United Kingdom Mathematics Trust

email [email protected]
web www.ukmt.org.uk

British Mathematical Olympiad Round 1 2023

Teachers are encouraged to distribute copies of this report to candidates.

Markers’ report
The 2023 paper
Olympiad marking
Both candidates and their teachers will find it helpful to know something of the general
principles involved in marking Olympiad-type papers. These preliminary paragraphs therefore
serve as an exposition of the ‘philosophy’ which has guided both the setting and marking of all
such papers at all age levels, both nationally and internationally.
What we are looking for are full solutions to problems. This involves identifying a suitable
strategy, explaining why your strategy solves the problem, and then carrying it out to produce
an answer or prove the required result. In marking each question, we look at the solution
synoptically and decide whether the candidate has a viable overall strategy or not. An answer
which is essentially a solution will be awarded near maximum credit, with marks deducted for
errors of calculation, flaws in logic, omission of cases or technical faults. On the other hand,
an answer which does not present a complete argument is marked on a ‘0 plus’ basis; up to
4 marks might be awarded for particular cases or insights. In general the logical structure of
the mark scheme aims to reflect the logical structure of the problem while rewarding correct
arguments more generously than correct calculations.
This approach is therefore rather different from what happens in public examinations such as
GCSE, AS and A level, where credit is given for the ability to carry out individual techniques
regardless of how these techniques fit into a protracted argument. It is therefore vital that
candidates taking Olympiad papers realise the importance of the comment in the rubric about
trying to finish whole questions rather than attempting lots of disconnected parts.
British Mathematical Olympiad Round 1 2023 Markers’ report

General comments
Candidates found this to be one of the most accessible BMO1 papers in recent years, with
impressive numbers making progress on three or more questions. Slightly gentler questions in
the fourth and fifth positions also gave the very best candidates time to think about the final
problem, with a number going on to solve it successfully. At the other end of the distribution,
it was good to see the vast majority of candidates getting into the 10− regime on at least one
question; only a handful of candidates did not seem to realise that full written solutions were
required. A few responses to the geometry question consisted of nothing other than a diagram
with every angle labelled as say 𝑥, 𝑦, 𝑥 + 𝑦, 90 − 𝑥 etc. This was particularly disheartening as
those candidates almost certainly understood the relevant geometry, but since they neglected to
explain the order in which the angles were found by defining one angle to be 𝑥 and then deducing
that some other angle must also be 𝑥 and so on, they could not be given any credit. Other
questions, notably 1, 5 and 6, naturally invited candidates to experiment with small examples
and form conjectures. It was good to see most scripts engaging with this experimentation in a
systematic way, and often spotting the correct patterns. However, this vital phase in solving
such problems is of little value on its own. To score highly candidates needed to support their
observations with arguments explaining why the observed patterns continued or the reasoning
about small cases could be generalised.
The 2023 British Mathematical Olympiad Round 1 attracted 1675 entries. The scripts were
marked in London (with some remote markers) from the 1st to the 3rd of December by a team
of: Eszter Backhausz*, Sam Bealing*, Jonathan Beckett, Jamie Bell*, Robin Bhattacharyya,
Damian Cheung, Laura Daniels, Stephen Darby, Chris Eagle, Ben Fairfax, Chris Garton,
Anthony Goncharov, Aleksander Goodier, Amit Goyal, Aditya Gupta, Ben Handley*, Stuart
Haring*, Jon Hart, Alexander Hurst, Ian Jackson*, Shavindra Jayasekera*, Vesna Kadelburg,
Thomas Kavanagh, Jeremy King, Patricia King*, David Knipe, Hayden Lam, Larry Lau, Rhys
Lewis, Warren Li, Sida Li, Thomas Lowe, Owen Mackenzie, Max Mackie, Sam Maltby, David
Mestel*, Ana Meta Dolinar, Harry Metrebian*, Oliver Murray, Joseph Myers, Daniel Naylor,
Martin Orr, Jenny Owladi, Preeyan Parmar*, Thomas Prince, Dominic Rowland, Heerpal
Sahota, James Sarkies, Amit Shah, Gurjot Singh, Geoff Smith*, Rob Summerson, Stephen Tate,
William Thomson, Velian Velikov, Tommy Walker Mackay, Zi Wang, Helen Xiaohui Chen,
Tianyiwa Xie, Lingde Yang, Harvey Yau. (An asterix shows that the marker was a problem
captain).
The problems were proposed by Geoff Smith, Geoff Smith, Sam Bealing, Dominic Yeo, Richard
Freeland, and Ben Handley, respectively.
In addition to the written solutions in this report, video solutions can be found here.

© 2023 UK Mathematics Trust www.ukmt.org.uk 2

British Mathematical Olympiad Round 1 2023 Markers’ report

Mark distribution
100

80

60
Frequency

40

20

0
0 10 20 30 40 50 60
Olympiad mark

The mean score was 20.9 and the median score was 21.
The thresholds for qualification for BMO2 were as follows:
Year 13: 43 marks or more.
Year 12: 42 marks or more.
Year 11: 39 marks or more.
Year 10 or below: 37 marks or more.

The thresholds for medals, Distinction and Merit were as follows:

Medal and book prize: 44 marks or more.
Distinction: 30 marks or more.
Merit: 13 marks or more.

© 2023 UK Mathematics Trust www.ukmt.org.uk 3

British Mathematical Olympiad Round 1 2023 Markers’ report

Question 1
An unreliable typist can guarantee that when they try to type a word with different letters,
every letter of the word will appear exactly once in what they type, and each letter will occur
at most one letter late (though it may occur more than one letter early). Thus, when trying
to type MATHS, the typist may type MATHS, MTAHS or TMASH, but not ATMSH.
Determine, with proof, the number of possible spellings of OLYMPIADS that might be
typed.

Solution
The answer is 28 = 256.
For a word with 𝑁 letters, the condition of the problem is equivalent to the statement that for
every positive integer 𝑛 < 𝑁, the 𝑛th letter of the word is typed somewhere in the first 𝑛 + 1
places.
In our problem, we take the letters of OLYMPIADS in order from the left and look at where
they might appear in the typist’s version. The letter 𝑂 must appear either first or second: 2
possibilities.
Next the letter 𝐿 must appear somewhere in the first three places. One of those three places has
already been taken by the 𝑂, so there are 2 remaining possibilities for the position of the 𝑂,
hence 2 × 2 = 4 possibilities for the positions of the 𝑂 and 𝐿.
If we have identified the positions of the first 𝑛 − 1 letters for some positive integer 𝑛 satisfying
2 < 𝑛 < 9, we know that in the typist’s version, the 𝑛th letter occurs somewhere in the first
𝑛 + 1 places, and 𝑛 − 1 of those have already been used for the first 𝑛 − 1 letters, so there are 2
possibilities for the position of the 𝑛th letter.
This continues until we have identified the position of the first 8 letters. The final 𝑆 must appear
in the one remaining position, wherever that is.
So the total number of ways that the typist could type OLYMPIADS is 28 = 256.

Remark
For every solution that considers possible locations at which a letter could be typed, there is
another solution that considers possible letters that could be typed at a particular location.
For example, observe that there are two possible letters that can be typed last — 𝐷 and 𝑆.
Having chosen one, then of the three possible letters that can be typed second-last — 𝐴, 𝐷 and
𝑆 — there are two remaining possibilities. Continue backwards through the locations until we
have selected a letter for each location.

Alternative
An alternative strategy is to relate the number of possibilities to the number of possible
spellings of a shorter word. Let 𝑓 (𝑛) be the number of valid spellings for a word of length
𝑛 (without repeated letters); then the problem is to find 𝑓 (9). Given a word 𝐴𝐵𝐶𝐷 · · · of
length 𝑛, consider all possible spellings of the shorter word 𝐵𝐶𝐷 · · · of length 𝑛 − 1. For

© 2023 UK Mathematics Trust www.ukmt.org.uk 4

British Mathematical Olympiad Round 1 2023 Markers’ report

each of the 𝑓 (𝑛 − 1) spellings of the shorter word, we can obtain two spellings of the original
word, by reinserting 𝐴 either at the front or at the second place. These spellings are all
different, because they differ either in the location of 𝐴 or in the location of at least one
letter from the shorter word. Furthermore, every spelling of the original word arises this way,
because removing 𝐴 from a valid spelling of the original word produces a valid spelling of the
shorter word. Therefore 𝑓 (𝑛) = 2 𝑓 (𝑛 − 1) for each 𝑛 ≥ 1. It is clear that 𝑓 (1) = 1, and so
𝑓 (9) = 2 · 𝑓 (8) = 22 · 𝑓 (7) = · · · = 28 · 𝑓 (1) = 28 . Alternatively, we could prove by induction
that 𝑓 (𝑛) = 2𝑛−1 , and then set 𝑛 = 9.
Alternative
A variation considers which letter is typed first. If the first letter of an 𝑛-letter word is typed
first then there are 𝑓 (𝑛 − 1) ways to type the remaining letters. If the 𝑘th letter is typed first,
then the first letter must be typed second, the second letter must be typed third, and so on,
until we find that the 𝑘 − 1th letter must be typed at position 𝑘. We have now typed the first 𝑘
letters in the first 𝑘 locations. The final 𝑛 − 𝑘 letters must therefore be typed in the final 𝑛 − 𝑘
locations, and there are 𝑓 (𝑛 − 𝑘) ways to do this. We therefore have
Õ
𝑛
𝑓 (𝑛) = 𝑓 (𝑛 − 𝑘)
𝑘=1

where 𝑓 (0) = 1 since there is 1 way to type no letters. As before, this recursion could be used
to directly find 𝑓 (9), or to prove by induction that 𝑓 (𝑛) = 2𝑛−1 for 𝑛 ≥ 1.

Alternative
A sophisticated strategy is to find a 1–1 correspondence (a bijection) between possible spellings
and possible subsets of a suitable set of size 8. Since there are 28 subsets of a set of size 8,
we can immediately deduce that there are 28 possible spellings. One such approach considers
subsets of the first 8 letters, and places each of those letters one place late. We then prove
that there is exactly one way to complete the spelling so that no other letter is typed late.
Suppose that the letters that are not typed late are the letters at positions 𝑗1 < · · · < 𝑗 𝑘 in the
original word. Letters 1, . . . , 𝑗 1 − 1 are each typed one place late and therefore occupy places
2 . . . , 𝑗1 . Since letter 𝑗 1 is not typed late it can only occupy the first position. Next, letters
𝑗 1 + 1, . . . , 𝑗 2 − 1 are each typed one place late, and so occupy positions 𝑗 1 + 2, . . . , 𝑗 2 . Since
letter 𝑗 2 is not typed late, there is only one position remaining for it, namely position 𝑗 1 + 1.
Arguing similarly for letters 𝑗 3 , . . . , 𝑗 𝑘 , we find that the only way to complete the spelling is to
place letter 𝑗ℓ at position 𝑗ℓ−1 + 1 for ℓ = 2, . . . , 𝑘, and that this produces a valid spelling.

MarkersŠ comments
The approach of iteratively choosing first where 𝑂 is typed, then 𝐿 and so on, was popular,
as was the variation that chooses which letter is typed last, and then which letter is typed
second-last, and so on. A complete solution must explain in which order the choices are being
made, since the choices cannot be made independently. It must also justify why there are
exactly 2 remaining choices at each stage. Many scripts lost marks for being unclear on one or
both of these points.
Students must also reason that at the last stage there is only 1 remaining choice. Some failed to
notice this, and concluded that the answer was 29 = 512. This is a logical flaw and loses marks.
© 2023 UK Mathematics Trust www.ukmt.org.uk 5
British Mathematical Olympiad Round 1 2023 Markers’ report

(Of course, students who correctly prove that the answer is 28 , but mistakenly believe that this
is equal to 512 are entirely forgiven, as are students who miscount the number of letters in the
word 𝑂 𝐿𝑌 𝑀 𝑃𝐼 𝐴𝐷𝑆 but otherwise have a correct argument.)
Many students found the number of possibilities for words of short lengths, correctly spotted
a pattern, and predicted from this pattern what the answer would be for a word of length 9.
Experimenting with small cases and formulating hypotheses is wise, and some marks were
available for doing this; but an essentially complete logical argument was required to get a
score beyond 3 marks.
Inductive approaches were also popular, and many were successful. Let there be 𝑓 (𝑛) legal
spellings of an 𝑛-letter word, then the common inductive steps were 𝑓 (𝑛) = 2 𝑓 (𝑛 − 1) for 𝑛 > 1
or 𝑓 (𝑛) = 𝑓 (𝑛 − 1) + 𝑓 (𝑛 − 2) + · · · 𝑓 (1) + 1 for 𝑛 > 1. These can be obtained by working
from inwards from either end of the word. The difficulty for markers was to make sure that
such a recurrence was obtained by a proper argument, and not just by pattern spotting.
There were various types of subset approaches. For example, you can argue that the actual
spelling of a word is determined by which of the first 𝑛 − 1 letters occur late. There are 2𝑛−1
subsets of a set of size 𝑛 − 1, so that yields a solution.
A few students tried systematically to enumerate all possible spellings. This was usually not
successful, either because some cases were missed, or because a pattern was claimed without
justification.
Several students observed that there are 9! spellings in total if we ignore the typist’s guarantee,
and then attempted to subtract the number of spellings that violate the typist’s guarantee. This
is a legitimate line of attack, but most candidates trying this approach made calculation errors.

© 2023 UK Mathematics Trust www.ukmt.org.uk 6

British Mathematical Olympiad Round 1 2023 Markers’ report

Question 2
The sequence of integers 𝑎 0 , 𝑎 1 , . . . has the property that for each 𝑖 ≥ 2, 𝑎𝑖 is either
2𝑎𝑖−1 − 𝑎𝑖−2 or 2𝑎𝑖−2 − 𝑎𝑖−1 .
Given that 𝑎 2023 and 𝑎 2024 are consecutive integers, prove that 𝑎 0 and 𝑎 1 are consecutive
integers.
(Note that 6 and 7 are consecutive integers, as are 7 and 6.)

Solution
We would like to prove by induction that for all 0 ≤ 𝑖 ≤ 2023, 𝑎𝑖 and 𝑎𝑖+1 are consecutive.
Base case: 𝑎 2023 and 𝑎 2024 are consecutive.
Inductive step: Suppose that for some 𝑖, 𝑎𝑖 and 𝑎𝑖+1 are consecutive.
If 𝑎𝑖+1 = 𝑎𝑖 + 1 then either 𝑎𝑖+1 = 𝑎𝑖 + 1 = 2𝑎𝑖 − 𝑎𝑖−1 , so 𝑎𝑖−1 = 𝑎𝑖 − 1 or 𝑎𝑖+1 = 𝑎𝑖 + 1 = 2𝑎𝑖−1 − 𝑎𝑖
so 𝑎𝑖−1 = (𝑎𝑖 + 1)/2, contradiction, as all terms are integers. So 𝑎𝑖−1 = 𝑎𝑖 − 1.
Similarly, if 𝑎𝑖+1 = 𝑎𝑖 − 1 then either 𝑎𝑖+1 = 𝑎𝑖 − 1 = 2𝑎𝑖 − 𝑎𝑖−1 , so 𝑎𝑖−1 = 𝑎𝑖 + 1 or
𝑎𝑖+1 = 𝑎𝑖 − 1 = 2𝑎𝑖−1 − 𝑎𝑖 so 𝑎𝑖−1 = (2𝑎𝑖 − 1)/2, contradiction, as all terms are integers. So
𝑎𝑖−1 = 𝑎𝑖 + 1.
In either event, 𝑎𝑖−1 and 𝑎𝑖 are consecutive.
As 𝑎 2024 and 𝑎 2023 are consecutive and we showed that if 𝑎𝑖 and 𝑎𝑖+1 are consecutive then so
are 𝑎𝑖−1 and 𝑎𝑖 , we have proved that 𝑎 0 and 𝑎 1 are consecutive.

Alternative
If 𝑎 0 and 𝑎 1 were equal, then all the terms of the sequence would be equal, which is not the
case. So we can assume 𝑎 0 and 𝑎 1 are distinct. As all 𝑎𝑖 are integers, |𝑎 1 − 𝑎 0 | ≥ 1. The
condition on the sequence can be expressed equivalently as the statement that for each 𝑖 ≥ 2:

𝑎𝑖 − 𝑎𝑖−1 = 𝑎𝑖−1 − 𝑎𝑖−2 or − 2(𝑎𝑖−1 − 𝑎𝑖−2 )

In either case |𝑎𝑖 − 𝑎𝑖−1 | ≥ |𝑎𝑖−1 − 𝑎𝑖−2 |. So

1 = |𝑎 2024 − 𝑎 2023 | ≥ |𝑎 2023 − 𝑎 2022 | ≥ · · · ≥ |𝑎 1 − 𝑎 0 | ≥ 1

This implies |𝑎 1 − 𝑎 0 | = 1, which is to say that 𝑎 0 and 𝑎 1 are consecutive.

MarkersŠ comments
Most successful candidates approached the problem by making 𝑎 2022 the subject of the two
relations. In doing so, they realised that only one of the relations results in an integer value and
also that this value and 𝑎 2023 must be consecutive. At this point, a carefully phrased induction
(applying the same ideas to 𝑎 2023 and 𝑎 2022 to find 𝑎 2021 , and so on) solved the problem. Some
candidates lost out on marks for not stating why only one of the relations can be used in every
step, not just the first one. Candidates were not penalised for phrasing their induction informally
but we recommend they look up the correct terminology.
Many candidates calculated options for 𝑎 2021 instead. This method needed a bit more algebra,
but worked similarly, except that it only proved that terms of the form 𝑎 2𝑖−1 and 𝑎 2𝑖 are

© 2023 UK Mathematics Trust www.ukmt.org.uk 7

British Mathematical Olympiad Round 1 2023 Markers’ report

consecutive. Some didn’t spot that 𝑎 0 and 𝑎 1 are not of this form, and so lost out on marks for
not discussing the last step needed from 𝑎 1 and 𝑎 2 to 𝑎 0 and 𝑎 1 .
Candidates considering the difference between consecutive terms were usually successful but
often lost marks as they presumed that |𝑎𝑖 − 𝑎𝑖+1 | ≤ 1 implied the terms were consecutive
without clearly excluding 0 and non-integer values.
Some candidates misunderstood the question and presumed that the same relation must be used
for the whole sequence, as opposed to a potential combination of both, others only considered
one. A few assumed that 𝑎 2023 and 𝑎 2024 are consecutive increasing, despite the note. These
mistakes stopped candidates from scoring highly.

© 2023 UK Mathematics Trust www.ukmt.org.uk 8

British Mathematical Olympiad Round 1 2023 Markers’ report

Question 3
Let 𝐴𝐵𝐶 be a triangle with ∠ 𝐴𝐶𝐵 < ∠𝐵𝐴𝐶 < 90◦ . Let 𝑋 and 𝑌 be points on 𝐴𝐶 and the
circle 𝐴𝐵𝐶 respectively such that 𝑋, 𝑌 ≠ 𝐴 and 𝐵𝑋 = 𝐵𝑌 = 𝐵𝐴. Line 𝑋𝑌 intersects the
circle 𝐴𝐵𝐶 again at 𝑍.
Prove that 𝐵𝑍 is perpendicular to 𝐴𝐶.

Solution

𝐴 𝑍

𝑋
2𝜃
𝜃

𝐵 𝐶
𝜃

To construct an accurate figure for this question we must note that 𝐴, 𝑋 and 𝑌 all lie on a circle
This means, amongst other things, that triangle 𝐴𝐵𝑋 is isosceles with apex 𝐵.
We let ∠𝑍 𝐵𝐴 = 𝜃 as shown.
We add segments 𝐵𝑋 and 𝐴𝑌 (shown in blue).
Now ∠𝑍𝑌 𝐴 = 𝜃 (Angles in the same segment in circle 𝐴𝐵𝑌 𝑍.)
Also ∠𝑋 𝐵𝐴 = 2𝜃 (Angle at centre 2× angle at circumference in circle 𝐴𝑌 𝑋.)
Thus 𝐵𝑍 is an (internal) angle bisector in the (isosceles) triangle 𝑋 𝐵𝐴, so 𝐵𝑍 is perpendicular
to 𝑋 𝐴.

Remark
The fact that in an isosceles triangle the angle bisector from the apex is an altitude can be
quoted at BMO level. It is easy to prove by checking that it splits the triangle into two triangles
that are congruent by SAS.

© 2023 UK Mathematics Trust www.ukmt.org.uk 9

British Mathematical Olympiad Round 1 2023 Markers’ report

Alternative
𝐴 𝑍
180◦ −𝛼−𝛿
𝛼

180◦ −𝛼−𝛿
𝛼
𝑋
𝛿

𝐵 𝐶
𝛿

We let ∠𝐵𝐴𝐶 = 𝛼 and ∠𝑋𝑌 𝐵 = 𝛿 as shown.

We add segments 𝐵𝑌 , 𝐵𝑋 and 𝐴𝑍 (shown in blue).
Now ∠𝐵𝑋𝑌 = 𝛿 (Isos triangle 𝑋 𝐵𝑌 .)
∠ 𝐴𝑋 𝐵 = 𝛼 (Isos triangle 𝐴𝐵𝑋.)
∠𝑍 𝑋 𝐴 = 180◦ − 𝛼 − 𝛿 (Straight line 𝑌 𝑋 𝑍.)
∠𝐵𝐴𝑍 = 180◦ − 𝛿 (Cyclic quad 𝐴𝐵𝑌 𝑍.)
So ∠𝑋 𝐴𝑍 = 180◦ − 𝛼 − 𝛿 (Angles at 𝐴.)
This means triangle 𝑋 𝑍 𝐴 is isosceles, so 𝑍 𝐴 = 𝑍 𝑋.
This, combined with the fact that 𝐵𝐴 = 𝐵𝑋 means 𝐴𝐵𝑋 𝑍 is a kite, so its diagonals, 𝐵𝑍 and
𝐴𝐶 are perpendicular.

Remark
The fact that diagonals of a kite a perpendicular can be quoted. It can be proved by first showing
the two non-isosceles halves of the kite are congruent using SSS, then using the fact that the
angle bisector of an isosceles triangle is an altitude as in the first solution.

© 2023 UK Mathematics Trust www.ukmt.org.uk 10

British Mathematical Olympiad Round 1 2023 Markers’ report

Alternative
𝐴 𝑍
𝛾
𝛾

𝛿
𝛾
𝐵 𝐶
𝛿

We let ∠ 𝐴𝐶𝐵 = 𝛾 and ∠𝑋𝑌 𝐵 = 𝛿 as shown.

We add segments 𝐵𝑌 , 𝐵𝑋 and 𝐴𝑍 (shown in blue).
Now ∠𝐵𝑋𝑌 = 𝛿 (Isos triangle 𝑋 𝐵𝑌 .)
∠ 𝐴𝑍 𝐵 = 𝛾 (Angles is same segment in circle 𝐴𝐵𝐶 𝑍.)
∠𝐵𝑍𝑌 = 𝛾 (𝐴𝐵 = 𝐵𝑌 and equal angles subtend equal arcs.)
∠𝐵𝐴𝑍 = 180◦ − 𝛿 (Cyclic quad 𝐴𝐵𝑌 𝑍.)
So ∠𝑍 𝑋 𝐵 = 180◦ − 𝛿 (Straight line 𝑍 𝑋𝑌 .)
Now triangles 𝐴𝐵𝑍 and 𝑋 𝐵𝑍 have two, and therefore three, angles in common. The have a
common side 𝐵𝑍 so are congruent by 𝐴𝑆 𝐴. This means 𝐴𝐵𝑋 𝑍 is a kite and we conclude as
before.

Remark
Having found the equal angles at 𝑍, the common side 𝐵𝑍 and the equal sides 𝐵𝐴 and 𝐵𝑋 it is
tempting to say that 𝐴𝐵𝑍 and 𝑋 𝐵𝑍 are congruent by ASS. However, this is not a congruence
condition. Indeed triangle 𝐵𝑌 𝑍 has the same common angle and sides.

MarkersŠ comments
Those who were most successful at this problem started by drawing a large, accurate diagram
with a compass and ruler as shown in the first solution. This helped them to make conjectures
around other equal angles and lengths which many were able to turn into a complete solution.
The solutions given above provide a useful template for writing up a BMO1 geometry problem.
Having two columns, one for angle equalities and one for the justification, ensures that each
result is clearly justified. Many lost substantial marks by failing to provide sufficient justification
for all of their steps, for example using the fact that equal chords subtend equal angles without
justification.

© 2023 UK Mathematics Trust www.ukmt.org.uk 11

British Mathematical Olympiad Round 1 2023 Markers’ report

Question 4
Find all positive integers 𝑛 such that 𝑛 × 2𝑛 + 1 is a square.

Solution
Let 𝑎 > 0 be an integer such that 𝑛 × 2𝑛 + 1 = 𝑎 2 . It follows that 2𝑛 𝑛 = (𝑎 − 1)(𝑎 + 1). The
integers 𝑎 − 1 and 𝑎 + 1 have the same parity and their product is even, so both must be even.
Moreover, since 𝑎 − 1 and 𝑎 + 1 are consecutive even integers, one of them is twice an odd
number, so the other divisible by 2𝑛−1 .
Thus 𝑎 + 1 ≥ 2𝑛−1 which means 𝑎 − 1 ≤ 2𝑛. This implies that 2𝑛 + 2 ≥ 𝑎 + 1 ≥ 2𝑛−1 .
We claim that this false for 𝑛 ≥ 5, so we only need to check 𝑛 = 1, 2, 3 and 4.
When 𝑛 = 5, we have 12 = 2𝑛 + 2 < 2𝑛−1 = 16, so the claim holds for 𝑛 = 5.
If we know that 2𝑘 + 2 < 2𝑘−1 for some positive 𝑘, we can consider:
𝑘 +2 𝑘 +2
2(𝑘 + 1) + 2 = (2𝑘 + 2) × < 2𝑘−1 × < 2𝑘−1 × 2 = 2 (𝑘+1)−1
𝑘 +1 𝑘 +1
so if the claim holds for 𝑘, it holds for 𝑘 + 1.
Checking 𝑛 = 1, 2, 3 and 4, we find that the values of 𝑛 × 2𝑛 + 1 are 3, 9, 25 and 65 respectively.
Hence 𝑛 = 2 and 𝑛 = 3 work and 𝑛 = 1, 4 do not.

MarkersŠ comments
This was quite an approachable question 4, and there were many good attempts at it. Many
rearranged and factorised using a difference of squares. Some deduced that the square must be
odd, so used a different designation for their unknown and simplified an alternative expression
before factorising.
Either approach led to an equation of the form 𝑛2𝑛−2 = 𝑚(𝑚 + 1) or something equivalent
to this. Many incorrectly deduced that 𝑛 and 2𝑛−2 must be 𝑚 or 𝑚 + 1 in some order. This
received little further credit as it misses the possibility that the factors of 𝑛 are split between
𝑚 and 𝑚 + 1. It was important to note the crucial point about 𝑚 and 𝑚 + 1 being coprime.
However, if 𝑛 = 𝑝𝑞 it is possible that 𝑝 | 𝑚 and 𝑞 | 𝑚 + 1. What can be deduced is that 𝑚 and
𝑚 + 1 cannot both contain factors of 2, so either 2𝑛−2 | 𝑚 or 2𝑛−2 | 𝑚 + 1.
From here, both cases need to be considered to complete the proof, and those who only
considered one were penalised. Some considered both and then used one having demonstrated
it as a worst-case to continue the proof, which was acceptable.
To finish the proof, it was necessary to bound a suitable expression to show equality was
impossible above that bound. Many noted that “exponentials beat linear expressions”, but that
is only true beyond a certain point, which needed to be noted and demonstrated. A more formal
inductive proof was the best approach. Those using graphs or considering rates of change
needed sufficient detail to get full credit.

© 2023 UK Mathematics Trust www.ukmt.org.uk 12

British Mathematical Olympiad Round 1 2023 Markers’ report

Lastly, all those values below the bound needed to be checked manually. There was little
penalty for failing to explicitly show this had been done and many just wrote the values which
worked, but it is good practice to show all of them to complete the solution.

© 2023 UK Mathematics Trust www.ukmt.org.uk 13

British Mathematical Olympiad Round 1 2023 Markers’ report

Question 5
An artist arranges 1000 dots evenly around a circle, with each dot being either red or blue.
A critic looks at the artwork and counts faults: each time two red dots are adjacent is one
fault, and each time two blue dots are exactly two apart (that is, they have exactly one dot in
between them) is another.
What is the smallest number of faults the critic could find?

Solution
Placing red then blue dots two at a time round the circle (. . . 𝑅𝑅𝐵𝐵𝑅𝑅𝐵𝐵 . . . ), gives 250 ‘red’
faults and no ‘blue’ faults.
It remains to prove that every possible configuration contains at least 250 faults.
In every five consecutive dots there must be at least one fault (if any of the middle 3 dots are
red, then either that dot has a red neighbour or both its neighbours are blue, both giving a fault;
if none of the middle 3 dots are red, then we have 2 blue dots that are 2 apart, giving a fault).
Now divide the 1000 dots into blocks of five consecutive dots, each block overlapping by one
dot at each end. There are 250 such blocks, each containing at least one fault. Since each fault
consist of at least two dots, and blocks overlap by at most one dot, it is impossible to count the
same fault in two blocks. Hence, there must be at least 250 faults.
A slight variation on this argument is to divide the 1000 dots into non-overlapping blocks of
four dots and argue that each such block either contains a fault, which we will call an internal
fault, or is 𝑅𝐵𝐵𝑅. In the latter case then the dot that comes directly after the block will be
the second dot in a red fault if it is red and will be the second dot in a blue fault if it is blue.
Either way we have a fault the first of whose dots is in our block, which we call an external
fault. Since each of the 250 blocks contributes either an internal or an external fault, there must
be at least 250 faults in any configuration.

Alternative
There are various alternative approaches to showing every configuration has at least 250 faults,
such as the following: without loss of generality, we may consider a configuration with the
minimum number of faults and no more than two consecutive dots the same colour. Indeed, if
there are three or more consecutive blue dots, replace one in the middle with a red dot (this
doesn’t add any faults, and might remove some faults made of two blue dots that are two apart);
if there are three or more consecutive red dots, replace one in the middle with a blue dot (this
removes two faults made of two adjacent red dots, adds at most two faults made of blue dots
two apart). This step increases the number of blocks of consecutive dots of the same colour, so
it must terminate after finitely many steps, at which point all such blocks have length at most 2.
Now that all blocks of consecutive dots the same colour have length 1 or 2, the number of faults
equals the number of such blocks of red dots (a block of two red dots has a fault in it; a block
of one red dot gives a fault of the two blue dots either side). Since there are at least 500 blocks,
of alternating colours, there are at least 250 blocks of red dots, so at least 250 faults.

© 2023 UK Mathematics Trust www.ukmt.org.uk 14

British Mathematical Olympiad Round 1 2023 Markers’ report

MarkersŠ comments
This question consisted of two parts: firstly finding a configuration of dots that has 250 faults
and secondly showing that any configuration of dots admits at least 250 faults. The first
part was very approachable for a question 5. Indeed, of the candidates that attempted this
question, approximately 80% managed to find the optimal configuration of 𝑅𝑅𝐵𝐵𝑅𝑅𝐵𝐵 . . .
repeated around the circle. Many candidates that did not stumble across this solution offered
configurations which repeated every three dots such as 𝐵𝑅𝐵𝐵𝑅𝐵 . . . or 𝑅𝐵𝐵𝑅𝐵𝐵 . . . with 333
or 334 faults; these received partial credit.
However, the majority of candidates struggled to prove the lower bound for the number of
faults is 250. Many assumed without justification that an optimal configuration must consist
of a repeating pattern and therefore any attempts at showing a lower bound did not apply to
all possible configurations of dots. Some candidates tried an approach via induction on the
number of dots but most of these attempts did not properly justify how the lower bound on the
number faults increases in the inductive step.
Several candidates made the promising observation that 𝑅𝐵𝐵𝑅 is the only sequence of four dots
that do not contain a fault and some went further to show that the sequence 𝑅𝐵𝐵𝑅 is always
followed by a fault. However, most candidates made an incorrect logical leap to assume that
since 𝑅𝐵𝐵𝑅 is the longest sequence without a fault, an optimal solution must repeat 𝑅𝐵𝐵𝑅.
Whilst it is true in this case that the optimal solution is 𝑅𝐵𝐵𝑅 repeated around the circle,
to show a lower bound on the number of faults, one must show that any configuration must
incur at least 250 faults. Note that candidates that stated claims regarding 𝑅𝐵𝐵𝑅 without full
justification were penalised.
In order to receive full marks for this approach, following an observation of the optimality of
𝑅𝐵𝐵𝑅, a candidate needed to divide the 1000 dots into 250 blocks of fours (non-overlapping)
or fives (overlapping) and conclude that there must be a fault that either begins in the block (in
the case of non-overlapping fours) or contained within the block (in the case of overlapping
fives); alternatively it was possible to base an argument on showing that counting around any
arrangement we must see a fault at least every four dots but for full credit this needed to be
explained clearly and carefully, which was rare.
Finally, for those who are interested, consider what the optimal solution would be for a general 𝑛
dot circle. Alternatively, consider the problem where we define a fault differently: for instance,
what if we have a fault when two blue dots are exactly three apart (instead of two)?

© 2023 UK Mathematics Trust www.ukmt.org.uk 15

British Mathematical Olympiad Round 1 2023 Markers’ report

Question 6
For some integer 𝑛 > 4 a convex polygon has vertices 𝑣 1 , 𝑣 2 , . . . , 𝑣 𝑛 in that cyclic order.
All its edges are the same length. It also has the property that the lengths of the diagonals
𝑣 1 𝑣 4 , 𝑣 2 𝑣 5 , . . . , 𝑣 𝑛−3 𝑣 𝑛 , 𝑣 𝑛−2 𝑣 1 , 𝑣 𝑛−1 𝑣 2 and 𝑣 𝑛 𝑣 3 are all equal.
For which 𝑛 is it necessarily the case that the polygon has equal angles?

Solution
It is necessary and sufficient that 𝑛 is odd.
If 𝑛 = 2𝑘 we can construct a counterexample as follows: begin with a regular 𝑘-gon (shown
in blue in the left hand figure, where 𝑘 = 6) and erect an isosceles triangle with angles
𝛼, 𝛼, 180 − 2𝛼 externally on each side. This creates an equilateral 2𝑘-gon whose external
angles are alternately, 2𝛼 and 360/𝑘 − 2𝛼 (shown in black in the figure).
If we choose 𝛼 small enough that 2𝛼 < 360/𝑘 − 2𝛼 this polygon will not be regular. It will also
be convex since its interior angles are 180−2𝛼 < 180 and 180− (360/𝑘 −2𝛼) < 180−2𝛼 < 180.
To check it satisfies the condition on the diagonals we note that each 𝑣 𝑖 -𝑣 𝑖+3 diagonal (shown
in red in the figure) is the side of triangle with one side joining consecutive vertices on the
2𝑘-gon, and another side joining two vertices of the 𝑘-gon we started with. The included angle
is equal to 180 − 360/𝑘 + 𝛼, so all such triangles are congruent by SAS.

𝑣 𝑖+1 𝑣 𝑖+4
𝑣 𝑖+2 2𝛼 𝑣𝑖 𝑣 𝑖+3
𝛼 𝛼
360
− 2𝛼 𝛽
𝑘
𝛼 𝑣 𝑖+2
𝑣 𝑖+3
𝛼

𝛽 𝑣 𝑖+1

𝑣𝑖

Now suppose 𝑛 is odd. Consider the vertices 𝑣 𝑖 , 𝑣 𝑖+1 , 𝑣 𝑖+2 , 𝑣 𝑖+3 , 𝑣 𝑖+4 (with subscripts mod 𝑛).
In the right hand figure the triangles 𝑣 𝑖 𝑣 𝑖+1 𝑣 𝑖+3 and 𝑣 𝑖+4 𝑣 𝑖+3 𝑣 𝑖+1 each have a black, a red and a
blue side. They are therefore congruent by SSS. We call the the (equal) angles opposite the
blue sides 𝛽. That is, we let ∠𝑣 𝑖+3 𝑣 𝑖+1 𝑣 𝑖 = ∠𝑣 𝑖+4 𝑣 𝑖+3 𝑣 𝑖+1 = 𝛽.
Since 𝑣 𝑖+1 𝑣 𝑖+2 𝑣 𝑖+3 is isosceles (with two black sides and one red) we may call its (equal) base
angles 𝛼. That is, we let ∠𝑣 𝑖+2 𝑣 𝑖+1 𝑣 𝑖+3 = ∠𝑣 𝑖+1 𝑣 𝑖+3 𝑣 𝑖+2 = 𝛼. Now it is clear that the angles in

© 2023 UK Mathematics Trust www.ukmt.org.uk 16

British Mathematical Olympiad Round 1 2023 Markers’ report

polygon at 𝑣 𝑖+1 and 𝑣 𝑖+3 are both equal to 𝛼 + 𝛽. This holds for every 𝑖.
Thus ∠𝑣 𝑛 𝑣 1 𝑣 2 = ∠𝑣 2 𝑣 3 𝑣 4 = · · · = ∠𝑣 𝑛−1 𝑣 𝑛 𝑣 1 = ∠𝑣 1 𝑣 2 𝑣 3 = · · · = ∠𝑣 𝑛−2 𝑣 𝑛−1 𝑣 𝑛 .
Therefore, all the angles of the polygon are equal.

MarkersŠ comments
As usual, Q6 proved to be a very challenging question, despite the solution being potentially
quite short. This might be because, as a ‘non-standard geometry’ problem, many students may
have found the required reasoning to be unfamiliar.
Many students managed to solve the 𝑛 = 5 case, but points were only given for statements that
apply more generally. Many other students were unsure about the exact meaning of ‘convex’.
The word was included in the problem statement in order to help students by eliminating any
strange diagrams that would otherwise need to be considered, but it did cause considerable
confusion.
Another common problem was students fixating on the value of 𝑛 mod 3, noticing that when
𝑛 ≡ 0 (mod 3) the given diagonals form into three disjoint 𝑛/3-gons. While true, this
observation didn’t actually lead to a solution and such scripts generally scored 0 unless there
were other observations that could lead to a solution.
Many other students were held back from full marks by the weakness in their handling of the
even case. Implicitly, a question such as this requires a proof that for all odd 𝑛 the polygon
must be regular, and that for all even 𝑛 it need not be regular. Concluding ‘therefore it must be
regular for all odd 𝑛’ is only half of the solution and will not score close to full marks. To score
a 10, the even case has to show that the construction is a) a polygon, b) convex, and c) satisfies
the length constraints. As a result, very few scripts scored full marks.
Finally, for those who are interested in this style of question, you may like to try solving it with
a different set of constraints. The problem as set could be called {1, 3} as ‘diagonals’ of step
size 1 and 3 are constrained. Can you solve {2, 3} or {1, 4}? To our knowledge, they are open
questions.

© 2023 UK Mathematics Trust www.ukmt.org.uk 17