Tetris NP

Tetris is Hard:
An Introduction to P vs NP
Based on Tetris is Hard, Even to

Approximate in COCOON 2003 by
Erik D. Demaine (MIT)
Susan Hohenberger (JHU)
David Liben-Nowell (Carleton)
Whats Your Problem?
A language is a set of strings:

PRIMES = {2, 3, 5, 7, 11, 13, 17, 19, 23, . . .}
We are interested in the following kind of decision problem:

Given x, is x PRIMES?
How much harder does this question get as x gets larger?
The Tetris Problem
So what do we mean by Tetris?

Offline Tetris: get to see entire piece sequence in advance.
Advanced level: the initial board is partially filled.
TETRIS = {"g, s# :
can play piece sequence s to clear entire gameboard g}
"g, s# TETRIS: given a gameboard and a piece sequence, can

we clear the entire board using the given sequence?
2
Instances of Decision Problems

Is 357667 PRIMES?
Is
"
TETRIS?
How much harder does this get as the gameboard gets bigger?
Find an algorithm A that decides if "g, s# TETRIS.
Measure resources consumed by A.
Measuring Resource Usage

How do we measure the resources consumed by A?
Be pessimistic.
Always consider the worst instance of a given size.
Resources: time, space.
Run A on all instances of size n on some computer.
Graph n versus the amount of each resource consumed
by A on the most consuming instance of size n.
5
12
200
x 10
180
10
160
140
120
100
80
4
60
40
2
20
Polynomial
10
10
12
14
16
18
Exponential
20
A Computer Theorists Worldview

Easy
Computable Uncomputable
Hard
Will 463 beat 363s median

on Quiz 3?
Will Windows crash when
I load this web page?
Can I color this map
with three colors?
Is this number prime?
Are all the numbers
in this list positive?
5
Complexity Classes
A complexity class is a set of languages.
A language L (PRIMES, TETRIS, ...) is a member of the
following complexity classes if there is an algorithm A ...
P
... deciding all questions x L in polynomial time.

e.g., for every x, time taken by A(x) is 10|x|3.
Games: JENGA
PSPACE ... deciding all questions x L in polynomial space.

Games: GO (in some countries), OTHELLO
EXPTIME
... deciding all questions x L in exponential time.

Games: CHESS, CHECKERS
Complexity Hierarchy
EXPTIME
PSPACE
!
Efficient Space
P
!
Efficient Time
P PSPACE EXPTIME
P &= EXPTIME
The Traveling Salesman Problem

Youre selling fair trade coffee beans.
Start at Deans Beans
World Headquarters in Dublin.
Visit (in any order):
Longs Bookstore, 1610 High St,
Cup of Joe, Easton Town Center,
446 Lane, Court House.
Drop off the leftovers in Dublin.
Can you do all deliveries in one hour?
What order should you use?
8
The Nave Solution to TSP

The nave method of solving TRAVELING-SALESMAN:
List all possible orderings of the cities.
Find the shortest ordering.
Check if it requires under one hour.
So TRAVELING-SALESMAN is in EXPTIME.
16 cities:
63 cities:
more orderings than the size of the US national debt.

more orderings than atoms in the known universe.
Can we do better?
9
Can we solve TSP efficiently??

Nobody has been able to:
give a faster solution to TRAVELING-SALESMAN.
prove that TRAVELING-SALESMAN needs exponential time.
So TRAVELING-SALESMAN might be in P!
But notice:
For a particular ordering of the coffeeshops, its easy to check
if the length of the trip is under one hour.
(Just add up the times.)
10
Guess and Check
Another way to measure the hardness of a language:

Suppose I claim that x L (and give you an argument to try to
convince you). How hard is it to check if my argument is right?
For example, for TETRIS: given "g, s#

I tell you all the moves m that Ill make in the game.
You check if moves m clear the gameboard g
using piece sequence s.
11
NP, continued
A language L is in NP if I can give you an argument A and
then you can check that A is right in polynomial time.
NP: Nondeterministic Polynomial Time
Nondeterminism: I give you an argument; you check it.
For some games, like CHESS, the move sequence might be
too long to check in polynomial time!
Summary: L P can be decided quickly.

L NP can be checked quickly.
12
Complexity Hierarchy
EXPTIME
PSPACE
!
NP
!
P
!
Chess, Checkers
Go, Othello, Sokoban,
Generalized Geography, Hex
Tetris, Mastermind,
Minesweeper, Sliding Blocks,
Traveling Salesman
Jenga, Im Thinking of a Number
P &= EXPTIME
P NP PSPACE EXPTIME
13
P vs. NP
NP&=P
We do not know which is true!
P=NP
One of the greatest unsolved problems in theoretical

computer science (and all of mathematics!).
Clay Mathematics Institute offers $1 million for solution.
Is it easier to verify a correct solution to a problem than it
is to solve the problem from scratch?
Huge impact on cryptography, airline scheduling, factory
layout, UPS truck packing, drug design, etc., etc., etc., etc.
14
NP-completeness
First major step on P vs. NP problem [Cook and Levin, 1970s]:
An NP-complete problem:
!
1. is in NP
2. is as hard as the hardest problem in NP.
NP
If we can quickly decide one

NP-complete problem, then we can
quickly decide them all. (P=NP)
If any problem is in NP and not P,
then every NP-complete problem
is in NP and not P. (P&=NP)
15
Technique #1: Divine Inspiration
Theorem [Cook/Levin]: SATISFIABILITY is NP-complete.
16
Technique #1: SAT is NP-complete
SATISFIABILITY = { formulas : there is a way of setting

the variables of so that is true }
e.g., x (x y) SATISFIABILITY.
(set x := true and y := false)
Theorem [Cook/Levin]: SATISFIABILITY is NP-complete.
16
Technique #2: Reduction

A reduction f is a mapping from one problem to another.
A
B
f
f
xA
if and only if
f (x) B.
Were interested
in efficient f .
If theres a quick algorithm for B, then theres one for A.

If theres no quick algorithm for A, then theres none for B.
B is as hard as A
17
NP-Completeness Summary
Theorem: SATISFIABILITY is NP-complete.

Cook/Levin.
Theorem: If: (1) A is NP-complete,

(2) B is in NP, and
(3) there is an efficient reduction from A to B
then B is NP-complete.
Karp.
18
Enough already! I came for the Tetris ...
To show that TETRIS is NP-complete:

Define efficient mapping m from instances of known
NP-Complete problem L to TETRIS.
Show that if x L, then m(x) TETRIS.
Show that if x
/ L, then m(x)
/ TETRIS.
19
From 3-PARTITION to TETRIS
3-PARTITION = {sets of integers that can be separated into

piles of three integers each, where each pile has the same sum}
Theorem [Garey and Johnson]: 3-PARTITION is NP-complete.
Our result:
There is an efficient reduction from 3-PARTITION to TETRIS.
20
Our Reduction ...

Given integers a1, . . . , a3s and the target sum T for each pile:
a1
initiator:
filler (a1 times):
6T + 22
terminator:
...
a2
initiator:
...
a3s
terminator:
Finally:
6s + 3
...
s,
, and
( 3T
2 + 5).
21
Reminder: NP-Completeness
To show that Tetris is NP-complete:
Define efficient mapping m from instances of

3-PARTITION to TETRIS.
Show that if x 3-PARTITION, then m(x) TETRIS.
Show that if x
/ 3-PARTITION, then m(x)
/ TETRIS.
22
For x L, we need m(x) TETRIS

Yes means yes.
If we have x 3-PARTITION, then need m(x) TETRIS.
Pile the integers according to their 3-PARTITION piles.
..
23
The yes case, continued

Piles each have the same sum, so they have the same height:
Each bucket is filled exactly to the top.
Remaining pieces will exactly clear entire gameboard.

24
To show that TETRIS is NP-complete:
Define efficient mapping m from instances of

3-PARTITION to TETRIS.
Show that if x
/ TETRIS.
25
Why this reduction?

We want to make sure that no means no.
cant clear any rows until the
...
have to completely fill each bucket

or its all over.
notches make it easy to get stuck.
26
For x
/ L, we need m(x)
/ TETRIS
No means no.
Step #1: if you dont (or cant) make the moves from two
slides ago, you get into one of the following configurations.
Step #2: if you get into one of these, youre hosed.
not div. by 4
27
To show that Tetris is NP-complete:
Define efficient mapping m from instances

of 3-PARTITION to TETRIS.
Show that if x
/ TETRIS.
Theorem: TETRIS is NP-complete.
28
Statement of Results
Our results actually apply for some other Tetris questions, too:
can we clear at least k rows?
can we place at least p pieces without losing?
can we place all pieces without filling a square at height h?
can we achieve at least t tetrises?
All of these are NP-complete.
29
Other Objectives and Inapproximability
Can strengthen our results further:

add a boatload of
pieces
to our sequence.
Can get into (and clear) lower
reservoir (using extra
pieces)
only if can clear the top using original.

Choosing a large reservoir yields
inapproximability results.
30
Other Work on Tetris
Theorem: A (sufficiently long) alternating

sequence of
s and
s will cause a loss.
[Brzustowski/Burgiel]
Implies that probability of a real Tetris game lasting infinitely
long is zero (regardless of players skill).
On the Tetris gameboard that you know and love.
(Our complexity results are as the gameboard grows.)
31
Conclusion and Open Questions

Many interesting problems are NP-complete.
(Traveling Salesman, Minesweeper, Tetris, Satisfiability, ...)
So what? Does P = NP?
Is it easier to verify that a solution is correct
than it is to come up with it from scratch?
Games are interesting because theyre hard (?)
Easy way to make a million bucks:
give an efficient algorithm for TETRIS!
32
Recap: Proving Membership in NP
SATISFIABILITY = { formula : there is a way of setting the

variables of so that is true }
Certificate: assignment to variables
Verifier: plugs in assignment and tests that is true
33
Recap: Proving Membership in NP

Certificate: assignment a to variables
Verifier: tests that is true with assignment a
CLIQUE = { graph G and integer k : there is a clique of size k
in G}
Certificate: nodes in the clique C
Verifier: tests that each node in C is connected to every other
node in C
34
Recap: Optimization vs Decision

Suppose P = NP.
How can you find a satisfying assignment in polynomial time?
35
Recap: Optimization vs Decision
CLIQUE = { graph G and integer k : there is a clique of size k

in G}
Suppose P = NP.
How can you find a k-clique in polynomial time?
36

Tetris NP

Uploaded by

Copyright:

Available Formats

Tetris NP

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Tetris NP

Uploaded by

Copyright:

Available Formats

Tetris is Hard:

Based on Tetris is Hard, Even to

Whats Your Problem?

A language is a set of strings:

We are interested in the following kind of decision problem:

The Tetris Problem

So what do we mean by Tetris?

"g, s# TETRIS: given a gameboard and a piece sequence, can

Instances of Decision Problems

Measuring Resource Usage

A Computer Theorists Worldview

Will 463 beat 363s median

... deciding all questions x L in polynomial time.

PSPACE ... deciding all questions x L in polynomial space.

... deciding all questions x L in exponential time.

The Traveling Salesman Problem

The Nave Solution to TSP

more orderings than the size of the US national debt.

Can we solve TSP efficiently??

Guess and Check

Another way to measure the hardness of a language:

For example, for TETRIS: given "g, s#

Summary: L P can be decided quickly.

One of the greatest unsolved problems in theoretical

If we can quickly decide one

Technique #1: Divine Inspiration

Theorem [Cook/Levin]: SATISFIABILITY is NP-complete.

Technique #1: SAT is NP-complete

SATISFIABILITY = { formulas : there is a way of setting

Theorem [Cook/Levin]: SATISFIABILITY is NP-complete.

Technique #2: Reduction

If theres a quick algorithm for B, then theres one for A.

Theorem: SATISFIABILITY is NP-complete.

Theorem: If: (1) A is NP-complete,

Enough already! I came for the Tetris ...

To show that TETRIS is NP-complete:

From 3-PARTITION to TETRIS

3-PARTITION = {sets of integers that can be separated into

Our Reduction ...

To show that Tetris is NP-complete:

Define efficient mapping m from instances of

For x L, we need m(x) TETRIS

The yes case, continued

Remaining pieces will exactly clear entire gameboard.

To show that TETRIS is NP-complete:

Define efficient mapping m from instances of

Why this reduction?

cant clear any rows until the

have to completely fill each bucket

To show that Tetris is NP-complete:

Define efficient mapping m from instances

Other Objectives and Inapproximability

Can strengthen our results further:

only if can clear the top using original.

Other Work on Tetris

Theorem: A (sufficiently long) alternating

s will cause a loss.

Conclusion and Open Questions