Is doubling a point irreversible?

multiply by the modular inverse of 2

Are there any irreversible operations that can be done on a point, that can also be done on the scalar?

e.g. f(k) = k', f(K) = K', and k'G = K'

not that I know of

I started collecting perf test results today. Is there anyone who likes data analysis and wants to take a crack at it? I will have a huge number of results (> 500 probably) that look like this:

```

CLSAG || Size (bytes): 1488 || batch size: 1 || inputs: 1 || outputs: 2 || rangeproof split: 0 || ref set size (2^4): 16

test_mock_tx<mock_tx::MockTxCLSAG> 1636129878 50 8685172.000000 8846827.000000 8775139.920000 8773286.000000 31912.109430 0.058095 8685172 8731500 8749842 8762462 8769708 8772274 8779010 8791740 8801171 8809290 8846827

```

If there are no takers, I can do it manually lol :)

UkoeHB: I can do it, but I need a bit of context. Such as: What do these numbers mean?

The first line is test parameters, the second line is test results (from `TimingsDatabase::instance` in the monero `src/common/timings.h` class).

size isn't a test parameter, it is just the tx size given the other parameters

UkoeHB: I see. It's summary statistics. What is the number of observations? 50?

right

can change if necessary; the stdev is really low on gingeropolous's test machine

The parameters are "batch size", "inputs", "outputs", "rangeproof split", and "ref set size"?

Also the tx type, CLSAG in this case

Ideally we'd want to test across multiple machines with various hardware and OS specs, I'd think. But this is a good first stab

and ref set decomposition (the 2^4 bit, i.e. n^m), which is relevant for other tx types

yeah

Ok I'll do it. Just post the data when ready :)

thank you!

UkoeHB: What are the research questions we are trying to answer with this data analysis, by the way?

I suppose it's "what are the tradeoffs between different parameters?", but maybe you can be more specific.

There are a couple things I am thinking about.

- comparisons between CLSAG, Triptych, Seraphis variants

- tradeoffs between size/verification cost for different design choices; these tradeoffs are multivariable - one choice may be relatively better at few in/outs (for example), while another is relatively better at many in/outs

- if there are design choices that are worse in all of size/verification/feature-set, then I want to eliminate those from consideration

Rather than 'research questions', these are more like 'engineering questions'. These performance numbers are meant to be useful for an engineer trying to decide what to implement. It is more a data-visualization effort than a conclusion-finding effort :).

Got it

I have some questions. For CLSAG are we talking BP+ namely after the HF? Also I presume ring size 16 if we assume there is where the consensus is going. It will also helpful to include a tx with 2 in and 2 out since it is the most common tx

Thanks

yes

yes to all