Let $p(x)$ and $f(x)$ be sufficiently smooth functions and $D=\frac{d}{dx}$. It is easy to show that $$e^{p(x)D}f(x)=f(e^{p(x)D}x).\tag{1}$$

If $p(x)=a \in \mathbb{R}$ , we have the shift operator as $e^{aD}f(x)=f(x+a)$.

I appreciate it if you could help me to get answers for the following questions:

First, if $p=D$, can we say that $$e^{DD}f(x)=f(x+D)~?\tag{2}$$

Update: The first question is answered below. No, LHS is a function and RHS is an operator.

Second, what is the relationship between $e^{p(x+D)D}f(x)$ and $e^{p(x)D}f(x)$? For instance:

$$e^{p(x+D)D}f(x)-e^{p(x)D}f(x)=(e^{p(x+D)D}-e^{p(x)D})f(x)= \qquad ?\tag{3}$$

$$\frac{e^{p(x+D)D}}{e^{p(x)D}}=e^{p(x+D)D} e^{-p(x)D}=\qquad ?\tag{4}$$

or is it possible to somehow simplify $e^{p(x+D)D}f(x)$?

Update: The problem above seems to be ambiguious. I decided to rephrase it as follows:

Assume that $p(x)$ ($p:\mathbb{R} \to \mathbb{R}$) in known numerically, is it possible to find a matrix representation of operator $p(x+D)D$? or is it possible to define $p(x+D)D$ at all?

  • 1
    $\begingroup$ I would think that $e^{DD} f(x)$ would work out to be function $\sum_{n=0}^\infty f^{(2n)}(x)/n!$, not an operator like $f(x + D)$. $\endgroup$ Apr 12, 2022 at 20:21
  • $\begingroup$ Good point Michael. $e^{DD}f(x)$ is a function but $f(x+D)$ is an operator. Thanks! $\endgroup$
    – Mirar
    Apr 12, 2022 at 20:32

2 Answers 2


People are liable to make mistakes in Heaviside operator calculus manipulations, as they forget symbols on the right. Here is a trick to reduce your $e^{DD}$ to the Weierstrass transform, namely convolution with a Gaussian.

Adapt the identity $$ e^{u^2}= \frac{1}{\sqrt{4\pi}}\int_{-\infty}^{\infty} \! dy ~ e^{-y^2/4} e^{-uy}, $$
to the formal one, $$ e^{D^2}= \frac{1}{\sqrt{4\pi}}\int_{-\infty}^{\infty} \! dy ~ e^{-y^2/4} e^{-yD}, $$ so, then, effectively, a linear combination of your starting Lagrange translation operator.

You then have $$ e^{D^2} f(x)= \frac{1}{\sqrt{4\pi}}\int_{-\infty}^{\infty}\!\! \! dy ~ e^{-y^2/4} e^{-yD} f(x)\\ = \frac{1}{\sqrt{4\pi}}\int_{-\infty}^{\infty}\!\! \! dy ~ e^{-y^2/4} f(x-y). $$

So your (2) is not even wrong.

You may learn more here and links therein...

(Soft) Addendum on comments When $[p,D]\neq 0$, you are not in Kansas anymore, and Lagrange's rewriting of the Taylor expansion fails dramatically and insidiously. Indeed, occasionally you see c-number arguments of functions shifted by operators, as in deformation quantization, but such operators commute with everything relevant and behave like c-numbers in that context. Your case, by contrast, is a recipe for grief.

Your proposed $\phi(x,t)=e^{t p(x+D)D} f(x)$, s.t. $\phi(x,0)= f(x)$ and $\phi(x,1)= e^{p(x+D) D }f(x)$ and $\partial_t \phi(x,t)= p(x+D)D \phi (x,t)$, is true, and is a standard Hausdorff move in CBH expansion procedures, as Lie algebra books cover. But I am not quite sure what you are really up to, so I couldn't comment. Your (4) is ill-defined, as the numerator and denominator don't commute.

As you are exploring your way, make sure your constructions agree with the known case in $e^{txD} f(x) = e^{t{d\over d\ln x}} f(e^{})= f(xe^t) $.

  • $\begingroup$ Thank you Cosmas. I found this answer helpful. I am aware of the Weierstrass transform and am mainly concerened with $p(x+D)$ operator. Based on the link above and the links therein, I see that $e^{p(x+D)D}f$ is the solution of $\frac{d}{dt}\phi(x,t)=p(x+D)D\phi(x,t)$, with $\phi(x,0)=f(x)$. Is my understanding correct? $\endgroup$
    – Mirar
    Apr 13, 2022 at 4:48
  • $\begingroup$ I mean the solution at $t=1$. I am wondering whether it would be possible to find a differential equation that $p(x+D)f$ is its solution (?) $\endgroup$
    – Mirar
    Apr 13, 2022 at 4:54
  • $\begingroup$ I updated the question @Cosmas. The intention is defining $p(x+D)D$ if $p(x)$ is known. $\endgroup$
    – Mirar
    Apr 13, 2022 at 17:35
  • $\begingroup$ If the power series of p(x) is known, p(x+D) is well defined and messy. You want a "normal-ordered" form with all D s to the right? Like $(x+D)^2= x^2+1 +2xD +D^2$ ? $\endgroup$ Apr 13, 2022 at 18:00
  • $\begingroup$ Yes, that is correct. The data that I have ($p$ and $f$) are numerically known. If I know the action of $p(x+D)D$ I can compute $e^{p(x+D)D}f$. What if we represent $p$ locally with cubic/quadratic spline? The highest degree is 3 or 2. Will it simplify the calculation? $\endgroup$
    – Mirar
    Apr 13, 2022 at 18:20
  1. In OP's equations it is ambiguous how far to the right the operator $D\equiv\frac{d}{dx}$ is acting. See e.g. this Phys.SE post for a similar ambiguity.

  2. Example. OP's eq. (1) is not true in general as an operator identity where $D$ acts all the way to the right. Take e.g. $p(x)=a$ and $f(x)=x^2$ in eq. (1). Then$^1$ $$LHS ~=~e^{p(x)D}f(x)~=~e^{aD}x^2~=~ (x+a)^2e^{aD}$$ while $$RHS~=~f(e^{p(x)D}x)~=~(e^{aD}x)^2~=~ (x+a)(x+2a)e^{2aD}$$

  3. Example. OP's eq. (2) is not true in general as an operator identity where $D$ acts all the way to the right. Take e.g. $f(x)=x$ in eq. (2). Then $$LHS ~=~e^{D^2}f(x)~=~e^{D^2}x~=~ (x+2D)e^{D^2}$$ while $$RHS ~=~f(x+D)~=~x+D$$

  4. Let's introduce a full stop notation "." to indicate how far $D$ acts to the right.

  5. Example. OP's eq. (1) is not true in general as a function identity if we insert a full stop "." to the right of each side. Take e.g. $p(x)=a$ and $f(x)=x^2$ in eq. (1). Then $$LHS ~=~e^{p(x)D}f(x).~=~e^{aD}x^2.~=~ (x+a)^2$$ while $$RHS~=~f(e^{p(x)D}x).~=~(e^{aD}x)^2.~=~ (x+a)(x+2a)$$

  6. We can formulate a correct version of OP's eq. (1) as follows: $$\begin{align} e^{p(x)D}f(x). ~=~&e^{p(x)D}f(x)e^{-p(x)D}. \cr ~=~&e^{p(x)D}f(x)e^{-p(x)D} \cr ~=~&f(e^{p(x)D}xe^{-p(x)D}) \cr ~=~&f(e^{p(x)D}xe^{-p(x)D}.) \cr ~=~&f(e^{p(x)D}x.) \end{align} \tag{1}$$

  7. Example. OP's eq. (2) is not true in general as a function identity if we insert a full stop "." to the right of each side. Take e.g. $f(x)=x^2$ in eq. (2). Then $$LHS~=~e^{D^2}f(x). ~=~e^{D^2}x^2.~=~ x^2+D^2x^2.~=~x^2+2$$ while $$RHS ~=~f(x+D).~=~(x+D)^2.~=~ (x+D)x.~=~x^2+1$$
    It doesn't help if we put the full stop in a different place, e.g. $$RHS ~=~f(x+D.)~=~(x+D.)^2~=~x^2$$

  8. There are similar issues with many of OP's other formulas.


$^1$ Notation. Note that one often implicitly identify a function, say, $p: x\mapsto p(x)$ with its corresponding left multiplication operator $m_p: g(x) \mapsto (m_p g)(x):=p(x)g(x)$, or a value $p(x)$ of the function.

  • $\begingroup$ In my understanding, if a differential operator acts on a real function ($f:\mathbb{R}\to\mathbb{R}$), it produces a function. In your example, $$e^{aD}x^{2}=(1+aD+\frac{1}{2}a^{2}D^{2}+...)x^{2}=x^{2}+2ax+a^{2}+ 0 + 0 + ... = (x+a)^{2}$$ Similarly, RHS is equal to $(x+a)^{2}$ because $e^{aD}$ acting on $x$, generates $x+a$. I am sorry, but I do follow your line of reasoning in this answer. Are you trying to say that the problem above is not correctly posed? $\endgroup$
    – Mirar
    Apr 13, 2022 at 4:13
  • $\begingroup$ I updated the answer. $\endgroup$
    – Qmechanic
    Apr 13, 2022 at 6:57
  • $\begingroup$ Thank you @Qmechanic. Unfortunately I am not familiar with the full stop "." notation. Do you mind pointing me to a reference? Additionally, do you believe that the problem, defined as it is, is ambiguous? Do you have any suggestion for improvement? $\endgroup$
    – Mirar
    Apr 13, 2022 at 7:20
  • $\begingroup$ Is the full stop notation unclear? $\endgroup$
    – Qmechanic
    Apr 13, 2022 at 10:40
  • $\begingroup$ Yes. And the main question is whether I should modify the question description to avoid ambiguity? $\endgroup$
    – Mirar
    Apr 13, 2022 at 11:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.