Update: This question is based on a wrong assumption, the problem has now been fixed. The problem was an incorrect register setting in the RTC. However the answers given regarding load capacitance for crystal oscillators may have value to people in general situations.
I have a prototype PCB of my own design. It was originally designed to use an ST brand "M41T81S" real time clock chip with a cheapo 12.5 pF watch crystal. Timing accuracy is not important (+/- 5 seconds per day drift is fine).
I have swapped out the old RTC and I'm now using a Microship brand "MCP7940N" instead, which is a pin-for-pin match for the original and has fully compatible voltages. The new chip works fine as far as the I2C communications goes. I can program the registers and read them out without problems.
The original design does not include load capacitors as the old chip didn't need them, it kept reliable time and didn't miss any clock pulses.
However the new chip does appear to have stricter requirements for the oscillator circuit and only sporadically responds to the crystal's signal (perhaps 1 in a million pulses is registered). Most of the time it does not "keep the time" at all. It effectively cannot see the crystal signal.
The scope reads a clean and stable 400mv p-p sine wave on the XTAL1 pin with the new RTC. I'm guessing this isn't suitable for the new chip, the datasheets for the xtal and new RTC do not specify the waveforms produced or required.
I admit that the new chip's datasheet says that "12.5 pF crystals are not recommended". Fair enough, my own stupid fault for buying these parts without properly checking for compatibility. Doesn't really matter in the long run.
Given that I've already made a dog's dinner of this prototype, is there a way that I can hack something together in the short term in order to get this new RTC to operate with the unsuitable 12.5 pF crystal and lack of load caps? It's only a prototype and definitely needs a re-spin of the layout for several other reasons besides this problem.
I figured I should change the crystal to a 6-9 pF as stated in the datasheet for the RTC but that still leaves me with nowhere to put the load caps. I've never had to mess much with crystal circuits before and could use some advice about just bodging this prototype to get this "proof of concept" board to function.
(If this turns out to be impossible then I still have the option of running the RTC from an external signal such as a microcontroller outputting a square wave, so all is not lost. I'm mainly asking this question out of curiosity rather than necessity)