At 01:20 PM 11/1/95 -0500, Michael Froomkin <froomkin@law.miami.edu> wrote:

I thought a property of Chaumian DigiCash was that a coin *had* to go back to the bank before it could be spent again.

No. The basic Chaum Digicash method looks like this: 1) Alice creates a number of a recognizable form (Chaum's 1985 CACM paper uses n1n2n3...n64n1n2n3....n64, i.e. a 64-bit number concatenated with itself). 2) Alice blinds the number and sends it to the bank (along with some request for withdrawing money from her account or payment in other coin or whatever.) 3) The bank signs the number and sends it back. 4) Alice unblinds the coin; now it's good, recognizably signed, and untraceable. 5a) Alice gives the coin to Bob, who deposits it; the bank records the coin number, and in case of double-spending, the first person to the bank wins. This is useful for on-line transactions, or off-line where everyone trusts each other. OR 5b) Alice gives the coin to Bob using a complicated cut&choose protocol that doesn't give away her identity if it's only used once, but if she also gives the same coin to Carol with the same protocol, Bob and Carol can identify Alice with probability 1 - 1/2**n, for some adequately large n. This is more work, but you can use it for off-line transactions where you don't trust Alice not to double-spend. The protocol doesn't say what to do to Alice if you catch her cheating; depending on the environment you can debit her account or sue her etc. 6) Bob now has a number, signed by the Bank of Foo, which he can either give to them to deposit or get cash or use for highway toll (if Foo is really the highway company) or give to somebody else to spend (which is a little messy in the cut&choose method.)

Logically, I can see at least four possibilities: 1) payee data is encoded onto the coin at time of payment, making it impossible for Carol to bank the coin. I see no evidence of this in the docs at the Digicash site, but I just rechecked quickly and may have missed it.

The basic protocol doesn't say anything about what a valid coin looks like; you could use the example in Chaum's paper or a long string followed by a checksum or whatever. You _could_ put the payee's name account number in the string as the 64-bit "random" number, or even put both payer and payee. The bank could insist on that sort of thing if they wanted. If I remember right, the version in the Digicash trial left you the choice of filling in a specific payee or using "@" for bearer-payable coins.

2) No payee data as such is encoded on the coin but it is marked "spent" to prevent multiple uses by payee to the detriment of payor.

The bank marks the coin spent upon deposit.

3) the Digicash software only allows you to send a "spent" coin to the bank. You have to hack the software to send the coin to Carol (do you have to break your own key?).

I don't know if their merchant-client software lets you do this or not, but it's just a matter of implementation, not protocol.

4) nothing in the DigiCash software or protocol prevents you from sending a coin to Carol so long as you trust Carol not to get you in trouble by misusing the coin in some way. That's why Chaum is interested in hardware based agents that would keep you from respending coins you receive.

Your problem isn't trusting Carol not to get you in trouble, it's trusting Alice not to spend the coin again. Hardware-based agents are interesting because they make it easier to enforce double-spending prevention in off-line systems, and to offer better anonymity because you've got more trust that the person didn't double-spend. Stefan Brands has done a lot of work on this. In on-line systems you can check whether a coin's been spent already by depositing it - the problem is that on-line systems aren't always convenient for many applications (e.g. newspaper machines), and the costs of communication for an on-line system may be higher than the cost of a sufficiently smart smart-card. #--- # Thanks; Bill # Bill Stewart, Freelance Information Architect, stewarts@ix.netcom.com # Phone +1-510-247-0664 Pager/Voicemail 1-408-787-1281 #---