I wonder why arsenic is an outlier

That seems to be because arsenic at atmospheric pressure sublimes; the lowest pressure at which it can be liquid is ~36 atm. mendeleev has 887 K as the boiling point (actually the sublimation point at 1 atm), and 1090 K as the melting point (which is accurate for 36 atm). I'll add a footnote.

If you’ve sorted by boiling point, why is there an outlier?

Ed Tufte, I suspect, would heartily approve of this presentation format.

@Tim See the footnote -- the melting point recorded by mendeleev for arsenic is greater than its boiling point. So the boiling point for arsenic is actually at the left of that line!

Note that the curve looks nice only because the elements have been sorted by boiling points. If you sorted by atomic number it would be a lot more ragged.

@CarlWitthoft: I agree it's a well done, beautiful graph. Tufte would probably still want to maximise the data-ink ratio. "Remove the yticks, remove the gray background, move the gridlines to the front of the graph, move the element symbols closer to the data...".

So tungsten is not only the element with the second-highest melting point, it's also the element with the highest boiling point!

@Sean I bought 1kg of tungsten a few months ago, on Ebay, as a toy. Every time I pick it up, I'm blown-away by the density. I cannot prepare my brain enough by saying "this will be surprisingly heavy for such a small object", and I laugh every time.

@EricDuminil Just for you, I redid the plot as Tufte might suggest (and also made some other improvements) (ggplot code embedded in the markdown source, if you're curious).

@zwol: I feel honoured, thank you very much! It looks even better. I've never used ggplot. The only thing I'd change would be to render the grid lines as white, so that they disappear from the background, but are only visible above the data. The labels would be above the gridlines. The gridlines could be shown every 1000K then.

@zwol: It took me a while to understand it isn't Python code. That's R, right? And I just realized the width of the colourful bands have a meaning. Very cool!

@EricDuminil Yeah, that's R code. I wrote a tiny Python script to extract the melting and boiling points of each element from mendeleev and write them to a CSV file and then I did the entire rest of the job with R. I honestly prefer Python, but ggplot is so powerful that faffing around with R's very, um, opinionated data model becomes worth it.

@EricDuminil Unfortunately, I don't know how to implement your "white gridlines above the data, labels above that" suggestion in ggplot. It too has opinions, and one of them is that the gridlines are below the data, always.

@zwol: Okay. I was impressed by how concise the code is. With Python/Matplotlib, it would be a least 20 lines of weird API calls.

@zwol why stop at uranium?

@yolo To match the other answer. This is really just a long footnote to that.

@zwol I mean, if there isn't really a scientific reason not to use elements above uranium I'd be glad to see it. The other answer's user seems to have disappeared.

@yolo There is a practical reason: many of the synthetic elements heavier than uranium have only ever been synthesized in tens-of-atoms quantities, and/or they have half-lives measured in microseconds or less -- not enough to measure a melting or boiling point.

@zwol why not manually disregard the many and work with the few that do for a more accurate analysis.

It'd be interesting to see this diagram, sorted by atomic number. Is there any pattern?