Perfect player expected return

[AK47]

Hi, anyone know where i can find this kind of informations

i would like to know what is the maximun expected value that a can be extract from a blackjack game using all informations(card counting) and by using optimal betting strategy from a infinite bankroll

example: what is the expected value of the optimal player in
AC rules
6 deck
Hit on soft 17
fair deck penetration (30%)
player can bet from 5$ to 100$

i assume is over 50% but what the value?

 

[KenSmith]

I also wouldn't describe 30% penetration as "fair". That's about as bad as it gets.

This wouldn't happen to be an online casino would it?

Regardless, I'm not sure exactly what number you are asking for. Once you mentioned 50%, you lost me. I'm guessing this game is not beatable if you have to play every hand. If you can sit out bad shoes, then it would be worth something, probably $1 or $2 per 100 hands would be my initial guess.
But it is just that, a guess.

 

[AK47]

ok , sorry about the penetration thing my estimate should be wrong i'm a newbie i didn't even played blackjack in a b&m casino because i'm still 20 anyway that not the point.

i just want to know if its exist any computed simulation of a perfect player against real b&m rules ?. This should be easy to comptute since computer play perfectly in blackjack.

i assusse that robot follow a optimal betting strategy and adjust strategy based on the perfect count

this value should give us a theorical maximum EV that we can use as a benchmark when comparing counting system.

if its not created yet (i'm sure someone think about that before) i will go on and create mine

 

[KenSmith]

The problem with calculating perfect play advantage with typical penetration is one of scale. The problem involves a massive number of possibilities. This same process of combinatorial analysis is used to determine the optimal basic strategy, but that task gets multiplied by many, many more combinations of cards if you want to deal multiple hands into the deck after shuffling.

Peter Griffin offered some educated guesses about how much perfect play would be worth in a single deck game, and I'll have to refer to his book to see what his results showed.

I don't think you'll ever see a brute-force calculation of perfect play value, even to 30% penetration. I don't think there's enough computing power on the planet to finish the task before we're all dust. There are certainly shortcuts available though, so perhaps an answer is out there somewhere.

It's an interesting question.

 

[AK47]

i was thinking about using monte carlo for estimating EV of the perfect player shoud be imaginable

 

[zengrifter]

Monte carlo doesn't compare to the current specialized state-of-art BJ simulation PC programs that exist today. But when you say "perfect play" it may mean something different to you than to us. Perfect play from our perspective could require a tandem-array AI-driven super computer. zg

 

[KenSmith]

Ah, yes, I think we're skinning two different cats here.

If your "perfect play" question means using a counting system perfectly to make decisions, then this question can be answered for any particular counting system quite easily, using any of the widely available simulation software.

I assumed you meant REAL perfect play, where every decision is mathematically perfect. Counting systems are merely an approximation to that.

For most of you reading along, you can drop out here, because now it gets more technical...

There's a challenge in between the easy and the impossible ones here, where you could use a different counting system for every individual decision. Since the 'effect of removal' of each rank of card is different for each decision, this would allow you to use the best possible counting system for each decision. Griffin's Theory of Blackjack includes what you need to create these hundreds of separate counting systems, but even that laborious task would not yield truly perfect play. The problem is that ANY card counting system based on effect of removal is an approximation to the real situations that arise during play.

The effect of removal of a single rank is actually impacted by the particular composition of the deck subset you're working with. This complexity can't be readily modeled by any linear process.