perm filename BKG[BKG,RWF] blob sn#856249 filedate 1988-04-21 generic text, type C, neo UTF8
COMMENT ⊗   VALID 00040 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00003 00002	BGMESS
C00005 00003	DOUBLG
C00015 00004	PAULM
C00026 00005	3chkrs
C00030 00006	dbgbkg.tex
C00036 00007	MARKET
C00039 00008	MARKET
C00042 00009	MARKET
C00044 00010	MARKET
C00045 00011	MARKET
C00046 00012	MARKET
C00047 00013	MARKET
C00048 00014	BGGUID
C00058 00015	ROLLS
C00063 00016	DVRSK1 TEX   
C00076 00017	ZEROSM
C00081 00018
C00084 00019
C00085 00020	ZEROSM
C00086 00021
C00087 00022	ZEROSM
C00088 00023
C00089 00024
C00090 00025
C00091 00026	ZEROSM
C00092 00027
C00093 00028	ZEROSM
C00094 00029
C00095 00030	ZEROSM
C00097 00031	ZEROSM
C00100 00032	ZEROSM
C00102 00033	UNHAPP
C00104 00034	UNHAPP
C00107 00035	UNHAPP
C00109 00036	UNHAPP
C00110 00037	PEOPLE.TEX
C00122 00038	PEOPLE
C00125 00039	PEOPLE.TEX
C00132 00040	PEOPLE.TEX
C00135 ENDMK
C⊗;
BGMESS


About your BG message of 6 October 82, at match point it's a very close decision
whether to put a second checker behind your 7-2 prime; I would do it because
the opponent may not be able to link on my 1-point.  If my prime were 10-5,
I would abstain; what I fear is that my opponent will link on the 3 or 4 point,
and run with 6-6 or 5-5 when I open up.  I estimate that one checker, closed
out, can win less than 5% of the time.  One expert says 3%.  A strong player
I know tried to get me to hit a second checker when I had an 8-3 prime. It's close.

What you call "games that are all defence" (I didn't know you were British),
they are called prime-vs-prime games.  They are notoriously difficult: when to
double, when to take, whether to split the back men, whether to flee or stay, etc.
As a rule of thumb, build your own prime very strong (e.g. 7-3, opponent with two
checkers on the ace) before making any escape play where you might be closed out.
Timing is everything.  Often I slot my bar point in situations like the one you
describe, if I don't think I can make it naturally; if not hit, I can double, or
cover and then double.
Directory for BKG
DOUBLG

Doubling Policy in the Racing Game (c) 1981
Bob Floyd

     Every basic backgammon book has a different formula for doubling  and
taking doubles when the game  has become a race  to bear off.  They  can't
all be right; maybe  all of them  are wrong.  I decided  to find the  most
ccnvincing evidence I  could for  correct doubling,  redoubling, and  drop
points, during the bearoff.

     First, I simplified the game to one  where no pips are wasted, as  if
each player had  only one checker  left on an  enormous backgammon  board.
Let's call this the  pip-count race, or PCR.   I programmed a computer  to
calculate, for  every pair  of  pip counts  in  the PCR,  the  expectation
(equity, fair settlement value) with X on roll for four situations: (1)  X
has the cube (2)  centered cube (3) O  has the cube (4)  cube not in  use.
From the equity with the  cube out of play, it  is also easy to  calculate
the raw probability of X winning the game, by the formula P=(E+1)/2.

     I found the doubling points summarized in Table 1.  For example, when
the player  on roll  has 80  pips  to go,  he should  double if  his  lead
(difference in pip  counts) is more  than 6.90,  at which point  he has  a
71.5% chance  of winning  if  the game  is rolled  out  to the  end.   The
fractional pip count lead of 6.90 is, of course, a fiction; it  represents
an interpolation between X's loss of equity on doubling with a 6-pip lead,
and his gain of equity on doubling with a 7-pip lead.  The last column  of
the table shows, then, that  X, with 80 pips to  go, doubles with a  7-pip
lead, redoubles with a 9-pip lead, and doubles O out with a 12-pip lead.

Table 1:  Doubling in the PCR

Pip count of 
player on roll		20	30	40	50	60	70	80

Lead in pips to
justify double		-1.43	 0.60	 2.21	 3.56	 4.71	 5.75	 6.90

Raw probability at
doubling point		 0.631	 0.661	 0.681	 0.693	 0.701	 0.707	 0.715

Lead in pips to
justify redouble	 -.83	 1.43	 3.29	 4.75	 5.95	 7.08	 8.06

Raw probability at
redoubling point	 0.655 	 0.687	 0.708	 0.719	 0.726	 0.731	 0.734

Lead in pips to
justify dropping	 2.58	 4.63	 6.42	 7.81	 9.03	10.12	11.21

Raw probability at
drop point		 0.766	 0.771	 0.777	 0.779	 0.780	 0.781	 0.782

     In a real bearoff, however, the underdog's chances are not as good as
in a PCR.  In a PCR, when X has 20 pips and O has 22, X has a 75.2% chance
to win.   In  a  real bearoff,  taking  as  X's position  a  composite  of
positions (A) and (B),




(A)				(B)





and taking as O's position a composite of (C) and (D),





(C)				(D)






X has a  76.1% chance  of winning.  The  difference is  that the  underdog
needs large doubletons to  catch up, but they  can't be used  efficiently.
If O rolls a 4-4 in position C, or a 5-5 in position D, he still needs two
more rolls to bear off.

     To correct  the  PCR  table  for use  in  real  backgammon  races,  I
investigated the difference in likelihood for the underdog to catch up  in
the PCR and in real races at a given pip count, using the same dice  rolls
for both.  (For connoiseurs of statistical methodology, I used  stratified
sampling to avoid having any one roll that plays inefficiently  occur  too
often or too seldom.)  If we say the  advantage to X is his lead in  pips,
plus four pips because he is on roll, then X needs only about 90% as  much
advantage in real backgammon as in the PCR to have a given probability  of
winning.  So,  the  table on  doubling  in the  PCR  should have  all  the
advantages scaled down by about 10%.  This gives Table 2, which I think is
a good guide to doubling in real backgammon positions where both sides are
well placed for the bearoff, like position (E).










(E)











Table 2:  Doubling in the Bearoff

Pip count of
player on roll		20	30	40	50	60	70	80

Lead in pips to
justify double		-1.69	 0.14	 1.59	 2.80	 3.84	 4.87	 5.81

Lead in pips to
to justify redouble	-1.15	 0.89	 2.56	 3.88	 4.96	 5.97	 6.85

Lead in pips to
justify dropping	 1.92	 3.77	 5.38	 6.63	 7.73	 8.71	 9.69

     Yes, I understand.  You don't want to memorize this table.  You  have
several things you'd rather do, such  as reporting your true winnings  for
the past year to the IRS, or telling your wife about your true losses,  or
standing in line to  buy an AMTRAK ticket,  or rereading your high  school
trigonometry book.  All right.  If the pip count of the player on roll  is
between 20 and 80 pips, the following formulas are accurate to within  one
pip.

	Drop if the opponent's lead is  more than P/8, where P  is
	his pip count.  Double if you are within four pips of  the
	opponent's drop point.  Redouble  if you are within  three
	pips of the opponent's drop point.

For example, in position (E), P  is 70; P/8 is a  bit under 9, and with  a
10-pip lead X should double and O should drop.  Move O's checker from  the
9-point to the 4-point, giving X a  5-pip lead, and X should double (5  is
more than P/8-4) but not redouble (5 is less than P/8-3).

(RWF: rewrite next paragraph)

     Be careful,  though;  if the  underdog  has a  position  where  large
doubletons don't play efficiently, like position (F), the leader should be
much more ready to double, and the trailer, to drop.






(F)






     In fact, when you are far behind, your checker play must assume  that
you will eventually roll a large doubleton.  In position (E), for example,
X has a 14-pip advantage: a ten pip lead and the roll.  Since the  average
roll is 8  pips, O  is still 6  pips behind  if he rolls  4-4.  While  O's
chance in position (E) is better than 20%, he has only a 5% chance to  win
the game  without rolling  a 6-6  or  5-5.  He  must therefore  strip  the
4-point and try to reach a position like (G)

(RWF: correct this example)





(G)				(H)








rather than (H), although if the race were close, (H) would be better.
PAULM

How Paul Magriel almost gave up backgammon, 
as the result of Wayne McClintock's misplayed ace.

by Bob Floyd
copyright 1982


It happened in Round 5  of the Black &  White Chicago Classic, in  October
1979.  Magriel  had  beaten  Gillman,  Anderson,  Yen,  and  Boyd.   Wayne
McClintock was hoping to  avenge his friend Bill  Boyd, who had been  tied
12-12 in a seventeen point match with Magriel until swept away by a gammon
with the cube at four.  The first  game had become a straight race; not  a
checker had been hit.  Both players were tired and making mistakes.  Twice
McClintock had failed  to double  in positions where  Magriel should  have
dropped.  Magriel had made a minor error or two in his usually  impeccable
checker play.  Then, in Position A, McClintock rolled 1-1, playing 23/off,
19/21.

		
				McClintock






		A






				Magriel

This play is a technical error; in case he rolls 1-2 next time, he  should
have played 23/off, 20/22, as we shall see later.  As the game  proceeded,
in Position B, Magriel  overlooked an obvious double  (he is 67% to  win),
and rolled a 6-3.  McClintock was then punished for misplaying the aces by
also rolling a 6-3.  Magriel won the point and later the match.

				McClintock






       		B
	





				Magriel

I like to imagine, though, a scenario in which McClintock is rewarded  for
the misplay of the double aces, and Magriel gives the correct double,  but
is punished for it.  In my  fantasy, the match continues from Position  B.
Magriel correctly doubles and rolls 1-2, grimaces, and plays 2/off, 4/3.







				McClintock






		C






				Magriel

McClintock sees  that he  is a  clear  favorite, and  that only  one  roll
permits Magriel to redouble.  At the match score, the redoubling  decision
is like that  in a money  game, so McClintock  correctly redoubles,  rolls
1-2, and without hesitation plays 21/24.

				McClintock






D					DD






				Magriel

Had McClintock played his aces correctly in Position A, the position would
now be DD, and Magriel could not have redoubled to eight.  In Position  D,
however, Magriel,  tilting  his  head sideways,  chewing  his  tongue  and
looking at the stars, estimates that cube handling should still be much as
in a money game, and  calculates that his average  gain of points will  be
1.30 if he  holds the cube,  1.43 if  he turns it.   Although fearing  the
redouble, he  turns the  cube to  eight.  McClintock,  envisioning a  grim
struggle if  Magriel should  bear  off for  eight  points, knows  he  must
nevertheless take.  After rechecking his arithmetic, he pulls the cube  to
his side of the board.  Magriel rolls 1-2, taking a checker off.

				McClintock







E					EE






				Magriel

Had McClintock played his aces correctly in Position A, the position would
have been EE, and McClintock, not having the cube, would simply roll.   In
Position E, however,  McClintock sees that  he is a  strong favorite,  and
that Magriel gets negligible value from possession of the cube at sixteen.
He places the cube, at sixteen, on the board.

Magriel estimates, using tables memorized years before, that a drop  would
leave him no  better than  a 17%  chance for  the match  against an  equal
opponent.  A take gives him a  34% chance.  He respects McClintock's  play
and trusts his own arithmetic, so, without rechecking, he slides the  cube
toward himself.  McClintock rolls and plays a 1-2.

				McClintock






F					FF





				Magriel

If McClintock had not misplayed his aces, the position would have been FF;
Magriel would have redoubled to  eight and McClintock would have  grumbled
and dropped.  In Position  F, Magriel sees that  the seventeenth point  to
win the match is  probably his on a  seventeen-to-one shot.  He  correctly
turns the cube to thirty two.

McClintock tries to imagine fighting Magriel  from a 0-16 score.  He  does
not know exactly what his chances for the match would be, but his estimate
is not far from the 7/10 of 1% Magriel is estimating.  He takes the  cube.
Perhaps he envisions  a pair  of dice coming  to rest  with small  numbers
showing.  Magriel  shakes  his cup  several  times, otherwise  showing  no
concern, and rolls a 1-2.

McClintock goes on to beat Barbara Glazer in  the semifinal and Roger  Low
in the final.  A  well known publisher asks  him to coauthor a  backgammon
book.  A collection of his favorite games, coauthored by George  Plimpton,
wins acclaim everywhere.  He subsequently lends his name to popular  books
on bridge, poker, blackjack, and cribbage.

Magriel, after offering McClintock a  firm handshake, murmurs a few  words
to an admiring  kibitzer, apparently  explaining why  McClintock took  the
cube at thirty-two.  He makes no mention of the double aces.  He writes  a
short note, perhaps a list of some  sort, which he gives to a young  woman
who has  followed the  game with  more than  average interest,  and  walks
toward the travel counter.  For a  long time, players ask if Magriel  ever
plays backgammon any  more.  As the  years pass, the  answer, as often  as
not, becomes ``Who?''

Author's note:  The game, as actually played, follows.  McClintock  should
have doubled before his twelfth and fifteenth rolls, and Magriel  probably
should have dropped.  18/15,  12/6 is a shade  better on Magriel's  eighth
roll.  Magriel's  play of  the  3-3 on  his  seventh roll  fascinates  and
mystifies me.

The fantasy, beginning with  Position B, represents  correct play by  both
sides for two players of  equal strength in the  first game of a  17-point
match; if the players  roll 1-2 five consecutive  times, the game  decides
the match.

The odds were at least 1889567 to 1 that Magriel would not give up backgammon.









Magriel					McClintock

1.	6-1	13/7, 8/7		5-4	1/10
2.	6-1	13,7, 6/5		6-5	10/16, 12/17
3.	6-3	8/5, 24/18		5-4	16/21, 17/21
4.	5-4	83, 7/3			4-3	1/8
5.	6-2	24/18, 6/4		6-3	8/17
6.	6-4	13/3			5-5	12/17 (4)
7.	3-3	18/12, 13/10, 7/4 (??)	5-3	17/22, 17/20
8.	6-3	18/9 (?)		2-1	17/20
9.	4-3	9/5, 13/10		3-1	17/20, 19/20
10.	3-3	7/4, 10/7, 12/6		5-2	17/19, 17/22
11.	5-4	10/5, 7/3		6-6	17/23 (2), 19/off (2)
12.	4-1	4/off, 3/2		3-2	22/off, 23/off (??)
l3.	6-4	6/off, 4/off		2-2	23/off, 21/off, 19/21
14.	5-4	5/off, 4/off		6-3	19/off, 22/off
15.	6-1	6/off, 5/4		1-1	21/off (??)
16.	6-1	6/off, 5/4		4-3	21/off, 20/23
17.	6-2	6/off, 2/off		2-1	23/off, 20/21
18.	3-1	3/off, 3/2		4-3	21/off, 20/23
19.	5-3	5/off, 3/off		1-1	23/off, 19-21 (?)
20.	6-3	4/off, 4/1 (???)	6-3
    Wins
3chkrs


Doubles, Redoubles, Takes, and Beavers: 
Three Checkers vs. Checkers on Ace Point

Suppose you have three checkers to  bear off, and your opponent has  three
or four on the ace point.  If your  chance to be off in two rolls is  more
than 0.90, he should drop your double. A pipcount of ten is marginal.  (In
the diagrams, p2 is X's chance to bear  off in two rolls; p is his  chance
to win the game if played to completion).














	A drop: p2=.92			A take: p2=.81,
					(1-3 and 1-4 miss completely.)

You should double or redouble with a good twelve-pip position:













	A redouble: p2=.76		Not a double: p2=.69
	     (P=.64)				(P=  )

The opponent should beaver all but one of the thirteen-pip doubles:













	Not a beaver: p2=.67		Beaver: p2=.64 (P=.54)

Now suppose you have three checkers to bear off and your opponent has five
or six on the ace  point. You should always double;  if your pip count  is
less than sixteen, your opponent should drop.














	A drop: p2=.49			A bare take: p2=.39,
	        p3=.96				     p3=.93
	    (P=.82)			     (P=.76)

What if the shoe is  on the other foot? If  your opponent is on roll  with
five or six checkers  on the ace point,  to take a double  you need a  .36
chance of being off in two rolls; a sixteen pip position is good enough:













	A drop				A take

The opponent should redouble at fifteen pips:













	A redouble			Not a double

You should beaver with a good 12 pip position:













	A beaver			Not quite a beaver

Summing up:	RD = Redouble, dropped
		RT = Redouble, taken
		DT = Double, but no redouble
		W  = Wait
		B  = Beavered
		?  = Depends on distribution

Your-three    	Your action	Opponent's	Your action
checker  	vs 3-4		action with	vs 5-6 
pip count	on ace		5-6 or ace	on ace

	
3-9		RD		B		RD
10		R(T?)		B		RD
11		RT		B		RD
12		R(?)T		B(?)		RD
13		W(B?)		W		RD
14		B		W		RD
15		B		RT		RD
16		B		RT		RT
17		B		RD		RT
18		B		RD		RT
dbgbkg.tex

\input basic
\input smacros
\ctrline{Doubling a Back Game}
\par \vskip .5in
\par
Division of risk works in your favor when your opponent is playing a well-timed
back game, so long as he does not own the cube.  Suppose that, if the game is played
out to the end, you would lose two thirds of the time, and win a gammon the
other third.  Without the cube, you would just break even.  If you have access
to the cube, however, you can play to a point in the bearoff at which one third
of your games will have been lost; at that point, you are equally likely to
lose the game and to win the gammon, so you have a near-perfect double; your
opponent has nothing to gain by taking, and if he drops, you win two thirds
of your games.

Let's estimate the correct point at which to double while bearing in
against a well-timed back game.  Suppose that your raw (no-cube) expectation
in a certain position is 0.2; that is, you will win a gammon 40\%\ of the time,
and lose the game 60\%\ of the time (assume the few backgammons and single games
cancel out).
Typically, you have a 30\%\ chance of being
lethally hit before your next roll.

If you play ten games to completion in the position you will lose three immediately,
three more later, and will gammon in four.  If you survive the next roll, you can
double your opponent out.
Suppose first that you retain the cube on your side.  In ten
games, your expectation is:

\ip{} Minus one point in three games where you are hit

\ip{} Plus one point in seven games where you double your opponent out\par\noindent
for a total of four points.
If instead you double immediately, your expectation is:

\ip{} Minus two points in six games

\ip{} Plus four points in four games\par\noindent
for a total of four points, also.
So the situation where you will win
gammons 40\%\ of the time, and lose the game 60\%\ of the time, is the break-even
point for doubling.  This conclusion, you should notice, does not depend on
whether the Jacoby Rule is in effect; no gammons will be won in undoubled
games in any event.

If you reach a situation where you will win gammons in more than two thirds
of your games, you are too good to double, and should play on, unless the
Jacoby Rule requires that you turn the cube to score the gammon.

If you own the cube, the arithmetic is a little different.  Suppose you will
still have a 20\%\ chance to win the game even if hit, and your chance to win
the gammon without being hit is 42\%\ .  By retaining the cube, your expectation
in one hundred games is:

\ip{} Minus one point in 24 games where you are hit and lose;
\ip{} Plus one point in 6 games where you are hit and win;
\ip{} Plus one point in 70 games where you double your opponent out next turn
\par\noindent
for a total of 52 points.
By doubling immediately, your expectation is:

\ip{} Minus two points in 58 games where you are hit and doubled out
\ip{} Plus four points in 42 games where you win a gammon\par\noindent
for a total of 52 points.
If you own the cube, a 42\%\ gammon chance is the breakeven point for redoubling.
\par\vfill\eject
Experiment suggests that the correct point to double, when bearing in
smoothly against a back game, is in positions like these:
\par \vskip 1in
\display{\tabskip 10pt plus 100pt minus 100pt
{\halign to 6.5in{\lft{#}	\lft{#}\cr
E	F\cr
\space{2in}
G	H\cr
}}}
\par\vskip .5in
\par \noindent
You should have no more than three points to clear against the 1-3 and 1-2
backgames, and only two points against the 2-3 backgame.

This analysis {\it only} holds in money play; in match play your opponent can
take away the value of your gammons by re-redoubling if you are fewer than eight
points from match.
\par\vfill\end
MARKET

			The Market

			Bob Floyd
			Copyright 1982

``If we don't double now, we'll lose our market''

You've heard it many times.  You've probably said it many times.  It's the
national anthem of backgammon; sing it and all differences of opinion  are
forgotten, everyone  stands  up and  salutes  the flag.  Still,  it  ain't
necessarily so. Next  time you hear  it, reply  ``Did you say  that if  we
don't double now, we'll lose some equity?''

Equity: your average winnings when the  position is played many times.  In
the long run, the player who  maximizes equity wins more money.  Sometimes
this means foregoing the use of the cube when you're a definite  favorite.
Often it means taking  the cube when  you have only a  25% chance to  win.
Sometimes it even means making a play that doesn't have the best chance to
win the game. Let's look  at the equity in  a late bearoff position  where
you can lose your market.













Are you a favorite here? Yes. Will your opponent take a double next  time?
No.  Should you double? No.

If this position is played out to the end, your chance of winning is  62%.
Out of 100 games, you will win 62 and lose 38, for a net gain of  62-38=24
times the stake, if your opponent doesn't redouble. But your opponent will
double you out if  you roll any  of your ten bad  numbers. This cuts  your
chance to  56%; if  you  double, your  net gain  in  100  games  is  still
only 56-44=12 times the doubled stake, or 24 times the initial stake.

If you wait, your winning  chance is more than  56%, because you will  now
double your opponent out in positions where if he owned the cube he  would
come back to beat  you. After average  rolls by both  players, you may get
this position:

MARKET

where you can redouble him out. The fraction of games you save this way is:
	17/36 (your rolls of 2-2, 3-4, etc.)
   X	30/36 (his non-doubletons)
   X    30/36 (your non-doubletons)
   X     6/36 (his doubletons)
= 0.055

You may also get this position:













You would have two ways  to lose this if you  couldn't turn the cube.  You
could roll 1-2 and be redoubled out,  with a chance of 

	8/36 (Your initial rolls of 1-3, 1-4, 1-5, or 1-6) 
   X 	30/36 (His non-doubletons) 
   X    2/36  (Your 1-2)
= 0.010 
or he  could come off with a  final doubleton, with a  chance
	8/36 (Your initial  rolls of 1-3,  etc.)  
   X 	30/36  (His non-doubletons) 
   X 	30/36 (Your average rolls: 1-3, 2-2, 5-6, etc.)  
   X 	5/36 (His doubletons) 
= 0.026

So, by keeping access to the cube, you will win   0.56+0.05+0.01+0.03=0.65
of your games.  Now in  100 games,  your net  gain is  65-35=30 times  the
initial stake, better than the  24 points you stand  to win if you  double
hastily.  At a  ten dollar stake,  you are giving  away sixty cents  every
time you give away  a centered cube  in this position.   If you owned  the
cube initially, your redouble costs you  $1.80, because you are giving  up
your chance to redeem yourself by  rolling a doubleton after your  initial
roll misses.

All right,  you  admit  you  shouldn't always  double  in  a  market-loser
situation, but you can't always stop to calculate the exact odds in a  hot
chouette.  We've thought of  that.  The rest of  this article gives  rules
for doubling, taking, and beavering in positions where only the player  on
roll can miss.

MARKET

Two Rolls

Everyone knows the answer in a one-roll position. In a two-roll  position,
double with fourteen bad numbers; redouble with twelve bad numbers








				X should not double
				(15 bad numbers)













				X should redouble
				(12 bad numbers)






The opponent should drop unless there are four bad numbers:













				O should barely take
				(4 bad numbers)






With fifteen bad rolls, a double should be beavered.






				O should beaver
				(thirteen bad numbers
				and many repeaters)






(I lost twenty-four  points in a  chouette by giving  a correct beaver  in
this position.  One of the  captain's dice came to  rest on a three.   The
other spun, spun some more, and then paired the first one.)


MARKET

Three Rolls

In a three-roll position, double with eight bad numbers; redouble with six
bad numbers.














				X should not double,
				(seven bad numbers plus too many repeaters)













				X should redouble
				(six bad numbers)






The opponent should take if there is at least one bad number:






				O has a bare take










MARKET

Four Rolls

In a four-roll position, the opponent has a take even if no numbers  miss,
and he will almost have a take next roll if each player takes two checkers
off, so  doubling,  and  especially redoubling,  should  be  conservative.
Double, but don't redouble, with two bad numbers:






				X should not double
				(four bad numbers)













				X should double but not redouble
				(two bad numbers)






Eleven bad rolls calls for a beaver:






				O should beaver
				(twelve bad numbers)



MARKET

Five Rolls or More

With more than five rolls  to go, don't double  if any roll misses;  don't
redouble at all.






				X should double but not redouble



MARKET

Summing Up

	Numbers of rolls:	1	2	3	4	5
	Bad rolls, to take
	   the cube:		9	4	2	0	0
	Bad rolls,
	   to redouble	       17      12	6	0	-
	Bad rolls, to double:  17      14       8	2	0
	Bad rolls, to beaver   19      15      13      12      11

A rule of thumb: On the last roll, double or redouble with any advantage.
For each additional roll remaining, you must have five fewer bad numbers
to double, and six fewer to redouble.

BGGUID




			A Killer Problem
			Bob Floyd
			Copyright 1982









					Should X double?
					He leads 9-7 in
					an 11-point match.


This artificial-looking position arose for  San Jose's Killer Joe  Glazier
in a recent Louisville tournament.  In a  money game, it is a double,  and
only a bare take because of the redouble equity; O's winning chance if the
game is played to completion is  24.96%.  At the match score, though,  the
redouble is  a  wicked  weapon.  Joe  doubled,  missed,  and  dropped  the
redouble. Surprisingly, this is  the right strategy  even though it  gives
him only a 53% chance to win the game, while waiting would give him a  75%
chance for the game.

First, a little background.  If X wins  one point, his  chance to win  the
match is 82%; O must win three consecutive games, unless he wins a  gammon
on the first or second. If X wins two points, he wins the match. If O wins
one point, X has a 59% chance for the match (there is argument about  this
figure; some place it as  low as 55%). If O  wins two points, the  chances
are equal. If O wins four points, X loses the match.

If X decides not to double, and then  misses, O must not double; to do  so
gives up an 18%  match equity when  he loses the game,  and gains no  more
than 9% match  equity when he  wins. O  should not give  even a  last-roll
double unless his  winning chance  is more  than 2  to 1.  O could  barely
double on the last roll with checkers on his 2- and 3-points.

So if X does not double, the game will be played out for one point, X will
win it almost exactly 3/4 of the time, and his match chance will be 
3/4 X .82 + 1/4 X .59 = 76.25%.

If X doubles and misses, he must  drop the redouble; it is better to  even
the match score at 9-9 than to play on for match as a slight underdog.  So
if X doubles, his match chance will be 19/36 + 17/36 X .50 = 76.39%. Joe's
double gained him an extra 1/7 of one percent chance of winning the match.
With that kind of precision  in his doubling, is  it any wonder that  they
call him Killer?

When you are two points away from match and your opponent is four or  more
away, a last-roll double is a powerful weapon. You should double with even
a 36% chance to win the game, as with checkers on the 2- and 6-points,  or
the 3- and 5-points. Your opponent should drop if your chance is more than
64%; checkers on the 1- and 5-points, or the 2- and 4-points, are marginal
drops.

On the other hand, before your  last roll, because your opponent can  give
the cube back and put the match on the line, your opponent will take  even
if you  have an  82% chance;  you should  not even  consider doubling  and
taking the redouble unless  your chance is 77%.  If each player has  eight
checkers on the  1-point, a  bare take  in a  money game,  you should  not
double at a  two-away-four-away match score;  in fact, mathematically  you
should wait until each player has just four checkers on the ace point.  In
practice, even a 5% chance that your opponent will drop justifies a double
with six checkers on each ace point.

When both players have most of  their checkers on the low-numbered  points
of the home board, and the player on  roll has a few numbers that waste  a
whole roll,  while his  opponent  can't miss,  what is  correct  doubling/
redoubling/taking/beavering strategy?

Two Rolls

Everyone knows the answer in a one-roll position. In a two-roll position,
don't double with fourteen bad numbers:






					X should not double






Double, but don't redouble, with twelve bad numbers:






					X should not redouble






The opponent should drop unless there are six bad numbers:






					O should take






With fifteen bad rolls, a double should be beavered.






					O should beaver
					(fourteen bad numbers
					and repeaters)




(I lost forty-eight  points in a  chouette by giving  a correct beaver  in
this position. One of  the opponent's dice  came to rest  on a three.  The
other spun, spun some more, and then paired the first one.)

Three Rolls

In a three-roll position, don't double with ten bad numbers:






					X should not double






Double, but don't redouble, with eight bad numbers:






					X should not redouble
					(six bad numbers plus repeaters)












					X should redouble
					(six bad numbers)





The opponent should drop unless there is at least one bad number:






					O has a bare take






Four Rolls

In a four-roll position, the opponent has a take even if no numbers  miss,
and he will almost have a take next roll if each player takes two checkers
off, so  doubling,  and  especially redoubling,  should  be  conservative.
Double, but don't redouble, with four bad numbers:






					X should not redouble






Redouble with two bad numbers:






					X should redouble






Eleven bad rolls calls for a beaver:






					O should beaver






Five Rolls or More

With more than five rolls  to go, don't double  if any roll misses;  don't
redouble at all.






					X should double but not redouble






Summing Up

	Number of rolls:	1	2	3	4	5
	Bad rolls, 
	   to redouble:        17      11	6	2	-
	Bad rolls, to double:  17      13	8	4	0
	Bad rolls, to take
	   the cube:		9	6	2	0	0
	Bad rolls, to beaver:  19      15      13      11

A rule of thumb: On the last roll, double or redouble with any  advantage.
For each additional roll remaining, you  must have four fewer bad  numbers
to double, and  five fewer to  redouble. On  the last roll,  take if  your
opponent has nine bad numbers.  For each additional roll remaining,  three
fewer bad numbers are needed to justify a take.
ROLLS

When both players have most of  their checkers on the low-numbered  points
of the home board, and the player on  roll has a few numbers that waste  a
whole roll,  while his  opponent  can't miss,  what is  correct  doubling/
redoubling/taking/beavering strategy?

Two Rolls

Everyone knows the answer in a one-roll position. In a two-roll position,
don't double with fourteen bad numbers:






					X should not double






Double, but don't redouble, with twelve bad numbers:






					X should not redouble






The opponent should drop unless there are six bad numbers:






					O should take






With fifteen bad rolls, a double should be beavered.






					O should beaver
					(fourteen bad numbers
					and repeaters)




(I lost forty-eight  points in a  chouette by giving  a correct beaver  in
this position. One of  the opponent's dice  came to rest  on a three.  The
other spun, spun some more, and then paired the first one.)

Three Rolls

In a three-roll position, don't double with ten bad numbers:






					X should not double






Double, but don't redouble, with eight bad numbers:






					X should not redouble
					(six bad numbers plus repeaters)












					X should redouble
					(six bad numbers)





The opponent should drop unless there is at least one bad number:






					O has a bare take






Four Rolls

In a four-roll position, the opponent has a take even if no numbers  miss,
and he will almost have a take next roll if each player takes two checkers
off, so  doubling,  and  especially redoubling,  should  be  conservative.
Double, but don't redouble, with four bad numbers:






					X should not redouble






Redouble with two bad numbers:






					X should redouble






Eleven bad rolls calls for a beaver:






					O should beaver






Five Rolls or More

With more than five rolls  to go, don't double  if any roll misses;  don't
redouble at all.






					X should double but not redouble






Summing Up

	Number of rolls:	1	2	3	4	5
	Bad rolls, 
	   to redouble:        17      11	6	2	-
	Bad rolls, to double:  17      13	8	4	0
	Bad rolls, to take
	   the cube:		9	6	2	0	0
	Bad rolls, to beaver:  19      15      13      11

A rule of thumb: On the last roll, double or redouble with any  advantage.
For each additional roll remaining, you  must have four fewer bad  numbers
to double, and  five fewer to  redouble. On  the last roll,  take if  your
opponent has nine bad numbers.  For each additional roll remaining,  three
fewer bad numbers are needed to justify a take.
DVRSK1 TEX   

\rm
\input jefmxs.tex
\def\today{\ifcase\month\or
  January\or February\or March\or April\or May\or June\or
  July\or August\or September\or October\or November\or December\fi
  \space\number\day, \number\year}

\line{\sevenrm dvrsk2.tex[v1,rfn] \today\hfill}


\centerline{The Principle of Divided Risk}
\vskip .125in
\centerline{Bob Floyd}
\vskip 12pt
\centerline{$\copyright 1982$}
\vskip .125in

When you play backgammon using the doubling cube, successful money management
is not just a matter of playing to win more games than your opponent.  It is
also a matter of winning games with the cube high, and losing them with the
cube low.

Suppose you have your choice of two plays in a certain position.  By making play A, 
you put the game on the line; you win (and double your opponent out on your next
turn) 47\%.\ of the time; you lose, and get doubled out , 53\%\ of the time.  
In a hundred plays of this position, you will lose six points.

By making play B, you divide your risk.  You run a 40\%\ risk of being doubled out
after your first play, and, if you survive, a 30\%\ chance of being doubled out 
after the second; if you get through both risks, you can claim the game.
Your chance to win the game is then only 60\%\ $\times 70$\%\ = 42\%; doesn't sound
very promising, does it?  Look again.  In a hundred games, your earnings will be:

\outlineone : Minus 1 point in 40 games, where you drop after one roll.

\outlineone : Minus 2 points in 18 games, where you double after the first roll and 
drop after the second.

\outlineone : Plus 2 points in 42 games, where you double after the first roll 
and go on to win.

\noindent
Your net is $-1\times 40-2\times 18+2\times 42= + 8$ points.
The play to win the game is play A; the play to win the money is play B.
This illustrates a general principle:

{\narrower\smallskip\noindent
{\bf If you have a certain total risk to run, and have choices 
about how to divide it between successive moves, try to divide it so as to 
give yourself a useful double at some stage.\smallskip}}

\noindent
I call this the principle of divided risk.  It can be used in many areas of
backgammon:  contact and non-contact bearoffs, holding games, backgames,
openings, escaping-checker games, probably any other kind of game you like.

\vskip .75in
(A)

\vskip .75in\par
\centerline{X to play 5-1}

\vskip 12pt

Position A is a dramatic illustration.
X had had a deuce-point game, and hit O's last loose checker.  He took a double 
after closing O out, if he hadn't already taken it; such a closeout is a
definite take for X, although probably a good double for O.  When X opened
up, O did not come in immediately, and X is planning to redouble when he
gets down to ten checkers, the theoretical last take point in such a bearoff.
Here he has rolled 5-1.  He could play 5/off, 5/4, and hope that O stays off
for two more rolls, or comes in and rolls small numbers.  The correct play,
however, is 5/off, l/off.  This play gives X equity of about +.25\%\ of the
cube; the cautious play gives him slightly negative equity.
\vfil\eject

\noindent
After the correct play, three possibilities occur:

\outlineone :(1) O comes in with a six (9 rolls).  Because X took off his 
odd checker, he had substantial chances to catch up in the race.

\outlineone :(2) O comes in with a five (11 rolls) hitting.  The position is 
nearly hopeless for X, but he wouldn't have had more than a 25\%\ chance with the
cautious play.

\outlineone :(3) O stays out (16 rolls).  Because he took an extra checker off, 
X has a powerful double, which is nearly as good as winning the game.  

Had X made the more cautious play, he would not be good quite enough to double, 
and by next roll, he would very likely be either so far beyond the doubling point 
that the cube gains him little, or else ahead by too little to double.
The bold play of bearing two checkers off reaches a good doubling position in two
ways.  If O stays out (42\%\ chance), X has an immediate double which may not even
be a take. If O comes in with a six, X rolls average numbers, and O fails to get
past the thirteen-point on his next roll (4\%\ chance), X again has a very good
double.

In this similar example from an actual game (Position B), 
X correctly played a 1-4 by taking two 
checkers off.  O fanned, X correcty doubled, O correctly took, and X won
eight points.
\vskip 1in

(B)
\vskip 1in

\centerline{X to play 4-1}

\vskip 12pt

In Position C, X has an ace to play in a non-contact bearoff.
\vskip 1in

(C)
\vskip 1in

\centerline{X to play a 1}
\vfil\eject

\noindent
The play more likely to win, by 100/46656=0.002, is 2/1.  Nevertheless,
X should play 3/2.  If X plays 2/1, he will either not be good enough to redouble
on his last roll (if he rolls a 1 or 2 next time), or he will have a virtual
claimer. If he plays 3/2, he has divided his risk almost perfectly when he rolls 
2-4, 2-5, 2-6, 3-4, 3-5, 3-6, 4-5, 4-6,or 5-6; he has a barely takable redouble.
As a result, 3/2 gives X higher average winnings, by 5\%\ of the value of the cube.
\vskip 1in

(D)
\vskip 1in

\centerline{X to play a 1}
\vskip 12pt
In Position D, X will be off in two rolls 45\%\ of the time after playing 2/1,
and only 43\%\ of the time after playing 6/5. Even after allowing for the
possibility of O rolling a doubleton and X rolling 5-5, 2/1 is the play to
win the game. Nonetheless, the strong cube action after X plays 6/5 and then rolls
2-2, 2-5, 2-6, 4-5, 4-6, or 5-6 gives the edge to 6/5.
\vskip 1in  

(E)
\vskip 1in

\centerline{X to play 5-4}
\vskip 12pt

In Position E, X has a 5-4 to play.  He can play it cautiously within his
board, hoping for a doubleton to clear the midpoint.  However, X has a nineteen pip
lead (after the roll)
which would allow him a powerful double after successfully breaking the
midpoint.  If X waits, he may have to pile up checkers wastefully on the lower
points in his board, losing his racing advantage.  By breaking the midpoint now,
13/9, 13/8, X allows O a 30\%\ chance to win, but in effect wins almost all of the
other 70\%\ of games by doubling on the next roll.  (Without X's access to the
cube, this play would give O nearly equal chances in the game, combining
hitting and racing chances.)
\vfil\eject

\noindent
[RWF: replace this by more appropriate material.]

In Position F (from Kent Goulding's Backgammon with the Champions series,
Vol. 1, Number 1, pg. 15; Magriel {\it vs} Eisenberg) Magriel, playing O, 
has a five remaining to play. 
If Eisenberg owned the cube, or would take the imminent double, Magriel might
well hit on the ace point to drive that checker forward.  In the actual game,
Magriel may have felt that Eisenberg would not take a double unless he hit,
so Magriel played 13/8, leaving only a 5-4 to hit.
This is a typical conservative risk-division play, where either play allows a
double-out if it goes well, so the smaller immediate risk is chosen.
\vskip 1in

(F)
\vskip 1in

\centerline{O to play a 5}
\vskip 12pt

\noindent
This is probably the most familiar appliation of the strategy.  When you will have
an untakable double, or a barely takable double, provided you are not hit, you 
should play to leave the fewest shots, even if another play gives you a sounder
position.

In Position G, with a 3-3 to play, X has the choice between 24/18, leaving
all eights to hit, and 24/12, leaving all twos to hit.  If the game is played
without the cube, or if O owns the cube, leaving the direct shot is better than
allowing likely repeated indirect shots.  However, if X has access to the cube,
he can double O if not hit, and should lower his immediate risk by playing
24/18, 6/3 (2).
The play 24/12 runs a 33\%\ risk.  The play 24/18, 6/3 (2) runs a total risk
of about 35\%\ , of which only 17\%\ is taken before he turns the cube.  (I have
ignored O's tiny racing equity.)
\vskip 1in

(G)
\vskip 1in
\centerline{X to play 3-3}

\vfil\end
ZEROSM


Backgammon is a two-person zero-sum game of perfect information, in  which
dice are used as a  random element. There is no  limit to the length of  a
backgammon game. Such a game can be represented by a directed graph, where
each node is one of:

(1) a maximizing node, with successor chosen by the maximizing player;

(2) a minimizing node, with successor chosen by the minimizing player;

(3) a randomizing node, with each successor chosen with a specified
    probability;

(4) a terminal node, with payoff determined by the rules of the game.

If the graph is finite and cycle-free, the game has an expected value; for
each non-terminal position, the expected value is the maximum, minimum, or
average of the expected value of  its successors, according to whether  it
is a maximizing, minimizing, or randomizing position.

Actual backgammon is not free of cycles, and allows positions which do not
have a well-defined expected  value, as we shall  see.  In practice,  most
positions have an expected value.  We shall treat the  game as if it  were
cycle-free.

The characteristic element of backgammon is the doubling of the stake. The
game is played initially for one point. Each player, at his turn, has  the
option of announcing a doubling of the stake (doubling); his opponent must
forfeit the previous stake, or agree to play for the doubled stake.  After
an initial double, the right to  double alternates between the players;  a
player may  not  double twice  unless  his  opponent has  doubled  in  the
interim. The terminal positions specify payoffs of the stake multiplied by
+ 1(``plain games''),  + 2(gammons),  or + 3(backgammons).  We shall  also
consider a simplification in which all payoffs are + 1 times the stake; we
shall call it bearoff, as it models the phase of backgammon in which  both
players are bearing checkers off the board.

For our purposes, a backgammon-like game is one with alternating plays and
a doubling rule,  with payoffs of  + 1,  + 2, or  + 3 times  the stake.  A
bearoff-like game is one with alternating plays and a doubling rule,  with
payoffs of + 1 times the stake.

A position is specified by three components:

(1) A checker position C

(2) a doubling status D, always a power of 2

(3) a right to double, R, which is + if only the maximizing player has
    the right to double first;- if only the minimizing player has the right
    right to double first; and + if either may double first. If D=1, R = +.
    If D>1, R = + or -.

At times we will consider versions of backgammon-like games where doubling
is not allowed; we shall say in this case that R = O.

A position has an expected value, E(C,D,R). For fixed C and R, E(C,D,R) is
proportional to D, so we will mainly use E(C,R)=E(C,D,R)/D.  It is obvious
that for fixed C,
E(C,-) ≤ E(C,+) ≤ E(C,+) and
E(C,-) ≤ E(C,O) ≤ E(C,+).
We shall also use E (C) for E(C,+), etc.

ZEROSM       


A simple game which approximates the backgammon bearoff is the random walk
of length n, where C is an integer in (0,n), and for 1≤C≤n-1 the successor
position is, with probability 1/2, C+1 or  C-1. If C=0, the payoff is  -1;
if C=n,  the payoff  is +1.  If the  random walk  game is  played  without
doubling, the probability that the maximizing player will win is C/n,  and
the expectation is 2(C/n)-1.

When n is a multiple of 5, the random walk game has a simple strategy. The
maximizing player should double if C≥0.8n,  and should decline a double if
C<0.2n; the minimizing player should double if C≤0.2n, and should  decline
a double if C>0.8n. This gives the expected values shown in the graph:













One can  show  the  correctness of  this  graph  by showing  that  if  the
maximizing player adopts  the strategy  above, he  guarantees an  expected
payoff at least as great as shown, and similarly the minimizing player may
guarantee an expected payoff no greater than shown.

Bearoff-like games

We shall show  some relations between  E-(C), E+(C), E+(C)  and E (C)  for
bearoff-like games.

Typically,  we  prove  linear  inequalities  among  them  by  mathematical
induction, and  define  particular bearoff-like  games  to show  that  all
relations satisfying these inequalities may actually occur.  We shall  use
P(C)=0.5(E (C)+1) as the probability  that the maximizing player wins when
doubling is disallowed.



ZEROSM       
ZEROSM











 
Proof: 

Since E+ ≥ E , E+ ≥ 2P-1, clearly E+ ≤ 1, since the  minimizing player has
has the option of declining  all doubles. The  minimizing player may adopt 
the strategy of declining doubles if P > 0.75, accepting if  P ≥ 0.75, and
and never  redoubling; an easy inductive proof then shows that E+ ≤ 4P/3-1.

The three extreme points are achieved by these three games:



















All other  points in  the shaded  region may  be achieved  by setting  the
probabilities α,β,  appropriately in this game:


















ZEROSM


















Proof: Use the relation E =2P-1 in the established relation between E+ and
P.













Proof: Analogous  to  the  proof  for  the  relation  between  E+  and  P,
interchanging max and min, and reversing the signs of the payoffs.

ZEROSM       



















Proof: Use the relation E =2P-1.

ZEROSM       
















Proof: As already seen, E+ ≥ E-, and  E+ ≤ 1. To show that E+ ≤ E +0.5, let  
the  minimizing  player  adopt  the  strategy  of  declining  doubles  when  
E- ≥ 0.5, accepting when E- < 0.5, and  never doubling; an  inductive proof
gives the result.  In essence, the arugment is  that even in the  positions 
where the maximizing player will double, he gains no more than half a point
point by having the ability to do so.

ZEROSM


The games below correspond to the extreme points in the graph of E+ vs. E-:














All other points  in the  graph may  be reached by  one of  the two  games
below, with appropriate values of α, β, , :

					(Case where E ≤ 0)





















					(Case where E > 0)
















ZEROSUM
ZEROSM


















Proof: -1 ≤ E+ ≤ +1;
	E+ ≤ E+ ≤ 4P/3-1;
	E+ ≥ E- ≥ 4P/3-1/3.

ZEROSM
















Proof: Obvious

ZEROSM
ZEROSM
















Proof: E+ ≥ E+.  To show that  E+ ≤ 4E+/3-0.5,  let the maximizing  player
adopt the policy of accepting doubles  if E+ ≥ -0.5, declining doubles  if
E+ <  -0.5, and  doubling  if 2E-  ≥ E+.   An  inductive proof  shows  the
inequality.

ZEROSM
















Proof: Analogous.

ZEROSM

Theorem: It is always correct to decline a double when E+ ≤ -0.5; when 
P ≤ 3/16; or when E ≤ -5/8.

Proof: If E+ ≤ -0.5, by accepting the double, one has expectation 2 x E+ ≤
-1; by  declining, one  has payoff  -1,  which is  no worse.  By  previous
results, if P ≤  3/16 or E ≤  -5/8, E+ ≤ -0.5.   The extreme situation  is
realized in this game:

























corresponding to this backgammon position:













Theorem: It is always correct to accept a double when E+≥-0.5; when P≥1/4;
or when E ≥-0.5.

Proof: Analogous

Theorem: It is always correct to double when E-≥0.5; when P≥13/16; when  E
≥5/8.
ZEROSM

Proof: When E-≥0.5,  it is  correct for the  opponent to  decline, so  the
expected value after  doubling must  be +1;  no larger  expected value  is
possible, because  the opponent  could  adopt a  policy of  declining  all
doubles. When P<13/16, the  game below shows that  it may be incorrect  to
double:


























Theorem: It is always correct to double with R=+, if there is  probability
≥0.5 of reaching a winning position (a terminal position with payoff  =+1)
before the next turn.

Proof: Let q≥0.5 be the probability of reaching a winning position. When a
winning position is reached, the  double has gained one point.  Otherwise,
doubling results in a  position C with expected  value 2E-(C) rather  than
E+(C). The minimum value  of 2E-(C)-E+(C) is -1,  achieved when P=0.  Thus
the gain of doubling isat least q-(1-q)=2q-1≥1.

Theorem: It is always correct to double with R=+ if there is a probability
≥0.6 of reaching a winning position before the next turn.

Proof: Let q≥0.6 be the probability of reaching a winning position. When a
winning position is reached, the redouble has gained one point. Otherwise,
the redouble results  in a position  C with expected  value 2E-(C)  rather
than E+(C).   The minimum  value of  2E-(C)-E+(C) is  -1.5, achieved  when
E+(C)=-0.5,  E-(C)=-1.0.   The  gain  of  having  redoubled  is  at  least
q-1.5(1-q)=2.5q-1.5, which is positive if q≥0.6.
ZEROSM

As applied to actual backgammon, in positions where gammons are not likely
it is always correct to redouble if  at least 22 of the 36 possible  rolls
result in a certain win; the position below is an example:















			X should redouble - barely.

Theorem: If you will never have another  chance to double, and R=+, it  is
always correct to double if P≥5/8.

Proof: By  doubling,  your  expectation  becomes  2E-.  By  waiting,  your
expectation is, in effect, E-. If P≥5/8,d E-≥0,d so 2E->E-.

Theorem: If you will never have another chance to redouble, and R=+, it is
always correct to double if P≥0.7.

Proof: By  doubling,  your  expectation  becomes  2E-.  By  waiting,  your
expectation is, in effect, E .

Since E-≥4E/3-1/3, 2E--E ≥5/3E -2/3, which is positive if E ≥0.4, i.e.  if
P≥0.7.

The following game, with α<0.6, shows that if P<0.7 it may be incorrect to
redouble even on your last turn:






























UNHAPP

			Bob Floyd
			Copyright 1983

If that unhappy  fraternity whose  only hope for  the current  game is  to
avoid being gammoned, if, I say, it agrees on anything, that would be  the
merit of crossovers into  a new quadrant.   Like the angels  contemplating
the torments of the damned from the palisades of paradise, those of us who
are seldom gammoned  may wonder, in  a spirit of  lazy curiosity,  whether
their belief is a sustaining and  nourishing one, or whether it is  merely
an opiate to  their pain, sustaining  an illusion that  they still have  a
measure of control over their fate.  How valuable, indeed, are crossovers,
and why, when at least one of them plays no special role in the definition
of the game, do they seem so important?

In extreme situations, taking  a crossover which leaves  an odd number  of
outside  crossovers  to  go  is  the  only  alternative  to  reliance   on
doubletons.  Naturally, in Position A,






(A)








	X to play a 1

we move 13/12, so that  most sixes bear a  checker off next time;  playing
the foolish 2/1 offers no hope but 3-3, 4-4, 5-5, or 6-6.

In Position B,







(B)	







	X to play a 2	
UNHAPP

X plays 14/12; his chances  of bearing off a  checker are still not  good,
but rolling 6-3 or better next time makes him a favorite; in fact, overall
he has a 42% chance to bear a  checker off in two rolls, and no play  that
requires a doubleton ever has  more than a 31% chance  to bear off in  two
rolls.  But what about positions where  there is one or more crossover  to
spare, like Position C,






(C)







	X to play a 1

where X can play 13/12, and still bear off in two turns with such mediocre
rolls rolls as 3-2, 5-1?  Even here, if  we look at every one of the  1296
ways four dice can  fall, the crossover  is superior by  1047 to 1015,  an
advantage of more than 2%.

The strongest clue to the value of the crossover
comes from  finding  the exceptional  situations  where the  crossover  is
wrong.  In Position D,






(D)








	X to play two 1's

the obvious 14/12 is worse than either 12/10 or 12/11, 14/13.  If we  look
ahead one roll, to avoid  depending on a last  roll doubleton X must  next
turn bring one checker home, and the other as far forward as possible.  If
he can play one  checker exactly to  his six point,  he saves the  maximum
possible number of pips to advance the other checker, so it is correct  to
maximize, by diversification, the likelihood of playing  exactly to the  6
point next turn.  That chance is  only 17/36 after 14/12, but 19/36  after
12/10, and also 19/36 after 12/11, 14/13.   (The play of  14/12  has other
drawbacks; a subsequent roll of  5-4  forces burying a checker deep in the
home board)
UNHAPP

Similarly, in Position E,







(E)








	X to play a 2

playing 11/9 is better than 13/11; again, the chance of bearing in exactly
to the six point is better with 11/9 (18/36 to 15/36).

We see  the crossover,  then, as  desirable when  it gives  the effect  of
diversification.  By increasing  the likelihood of  bearing in exactly  to
the sixpoint next turn, we reduce the likelihood of being forced to  waste
pips later.  Crossing to a new  quadrant always increases the chance  that
the moved  checker  will  play to  the  six  point next  turn.   From  the
opponent's home board, the chance of playing  to the six point is at  most
1/36.  From his outer board, it varies from 2/36 to 6/36.  From one's  own
outer board, the chance is at least 11/36.

This analysis suggests that a higher rule for running off the gammon  than
taking crossovers is the rule of  maximizing numbers that play to the  six
point.  Avoid positions like F,






(F)








with outside checkers  six pips  apart; they duplicate  both good  numbers
(such as 6-2 and 2-2 in Position F) and bad ones (any roll containing a  1
in Position F).
UNHAPP

When the positiions of other checkers are considered, it may be better  to
move a checker within a board than take a crossover.  In Position (G),






(G)








	X to play a 1

to play 13/12 takes away 5-2 as a roll which plays to the 6-point; to play
12/11 adds 5-3.  In Position H,






(H)








	X to play a 3

playing 20/17 adds no number which  plays to the six point; playing  16/13
adds 4-3.

(RWF: check out this position)
PEOPLE.TEX


\input basic
\input smacros
\ctrline{Advanced Gammon Avoidance}
\ctrline{Bob Floyd}
\ctrline{$\copyright$ 1982}
\par\vskip .5in\par
People who have given up on the game, and are racing home to save the gammon,
usually play very fast, as though there were no element of skill in bringing
the checkers home.  Actually, there are many subtle issues in escaping the
gammon when you have three or more outside checkers, and many experts misplay
these races.  This article explores some of the subtleties.

I assume you are familiar with the basics, as in Magriel's Chapter 23.  When
you can, you bear in exactly to your six point; you take crossovers and diversify
when you can; when you cross over into a new board, you try to avoid going
deep into that board; you ordinarily don't waste pips inside your home board
without a strong reason.

When analyzing a gammon-avoidance position, you should first try to estimate how
likely the gammon is.  Your correct play depends very much on whether you can be
optimistic or not.  Estimate how many rolls it will take your opponent to bear
off.  To avoid being gammoned, you must take a checker off in one roll fewer.
If you expect your opponent to be off in four rolls, you will have three.  You
can reasonably expect to do so if you need no more than five outside crossovers,
and about twenty as an outside pip count.  With more than five crossovers, you
must depend on rolling a doubleton; with exactly five, you must depend on not
missing a subsequent crossover.

If either the pip-count or the crossover count dictates pessimism, you must plan
to use large doubletons effectively. However, except for one-roll situations,
this is seldom best done by stacking your checkers on one point; you should be
sure that average rolls play well while you wait for the saving 5-5.

If the counts allow optimism, you should take precautions against rolling small
numbers; if you expect to have three more turns, for example, and you will need
four outside crossovers, try to play so that you will not miss
every time you roll an ace; otherwise even rolling 1-6 twice will destroy you.

In intermediate situations, you must balance many conflicting needs.  You should
try to make crossovers where possible, but you should also try to protect yourself
against rolling specific bad numbers (normally 1s) or never rolling a specific
needed number (usually a 6), and you should try not to waste pips inside your
home board, because wasted pips return to haunt you as later missed crossovers.
As a general rule, you should consider wasting pips to get down to an odd number
of outside crossovers, but not to an even number; there are exceptions, though.
Most of the positions that follow are intermediate-to-optimistic; they call for
subtle play to guard against certain unobvious dangers.

In Position A, X has a 2 to play.
\par\vskip 1in\par
(A)
\par\vskip 1in
\ctrline{X to play a 2}
\par\vskip 12pt
\par \noindent
He has, in all likelihood, another two rolls to bear off.  Nothing he does with
a 2 can affect his chance to be off in one roll, so he need only look at his
two-roll chances.  When this position arose in a chouette, the captain
automatically began to play 7/5 for the crossover.  I stopped him, noticing
that his play left the number of crossovers even, so that missing a crossover
now would not matter if we didn't miss again.  After 9/7, X will not miss on
his first roll, and seems unlikely to on his second.  I started looking for
bad sequences of rolls for either play.
When crossovers are important, two aspects of a position (other than just not
getting enough pip count) are likely to make you miss:

\ip{(1)} You may not have enough checkers on the 1, 7, 13, and 19 points to
cross over with all the 1's you will roll.

\ip{(2)} You may have too many checkers on the 12, 18, and 24 points, and not be
able to roll enough sixes to bring them into a new board each time they move.
\par \noindent
Here only the first danger is present.  The needed pip count is only eight or
nine, easy to roll in two turns.
I looked at rolls containing ones.  The play of 9/7 looked safe even if both
numbers contained a one.  The first roll could be played 7/6, 9/?; the second,
7/6, ?/off.  Only 1-2 seemed to offer difficulties.  The play of 7/5, if
followed by any ace, would leave X with a checker in the 8 or 9 point and a
gap on the ace point, so that another ace would miss again.  We played 9/7,
and sighed in relief when our next roll was 1-6.  We had only two bad rolls,
(1-2 and 2-1), to worry about, rather than eleven.
Later, I looked at every two-roll sequence in this position; there is no
sequence that would have made us regret playing 9/7, while any of 1-1, 1-3, 1-4,
1-5, and 1-6 followed by another number of this same set misses if we play 7/5.

The moral:  when you are a favorite to escape the gammon, look at your worst
rolls, especially those containing aces.  If you don't have several checkers
which can cross over with aces, you are at risk of missing a crossover.
If you also have no checkers on the 2, 8, 14, and 20 points, as in position A,
you run the risk of missing twice in a row on rolls containing aces.

At the other end of the spectrum, some positions will miss unless you roll one
or more sixes, even though the total pip count is modest.  In Position B, X
has a 3-2 to play.
\par\vskip 1in\par
(B)
\par\vskip 1in
\ctrline{X to play 3-2}
\par\vskip 12pt
\par\noindent
The obvious play is 15/12, 8/6.  This play wastes no pips, bears in perfectly
to the 6-point, achieves two crossovers, and gets another checker ready to
bear in.  It is also wrong.  If X makes this play, he will fail to be off in
three rolls unless he rolls a six or a doubleton.  He should rather bear in
both checkers from the 8-point.  He will still be off the gammon in three
rolls if he rolls a six and any other number which is three or greater.  He
will also escape the gammon if he rolls two fives, or a four and a five, even
with no sixes.  The chance of missing with 15/12 is about $({20/36})↑3=17$\%,
since the chance that a roll will be a singleton with no sixes is twenty
out of thirtysix.  The chance of missing after bearing the checkers in from
the eight and nine-points, while harder to estimate, is actually about 6\%.
Even if the checker were on the 16 point, with a 4 to play, it is slightly
better to avoid being dependent on rolling a doubleton or a six in the
next three rolls.
\par\vfill\eject
Position C, from Magriel's {\it Backgammon}, pg. 79, 
is a close analogue of Position B.
\par\vskip 1in\par
(C)
\par\vskip 1in\par
\ctrline{X to play 4-3}
\par\vskip 12pt
\par \noindent
Magriel's play of 16/12, 9/6 is not as good as 9/6, 9/5, which would have
escaped the gammon in the game from which the position is taken.
Again, the reason is the
dependency on sixes, in a position where there are no crossovers to spare.
(The difference in the probability of bearing a checker off in three more
rolls is only a few percent.)

PEOPLE

Position D is also from Magriel, pg. 80, and is also misplayed there.
\par \vskip 1in\par
(D)
\par\vskip 1in\par
\ctrline{X to play 5-2}
\par\vskip 12pt
\par \noindent
Here O is at least 50-50 to be off in three more rolls, so X must plan to be
off in two.  Barring doubletons, he can only do this by bringing two checkers
in.  Magriel's play of 12/7, 8/6, wasting no pips, only has about a 28\%  chance
to escape the gammon in two more rolls, while 9/4, 8/6 has about a 53\%  chance.
When you have an odd number of crossovers remaining it is usually {\it essential}
to take two crossovers per roll if you can, even at the cost of wasting pips.
\par\vfill\eject
When you come down
to one last outside checker  and you have not slotted
the one-point, you may have one of these positions:
\par \vskip 1in
\display{\tabskip 10pt plus 100pt minus 100pt
{\halign to 6.5in{\lft{#}	\lft{#}\cr
E8	E9\cr
\space{2in}
E11	E12\cr
}}}

If the outside checker is on the 8 or 9 point, you will wish you had played
2/1.  If it is on the 11 or 12 point, you will be glad you didn't.  On the
7 or 10 point, it makes no difference.  Until you can predict where the last
checker is likely to be, hold off on slotting inside points.  Position F
(from Deyong's Playboy Book of Backgammon, Diagram 5-15), exemplifies this.
\par \vskip 1in\par
(F)
\par\vskip 1in\par
\ctrline{X to play a 1}
\par\vskip 12pt
\par \noindent
Deyong's play, 2/1, escaped the gammon in four rolls once {\it less} often
in 36 trials, using the same dice rolls for all positions, than 16/15 or 9/8.
\par\vfill\eject
PEOPLE.TEX

When there is a single outside checker, it is occasionally right to slot the
ace point before the next-to-last roll.  In Position G:
\par \vskip 1in
(G)
\par\vskip 1in
\par \ctrline{X to play a 1}
\par \noindent
If X mistakenly plays 17-16, he is a shade more likely to be on the 8- or 9-point
next time (rolls of 3-5, 2-5, 3-4) than on the 11- or 12-point (rolls of
2-4, 2-3).  
For similar reasons, he should slot the ace-point if he is on the
14 or 15 points (but not on the 16, because of the value of being off in one more
roll with 3-3).

In position (H), X, who had almost given up, rolls a 3-3.  How many checkers
should he bring in?  Calculation shows that his chance to be off in two more
rolls after playing 8/5 two, three, or four times is, respectively 55\%, 66\% 
and 62\%.   How do we explain these figures?  How could we find the correct
play over the board?
\par \vskip 1in\par
(H)
\par\vfill\eject  
To bring in none or one of the checkers on the 8-point leaves too many
crossovers, unless X rolls another doubleton.  To bring two in and play
18/12 creates the double dangers of rolling a 1 on the next roll and of not
rolling a six in the next two rolls.  So the candidate plays are to bring in
3 or 4,
\par \vskip 2.5in
\display{\tabskip 10pt plus 100pt minus 100pt
{\halign to 6.5in{\lft{#}	\rt{#}\cr
(H3)	(H4)\cr
}}}
\par\vskip 12pt
\par \noindent
resulting in positions H3 and H4.  If the numbers on the next roll
total nine or more, X is virtually sure to be off on the following roll, in either
position.  If the numbers total less than six, they require a subsequent
doubleton, except for 3-2.  We need only look in detail at the remaining
numbers:

1-5 and 1-6 substantially favor H4.

2-3, 2-4, 2-5, and 2-6 substantially favor H3.

3-4 and 3-5 are indifferent, or nearly so.
\par\noindent
The preponderance favors H3, although they are close.  The most important
decision, though, is to reject bringing in just two checkers.
\par\noindent
In a longer race to get off the gammon, it becomes important to know whether
or not you are favored.  Uusually, in these situations, you can count pairs
of checkers as rolls for your opponent, and pairs of crossovers or eights of
pips (whichever is larger) as rolls for you.  If you are favored, it becomes
most important to follow traditional principles:  diversify in your outfield,
don't waste pips, take crossovers, bear in to your six-point.  If you are
significantly behind, plan to take advantage of large numbers and especially
the large doubtletons; play your outside checkers to the 18, 16, 12, and 11
points.  If the race is fairly even, correct play depends on whether you are
pip-count-bound or crossover-bound.

If pip-count-bound, assume that to have a chance you must roll 4's, 5's, and
6's anyway, so place checkers on the high points of your outfield.  If
crossover-bound, avoid creating positions where too many aces or not enough
sixes will hurt you.  Either way, in most races where you have a medium
chance, desperation plays of stacking several checkers on a single point
or bearing deep into your home board are very suspect.
Remember, when you are tempted to waste pips to save a crossover, that a bad
pip count can be expected to turn into missed crossovers later.
\par\vfill\eject
{\topinsert{\vskip 2.5in}}
\par \ctrline{X to play a 3}
\par \noindent
Your chances are only moderate, because of the pip count, but you have a crossover
to spare.
The crossover 15-12 creates insurance against a later miss.  
Even rolling a subsequent ace and
no six is not necessarily catastrophic.  To play 8-5 would aggravate your
pip-count problem.
\par \vskip 2.5in
\par \ctrline{X to play a 3}

The crossover is essential, but 15/12 makes you dependent on a six.  You are
favored, but you need every crossover. Play 8/5.
\par\vskip 2in
\par \ctrline{X to play a 1}
\par\vfill\eject
\par \noindent
Diversification is a mistake here; 4-1, 5-1, and 6-1 next roll are three times
more likely than a 3-2.  You are favored, but you need every crossover.
Guard against aces when your chances are good; play 8/7.

PEOPLE.TEX

\par\vskip 2.5in
\ctrline{X to play a 1}

Here you have no crossovers to spare, and a pip count which may lead to a
later miss.
You will need a good number of pips to be off in three rolls however you play,
so assume that you will be rolling a six and no aces; play 11-10 and pray.
If you roll a 4 next time, this play will save you two pips.
\par\vskip 2.5in
\par
\ctrline{X to play a 1}
\par\vfill\eject
Even though you are getting desperate here, don't play 10-9 to stack up for
double 3's; if you do get them next roll, you will still be an underdog to escape
the gammon.  A play of 12-11 leaves you favored to escape after 4-4, 5-5, or 6-6.
Another play to consider is 8/7, in case you get an ace on the next roll
followed by doubletons, but it doesn't help unless you also get a six.

The gains from each play are these combinations:

\display{\tabskip 10pt plus 100pt minus 100pt
{\halign to 6.5in{\lft{#}	\lft{#}	\lft{#}	\lft{#}\cr

Play	First roll	Second roll	Total\cr

12-11	5-2, 5-3	4-4, 5-5\cr
	5-4	3-3, 4-4, 5-5\cr
	4-4	5-1, 5-2, 5-3, 5-4	43 (in 1296)\cr
	5-5	twenty-one rolls\cr
\space{20 pt}
10-9	6-2, 6-3	3-3\cr
	3-3	6-1, 6-2, 6-3, 6-4, 6-5	14\cr
\space{20 pt}
8-7	1-6	4-4, 5-5, 6-6\cr
	1-1, 1-3, 1-4, 1-5	6-6	13\cr
}}}
Notice that X gains more (14 chances in 1296) on 12/11 by not missing on an
initial 5-2, 5-3, or 5-4 than he does on 8/7 by rolling a 3-3 (only 10 in 1296).

The moral: even when your position is desperate, if there is more than one
roll remaining you should prepare for normal rolls rather than for a
specific doubleton.
\par \vfill \end