In practice, you should be able to tell that a table is a group table by matching it with a group that you already know. Assuming that you don't look at tables with more than six elements, here are the possible groups:
- Cyclic groups: $C_2, C_4,\dots,C_6$
- The Klein four group $V_4$
- The symmetric group $S_3$.
So if you know what the multiplication tables of these groups look like, you'll be in good shape.
It's also good to know how to eliminate possibilities.
Here are three necessary criteria for a table to be a group multiplication table. When I say necessary, I mean that these can only tell you if the table is not a group table. In other words, if these criteria are satisfied, it is nonetheless possible that the table is not a group table.
Say the elements to be multiplied are $a_1,a_2,\dots,a_n$ (in that order). Then
- There is a row $R$ that reads $a_1, a_2,\dots, a_n$ (in that order), and a column $C$ that reads $a_1,a_2,\dots,a_n$ (in that order). Edit: $R$ and $C$ must intersect on the diagonal.
- Every row contains each of $a_1,a_2,\dots,a_n$ exactly once.
- Every column contains each of $a_1,a_2,\dots,a_n$ exactly once.
Condition 1 comes from the existence of an identity element, and conditions 2 and 3 come from the existence of inverses. (Having inverses basically means that we can always undo multiplication, but if $ax=bx$ for $a\neq b$ -- i.e., if a row has duplicates -- then we can't undo multiplication by $x$. I can give more details if you're interested.)
The condition I haven't mentioned yet is associativity. If I add the condition
- The multiplication is associative.
then conditions 1 - 4 become necessary and sufficient, which means that conditions 1 - 4 are true if and only if the table is a multiplication table.
No reasonable professor will ever ask you to verify that a multiplication table is associative. There are algorithms to do so, which this MSE post mentions, but they're beyond the scope of an introductory algebra class. If conditions 1 - 3 hold and the table is small, it's easier to see if the table matches the multiplication table of one of the groups I listed above.
FranzNietzsche gave a very nice hint in the comment above, but I was too slow to immediately grasp the idea. Also, I found a neat sketch of proof on this webpage. Since no one has answered, I shall just answer myself.
As I described in the original question, property 1 implies every equation of the form $ax=b$ or $ya=b$ has a solution (note that property 1 indeed implies every element in $G$ has a left inverse and a right inverse, but they are not necessarily the same). Then it suffices to show $G$ is a semigroup. Property 1 ensures the operation is defined and well-defined, hence it is left to show associativity.
First consider a rectangle in the multiplication table whose one vertex is $1$ given by $(1,1)$. Choose $x$ in the row, and $y$ in the column. Then the fourth vertex is naturally $yx$.
$$\begin{array}{c|c c}
&1&x\\
\hline
1&1&x\\
y&y&yx\\
\end{array}$$
Now if we pick any other rectangle in the multiplication table, with $1$ as one vertex (not necessarily given by $(1,1)$), $x$ as the other horizontal vertex, and $y$ as the other vertical vertex, then property 2 implies the fourth vertex has to be $yx$, since no matter where it is, this rectangle should be the same as the one noted above, in view of property 2.
$$\begin{array}{c|c c c c c}
&1&&&&\\
\hline
1&&&&&\\
\\
&&&1&x&\\
&&&y&z&\\
&
\end{array}$$
If $z=1$, then trivially we have $(xy)z=xy=x(yz)$. Hence let's suppose $z\neq 1$. Consider the following table:
$$\begin{array}{c|c c c c c}
&1&&y&&\\
\hline
1&1&&y&&v\\
\\
a&&&1&&z\\
\\
x&x&&xy&&w
\end{array}$$
In this case, we assumed $xy\neq 1$, so $ay=1$ for some $a\neq x$. In the rectangle $y-1-z-v$ (top right), we must have $v=yz$ by the previous observation. In the rectangle $1-xy-w-z$ (bottom right), we similarly have $w=(xy)z$. But in the rectangle $1-x-w-v$ (peripheral), we have $w=xv=x(yz)$. Hence $(xy)z=x(yz)$ in this case.
The case where $xy=1$ is easier. Consider:
$$\begin{array}{c|c c c c c}
&1&&y&&\\
\hline
1&1&&y&&v\\
\\
x&x&&1&&z\\
\\
\end{array}$$
Then in the rectangle $y-1-z-v$ (right), we have $v=yz$. Then in the rectangle $1-x-z-v$ (peripheral), we have $z=xv=x(yz)$. But since $xy=1$, we have $z=1\cdot z=(xy)z$. Hence we again have $(xy)z=x(yz)$.
Associativity is proven. Thus $G$ is a semigroup, and it follows that $G$ is a group.
Best Answer
Without associativity, you are right that the statement is false. In fact,