Manhattanhenge: Loving Wolfram Alpha

A coworker posted something today about Manhattanhenge occuring on May 30 of this year.  That’s when sunset aligns with the east-west grid of Manhattan.  The streets are offset from true east-west by 28.9 degrees, according to Wikipedia.  So I went to Wolfram Alpha to figure out the solar azimuth on that day.  It said 30 degrees 30 minutes.  Wait a second, that’s not 28.9 degrees!  So initially I was confused.  However, you’ll note that it indicates an altitude of -1 degrees.  Because of refraction, the sun appears higher than it actually is, in this case about a degree.  Since the sun rises and sets at an angle to the horizon, the azimuth is also effected.  Furthermore, sunset is the point at which the trailing edge vanishes.  For dating Manhattanhenge, you probably want the leading edge so that it sits right on top of the street.  The angular diameter of the sun is 31 arc-minutes.  So you want the date when the azimuth is at 28.9 degrees and the altitude is about 30 minutes.  That gives you May 30th at 8:13pm.  It gives 29 degrees, close enough to 28.9.  However, the azimuth is +30 minutes.  Visual sunset is closer to an altitude of -1 degrees, so I’d think the correct altitude to look for would be -30 minutes.  However the astronomers that gave the 30th as the date perhaps used astronomical sunset instead of visual sunset.  

The Hayden Planetarium gives 8:17pm as the time of Manhattanhenge.  For that date and time, Wolfram Alpha provides an altitude of -10 minutes, and an azimuth of 29 degrees 40 minutes.  Perhaps the altitude of -10 minutes is a better number for when the leading edge of the sun will hit the street.

Anyhow, you can get pretty close with just Wolfram Alpha and some simple geometry.  I’m really loving that site.

—-
update:  I found a better number for the shift in the apparent position of the sun at sunset, about 35 minutes.   The -1 degree altitude reported at sunset by WA is off, or perhaps it has been rounded.  If you back out the diameter of the sun, which is 31 minutes, that’s pretty close to the numbers WA gives for Hayden Planetarium’s date and time.  So with a bit of extra knowledge, you can accurately find the date and time of Manhattanhenge from Wolfram Alpha.  It won’t solve for it, or at least I can’t figure out how to ask it for dates and times where the azimuth is 28.9 and the altitude is -5 or -10 minutes, but you can at least do a quick search around likely dates to find the right numbers.

The Wolfram Alpha Launch

Wolfram Alpha launched on Friday evening.  I watched a bit of the video, it was an interesting look at their launch process and hardware.  They have 6 colo’s, the largest of which has around 4,000 cores and a few hundred terabytes of data, with fairly high density 32-core 2U units from Dell.  Pretty big guns for a launch, so they mean business.  Necessary perhaps because of the press they’ve received, which is way over the top.  At this moment, it’s actually the first link on the Drudge Report, claiming that they’re challenging Google and have “all the answers“.  Wolfram isn’t claming to have all the answers, but they can answer a few that Google doesnt.  And they generally answer with hard data.  I like hard data.

First example, the word “thought”.   It gives the word frequency in spoken and written language, broader terms, narrower terms, a visual synonym network, the percent of the .  Beautiful!  Except it neglects the past participle and only lists the noun.  If you search for the word “ran”, it gives you a page about the drug ranitidine, but as a link if you wanted the word.  

Next example, it can factor fairly large numbers.  63 digits!  That’s an NP-hard problem.  And you can do many of the things you might do with Mathematica, like take the Fourier transform of something.  And conveniently, you can click on an equation and get a plaintext version that you can paste back into the query bar.  So consider it Mathemetica-Lite.

In chemistry, it has various thermodynamic properties, will calculate the vapor pressure at different temperatures, and boiling points at various pressures.  You can compare silicon dioxide and silicon nitride.  It assumes you are interested in quartz, but you can correct that assumption.  Its easier than looking up something in the CRC or Merck, if folks still do that these days.  In physics, its a bit limited.  Nothing terribly useful beyond being a light front-end for mathematica, although it offers data on particles.  It won’t solve any abstractly defined problems, but it will tell you the stopping power of various materials for beta radiation.  No data for hot neutrons, alpha particles, or photos it seems.    

The financial information is great — return vs. volatility, alpha, beta, R-square.  You can compare the GDP between the United States and Japan.   

Looking at the long list of examples, and all of the fields they cover, its an impressive assembly of data.  The interface is brittle, you have to learn how to format the questions correctly, and its still spotty in areas.  Its clearly not a general-purpose search engine, so its a pity some press outlets have decided they want to be a Google-killer.  

Other reviews have been mixed.  I think it will prove to be a useful research tool for a lot of people. The interface frequently makes the wrong assumption about what you’re looking for, and its far from complete.  That’s apt to improve, but I think still there will be a significantly steeper learning curve than Google.  You need to learn how to write good queries for the particular domain you’re working in, and until you do your results will be spotty.

I’m not sure how they’re going to make money. Its probably not something advertisers are going to like.  Perhaps it will help with sales of Mathematica.  Clearly they have a lot of data locked up in their servers, and you can only get at it in tiny slices at a time.  They could make good money licensing fuller access to that data, even if it were still housed remotely.  Take for example the GDP data, or the census data.  Imagine if you could launch a local version of Mathematica, then call up an array representing that data, and work it into a model.  They have a few hundred terabytes?  All of that could be at your fingertips.