Let $\{T(t),t\ge 0\}$ be a $C_0$ semigroup on a Hilbert space $X$, does that exist a larger Hilbert space $Y$ such that $X\subset Y$, and $T(t)$ extend to a $C_0$ group $T'(t)$ (so $t<0$ make sense now) on $Y$?

Edit: Here "extend" means that $(T(t)x, y)=(T'(t)x, y)$ for all $x, y\in X$, which is weaker than its usual meaning. If the semigroup $T(t)$ is a contraction, then the conclusion is true, see Theorem 8.1 (p.29) in "Harmonic analysis of operators on Hilbert spaces" written by Sz. Nagy, C. Foias, H. Bercovici and L. Kérchy.

  • $\begingroup$ For something like the heat semigroup on $L^2(\mathbb{R})$, it's difficult to imagine what that would look like. $\endgroup$ Oct 11, 2015 at 15:47
  • $\begingroup$ More specifically, $e^{\Delta}$, say, has an unbounded inverse on $X=L^2$, and making the Hilbert space larger cannot fix this. $\endgroup$ Oct 11, 2015 at 17:08
  • 1
    $\begingroup$ There are well known examples of semigroups for which T(t) is not injective for sufficiently large t. Clearly such a semigroup cannot be extended to a group on any space. $\endgroup$ Oct 11, 2015 at 21:05
  • $\begingroup$ @NateEldredge , I also feel hard to believe at the beginning. But if the semigroup is a contraction (e.g. the heat semigroup), then there do exist such an "extension", see Thm 8.1 in the book I mentioned above $\endgroup$
    – Tomas
    Oct 12, 2015 at 12:37
  • 1
    $\begingroup$ I cannot read @MichaelRenardy's mind. But for example you can consider for $T$ the nilpotent left-shift on $X=L^2(0,1)$. Then $T(t)=0$ for all $t\geq 1$, and you cannot extend this to a group in the ordinary way. However, in this example the left-shift on $Y=L^2(\mathbb{R})$ is indeed an extension of $T$ in the sense you defined in your edit. $\endgroup$
    – gsa
    Oct 13, 2015 at 13:22

1 Answer 1


An operator $A$ is a group generator if and only if both $A$ and $-A$ are semigroup generators. So under your assumptions, if we denote by $\tilde{A}$ the generator of the extension $\{\tilde{T}(t),t\ge 0\}$ of $\{T(t),t\ge 0\}$ to $Y$, then $-\tilde{A}$ would be a semigroup generator, too.

If by "extend" you mean the usual thing (viz, that $X$ is left invariant under $\{\tilde{T}(t),t\ge 0\}$ and the restriction to $X$ of $\{\tilde{T}(t),t\ge 0\}$ is $\{T(t),t\ge 0\}$ again), then $X$ would be left invariant under $\{T(t),t\le 0\}$, too, and the generator of this semigroup would necessarily be the part in $X$ of $-\tilde{A}$. So, the analytic semigroup $\{T(t),t\ge 0\}$ would actually imbed into a group $\{T(t),t\in \mathbb R\}$ on $X$ as well.

Now, an analytic semigroup on a Banach space $X$ can imbed into a group if and only if $X$ is finite dimensional its generator is bounded.

  • $\begingroup$ I guess I'm misinterpreting something, but your last claim puzzles me: why doesn't the semigroup $e^{it\Delta}$, $t\ge 0$, on $L^2$ embed in the group $e^{it\Delta}$, $t\in\mathbb R$ ? $\endgroup$ Oct 11, 2015 at 18:19
  • 2
    $\begingroup$ Yeah, I don't think that can be what you mean. More simply, what about the semigroup $T(t) x = e^{t} x$, whose generator is $I$? That's an analytic semigroup that embeds in a group, no matter what $X$ is. $\endgroup$ Oct 11, 2015 at 19:00
  • $\begingroup$ @NateEldredge: You are right, I have edited. $\endgroup$ Oct 12, 2015 at 7:29
  • $\begingroup$ @ChristianRemling: But $e^{it\Delta}$ is not analytic! $\endgroup$ Oct 12, 2015 at 7:37
  • $\begingroup$ Thanks, here, "extend" is a little bit weaker than the usual meaning, see my edit above. I'm curious to know that if this also implies that the generator of the "smaller" semigroup is part of $-\tilde{A}$. $\endgroup$
    – Tomas
    Oct 13, 2015 at 3:05

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.