2

Consider attempting to derive the distance formula

$$d((x_1, y_1), (x_2, y_2)) = \operatorname{arcosh} \left( 1 + \frac{ {(x_2 - x_1)}^2 + {(y_2 - y_1)}^2 }{ 2 y_1 y_2 } \right)$$

from the metric of the Poincare upper half-plane model

$$(ds)^2 = \frac{(dx)^2 + (dy)^2}{y^2}$$

See e.g. here. Though, from what I can see, this is not a straightforward calculation, since it involves solving the geodesic equation.

The problem I have is, what if I need a distance formula for a hyperbolic metric such as from this post, for some $t \in [0,1]$,

$$e^{2ty}(dx)^2+(dy)^2$$

I can no longer fall back on the standard theory of the distance in hyperbolic space to derive the first formula above, since the space no longer has sectional curvature $-1$. Can a distance formula be given in a similar way for $t \in [0,1]$? Would I need to solve the geodesic equation from scratch?

I would have just integrated along a geodesic in the usual way to find the arc length, given the metric as a function of $t$, but even the simple case $t=1$ seems intractable.

apg
  • 2,797
  • 1
    $K=1/2$ is, of course, a typo. When you say “the simple case,” do you mean $t=0$? See Section 3.2 of my differential geometry text (linked in my profile) for the usual computations. I suspect it’s not that hard to adapt them. – Ted Shifrin Jun 02 '22 at 01:09
  • 1
    For constant negative section curvature $K$, you can use the metric $$ \sqrt{-K}\frac{(dx)^2+(dy)^2}{y^2}$$ From this, it is evident that distances simply rescale by a factor of $\sqrt{-K}$. – Deane Jun 02 '22 at 01:52
  • 2
    You don't mention it, but, would you be satisfied with a distance formula that is described by an entirely different method than either of the ones you have mentioned in your post? I ask because the simplest derivation of a distance formula is to first write down an isometry between the two models of the hyperbolic plane, and then use that isometry to transport the distance formula from one model to the other. – Lee Mosher Jun 03 '22 at 12:34
  • Yes thank you, I am trying to build a random graph built from random points of this space, but need distances between points to get the probability of edges between those points. But basically have no ideas what this distance is. – apg Jun 03 '22 at 14:19
  • Related: https://math.stackexchange.com/q/160338/11127 – Qmechanic Feb 29 '24 at 18:06

2 Answers2

3

The metric $$ds^2 = e^{2ty}dx^2 + dy^2$$ has constant curvature $-t^2$, which goes from $0$ to $-1$ as $t$ goes from $0$ to $1$. After a few moments' work, for $t\ne 0$, I don't see obvious geodesics other than vertical lines. So I don't see a distance formula in general.

Now, for $t=1$, the metric $e^{2y}dx^2 + dy^2$ is isometric to the usual hyperbolic metric $(du^2+dv^2)/v^2$ on the upper half-plane. The mapping $u=x$, $v=e^{-y}$ is a global isometry. So you can transform the geodesics, and hence the distance formula, back to the plane with this metric of curvature $-1$.

apg
  • 2,797
Ted Shifrin
  • 115,160
  • Ok so we have constant curvature $-t^2$ when you lose that $y^2$ in the denominator (otherwise it has non-constant curvature). I think I can solve the geodesic equation to obtain the geodesics, then find their length using numerical integration? The OP's question is essentially about "how do I get distances from the metric", and I suppose I've been able to answer that as "its not always possible" and "you first need to get geodesics via the geodesic equation", then integrate to find their length, but all this information is helpful, thank you. – apg Jun 03 '22 at 09:17
  • Could you point out a reference for finding a distance formula for the case =1 (perhaps in your book)? Starting with just the metric, I can't find anything online, or even on stack exchange. – apg Jun 03 '22 at 09:31
3

EDIT: Fixed the scaling of the distance function: $d_h = d/t$ and not $td$.

It is not hard to verify that with respect to the metric $$ g = e^{2ty}\,dx^2 + dy^2, $$ the vertical lines are geodesics and the horizontal lines are curves with curvature $-t$ orthogonal to the vertical geodesics. On the other hand, with respect to the metric $$ h = \frac{du^2+dv^2}{t^2v^2} $$ the vertical lines are geodesics and the horizontal lines are curves with curvature $t$ orthogonal to the vertical geodesics. Both metrics have constant Gauss curvature $-t^2$.

Therefore, there is an isometry of the form $(x,y)\mapsto (u(x), v(y))$, which simply reparametrizes the horizontal and vertical lines. If we pull back the metric $h$ using this map, we get \begin{align*} \frac{du^2+dv^2}{t^2v^2} &= \frac{(u'(x))^2\,dx^2 + (v'(y))^2\,dy^2}{(tv(y))^2} \end{align*} Therefore, the map is an isometry if \begin{align*} \left(\frac{u'(x)}{v(y)}\right)^2 &= t^2e^{2ty}\\ \left(\frac{v'(y)}{v(y)}\right)^2 &= t^2 \end{align*} One possible solution is $(u,v) = (tx, e^{-ty})$. Double checking this, \begin{align*} \frac{du^2+dv^2}{t^2v^2} &= \frac{t^2dx^2+ (-te^{-ty}\,dy)^2}{t^2e^{-2ty}}\\ &= e^{2ty}\,dx^2 + dy^2. \end{align*}

$\newcommand\arccosh{\operatorname{arccosh}}$ According to Wikipedia (https://en.wikipedia.org/wiki/Poincar\'e_half-plane_model#Distance_calculation), the distance between two points $(u_1,v_1)$ and $(u_2,v_2)$ in the upper half-plane model of hyperbolic space is $$ d((u_1,v_1),(u_2,v_2)) = \arccosh\left(1 + \frac{|(u_1-u_2,v_1-v_2)|^2}{2v_1v_2}\right). $$ Rescaling this, we see that $$ d_h(u_1,v_1),(u_2,v_2)) = \frac{1}{t}\arccosh\left(1 + \frac{|(u_1-u_2,v_1-v_2)|^2}{2v_1v_2}\right). $$ Therefore, with respect to the metric $g$, the distance between two points $(x_1,y_1)$ and $(x_2,y_2)$ is $$ d_g((x_1,y_1),(x_2,y_2)) = \frac{1}{t}\arccosh\left(1 + \frac{|(t(x_1-x_2),e^{-ty_1}_1-e^{ty_2})|^2}{2e^{-t(y_1-y_2)}}\right). $$

Deane
  • 7,582
  • Thank you, this is what I was looking for, so I suppose one obtains these distance formulas via methods like this, rather than directly deriving it from the metric in a procedural manner. That's very good to know. Thanks again! – apg Jun 05 '22 at 13:06
  • 1
    In differential geometry, there are usually at least three different ways to do a calculation. Usually one of them is much easier than the others. It also usually helps to understand the geometric picture. Once I understood that it was an orthogonal system of geodesics and horocycles, it became clear what to do. – Deane Jun 06 '22 at 00:05