Pome

17.16.23 Negative directions of the raw action124D4C83F5E6

On this Hopf background, constant Perron translations are stationary. Radial variations detect the failure of minimization. Every smooth two-periodic function of $r$ defines a smooth function on $X$, and for a radial path $u(t,r)$,

\[ \omega +\mathrm{dd}^{\mathrm{c}}u=(1+u_ r)\alpha _ n+(1+u_{rr})\beta _ n. \]

For $\langle F\rangle =\frac12\int _0^2F(r)\, \mathrm dr$, fiber integration therefore gives

905CCA72845B 17.16.24
\begin{equation} \mathscr S(u)=n\int _0^1 \left\langle u_ t^2(1+u_ r)^ d(1+u_{rr})\right\rangle \mathrm dt. \end{equation}

Changing the local base section shifts the start of the full period and leaves the average unchanged.

Theorem 17.16.25.944944D6FF68 For every nonzero constant $c$, the full Perron segment $u_ t=ct$ is stationary for the raw action, whose Hessian has an infinite-dimensional negative subspace. The infimum of raw lengths over all smooth strictly positive paths from $0$ to $c$ is strictly smaller than $d_ B(0,c)=\sqrt{2n}|c|$.

Proof. The constant path is basic, equals the chord, and is the full Perron segment by corollary 17.16.14. Its stationarity also follows from proposition 17.16.22. Let $w(t,r)$ be smooth, two-periodic in $r$, and zero at $t=0,1$. The paths $ct+\varepsilon w$ have positive spatial forms for sufficiently small $|\varepsilon |$. Expanding (17.16.24) to second order gives

\[ 2n\int _0^1\left\langle w_ t^2+2cw_ t(dw_ r+w_{rr}) +c^2\left(\frac{d(d-1)}2w_ r^2+dw_ rw_{rr}\right) \right\rangle \mathrm dt. \]

The integral of $w_ rw_{rr}$ over a period vanishes, while $\int _0^1\langle w_ tw_{rr}\rangle \mathrm dt =-\frac12[\langle w_ r^2\rangle ]_0^1=0$. Completing the square gives the Hessian

F20F3B72C244 17.16.26
\begin{equation} 2n\int _0^1\left\langle (w_ t+dcw_ r)^2-\frac{d(d+1)}2c^2w_ r^2 \right\rangle \mathrm dt. \end{equation}

For a positive integer $k$, take

\[ w_ k(t,r)=\sin (\pi t)\cos \bigl (\pi k(r-dct)\bigr ). \]

Then $w_{k,t}+dcw_{k,r}=\pi \cos (\pi t)\cos (\pi k(r-dct))$, and (17.16.26) is

\[ \frac{n\pi ^2}2\left(1-\frac{d(d+1)}2c^2k^2\right). \]

It is negative for all sufficiently large $k$. Distinct modes and their derivatives are orthogonal under the $r$ average, so the quadratic form is negative on every nonzero finite combination of these modes. Each combination gives a positive path for sufficiently small variation parameter and has action strictly below $nc^2$. Its raw length satisfies $L^2\leq 2\mathscr S$ on the unit interval, so it is strictly below $\sqrt{2n}|c|$. $\square$

These competing paths need not be subgeodesics. The strict length inequality does not establish definiteness or completeness of the unrestricted length infimum. For $n=3$ and $c=1$, a concrete path is

\[ u_\varepsilon (t,r)=t+\varepsilon \sin (\pi t)\cos \bigl (\pi (r-2t)\bigr ). \]

Its spatial coefficients are at least $1-\pi |\varepsilon |$ and $1-\pi ^2|\varepsilon |$. Its exact action is

00E6402ACAB5 17.16.27
\begin{equation} \mathscr S(u_\varepsilon ) =3-\frac{3\pi ^2}2\varepsilon ^2+\frac{111\pi ^4}{64}\varepsilon ^4. \end{equation}

Indeed, put $f=\sin (\pi t)$, $p=f'$, and $\vartheta =\pi (r-2t)$. Substitute $u_ t=1+\varepsilon (p\cos \vartheta +2\pi f\sin \vartheta )$, $u_ r=-\varepsilon \pi f\sin \vartheta $, and $u_{rr}=-\varepsilon \pi ^2f\cos \vartheta $ into (17.16.24). Odd powers have zero periodic average. The averaged quadratic coefficient is $(p^2-3\pi ^2f^2-2\pi ^2fp)/2$, and the quartic coefficient is $\pi ^2f^2(p^2+6\pi ^2fp+12\pi ^2f^2)/8$. The identities

\[ \begin{gathered} \int _0^1fp\, \mathrm dt=\int _0^1f^3p\, \mathrm dt=0, \\ \int _0^1f^2\, \mathrm dt=\frac12, \quad \int _0^1p^2\, \mathrm dt=\frac{\pi ^2}2,\\ \int _0^1f^4\, \mathrm dt=\frac38, \quad \int _0^1f^2p^2\, \mathrm dt=\frac{\pi ^2}8 \end{gathered} \]

give (17.16.27). Taking $\varepsilon =1/100$ gives a positive path with action $2.9985212487\ldots <3$.

Cite this page

Comments

Corrections, questions, and mathematical discussion are welcome. Sign in with GitHub to comment. Comments are public.