Calculations are unimportant. The results depend on different factors of the LED and the battery which can't be known until you do it and measure the result.The answer depends critically on what you meant by "in theory".
A battery could mean an ideal voltage source, or an ideal voltage source + internal resistance; A LED could be modelled as an ideal voltage source, or an ideal voltage source + dynamic resistance (which may be non-linear).
So a simplistic model for a battery + led + current limiting resistor would have a current of
I = (Vb - Vfwd) / (Rb + Rled + R), whereby Vb is the battery's voltage, Vfwd is the LED's voltage drop, Rb is the battery's internal resistance, Rled is the led's dynamic resistance and R is the resistance of the current limiting resistor.
As R -> zero, I -> (Vb - Vfwd) / (Rb + Rled). Obviously, it does not go to infinity if Rb + Rled is non-zero.
But here lies the problem with not using a current limiting resistor: if Rb + Rled is small (if for example it is powered by a set of rechargeable battery, or Li-on battery, or a regulated power source (Rb -> 0)), any small changes in Vb - Vfwd can cause large changes in I;
Worst yet, as the diode heats up, Vfwd decreases, causing I to go up further -> more heat -> a vicious cycle.
Two ways to control that:
1) A large Rb + Rled + R is a negative feedback mechanism in that case. Or
2) a negative feedback loop where you lower Vb as I goes up (or Vfwd goes down).
The 1st approach is typically used for signaling LEDs - simple but inefficient - and the 2nd approach is typically used for power LEDs.