A fifth or sixth workaround (Version 14.1 and higher):
SetAttributes[y, NonThreadable]; NDSolve[{y'[t] == -{1, 0} y[t], y[0] == {1, 2}}, y, {t, 0, 1}] (* {{y -> InterpolatingFunction[{{0., 1.}}, "<>"]}} *)
This feels somewhat embarrassing to me:
In[7]:= Assuming[{y \[Element] Vectors[2]}, {1, 0}*y] Out[7]= {y, 0}
or this
In[11]:= Assuming[{y \[Element] Vectors[2]}, 0*y] Out[11]= 0
One more workaround:
y = ArraySymbol["y", {2}]; NDSolve[{y'[t] == -{1, 0} y[t], y[0] == {1, 2}}, y, {t, 0, 1}]
Then how can you explain David's point?
Setting the 0 term to a very small nonzero number appears to work:
No, y'[t] == {-y[t], 0} is a well-defined equation, because 0 can have any dimension. The point here is, the first equation by its form is a vector differential equation, so it should work in the same way as the second equation does, but the second works while the first doesn't. Also, as david noticed, it isn't a consistent behavior. So I think I should file a bug report, but how (if possible) to do that?
