The slew rate of a 741 (0.5V/us typical) is inadequate for high frequencies. A +/-10V sine wave output at 200kHz requires a slew rate about 25x higher than a 741 is capable of.
Try a JFET op-amp such as a TL081 or LF351.
Startup can come from several things
In general, there is going to be a source out there for the initial excitation of an oscillator's feedback loop.
- The power-on ramp could get coupled into the output of the amplifier slightly, providing an edge that excites the feedback network
- Some noise on the output lead could excite the feedback network at the right frequency to pass through to the input
- Some noise could show up at the op-amp input and be amplified to be filtered down to the right frequency by the feedback network
Keep in mind that everything, even a humble resistor, has inherent noise sources when you're doing this, by the way!
The AC signal can be coupled back just fine for feedback
While DC can't pass through C2, AC can get through it just fine, and that's what matters for oscillator feedback.
As to that lamp? It's an AGC system
The incandescent lamp in the Wien-bridge oscillator is a gain control mechanism -- at startup, it's cold and offers a low resistance, causing the amp to have high gain and amplify tiny noises greatly. As it heats up from its own current draw, its resistance increases, causing the gain of the amp to drop significantly and stabilizing the oscillator at an even output level.
Likewise, if the output increases, the voltage across the lamp will increase due to voltage divider action, causing the lamp to heat up more and the gain to drop. If the output decreases, the voltage across the lamp will decrease, allowing the lamp to cool slightly and the gain to rise. It is a highly elegant system that is actually quite annoyingly hard to replicate using solid-state parts.
Best Answer
I think the DDS is the way I'd do it but, if you want a straight 5V logic solution you can use a 4060 logic chip. Here's one - the device uses a reference frequency from a crystal (or RC network) and offers a frequency divided down version - it's a 14-stage binary ripple counter and if you use a 1MHz clock you can easily derive 976.56Hz.
Next, if you are not happy with a square wave you can apply low-pass filtering to extract the fundamental frequency. Depending on how much low-pass filtering you do determines how pure the sinewave is. Here is a good application note about turning squares into sines. None of it is rocket science hence I'm letting the links speak for themselves.