> You can draw a strait line in China with 6% of the population on the larger half, and 94% of the population on the smaller half.

America's most populous county, Los Angeles County, is approximately the same. The northeastern part is mostly very hot high desert or national forest land (severe building restrictions). By contrast, the southwestern part is coastal and much more suitable for human habitation.

http://maps.latimes.com/neighborhoods/population/density/nei...

In general, Californians live in the third of the state along the Pacific coast.

You could probably do the same in Canada fwiw.

Technically you could draw such a straight line in any country.

You could draw a straight line across any country that divides the population into 6%/94%, but I doubt there are many countries where the 94% side would be smaller than the 6% side for any line.

This is such an interesting idea! Lots of potential.

Given n (city,population, area) tuples for country C, find 2 cities, the line thru which slices C into two land masses M and N. M would then have k tuples and N would have n-k, such that the ratio of aggregate population of k tuples to the aggregate population of n-k would be 90+delta:10-delta, for delta in interval [1,9], whilst simultaneously you have the ratio of the aggregate area of the k tuples to the aggregate area of n-k working out to 1+nabla:1-nabla, for nabla in [0,0.5].

No idea what the optimal algorithm would be, short of brute force so O(n^2).

This is the sort of DS project a HuffPo would run - you can slice up countries not just based on 94:6 population but also crime rate, education, victims of homicide, what have you; all you need in tuples with larger arity.

"tuples with larger arity" is just computer scientist speak for "take more things into account".

Given the generality and popularity of sweepline algorithms in computational geometry, I'm wondering where the problem is here. Can't you just move a sweepline across a given territory in any direction and get a result? If you want maximum outrage (largest value of X in smallest area), I agree that's more complex and I'm not sure how you would prove an optimum result.

Given the city-local nature of most population density, I'd imagine a grow-out-from-cities + line fit model would work fairly well at lower computational complexity.

It'd be curious what different countries look like in terms of gradient decent difficulty on population density. I'd imagine most look fairly similar (densely populated regional urban areas surrounded by rural), but I imagine there are exceptions to some degree out there?

There are not that many city's in any given country so N^2 is probably not a big deal. However, you can get approximate results by drawing a line at every angle through a country and then moving that line up and down. Then just look for city's near any good candidate lines.

> drawing a line at every angle through a country and then moving that line up and down.

This part is quite messy. For every given (c1,c2) city pair, you'd get 1 straight line that connects c1 to c2. Then for every city ck, you need info on whether ck is above that line or below - you can compute that by point-slope geometry with the lat-long coordinates ( given point p(x,y) & line l with some slope, is p above l or below ).

If you want to poke around, the dataset is here: http://seer.cancer.gov/popdata/

Consider a country with utterly uniform population density. It is trivially false that you can draw "a straight line with 6% of the population on the larger half, and 94% of the population on the smaller half" in such a country. I would be very surprised if all countries in the world were far enough from "uniform population density" for your statement to apply to them.

But you couldnt draw a straight line that has half the land area on each side yet 94/6% split of population.

True enough. Though Canada has more land than China and less than 3% of China's population. So, it's not nearly as big a deal.

Is it really uninhabitable? The territories in the north maybe too harsh, but with global warming helping too I think the southern half is quite alright, which is huge. It's all forest, lots of lakes, no lack of water and not too high, and no desert.

On the western half:

The 'low' areas have minimal rainfall and are 1KM over sea level. http://www.chinamaps.org/china/china-map-of-precipitation-an...

The high areas get a little more rainfall, but are mostly 4-5+km over sea level. http://www.china-food-security.org/data/maps/dem/dem_h.htm

Combine them and you can see why the population map looks the way it does. http://www.china-mike.com/china-travel-tips/tourist-maps/chi...

Uhm... Canada, not China. I responded to the comment that said

    > You could probably do the same in Canada fwiw.

I would not ask about China, I've seen enough documentaries about their deserts and high plateaus. Some of them about the silk road, which was not an easy one because of that geography.

From what I see, there's a pretty good correlation between the population density map (https://www.britannica.com/place/Canada/images-videos/Popula...) and the portion of Canada at least sporadically covered by permafrost. (http://www.enr.gov.nt.ca/state-environment/13-permafrost)

In addition to the rather obvious climate this implies (eg, rather cold), from what I understand, building anything on permafrost is very challenging.

"sporadically covered by permafrost" - I think you're confused as to the permanency of permafrost.

It's no more uninhabitable than the uninhabitable areas of China. But it is non-arable.

Most of the eastern half of Canada north of a certain is "Canadian shield" and is granite shield rock with little if any topsoil.

The other half is boreal forest and some of it is indeed farmed to a fairly high latittude (central Alberta where I'm from) but once you get up high enough the growing degree days are not sufficient to grow a lot of crops. It's good for grazing if that. And also fairly arid in spots.

And the rest is mountains, or arctic.

Canada and China are roughly the same size.

Tibetless China is still a lot bigger than India and doesn't have that 94% property. The spirit of the GP's comment holds, I think.

China:

  ~94% of 1.357b = 1.28 billion population. 
  ~40% of 3.7 mill = 1.5 million square miles.
  = ~0.85 billion people per million square miles.

India

  1.252 billion people on
  1.269 million mi
  =~ 1 billion people per million square miles.

So, India is around 15% denser. Of course you can avoid the strait line cut and push China's density much higher, but the point is they are not that far off where people actually live.

Where are you getting the 40% figure from? Are you saying that Tibet(-ish) is 60% of the land area of China?

Eyeballing: http://www.undertheraedar.com/2012/06/population-of-china.ht.... Tibet is part of that side with 6% of the population, but rainfall is a major issue limiting population over a much larger area.