Summary:
Yes "polarised" aluminum "wet electrolytic" capacitors can legitimately be connected "back-to-back" (ie in series with opposing polarities) to form a non-polar capacitor.
C1 + C2 are always equal in capacitance and voltage rating
Ceffective = = C1/2 = C2/2
Veffective = vrating of C1 & C2.
See "Mechanism" at end for how this (probably) works.
It is universally assumed that the two capacitors have identical capacitance when this is done.
The resulting capacitor with half the capacitance of each individual capacitor.
eg if two x 10 uF capacitors are placed in series the resulting capacitance will be 5 uF.
I conclude that the resulting capacitor will have the same voltage rating as the individual capacitors. (I may be wrong).
I have seen this method used on many occasions over many years and, more importanttly have seen the method described in application notes from a number of capacitor manufacturers. See at end for one such reference.
Understanding how the individual capacitors become correctly charged requires either faith in the capacitor manufacturers statements ("act as if they had been bypassed by diodes" or additional complexity BUT understanding how the arrangement works once initiated is easier.
Imagine two back-to-back caps with Cl fully charged and Cr fully discharged.
If a current is now passed though the series arrangement such that Cl then discharges to zero charge then the reversed polarity of Cr will cause it to be charged to full voltage. Attempts to apply additional current and to further discharge Cl so it assumes incorrect polarity would lead to Cr being charge above its rated voltage. ie it could be attempted BUT would be outside spec for both devices.
Given the above, the specific questions can be answered:
What are some reasons to connect capacitors in series?
Can create a bipolar cap from 2 x polar caps.
OR can double rated voltage as long as care is taken to balance voltage distribution. Paralleld resistors are sometimes used to help achieve balance.
"turns out that what might LOOK like two ordinary electrolytics are not, in fact, two ordinary electrolytics."
This can be done with oridinary electrolytics.
"No, do not do this. It will act as a capacitor also, but once you pass a few volts it will blow out the insulator."
Works OK if ratings are not exceeded.
'Kind of like "you can't make a BJT from two diodes"'
Reason for comparison is noted but is not a valid one. Each half capacitor is still subject to same rules and demands as when standing alone.
"it is a process that a tinkerer cannot do"
Tinkerer can - entirely legitimate.
So is a non-polar (NP) electrolytic cap electrically identical to two electrolytic caps in reverse series, or not?
It coild be but the manufacturers usually make a manufacturing change so that there are two Anode foils BUT the result is the same.
Does it not survive the same voltages?
Voltage rating is that of a single cap.
What happens to the reverse-biased cap when a large voltage is placed across the combination?
Under normal operation there is NO reverse biased cap. Each cap handles a full cycle of AC whole effectively seeing half a cycle. See my explanation above.
Are there practical limitations other than physical size?
No obvious limitation that i can think of.
Does it matter which polarity is on the outside?
No. Draw a picture of what each cap sees in isolation without reference to what is "outside it. Now change their order in the circuit. What they see is identical.
I don't see what the difference is, but a lot of people seem to think there is one.
You are correct. Functionally from a "black box" point of view they are the same.
MANUFACTURER'S EXAMPLE:
In this document Application Guide, Aluminum Electrolytic Capacitors bY Cornell Dubilier, a competent and respected capacitor manufacturer it says (on age 2.183 & 2.184)
If two, same-value, aluminum electrolytic capacitors
are connected in series, back-to-back with the positive
terminals or the negative terminals connected, the
resulting single capacitor is a non-polar capacitor with
half the capacitance.
The two capacitors rectify the
applied voltage and act as if they had been bypassed
by diodes.
When voltage is applied, the correct-polarity capacitor gets the full voltage.
In non-polar aluminum electrolytic capacitors and motor-start aluminum electrolytic capacitors a second anode foil substitutes for the cathode foil to achieve a non-polar capacitor in a single case.
Of relevance to understanding the overall action is this comment from page 2.183.
While it may appear that the capacitance is between
the two foils, actually the capacitance is between the
anode foil and the electrolyte.
The positive plate is the
anode foil;
the dielectric is the insulating aluminum
oxide on the anode foil;
the true negative plate is the
conductive, liquid electrolyte, and the cathode foil
merely connects to the electrolyte.
This construction delivers colossal capacitance
because etching the foils can increase surface area
more than 100 times and the aluminum-oxide dielectric is less than a micrometer thick. Thus the resulting
capacitor has very large plate area and the plates are
awfully close together.
ADDED:
I intuitively feel as Olin does that it should be necessary to provide a means of maintaining correct polarity. In practice it seems that the capacitors do a good job of accommodating the startup "boundary condition". Cornell Dubiliers "acts like a diode" needs better understanding.
MECHANISM:
I think the following describes how the system works.
As I described above, once one capacitor is fully charged at one extreme of the AC waveform and the other fully discharged then the system will operate correctly, with charge being passed into the outside "plate" of one cap, across from inside plate of that cap to the other cap and "out the other end". ie a body of charge transfers to and from between the two capacitors and allows net charge flow to and from through the dual cap. No problem so far.
A correctly biased capacitor has very low leakage.
A reverse biased capacitor has higher leakage and possibly much higher.
At startup one cap is reverse biased on each half cycle and leakage current flows.
The charge flow is such as to drive the capacitors towards the properly balanced condition.
This is the "diode action" referred to - not formal rectification per say but leakage under incorrect operating bias.
After a number of cycles balance will be achieved. The "leakier" the cap is in the reverse direction the quicker balance will be achieved.
Any imperfections or inequalities will be compensated for by this self adjusting mechanism.
Very neat.
100 µF is really pushing the limit for ceramic caps. If your voltages are low, as a few volts to 10 or maybe 20 volts, then paralleling multiple ceramics may be reasonable.
High capacitance ceramic caps have their own set of advantages and disadvantages. The advantages are much lower equivalent series resistance and therefore much higher ripple current capability, usefulness to higher frequencies, less heat sensitivity, much better lifetime, and in most cases better mechanical ruggedness. They have their own problems too. The capacitance can degrade significantly with voltage, and the denser (more energy storage per volume) ceramics exhibit piezo effects often called "microphonics". In just the wrong circumstance, this can lead to oscillation, but that is rare.
For switching power supply applications, ceramics are usually a better tradeoff than electrolytes unless you need too much capacitance. This is because they can take much more ripple current and heat better. The lifetime of electrolytes is severely degraded by heat, which is often a problem with power supplies.
You don't need to derate ceramics as much as electrolytes because the lifetime of ceramics is much larger, to begin with, and is much less a function of the applied voltage. The thing to watch out for with ceramics is that the dense ones are made from a material that is non-linear, which shows up as a reduced capacitance at the higher ends of the voltage range.
Added about microphonics:
Some dielectrics physically change size as a function of the applied electrical field. For many, the effect is so small that you don't notice and it can be ignored. However, some ceramics exhibit a strong enough effect that you can eventually hear the resulting vibrations. Usually, you can't hear a capacitor by itself, but since these are soldered fairly rigidly to a board, the small vibrations of the capacitor can cause the much larger board to also vibrate, especially at a resonant frequency of the board. The result can be quite audible.
Of course, the reverse works too since physical properties generally work both ways, and this one is no exception. Since applied voltage can change the dimensions of the capacitor, changing its dimensions by applying stress can change its open-circuit voltage. In effect, the capacitor acts as a microphone. It can pick up the mechanical vibrations the board is subjected to, and those can make their way into the electrical signals on the board. These types of capacitors are avoided in high sensitivity audio circuits for this reason.
For more information on the physics behind this, look up properties of barium titanate as an example. This is a common dielectric for some ceramic caps because it has desirable electrical properties, particularly fairly good energy density compared to the range of ceramics. It achieves this by the titanium atom switching between two energy state. However, the effective size of the atom differs between the two energy states, hence the size of the lattice changes, and we get physical deformation as a function of applied voltage.
Anecdote: I recently ran into this issue head-on. I designed a gizmo that connects to the DCC (Digital Command and Control) power used by model trains. DCC is a way to transmit power but also information to specific "rolling stock" on the tracks. It is a differential power signal of up to 22 V. Information is carried by flipping the polarity with specific timing. The flipping rate is roughly 5-10 kHz. To get power, devices full wave rectify this. My device wasn't trying to decode the DCC information, just get a little bit of power. I used a single diode to half wave rectify the DCC onto a 10 µF ceramic cap. The droop on this cap during the off half-cycle was only about 3 V, but that 3 Vpp was enough to make it sing. The circuit worked perfectly, but the whole board emitted a quite annoying whine. That was unacceptable in a product, so for the production version, this was changed to a 20 µF electrolytic cap. I originally went with ceramic because it was cheaper, smaller, and should have a longer life. Fortunately, this device is unlikely to be used at high temperatures, so the lifetime of the electrolytic cap should be a lot better than its worst case rating.
I see from the comments there is some discussion about why switching power supplies sometimes whine. Some of that could be due to the ceramic caps, but magnetic components like inductors can also vibrate for two reasons. First, there is force on each bit of wire in the inductor proportional to the square of the current thru it. This force is sideways to the wire, making the coil vibrate if not held in place well. Second, there is a magnetic property similar to the electrostatic piezo effect, called magnetostriction. The inductor core material can change size slightly as a function of applied magnetic field. Ferrites don't exhibit this effect very strongly, but there is always a little bit, and there can be other material in the magnetic field. I once worked on a product that used the magnetostrictive effect as a magnetic pickup. And yes, it worked very well.
Best Answer
You should consider the reduction in capacitance that ceramic capacitors exhibit under bias. This is particularly true of the type of capacitor or capacitors that you will need to get to 100uF/16V.
For example, take this 22uF/16V X7R capacitor:
As you can see, at 12V, the capacitance has dropped to about 9uF, so you would need 11 of them to get 100uF with 12VDC bias, rather than the 5 you might have thought.
Electrolytic capacitors have stable value regardless of bias voltage.
The other possible consideration is that if you have something like a voltage regulator connected to the capacitor, the ESR may be too low when using the ceramic capacitors. Normally that might be considered an advantage, but some regulators are not stable (i.e. may oscillate at high frequency under some load, line or temperature conditions) if the ESR is too low. You can simply put a small resistor in series to mimic the electrolytic cap if that is the case.