{Jt}P=P^{-1}e^{Tt}P$$

From Theorem 9.15 (Jordan’s Extension), $T$ is the diagonal elements. Then,

$$A=\begin{pmatrix}1&1\\0&2\end{pmatrix}\quad\Rightarrow\quad T=\begin{bmatrix}1&0\\0&2\end{bmatrix}.$$

MISTAKE ↓ We have two $P$’s …

Now, if $A=PTP^{-1}$, we do some basic linear algebra arithmetic.

$$\begin{gathered} A=PTP^{-1}\quad\Rightarrow\quad[A=P(PT^{-1})^{-1}]\quad\Rightarrow\quad[A=P(PT^{-1})^{-1}](PT^{-1})\\ \Rightarrow\quad A(PT^{-1})=P(PT^{-1})^{-1}(PT^{-1})\quad\Rightarrow\quad A(PT^{-1})=PI\\ \Rightarrow\quad A(PT^{-1})=P \end{gathered}$$

END MISTAKE


Backtracking: Pg. 721 “Because the Jordan form is a triangular matrix, then $T=J$ gives an algorithm for generation of columns in matrix $P$.”

Now, the book appears to be pointing to using a software program—as this is for engineers—so, reference P.721 “Computer algebra systems can find matrices J and P in the Jordan form of matrix A. With limitations, there is a constructible matrix P for Jordan’s two theorems 9.14 and 9.15. See page 898 for a maple example.”

2   Professor Jonathan D. Tullis
November 2026 | Volume 1

Before I go forward, I would like to know what maple is. I will not use the internet; I will use the book. I used duckduckgo.com because it doesn’t use A.I. – Maple is just an old program like MATLAB…

This textbook appears to want us to use Maple to find $P$. From Pg. 898, we see the following code [the following is a direct screenshot of Pg. 898.]

Jordan Decomposition using maple

The matrix

$$A=\begin{pmatrix}4&-2&5\\-2&4&-3\\0&0&2\end{pmatrix}$$

has a Jordan decomposition

$$A=PJP^{-1}=\begin{pmatrix}1&4&-7\\-1&4&1\\0&0&4\end{pmatrix}\begin{pmatrix}6&0&0\\0&2&1\\0&0&2\end{pmatrix}\begin{pmatrix}\frac14&1&-\frac74\\-\frac14&1&\frac14\\0&0&1\end{pmatrix}$$
# Maple, Find Jordan Form of matrix A
 A := Matrix([[4, -2, 5], [-2, 4, -3], [0, 0, 2]]);
 factor(LinearAlgebra[CharacteristicPolynomial](A,lambda));
 # Answer == (lambda-6)*(lambda-2)^2
 J,P:=LinearAlgebra[JordanForm](A,output=[’J’,’Q’]);
 zero:=A.P-P.J; # zero matrix expected

Now, this is a special case where we would use A.I. unless otherwise specified by your professor to use an alternative software. Since I don’t have any available, I will use ChatGPT by asking it to use the above image with our matrix $A$, to output $P$ for us.


TIME TRAVELING MATH

From ChatGPT, we got Maple:

MathMonthly.com   3
The Math Monthly Journal

Using $A=\left(\begin{smallmatrix}1&1\\0&2\end{smallmatrix}\right)$, Maple’s transformation matrix is

Plain text

     [1   1]
P := [     ]
     [0   1]

Its columns are eigenvectors for 1 and 2, respectively, giving $AP=PJ$ with $J=\operatorname{diag}(1,2)$.

Now, we have $P=\begin{bmatrix}1&1\\0&1\end{bmatrix}$. Since, this is a $2\times 2$ matrix, we can just compute the inverse directly rather than trying to find something crazy.
MISTAKE by ChatGPT ↑ ($P$ is actually $P^{-1}$)


MISTAKE BEGINS ↓

$$A=\begin{bmatrix}a&b\\c&d\end{bmatrix}\quad\Rightarrow\quad A^{-1}=\frac{\operatorname{adj}(A)}{\det(A)}=\textcolor{red}{\frac{\begin{bmatrix}1&0\\-1&1\end{bmatrix}}{(1)(1)-(1)(0)}=\begin{bmatrix}1&0\\-1&1\end{bmatrix}\;\text{WRONG.}}$$

Therefore,

$$A=\begin{pmatrix}1&1\\0&2\end{pmatrix}\quad\Rightarrow\quad PTP^{-1}=\begin{bmatrix}1&1\\0&1\end{bmatrix}\begin{bmatrix}1&0\\0&2\end{bmatrix}\begin{bmatrix}1&0\\-1&1\end{bmatrix}.$$

Now, we continue:

$$e^{\textcolor{red}{A}t}=P^{-1}e^{Jt}P=P^{-1}e^{Tt}P\quad\Rightarrow\quad P^{-1}e^{Tt}P=\begin{bmatrix}1&0\\-1&1\end{bmatrix}e^{\left[\begin{smallmatrix}1&0\\0&2\end{smallmatrix}\right]t}\begin{bmatrix}1&1\\0&1\end{bmatrix}.$$

We extract,

$$e^{\left[\begin{smallmatrix}1&0\\0&2\end{smallmatrix}\right]t}$$

and simplify. We have the provided formulae so we can go ahead and crunch this out.

4   Professor Jonathan D. Tullis
November 2026 | Volume 1
$$\begin{aligned} e^{\left[\begin{smallmatrix}1&0\\0&2\end{smallmatrix}\right]t} &=\sum_{n=0}^{\infty}\begin{bmatrix}1&0\\0&2\end{bmatrix}^{n}\frac{t^n}{n!} =\sum_{n=0}^{\infty}\begin{bmatrix}1^n&0\\0&2^n\end{bmatrix}\frac{t^n}{n!} =\sum_{n=0}^{\infty}\begin{bmatrix}1^n\frac{t^n}{n!}&0\\0&2^n\frac{t^n}{n!}\end{bmatrix}\\ &=\begin{bmatrix}\displaystyle\sum_{n=0}^{\infty}1^n\frac{t^n}{n!}&0\\0&\displaystyle\sum_{n=0}^{\infty}2^n\frac{t^n}{n!}\end{bmatrix} =\begin{bmatrix}e^t&0\\0&e^{2t}\end{bmatrix}. \end{aligned}$$

Next,

$$\begin{gathered} \begin{bmatrix}1&0\\-1&1\end{bmatrix}e^{\left[\begin{smallmatrix}1&0\\0&2\end{smallmatrix}\right]t}\begin{bmatrix}1&1\\0&1\end{bmatrix} =\begin{bmatrix}1&0\\-1&1\end{bmatrix}\begin{bmatrix}e^t&0\\0&e^{2t}\end{bmatrix}\begin{bmatrix}1&1\\0&1\end{bmatrix} =\begin{bmatrix}e^t&e^{2t}\\-e^t&e^{2t}\end{bmatrix}\begin{bmatrix}1&1\\0&1\end{bmatrix}\\ =\begin{bmatrix}e^t&e^t+e^{2t}\\-e^t&-e^t+e^{2t}\end{bmatrix}. \end{gathered}$$

MISTAKE ↑ or TYPO


$$\begin{pmatrix}1&0\\-1&1\end{pmatrix}\begin{pmatrix}e^t&0\\0&e^{2t}\end{pmatrix}\begin{pmatrix}1&1\\0&1\end{pmatrix} =\boxed{\begin{pmatrix}e^t&e^t\\-e^t&e^{2t}-e^t\end{pmatrix}}$$

I secondly asked ChatGPT WITHOUT ASKING FOR A SOLUTION

NOTE I want to point out that, this form is more commonly referred to in present time as “Diagonalization.” I had ChatGPT Diagonalize $A$.

$$A=PJP^{-1}=\begin{bmatrix}1&1\\0&1\end{bmatrix}\begin{bmatrix}1&0\\0&2\end{bmatrix}\begin{bmatrix}1&-1\\0&1\end{bmatrix}$$

***SO FREAKING IMPORTANT***

What I did, was not write out the formula for the invers. I did it in my head.

MathMonthly.com   5
The Math Monthly Journal
$$A=\begin{bmatrix}a&b\\c&d\end{bmatrix}\quad\Rightarrow\quad A^{-1}=\frac{\operatorname{adj}(A)}{\det(A)}=\frac{\begin{bmatrix}d&-b\\-c&a\end{bmatrix}}{(a)(d)-(b)(c)},\qquad ad-bc\ne0.$$

The typo I made was flipping the $b$ and $c$. All because I was an arrogant egotistical cocky douchebag that thought he could do basic multiplication in his head.

Now, we have redo the whole thing from that point. I leave this to you do.

Going back

Correct: Starting from

$\displaystyle\begin{bmatrix}1&0\\-1&1\end{bmatrix}\begin{bmatrix}e^t&0\\0&e^{2t}\end{bmatrix}\begin{bmatrix}1&1\\0&1\end{bmatrix}$ $\displaystyle\quad\Rightarrow\quad\begin{bmatrix}1&-1\\0&1\end{bmatrix}\begin{bmatrix}e^t&0\\0&e^{2t}\end{bmatrix}\begin{bmatrix}1&1\\0&1\end{bmatrix}=\begin{pmatrix}e^t&e^t-e^{2t}\\0&e^{2t}\end{pmatrix}$

Again: write out the formula. I put the inverse on the wrong side. I wrote the original formula with the inverse on the wrong side.

$$Pe^{Jt}P^{-1}=\begin{bmatrix}1&1\\0&1\end{bmatrix}\begin{bmatrix}e^t&0\\0&e^{2t}\end{bmatrix}\begin{bmatrix}1&-1\\0&1\end{bmatrix}=\begin{pmatrix}e^t&e^{2t}-e^t\\0&e^{2t}\end{pmatrix}$$

Our current issue is that the formula $e^{\textcolor{red}{A}t}=P^{-1}e^{Jt}P$ conflicts with the answer.

I am currently confused because the inverse needs to be on the opposite side of the provided equation in order to match exactly.

I will use diagonalization to check something. I request to diagonalize the original matrix, $A$.

$$\begin{pmatrix}1&1\\0&2\end{pmatrix}=\begin{bmatrix}1&1\\0&1\end{bmatrix}\begin{bmatrix}1&0\\0&2\end{bmatrix}\begin{bmatrix}1&-1\\0&1\end{bmatrix}.$$

There appears to be a conflict with the author’s formula, $e^{\textcolor{red}{A}t}=P^{-1}e^{Jt}P$.

Due to my experience with this problem in other courses using diagonalization, I believe that the author’s formula is a typo. That it should be $e^{\textcolor{red}{A}t}=Pe^{Jt}P^{-1}$.

6   Professor Jonathan D. Tullis
November 2026 | Volume 1

However, we know from linear algebra, if $A$ is diagonalizable, then

$$A=PDP^{-1}=P^{-1}DP.$$

Now, I question: Was ChatGPT’s calculation correct? Let us do it ourselves.

$$\begin{aligned} e^{\textcolor{red}{A}t}=P^{-1}e^{Jt}P &=\begin{bmatrix}1&-1\\0&1\end{bmatrix}\begin{bmatrix}e^t&0\\0&e^{2t}\end{bmatrix}\begin{bmatrix}1&1\\0&1\end{bmatrix} =\begin{bmatrix}e^t&-e^{2t}\\0&e^{2t}\end{bmatrix}\begin{bmatrix}1&1\\0&1\end{bmatrix}\\ &=\begin{bmatrix}1&-e^{2t}\\0&e^{2t}\end{bmatrix}\begin{bmatrix}1&1\\0&1\end{bmatrix} =\begin{bmatrix}1&e^t-e^{2t}\\0&e^{2t}\end{bmatrix}. \end{aligned}$$

Now, the issue we have is why? Did we get the wrong $P$? Or was the $P^{-1}$ actually the $P$ and that is the difference between the Jordan Form and diagonalization?


Going back a little further to the Jordan form.

$$e^{\textcolor{red}{A}t}=P^{-1}e^{Jt}P\qquad\leftrightarrow\qquad J=P\textcolor{red}{A}P^{-1}$$

From the book, we know $T=J$. Thus,

$$J=\begin{bmatrix}1&0\\0&2\end{bmatrix}=P\textcolor{red}{A}P^{-1}=\begin{bmatrix}1&1\\0&1\end{bmatrix}\begin{bmatrix}1&1\\0&2\end{bmatrix}\begin{bmatrix}1&-1\\0&1\end{bmatrix}=\begin{bmatrix}1&2\\0&2\end{bmatrix}\ne J.$$

This is an issue, because we used ChatGPT to find the “Maple” calculation. However, it doesn’t line up with the formula.

I now asked ChatGPT to give us the Jordan form from the provided relation since we do not have access to Maple.

Using your convention $J=PAP^{-1}$,

$$\boxed{J=\underbrace{\begin{pmatrix}1&-1\\0&1\end{pmatrix}}_{P}\underbrace{\begin{pmatrix}1&1\\0&2\end{pmatrix}}_{A}\underbrace{\begin{pmatrix}1&1\\0&1\end{pmatrix}}_{P^{-1}}=\begin{pmatrix}1&0\\0&2\end{pmatrix}}$$
MathMonthly.com   7
The Math Monthly Journal

$\mathcal{F}_3$ [Finalize]

Thus, our issue at hand is simply not having access to the correct software. However, we are able to conclude with ChatGPT’s mistakes and its corrections on basic arithmetic that our provide $P$, is actually $P^{-1}$.

Thus,

$$e^{At}=\textcolor{red}{P^{-1}}e^{Jt}P=\textcolor{red}{\begin{bmatrix}1&1\\0&1\end{bmatrix}}\begin{bmatrix}e^t&0\\0&e^{2t}\end{bmatrix}\begin{bmatrix}1&-1\\0&1\end{bmatrix}=\begin{pmatrix}e^t&e^{2t}-e^t\\0&e^{2t}\end{pmatrix}.$$
8   Professor Jonathan D. Tullis

AI Summarization of the Video

Author Jonathan David revisits Exercise 11.4.15 and traces two issues in the first attempt. The first is a misplaced negative entry when taking a $2\times2$ inverse. The second is a switch in what the symbol $P$ denotes between the ChatGPT responses and the relation used in the write-up.

  • Write out the inverse. The off-diagonal entries $-b$ and $-c$ stay in their respective positions in the adjugate. Writing out the formula makes the sign and placement easy to check.
  • Check the similarity relation directly. With $J=PAP^{-1}$, the correct $P$ has $-1$ in the upper-right entry. Substituting the first proposed matrix for $P$ produces an upper-right entry of $2$ in $PAP^{-1}$, while $J$ is diagonal.
  • Carry the same convention into the exponential. For $J=PAP^{-1}$, the identity is $e^{At}=P^{-1}e^{Jt}P$. The upper-right entry is $e^{2t}-e^t$.

The video uses the conflicting AI outputs to show why each result should be checked against the defining equation before it goes into the next line of a solution.

0 0 votes
Article Rating
Subscribe
Notify of
0 Comments
Oldest
Newest Most Voted
0
Would love your thoughts, please comment.x
()
x