That being said, in Japan it’s actually common to have two fields, one for the name in kanji (the “nice” name), and one in katakana (restricted, 8-bit phonetic alphabet, which earlier/older computer systems used and some probably still use).
Those are translations coming from professionals who swore an oath. Don’t try to do it with code.
Also, USCIS usually has an "aliases" field on their forms, which would be a good place to put German government misspellings.
I know someone that still doesn't know whether they have a middle name as far as american authorities are concerned.
Coupled with "two last names" and it gets really messy, really quickly.
Purchases names don't match the CC name.
Bank statements are actually "for another person".
Border crossings are now extra spicy.
And "pray" that your name doesn't resemble a name in some blacklist.
I think these name problems are pretty common everywhere, though: even in Korea, where bureaucratic tasks are quite streamlined compared to Europe, foreigners whose names are too long get dashes in their official ID. Which official name goes into bank accounts and stuff. Pity the poor souls whoese bank ID doesn't match the telecom ID used for online and mobile ID verification during transactions...and during taxes.
The real problem here is that transliteration rules depend on the source language, and you often don't have the context to distinguish that (e.g. the name can be spelled identically in Russian, Bulgarian, and Ukrainian, yet transliterations will be different). And even where you do have sufficient context, this still requires the system to be able to handle any source language that can be potentially thrown at it, which is just about anything. So it never gets implemented properly in practice.
Debit cards a pretty common example of this. I believe you can only have ASCII in the cardholder name field.
No you can't.
Add: Okay, you need to know why. I'm right here a living breathing person with a government id that has the same name scribed in two scripts side by side.
There is an algorithm (blessed by the same government that issued said it) which defines how to transliterate names from one to another, published on the parliament web site and implement in all the places that are involved in the id issuing business.
The algorithm will however not produce the outcome you will see on my id, because me, living breathing person who has a name asked nicely to spell it the way I like. The next time I visit the id issuing place, I could forget to ask nicely and then I will have two valid ids (no, the old one will not be marked as void!) with three names that don't exactly match. It's all perfectly fine, because name as a legal concept is defined in the character set you probably can't read anyway.
Please, don't try be smart with names.
But even if the user is entering both, warning them that the transliteration doesn't match and letting them continue if they want is something that pays for itself in support costs.
Human systems aren't always interested in avoiding or fixing defects.
And please please please, don't try to be smart and detect the cultural context from the character set before automatically translating it to another character set. It will go wrong and you will not notice for a long time, but people will make mean passive aggressive screenshots of your product too.
My bank for example knows my legal name in Cyrillic, but will not print it on a card, so they make best-effort attempt to transliterate it to ASCII, but make it editable field and will ask me to confirm this is how I want it to be on a card.
Where I live, no less than 3 jurisdictions have a say about my "legal" name, and their laws do not require them to match. At one point, one jurisdiction had two different "legal" names for me, one a typo by my standards, but AFAICT, both equally valid.
There's no solution here, AFIACT, it's just evidence towards why computers cannot be accountability sinks.
For instance, in many Japanese forms, there are dedicated fields for the name and the pronunciation of the name. There are possibly multiple ways to read a name (e.g. 山崎 is either やまざき or やまさき). It is better to ask the person "how to read your name?" rather than execute code to guess the reading.
As for transliteration, it's best to avoid if possible. If not possible, then rely on international standards (e.g. Japanese has ISO 3602 and Arabic has ISO 233-2). When international standards don't exist, then fall back to "context-dependent" standards (e.g. in Taiwan, there are several variants of Pinyin. Allow the user to choose the romanization that matches their existing documentation).
Automatic transliteration seems to be very dangerous (wrong name on bank accounts, for instance), and not always feasible (some unicode characters have more than one way of being transliterated).
Should we apologize to the user, and just ask the user twice, once correctly, and once for the bad computer systems? This seems to be the only approach that both respects their spelling, and at the same time not creating potential conflict with other systems.