Invariants, Monovariants, and Extrema

Matthew Rognlie

Department of Mathematics
Duke University

September 9, 2009
Invariants Monovariants Extrema

Why Invariants?

Common Questions
Say that we investigating some process, which has a starting state S and
a sequence of possible subsequent transformations. We may want to
know:
Invariants Monovariants Extrema

Why Invariants?

Common Questions
Say that we investigating some process, which has a starting state S and
a sequence of possible subsequent transformations. We may want to
know:
• Is it possible to reach a given state?
Invariants Monovariants Extrema

Why Invariants?

Common Questions
Say that we investigating some process, which has a starting state S and
a sequence of possible subsequent transformations. We may want to
know:
• Is it possible to reach a given state?
• Find all reachable states.
Invariants Monovariants Extrema

Why Invariants?

Common Questions
Say that we investigating some process, which has a starting state S and
a sequence of possible subsequent transformations. We may want to
know:
• Is it possible to reach a given state?
• Find all reachable states.
• Find all reachable end states (states where no more transformations
are possible).
Invariants Monovariants Extrema

Why Invariants?

Common Questions
Say that we investigating some process, which has a starting state S and
a sequence of possible subsequent transformations. We may want to
know:
• Is it possible to reach a given state?
• Find all reachable states.
• Find all reachable end states (states where no more transformations
are possible).
• Will the process inevitably converge to some end state?
Invariants Monovariants Extrema

Why Invariants?

Common Questions
Say that we investigating some process, which has a starting state S and
a sequence of possible subsequent transformations. We may want to
know:
• Is it possible to reach a given state?
• Find all reachable states.
• Find all reachable end states (states where no more transformations
are possible).
• Will the process inevitably converge to some end state?
• Is there periodicity?
Invariants Monovariants Extrema

Why Invariants?

Common Questions
Say that we investigating some process, which has a starting state S and
a sequence of possible subsequent transformations. We may want to
know:
• Is it possible to reach a given state?
• Find all reachable states.
• Find all reachable end states (states where no more transformations
are possible).
• Will the process inevitably converge to some end state?
• Is there periodicity?
If we want to answer any of these questions, it’s likely that we will make
use of invariants (or, as we’ll discover later, monovariants)!
Invariants Monovariants Extrema

The Basic Idea

When we are examining a process that involves repeated transformations,

we should explore what does not change. This is often critical to
understanding the properties of the process. Some simple examples:
Invariants Monovariants Extrema

The Basic Idea

When we are examining a process that involves repeated transformations,

we should explore what does not change. This is often critical to
understanding the properties of the process. Some simple examples:
• In classical physics, the total energy of a closed system is invariant.
Invariants Monovariants Extrema

The Basic Idea

When we are examining a process that involves repeated transformations,

we should explore what does not change. This is often critical to
understanding the properties of the process. Some simple examples:
• In classical physics, the total energy of a closed system is invariant.
• The parity of a number (whether it is odd or even) is invariant to
multiplication by an odd number.
Invariants Monovariants Extrema

The Basic Idea

When we are examining a process that involves repeated transformations,

we should explore what does not change. This is often critical to
understanding the properties of the process. Some simple examples:
• In classical physics, the total energy of a closed system is invariant.
• The parity of a number (whether it is odd or even) is invariant to
multiplication by an odd number.
• Repeatedly sum the digits of a number to get a new number. For
instance, we may find:

27708 ! 24 ! 6 ! 6 . . .

The value of our number mod 9 is invariant.

Invariants Monovariants Extrema

A Slightly More Involved Example

Suppose n is odd, and the numbers 1, 2, . . . , 2n are written on the
blackboard. I repeatedly pick two arbitrary numbers on the blackboard a
and b, erase them, and write |a b| instead. This continues until only
one integer is left. Is the number at the end odd or even?
Invariants Monovariants Extrema

A Slightly More Involved Example

Suppose n is odd, and the numbers 1, 2, . . . , 2n are written on the
blackboard. I repeatedly pick two arbitrary numbers on the blackboard a
and b, erase them, and write |a b| instead. This continues until only
one integer is left. Is the number at the end odd or even?
Once we realize the sum of numbers on the blackboard mod 2 is
invariant, this is easy!
Invariants Monovariants Extrema

A Slightly More Involved Example

Suppose n is odd, and the numbers 1, 2, . . . , 2n are written on the
blackboard. I repeatedly pick two arbitrary numbers on the blackboard a
and b, erase them, and write |a b| instead. This continues until only
one integer is left. Is the number at the end odd or even?
Once we realize the sum of numbers on the blackboard mod 2 is
invariant, this is easy!
Indeed, for any a and b, a + b ⌘ a b (mod 2). Thus when I replace two
numbers by their di↵erence, the parity of the sum never changes. All we
need to do now is calculate whether the sum 1 + 2 + . . . + 2n is odd or
even:
X2n
2n(2n + 1)
k= = n(2n + 1)
2
k=0

Since n is odd, the rightmost expression is odd as well, and we can

conclude that regardless of the order in which I pick and replace the
numbers, the final one will be odd. (Like many examples in this
presentation, this is taken from Problem Solving Strategies by Arthur
Engel, our recommended text for the course.)
Invariants Monovariants Extrema

Another Contrived Example

Problem
A dragon has 100 heads. A knight can cut o↵ 15, 17, 20, or 5 heads,
respectively, with one blow of his sword. In each of these cases, 24, 2, 14,
or 17 heads grow on its shoulders. If all heads are blown o↵, the dragon
dies. Can the dragon ever die?
Invariants Monovariants Extrema

Another Contrived Example

Problem
A dragon has 100 heads. A knight can cut o↵ 15, 17, 20, or 5 heads,
respectively, with one blow of his sword. In each of these cases, 24, 2, 14,
or 17 heads grow on its shoulders. If all heads are blown o↵, the dragon
dies. Can the dragon ever die?

Proof.
At first glance, this problem is convoluted and intractable. Once we hit
upon the idea of using invariants, however, it becomes trivial. We note:

(24 15) ⌘ (2 17) ⌘ (14 20) ⌘ (17 5) ⌘ 0 (mod 3)

Invariants Monovariants Extrema

Another Contrived Example

Problem
A dragon has 100 heads. A knight can cut o↵ 15, 17, 20, or 5 heads,
respectively, with one blow of his sword. In each of these cases, 24, 2, 14,
or 17 heads grow on its shoulders. If all heads are blown o↵, the dragon
dies. Can the dragon ever die?

Proof.
At first glance, this problem is convoluted and intractable. Once we hit
upon the idea of using invariants, however, it becomes trivial. We note:

(24 15) ⌘ (2 17) ⌘ (14 20) ⌘ (17 5) ⌘ 0 (mod 3)

The knight’s gallantry can never change the number of heads of the
dragon mod 3. Since we start at 100 ⌘ 1 (mod 3), we can never get to
0. The dragon lives!
Invariants Monovariants Extrema

One Final Example

Problem
The integers 1, . . . , n are arranged in any order. In one step, you may
switch any two neighboring integers. Prove that you can never reach the
initial order after an odd number of steps.
Invariants Monovariants Extrema

One Final Example

Problem
The integers 1, . . . , n are arranged in any order. In one step, you may
switch any two neighboring integers. Prove that you can never reach the
initial order after an odd number of steps.

Proof.
Let the sequence of integers be a1 , . . . , an . For any indices i < j, let bij
equal 0 if ai < aj and 1 if aj < ai . In other words, bij = 1 if ai and aj are
out of order. WeP assert that switching two neighboring integers will
change the sum i<j bij . Indeed, if we switch integers ak and ak+1 , the
only value of bij that changes is bn,n+1 , which either goes from 0 to 1 or
1Pto 0. In any case, its parity changes, and thus the parity of the sum
i<j bij changes. After an odd number of steps, the parity of this sum
cannot be the same as the initial parity, and we cannot be at the initial
order.
Invariants Monovariants Extrema

Some Observations

What We Saw
Invariants Monovariants Extrema

Some Observations

What We Saw
• These problems were pretty easy! Not all problems using invariants
are so simple, but once you realize the right approach you’re often
close to done.
Invariants Monovariants Extrema

Some Observations

What We Saw
• These problems were pretty easy! Not all problems using invariants
are so simple, but once you realize the right approach you’re often
close to done.
• Arguments using parity and modular arithmetic in general are very
commonly associated with invariants.
Invariants Monovariants Extrema

Some Observations

What We Saw
• These problems were pretty easy! Not all problems using invariants
are so simple, but once you realize the right approach you’re often
close to done.
• Arguments using parity and modular arithmetic in general are very
commonly associated with invariants.
• Taking sums and exploiting symmetry is a good way to design an
invariant.
Invariants Monovariants Extrema

Some Observations

What We Saw
• These problems were pretty easy! Not all problems using invariants
are so simple, but once you realize the right approach you’re often
close to done.
• Arguments using parity and modular arithmetic in general are very
commonly associated with invariants.
• Taking sums and exploiting symmetry is a good way to design an
invariant.
• We’re not limited to looking for values that literally never change.
Instead, we often want values that change in a predictable way. In
the last example, for instance, we found a sum that alternated
between odd and even parity with each step.
Invariants Monovariants Extrema

What is a Monovariant?

Many problems require a generalization of the idea of an “invariant.”

Even if we cannot identify some function of the state of a process that
never changes, we may be able to identify a function that always changes
in the same direction. In fact, this information can be invaluable.
Consider the following fact:
Invariants Monovariants Extrema

What is a Monovariant?

Many problems require a generalization of the idea of an “invariant.”

Even if we cannot identify some function of the state of a process that
never changes, we may be able to identify a function that always changes
in the same direction. In fact, this information can be invaluable.
Consider the following fact:
If there is some positive integral function that decreases at each
step of a process, the process must eventually terminate.
Invariants Monovariants Extrema

What is a Monovariant?

Many problems require a generalization of the idea of an “invariant.”

Even if we cannot identify some function of the state of a process that
never changes, we may be able to identify a function that always changes
in the same direction. In fact, this information can be invaluable.
Consider the following fact:
If there is some positive integral function that decreases at each
step of a process, the process must eventually terminate.
This is trivial: if a positive integral function starts at n and decreases
with each step, the process certainly cannot continue for more than n 1
steps. Yet despite its apparent obviousness, this principle is perhaps the
most powerful way for us to draw conclusions about how a process
behaves.
Invariants Monovariants Extrema

First Example

Problem
In the Senate of Kazakhstan, each member has at most three enemies. A
member cannot be his own enemy, and enmity is mutual. Prove that the
Senate can be divided into two factions such that each Senator has at
most one enemy within his faction.
Invariants Monovariants Extrema

First Example

Problem
In the Senate of Kazakhstan, each member has at most three enemies. A
member cannot be his own enemy, and enmity is mutual. Prove that the
Senate can be divided into two factions such that each Senator has at
most one enemy within his faction.

Proof.
First, we separate the members arbitrarily into two factions. Let H be the
sum of all the enemies each member has in his own faction. Suppose one
member (let’s call him Bob) has at least two enemies in his own faction.
Then if Bob switches factions, H will decrease. Let this process continue
for all members in the same situation as Bob. Since H is a positive
integral function that decreases at each step of the process, the process
must terminate. At this point, no Senator can have more than one
enemy in his own faction, because otherwise the process (by definition)
would not have terminated. Thus we have found the desired division of
Senators.
Invariants Monovariants Extrema

Second Example: from the 2008 Putnam contest!

Problem
Start with a finite sequence a1 , a2 , . . . , an of positive integers. If possible,
choose two indices j < k such that aj does not divide ak , and replace aj
and ak by gcd(aj , ak ) and lcm(aj , ak ), respectively. Prove that if this
process is repeated, it must eventually stop.
Invariants Monovariants Extrema

Second Example: from the 2008 Putnam contest!

Problem
Start with a finite sequence a1 , a2 , . . . , an of positive integers. If possible,
choose two indices j < k such that aj does not divide ak , and replace aj
and ak by gcd(aj , ak ) and lcm(aj , ak ), respectively. Prove that if this
process is repeated, it must eventually stop.

Proof.
First, note that the product a1 a2 . . . an is invariant, because
aj ak = gcd(aj , ak ) lcm(aj , ak ). Now, each element in the sequence is
bounded from above by this product. The last element can never
decrease, because it is only replaced by its least common multiple with
another integer. Thus eventually it reaches its maximum value and
becomes fixed. After this happens, the second-to-last element will never
decrease; it also eventually reaches its maximum and becomes constant,
and so on, until the entire sequence is fixed.
Invariants Monovariants Extrema

Multiple Ways to Solve a Problem

On the actual contest, I solved the previous example using a di↵erent

monovariant: the sum of all elements in the sequence. A little algebra
proves:
Invariants Monovariants Extrema

Multiple Ways to Solve a Problem

On the actual contest, I solved the previous example using a di↵erent

monovariant: the sum of all elements in the sequence. A little algebra
proves:
aj 6= ak =) gcd(aj , ak ) + lcm(aj , ak ) < aj + ak
Invariants Monovariants Extrema

Multiple Ways to Solve a Problem

On the actual contest, I solved the previous example using a di↵erent

monovariant: the sum of all elements in the sequence. A little algebra
proves:
aj 6= ak =) gcd(aj , ak ) + lcm(aj , ak ) < aj + ak
This is often a nice feature of problems that require monovariants: there
isn’t a single “correct” monovariant that you need to find before you
make any progress. Instead, there may be many candidates, di↵ering in
elegance and ease but not in substance. Just remember that most good
monovariants will exploit symmetry, and work out some examples by
hand to see if any pattern jumps out.
Invariants Monovariants Extrema

A Scary Monovariant
The following monovariant appears in the solution to a problem from the
1986 International Mathematical Olympiad:
Invariants Monovariants Extrema

A Scary Monovariant
The following monovariant appears in the solution to a problem from the
1986 International Mathematical Olympiad:

S(v , w , x, y , z) =|v | + |w | + |x| + |y | + |z| + |v + w | + |w + x| + |x + y |

+ |y + z| + |z + v | + |v + w + x| + |w + x + y |
+ |x + y + z| + |y + z + v | + |z + v + w | + |v + w + x + y |
+ |w + x + y + z| + |x + y + z + v | + |y + z + v + w |
+ |z + v + w + x|
Invariants Monovariants Extrema

A Scary Monovariant
The following monovariant appears in the solution to a problem from the
1986 International Mathematical Olympiad:

S(v , w , x, y , z) =|v | + |w | + |x| + |y | + |z| + |v + w | + |w + x| + |x + y |

+ |y + z| + |z + v | + |v + w + x| + |w + x + y |
+ |x + y + z| + |y + z + v | + |z + v + w | + |v + w + x + y |
+ |w + x + y + z| + |x + y + z + v | + |y + z + v + w |
+ |z + v + w + x|

Note that this isn’t quite as scary as it first appears: if v , w , x, y , z are

points on a pentagon, then this is just the sum of the absolute values of
sums of individual vertices, pairs, triples, and foursomes. (The original
problem involved a procedure applied to integers at the vertices of a
pentagon.) In fact, this is a good illustration of how sophisticated
symmetries that reflect the underlying structure of a problem can
produce impressive results. Some monovariants are simple and elegant,
but don’t be afraid to try something a little more unwieldy. Experiment!