How to convince myself (imagine) that $\Bbb S^1$-action on $\Bbb S^3$ fixes a circle of sphere? 
How to convince myself (imagine) that $\Bbb S^1$-action on $\Bbb S^3$ fixes a circle of sphere?

Due to this comment of Jason DeVito, it is easy to see that action of $\Bbb S^1$ on $\Bbb S^3\subset \Bbb C^2$ defined by $z*(w_1,w_2)=(zw_1,w_2)$ fixes the entire circle $\{(0,w):|w|=1\}\subset\Bbb S^3\subset \Bbb C^2$.
But I can't imagine it, because the common picture of action in my mind is that an circle action is a kind of rotation, so it has a rotation axis and spinning around this axis can fix at most 2 point. Is it possible that the axis of rotation is not a line?

Now, how can I think about this action geometrically? $z*(w_1,w_2)=(zw_1,\bar zw_2)$.

Edit: My understanding of last action is that: one side of $\Bbb S^3$ is spinning clockwise and other side is spinning counterclockwise (in different plane from first action) and these actions effect on the middle of sphere and it become scarious and kink in middle, Like cylinder if we spin the boundaries of it in different directions it become kink in middle like screw.
 A: One would not expect a rotation in $\mathbb C^2 \approx \mathbb R^4$ to have an "axis of rotation" which is a line, i.e. something of real dimension $1$. On the other hand, one would expect the "axis of rotation" to have real codimension $2$, which it does: the entire plane $w_1=0$ is fixed. And when you intersect that plane with $S^3$ you get a circle that is fixed.
If you want to visualize this example, it can be done using the fact that $S^3$ is the one-point compactification of $\mathbb R^3$, which I'll write as $S^3 = \mathbb R^3 \cup \{\infty\}$. In this model, one can visualize the circle of fixed points as the unit circle in the $x,y$-plane:
$$\{(x,y,0) \mid x^2 + y^2 = 1\}
$$
Outside of this circle of fixed points, every other orbit of the action is a circle, and one can visualize these circle orbits in $\mathbb R^3 \cup \{\infty\}$ using $(r,\theta,z)$ cylindrical coordinates, as follows. One of the circle orbits is $\text{$z$-axis} \cup \{\infty\}$. Then, for each constant angle $\theta_0$, the half plane $\theta = \theta_0$ pierces the fixed circle at the single point $P(\theta_0)$ with coordinates $(r,\theta,z)=(1,\theta_0,0)$, the boundary edge of that half plane is the $z$-axis which is an orbit, and the rest of the half-plane is foliated by a family of circle orbits which approach that single point in one direction getting smaller and smaller, and which approach the $z$-axis in the other direction getting larger and larger (in the hyperbolic metric $\frac{dr^2+dz^2}{r^2}$ on this half-plane, these are the concentric circles centered on $P(\theta_0)$).
A: For me, the way I think about rotations is a consequence of the maximal torus theorem for $\mathrm{SO}(n)$.  Namely, given any $A\in \mathrm{SO}(n)$ (i.e., a rotation of $\mathbb{R}^n$ which fixes $0$), there is some basis of $\mathbb{R}^n$ with the property that in this basis, $A$ consists of a bunch of regular $2$-dimensional rotation blocks.
More precisely, writing $R(\theta) = \begin{bmatrix} \cos\theta & -\sin\theta \\ \sin\theta & \cos\theta\end{bmatrix}$ for the standard counterclockwise rotation matrix, there is always an orthonormal basis of $\mathbb{R}^n$ in which $A$ takes the block diagonal form $$A=\begin{cases} \operatorname{diag}\Big(R(\theta_1), R(\theta_2),..., R(\theta_{n/2})\Big) & n \text{ even}\\ \operatorname{diag}\Big(R(\theta_1), R(\theta_2),..., R(\theta_{(n-1)/2},1)\Big) & n \text{ odd}\end{cases}.$$
This indicates that rotations are fundamentally two-dimensional ideas which are then bootstrapped to higher dimensions.  In fact, it gives a recipe for constructing all rotations of $\mathbb{R}^n$:  Pick any $2$-plane and rotate it a bit.  In the orthogonal complement, pick any $2$-plane and rotate it.  In the orthogonal complement of these two $2$-planes, pick any $2$-plane and rotate it, etc.
Thinking about $\mathbb{R}^3$ for a moment, a rotation in the $xy$-plane doesn't change the distance from a point in the $xy$ plane to any point in the $z$-axis.  In fact, a rotation in the $xy$ plane has no effect on the $z$ axis.  The above decomposition indicates that this idea propagates to higher dimensions.  For example, in $\mathbb{R}^4$ (with coordinates, say, $(x,y,z,t)$) a rotation in the $xy$ plane doesn't change the distance from a point in the $xy$ plane to a point in the $zt$ plane.
This is why, for example, your action on $\Bbb S^3$ can rotate two things in opposite directions.  It's hard to visualize, but a rotation in the $xy$-plane has no effect on the $zt$-plane, so no "twisting" of $\Bbb S^3$ occurs in your action.
On the other hand, for your cylinder action, note that your action is not a rotation of $\mathbb{R}^3$ restricted to the cylinder, so none of the above applies.  In fact, I wouldn't call your action on the cylinder a rotation.  It is a rotation on each boundary component, but who knows what it is in between!
