DR

Not 24

Mar 07, 2026

Unnecessary proofs

[NOTE] This post is especially long with a lot of proofs, to make reading easier proofs can be collapsed by clicking on the QED square at the bottom of the environment.

Background

Consider the following game, loosely inspired by 24. Let's say we have 4 people. Each person names a digit between 1 and 9 (inclusive)1. Given only the operations {+,,×,/,,()2}\{+, -, \times, \cdot / \cdot, \sqrt{\cdot}, (\cdot)^2 \}2 (and of course parentheses to dictate order), and the constraint that each digit must be used exactly once, construct a valid equation that evaluates to 10. If a digit is named more than once, it must be used exactly as many times as it was named.

This is simple enough, but what if I want to play with more people? We may also ask is it ever possible to get a set of kk numbers such that it is not possible to get to 10? I've been thinking about this problem since November and I've made some progress, but I thought it'd be interesting to sorta document how this goes.

Statement and Strategy

Statement

It will likely be helpful to formally state out exactly what we're looking for. Given a multiset of digits D[9]kD \in [9]^k, and the operations list M{+,,×,/,,()2}M \doteq \{+, -, \times, \cdot / \cdot, \sqrt{\cdot}, (\cdot)^2\} (where we note that \sqrt{\cdot} is defined only for integer perfect squares), is there some valid expression in which we use all the kk digits exactly once (and use operations as many times as we wish) that will evaluate to 10?

We then seek the minimum kk such that the above is always true regardless of the multiset.

Strategy

Our strategy will be to construct increasingly tight upper and lower bounds until they meet (sort of how the Rubik's cube God's number proof was done). I also really want to do this without computer assistance, but we'll see how feasible that is.3

My first crack at this (roughly November to January) involved trying to construct some sort of homogeneity lemma (basically that a multiset where every digit is the same is the hardest case to solve). I still think this is right (particularly for 7 since it is the largest prime that is coprime to 10), but I cannot figure out how to prove it so for now we'll just leave it.4

Answer

The final answer is that the minimum kk is 5.

k18k \leq 18 (sometime in Dec/Jan)

Claim 1 (Achievability for k18k \geq 18). For every k18k \geq 18, we can construct 10 according to the rules of the game.

Before we proceed with the proof, it'll be helpful to introduce some notation and set things up.

Let our mutliset of digits be denoted D[9]kD \subset [9]^k. Let cdc_d denote the multiplicity of each digit dDd \in D so i=19cd=k\sum_{i=1}^9 c_d =k. Note that we can construct 10 from 4 1s: (1+1+1)2+1=10. (1+1+1)^2 + 1 = 10.

We may easily obtain 11 via d/dd/d, which consumes 2 occurrences of the digit dd. Therefore, to build 4 1s of the form d/dd/d, we need 4 disjoint equal pairs among the digits. Define the number of disjoint equal pairs available in DD by P(D)d=19cd2. P(D) \doteq \sum_{d=1}^9 \left\lfloor \frac{c_d}{2} \right\rfloor.

Now, we introduce the following lemma for convenience:

Lemma 2 (Pair-count lower bound). If k16k \geq 16, then P(D)4P(D) \geq 4 for every multiset d[9]kd \subseteq [9]^k.

Proof of Lemma 2.

Suppose P(D)3P(D) \leq 3. Then each digit contributes at most cd2\left\lfloor \frac{c_d}{2}\right\rfloor pairs, so cd2cd2+1 c_d \leq 2 \left\lfloor \frac{c_d}{2}\right\rfloor + 1 for each dd. Summing over d=1,,9d = 1, \dots, 9, k=i=19cdi=19(2cd2+1)=2P(D)+923+9=15. k = \sum_{i=1}^9 c_d \leq \sum_{i=1}^9 \left(2\left\lfloor \frac{c_d}{2}\right\rfloor + 1 \right) = 2P(D) + 9 \leq 2\cdot 3 + 9 = 15. Thus, if k16k \geq 16, it is impossible to have P(D)3P(D) \leq 3; hence P(D)4P(D) \geq 4.

Now we may prove the original claim.

Proof of Claim 1.

By the above lemma, for k16k \geq 16 we may choose four disjoint pairs (a,a),(b,b),(c,c),(d,d)(a,a), (b,b), (c,c), (d,d) (not necessarily unique digits) and form aa=bb=cc=dd=1, \frac{a}{a} = \frac bb = \frac cc = \frac dd = 1, thus we can reach 10. To guarantee we can use all remaining digits without affecting the value 10, it suffices to have an additional equal pair (e,e)(e,e) disjoint from the 8 digits already used, because then we can form ee=0e-e=0 and multiply the entire leftover expression by 0.

So we want P(D)5P(D) \geq 5. From the same inequality as used in the lemma, k2P(D)+9. k \leq 2 P(D)+9. Plugging in our desired P(D)1P(D)-1 (same logic as lemma), we have that k24+9=17k \leq 2 \cdot 4 + 9 = 17, so to ensure P(D)5P(D) \geq 5, we require k18k \geq 18, as claimed.

k5k \geq 5 (Feb 05)

We show this explicitly by counterexample.

Claim 3 (Counterexample for k=4k=4). Given the operations in MM and the multiset {7,7,7,7}\{7,7,7,7\}, we can never construct 10.

Proof of Claim 3.

Intuition. The only way to use exactly 4 leaves is via a full binary tree with 4 leaves. There are only 2 root shapes: (2,2)(2,2) and (1,3)(1,3). The delicate point is that unary operations \sqrt{\cdot} and ()2(\cdot)^2 may be applied at internal nodes without consuming digits, so we must account for them completely. We reduce the problem to a finite target-exclusion check.

Unary closure

Define the unary-closure operator on a set SS by cl(S)S{x2:xS}{x:xS and x is a perfect square integer}. cl(S) \doteq S \cup \{x^2 : x \in S\} \cup \{\sqrt{x} : x \in S \text{ and } x \text{ is a perfect square integer}\}. Since unary operations act only on the value already produced at a node, the set of values attainable by any fixed subtree is closed under cl()cl(\cdot).

Classify all 2-leaf values from {7,7}\{7,7\}

Let S2S_2 denote the set of all values obtainable from exactly two leaves {7,7}\{7,7\} using operations in MM. We claim S2{0, 1, 14, 49, 98, 196, 2401, 1/7, 1/49}. S_2 \supseteq \{0,\ 1,\ 14,\ 49,\ 98,\ 196,\ 2401,\ 1/7,\ 1/49\}.

Justification:

  • Using only the binary operations on raw leaves gives 77=07-7=0, 7/7=17/7=1, 7+7=147+7=14, 7×7=497\times 7=49.
  • Applying ()2(\cdot)^2 to 1414 and 4949 yields 196196 and 24012401. Applying \sqrt{\cdot} to 196196 and 24012401 returns 1414 and 4949; no other square roots are defined from two 7s.
  • By allowing a unary operation on a single leaf before the binary root, we may use 4949 as a leaf value and form 7/49=1/77/49 = 1/7. Squaring yields (1/7)2=1/49(1/7)^2 = 1/49.

(We will only need the displayed subset of small values in the later finite checks.)

Reduce the 4-leaf problem to two root shapes

Every full 4-leaf binary tree has root of type (2,2)(2,2) or (1,3)(1,3). Concretely, any 4-leaf expression is of one of the following forms (up to swapping left/right): E=U(AB)with A,B 2-leaf subexpressions, E = U(A \circ B) \quad \text{with } A,B \text{ 2-leaf subexpressions}, or E=U(7T) or U(T7)with T a 3-leaf subexpression, E = U(7 \circ T) \ \text{or}\ U(T \circ 7) \quad \text{with } T \text{ a 3-leaf subexpression}, where {+,,×,/}\circ \in \{+,-,\times,\cdot/\cdot\} and U()U(\cdot) denotes an (optional) finite composition of unary operations applied at the root (each application must be defined).

Unary operations at the root force a finite set of binary targets

The only unary operations are xx2x \mapsto x^2 and xxx \mapsto \sqrt{x} (when defined). Since 1010 is neither a perfect square nor the square root of a perfect square integer other than 100, the only way a root unary chain can produce 10 is if the binary root value lies in {10, 10, 100}. \{10,\ -10,\ 100\}. (x2=10x^2=10 has no rational solution; x=10\sqrt{x}=10 forces x=100x=100; and x2\sqrt{x^2} can introduce an absolute value, hence 10-10.)

Thus, to show impossibility, it suffices to show that in either root shape, the binary root can never evaluate to any of 10,10,10010,-10,100.

Exclude root shape (2,2)(2,2)

In this shape, A,BS2A,B \in S_2. Therefore the binary root lies in the finite set {ab:a,bS2, {+,,×,/}}. \{a\circ b : a,b \in S_2,\ \circ \in \{+,-,\times,\cdot/\cdot\}\}. A direct check shows this set does not contain 1010, 10-10, or 100100. Hence the (2,2)(2,2) shape cannot yield 10.

Exclude root shape (1,3)(1,3)

In this shape, the binary root is 7T7 \circ T or T7T \circ 7. To have the binary root in {10,10,100}\{10,-10,100\}, TT must satisfy one of finitely many equations, giving the forced target set T{3, 3, 17, 17, 93, 93, 107, 10/7, 10/7, 100/7,100/7, 7/10, 7/10, 7/100, 7/100, 70, 70, 700, 700}. \begin{aligned} \mathcal{T} \doteq \{&3,\ -3,\ 17,\ -17,\ 93,\ -93,\ 107,\ 10/7,\ -10/7,\ 100/7,\\ &-100/7,\ 7/10,\ -7/10,\ 7/100,\ -7/100,\ 70,\ -70,\ 700,\ -700\}. \end{aligned} (These come from solving 7T=y7 \circ T = y and T7=yT \circ 7 = y for y{10,10,100}y \in \{10,-10,100\} and {+,,×,/}\circ \in \{+,-,\times,\cdot/\cdot\}.)

Now, any 3-leaf expression necessarily contains a 2-leaf subtree; evaluating that subtree first yields some xS2x \in S_2, and then TT is obtained by combining xx with the remaining leaf 77 using one binary operation, followed by unary closure: Tcl({x7, 7x:xS2, {+,,×,/}}). T \in cl(\{x \circ 7,\ 7 \circ x : x \in S_2,\ \circ \in \{+,-,\times,\cdot/\cdot\}\}). Therefore, it suffices to check the finite set on the right-hand side and verify it is disjoint from T\mathcal{T}. This direct check succeeds: no element of T\mathcal{T} appears.

Hence the (1,3)(1,3) shape cannot yield a binary root in {10,10,100}\{10,-10,100\}, and therefore cannot yield 10 after any root-level unary operations.

Neither root shape (2,2)(2,2) nor (1,3)(1,3) can yield a binary root value in {10,10,100}\{10,-10,100\}, hence no valid expression using exactly four 7s can evaluate to 10. This proves the claim.

k16k \leq 16 (Feb 10)

Claim 4 (Achievability for k16k \geq 16). For every k16k \geq 16, we can construct 10 according to the rules of the game.

The k18k \leq 18 proof earlier "wastes" 2 digits to manufacture a spare equal pair (e,e)(e,e) so we can build ee=0e-e=0 and discard all remaining digits. At k=16k=16 we cannot always guarantee a 5th equal pair, so we instead manufacture a duplicate-free zero using the unary operations.

We retain the pair-count notation from before. Let our multiset of digits be D[9]kD \subset [9]^k, let cdc_d denote the multiplicity of each digit d[9]d \in [9], and define P(D)d=19cd2. P(D) \doteq \sum_{d=1}^9 \left\lfloor \frac{c_d}{2} \right\rfloor.

By Lemma 2, if k16k \geq 16 then P(D)4P(D) \geq 4. Thus, for k=16k=16 we may always select 4 disjoint equal pairs and form 4 1s: aa=bb=cc=dd=1, \frac{a}{a} = \frac{b}{b} = \frac{c}{c} = \frac{d}{d} = 1, and hence we can build T(aa+bb+cc)2+dd=10. T \doteq \left(\frac{a}{a}+\frac{b}{b}+\frac{c}{c}\right)^2+\frac{d}{d} = 10. This uses exactly 8 digits (4 pairs). The remaining issue is: how do we force the remaining 8 digits to be used exactly once without changing the value?

We can resolve this via the following observation: when k=16k=16 and P(D)=4P(D)=4, the multiset must contain all but at most one digit from [1,9][1,9]. In particular, it must contain either {2,4}\{2,4\} or {3,9}\{3,9\}, which gives a zero without requiring duplicates.

Lemma 5 (Forced zero gadget at k=16k=16). Let D[9]16D \subseteq [9]^{16} and suppose P(D)=4P(D)=4. Then at least one of the following holds: (i) digits 2 and 4 both occur in DD, or (ii) digits 3 and 9 both occur in DD. Hence, DD admits a duplicate-free zero: Z{422, 932}, Z \in \{4-2^2,\ 9-3^2\}, and hence Z=0Z=0.

Proof of Lemma 5.

Write cd=2cd2+rd,rd{0,1}. c_d = 2\left\lfloor \frac{c_d}{2}\right\rfloor + r_d, \quad r_d \in \{0,1\}. Summing over d=1,,9d=1,\dots,9, 16=d=19cd=2P(D)+d=19rd=8+d=19rd, 16 = \sum_{d=1}^9 c_d = 2P(D) + \sum_{d=1}^9 r_d = 8 + \sum_{d=1}^9 r_d, so d=19rd=8\sum_{d=1}^9 r_d = 8. Since rd=1r_d=1 implies cdc_d is odd (hence 1\geq 1), this means that at least 8 distinct digits appear in DD, i.e. at most one digit is missing from {1,,9}\{1,\dots,9\}.

Now consider the set {2,3,4,9}\{2,3,4,9\}. If 2 is missing then 3,4,9 are present, so in particular {3,9}D\{3,9\} \subset D. If digit 3 is missing then {2,4}D\{2,4\} \subset D. Similarly, if 4 is missing then {3,9}D\{3,9\} \subset D, and if 9 is missing then {2,4}D\{2,4\} \subset D. If none are missing, then both {2,4}\{2,4\} and {3,9}\{3,9\} are present. Thus at least one of {2,4}\{2,4\} or {3,9}\{3,9\} occurs, and we can form 422=04-2^2=0 or 932=09-3^2=0 using each digit exactly once.

Lemma 6 (Zero gadget can be chosen without destroying the four pairs). Let D[9]16D \subseteq [9]^{16} and suppose P(D)=4P(D)=4. Then there exists a choice of Z{422, 932} Z \in \{4-2^2,\ 9-3^2\} such that the digits used to build ZZ can be reserved while still leaving at least four disjoint equal pairs among the remaining digits. In particular, after reserving the digits of ZZ, we still have enough pairs to build T=(aa+bb+cc)2+dd=10. T = \left(\frac{a}{a}+\frac{b}{b}+\frac{c}{c}\right)^2+\frac{d}{d} = 10.

Proof of Lemma 6.

Since P(D)=4P(D)=4 and k=16k=16, write cd=2cd2+rd,rd{0,1}. c_d = 2\left\lfloor \frac{c_d}{2}\right\rfloor + r_d,\quad r_d\in\{0,1\}. As in Lemma 5, summing gives d=19rd=8\sum_{d=1}^9 r_d = 8. Hence exactly one digit has even multiplicity; call it ee. (All other digits have odd multiplicity, hence at least 1.)

Now note the following fact: if cdc_d is odd, then reserving a single copy of dd does not decrease the number of available pairs contributed by dd, since cd12=cd2when cd is odd. \left\lfloor \frac{c_d-1}{2}\right\rfloor = \left\lfloor \frac{c_d}{2}\right\rfloor \quad \text{when } c_d \text{ is odd}. The only way reserving digits can decrease P(D)P(D) is if we reserve a copy of the unique even-multiplicity digit ee.

By Lemma 5, at least one of the gadgets is available; choose Z{422,932}Z \in \{4-2^2, 9-3^2\} so that reserving its digits leaves four disjoint equal pairs. Choose the zero gadget ZZ so that it does not use the digit ee:

  • If e{2,4}e \in \{2,4\}, choose Z=932Z = 9-3^2.
  • If e{3,9}e \in \{3,9\}, choose Z=422Z = 4-2^2.
  • If e{2,3,4,9}e \notin \{2,3,4,9\}, choose whichever of the two gadgets is available.

Because at most one digit is missing from {1,,9}\{1,\dots,9\}, in the first two bullets the "other" gadget is guaranteed to be available: if (say) e=2e=2 then digit 2 is present (even multiplicity) and so the only possible missing digit is not both 3 and 9, hence {3,9}D\{3,9\}\subset D; similarly for the other cases.

With this choice, the digits reserved for ZZ come only from odd-multiplicity digits, so reserving them does not reduce P(D)P(D). Therefore there remain at least four disjoint equal pairs available to build T=10T=10.

We can now finish our original k16k \leq 16 construction.

Proof of Claim 4.

Fix k16k \geq 16. If P(D)5P(D) \geq 5, then we revert to the k18k \leq 18 proof: use 4 pairs to build T=10T=10 and use the 5th pair to build ee=0e-e=0, absorbing all remaining digits.

So assume k=16k=16 and P(D)=4P(D)=4. By Lemma 6, we can form a duplicate-free zero gadget Z{422, 932}Z \in \{4-2^2,\ 9-3^2\} with Z=0Z=0 (choose whichever is available in DD).

Independently, since P(D)=4P(D)=4, we can select 4 disjoint equal pairs and construct T=(aa+bb+cc)2+dd=10. T = \left(\frac{a}{a}+\frac{b}{b}+\frac{c}{c}\right)^2+\frac{d}{d} = 10. Let RR be any expression using every remaining unused digit exactly once (e.g. their product in any parenthesization). Then T+ZR=10 T + Z\cdot R = 10 uses every digit exactly once and evaluates to 10.

k14k \leq 14 (Feb 28)

Claim 7 (Achievability for k14k \ge 14). For every k14k \ge 14 and every multiset D[9]kD \subset [9]^k, there exists a valid expression using each digit exactly once that evaluates to 1010.

The main improvement over the k16k \le 16 argument is that we no longer insist on manufacturing 1 exclusively from equal pairs (d,d)(d,d). We can also obtain 1 from adjacent digits: ab=1whenever ab=1, \left| a-b \right| = 1 \quad \text{whenever } \left| a-b \right|=1, or, if we wish to avoid ordering, (ab)2=1. (a-b)^2 = 1. Thus any disjoint pair (x,y)(x,y) with xy1\left| x-y \right| \le 1 can be converted into a 1.

This reduces the combinatorics of “how many 1s can we force?” to a matching problem on a path.

Let cic_i denote the multiplicity of digit ii. Write ci=2ci2+ri,ri{0,1}. c_i = 2\left\lfloor \frac{c_i}{2} \right\rfloor + r_i, \qquad r_i \in \{0,1\}.

Define:

Interpret the digits as tokens placed on the path 1239. 1 - 2 - 3 - \cdots - 9. A unit-producing pair is either:

Let U(D)U(D) denote the maximum number of disjoint unit-producing pairs extractable from DD.

After removing all equal pairs (giving P(D)P(D) pairs), the remaining tokens lie on the vertices with ri=1r_i=1, at most one per vertex. Among these, we may additionally match adjacent vertices along the path.

Let A(D)ν(P9[{iri=1}]), A(D) \doteq \nu\big(P_9[\{i \mid r_i=1\}]\big), where ν(G)\nu(G) denotes the matching number of a graph GG; the number of pairwise disjoint edges in GG. Then

U(D)P(D)+A(D). U(D) \geq P(D) + A(D).

Lemma 8 (Four unit pairs from 12 digits). If D=12\left| D \right| = 12, then U(D)4U(D) \ge 4.

Proof of Lemma 8.

Since 12=i=19ci=2P(D)+R(D), 12 = \sum_{i=1}^9 c_i = 2P(D) + R(D), we have R(D)=122P(D). R(D) = 12 - 2P(D).

We argue by cases.

Case 1: P(D)4P(D) \ge 4. Then U(D)P(D)4U(D) \ge P(D) \ge 4.

Case 2: P(D)=3P(D) = 3. Then R(D)=6R(D) = 6. So six vertices of the path carry an odd token.

But the path on nine vertices has independence number 55; hence any subset of six vertices must contain at least one adjacent pair. Therefore A(D)1A(D) \ge 1, and U(D)3+1=4. U(D) \ge 3 + 1 = 4.

Case 3: P(D)2P(D) \le 2. Then R(D)8R(D) \ge 8.

Now we claim any induced subgraph of the 9-vertex path on at least 7 vertices has matching number at least 2.

Indeed, if a subset SS of vertices satisfies ν(P9[S])1\nu(P_9[S]) \le 1, then the induced subgraph consists of isolated vertices together with at most one component containing an edge. In a path, such a component has size at most 3. The remaining vertices must be isolated and therefore pairwise nonadjacent; there can be at most 4 of these. Hence S3+4=6\left| S \right| \le 3 + 4 = 6. Contrapositively, S7\left| S \right| \ge 7 implies ν(P9[S])2\nu(P_9[S]) \ge 2.

Applying this with S=R(D)8\left| S \right| = R(D) \ge 8, we obtain A(D)2A(D) \ge 2, and therefore U(D)P(D)+A(D)2+2=4. U(D) \ge P(D) + A(D) \ge 2 + 2 = 4.

All cases give U(D)4U(D) \ge 4.

The bound is tight at 11 digits: for example, D={5,5,5,5,5,5,5,1,3,7,9} D = \{5,5,5,5,5,5,5,1,3,7,9\} admits only three disjoint unit pairs. (We may note that we can construct a 4th 1 e.g. via (97)/(31)(9-7)/(3-1), but figuring out how to formalize this is too much work; I'll likely revisit it later.)

Proof of Claim 7.

Let k14k \ge 14. Since there are only 9 digit types, some digit ee appears at least twice. Reserve two copies and form Zee=0. Z \doteq e - e = 0.

Remove these two digits; the remaining multiset DD' has size D=k212. \left| D' \right| = k - 2 \ge 12.

By Lemma 8, we have U(D)4U(D') \ge 4. Hence we may extract 4 disjoint unit pairs from DD' and convert them into 4 1s.

Using these four 1s, construct T(1+1+1)2+1=10. T \doteq (1+1+1)^2 + 1 = 10.

Let RR be any expression using all remaining unused digits exactly once (for instance, their product in any parenthesization). Then T+ZR=10 T + Z \cdot R = 10 uses every digit exactly once and evaluates to 1010.

k11k \le 11 (Mar 01)

Claim 9 (Achievability for k11k \ge 11). For every k11k \ge 11 and every multiset D[9]kD \subset [9]^k, there exists a valid expression using each digit exactly once that evaluates to 1010.

We shall keep the same “absorb leftovers with a 00” architecture as before, but sharpen the internal 1010-kernel: instead of forcing four disjoint unit pairs (to make four 11s), we show that any multiset of size 99 already contains a small kernel for 1010. This drops the global threshold by 2.

We again view the digits as tokens on the path 1239. 1 - 2 - 3 - \cdots - 9.

A unit-producing pair is defined the same as before.

A nine-producing gadget is any of:

The main kernel identity is 10=9+1, 10 = 9 + 1, so it suffices to build a nine-gadget and a unit-pair disjointly.

We shall want a crude extremal identity for unit pairs, stated below.

Lemma 10 (Unit pair from six digits). If D6\left| D \right| \ge 6, then DD contains a unit-producing pair.

Proof of Lemma 10.

If some digit repeats, we have an equal pair (i,i)(i,i) and hence 1=i/i1=i/i. Otherwise all digits are distinct. But the independence number of the 9-vertex path is 55, so any set of 66 distinct vertices contains an adjacent pair (i,i+1)(i,i+1). Then 1=(i(i+1))21=(i-(i+1))^2.

We will also need a 2-value reduction.

Lemma 11 (Two-value support at size 9 gives 10). Let D[9]9D \subset [9]^9. If DD is supported on at most two digit-values, then DD admits a valid expression using some of the digits of DD that evaluates to 1010.

Proof of Lemma 11.

If DD is supported on at most two values, write D={xcx,ycy},cx+cy=9. D = \{x^{c_x}, y^{c_y}\}, \qquad c_x + c_y = 9. Then the number of disjoint equal pairs is P(D)=cx2+cy2. P(D) = \left\lfloor \frac{c_x}{2} \right\rfloor + \left\lfloor \frac{c_y}{2} \right\rfloor. For any integers cx,cy0c_x,c_y \ge 0 with cx+cy=9c_x+c_y=9, we have P(D)4P(D) \ge 4; the minimum occurs at 1+81+8 or 2+72+7, giving 0+4=40+4=4 or 1+3=41+3=4.

Thus we can always select four disjoint equal pairs and form four 11s: aa=bb=cc=dd=1. \frac aa = \frac bb = \frac cc = \frac dd = 1. Then (1+1+1)2+1=10 (1+1+1)^2 + 1 = 10 gives a 1010-kernel.

Now we will require a kernel lemma.

Lemma 12 (A 10-kernel from 9 digits). For every multiset D[9]9D \subset [9]^9, there exists a valid expression using some of the digits of DD (each at most once) that evaluates to 1010.

Proof of Lemma 12.

We proceed by successive reductions. Throughout, we shall ignore unused digits.

Immediate 2-leaf kernels. If DD contains any of (1,9),(2,8),(3,7),(4,6)(1,9),(2,8),(3,7),(4,6) or contains two 55s, then we can form 1010 by addition and are done. Hence assume:

  • no complement-to-10 pair occurs in DD, and
  • c51c_5 \le 1.

Singleton nine gadgets. If 9D9 \in D, we have 99. If 3D3 \in D, we have 9=329=3^2. In either case, removing that one digit leaves 8 digits. By Lemma 10, the remaining digits contain a disjoint unit pair producing 11. Hence 10=9+110=9+1. So assume: 3Dand9D. 3 \notin D \quad \text{and} \quad 9 \notin D.

2-leaf nine gadgets. If DD contains any sum-to-9 pair among (1,8),(2,7),(4,5)(1,8),(2,7),(4,5), then 9=x+y9=x+y uses two digits, leaving 7 digits; by Lemma 10, these yield a disjoint 11, hence 10=9+110=9+1. Similarly, if DD contains a distance-3 pair (i,i+3)(i,i+3), then 9=(i(i+3))29=(i-(i+3))^2 and again the remaining 7 digits yield a disjoint 11. Hence we may assume:

  • none of (1,8),(2,7),(4,5)(1,8),(2,7),(4,5) occurs in DD, and
  • no distance-3 pair (i,i+3)(i,i+3) occurs in DD.

mod-3 chain collapse. The distance-3 edges partition the digits into chains: 147,258,369. 1-4-7, \qquad 2-5-8, \qquad 3-6-9. The absence of any distance-3 pair implies that the support of DD restricts on each chain to an independent set of that 3-vertex path. Thus:

  • on 1471-4-7, the allowed occupied subsets are ,{1},{4},{7},{1,7}\emptyset, \{1\}, \{4\}, \{7\}, \{1,7\},
  • on 2582-5-8, the allowed occupied subsets are ,{2},{5},{8},{2,8}\emptyset, \{2\}, \{5\}, \{8\}, \{2,8\},
  • on 3693-6-9, since 3,9D3,9 \notin D, the only possibilities are \emptyset and {6}\{6\}.

Now use the earlier exclusions:

  • {2,8}\{2,8\} is forbidden because it is a complement-to-10 pair,
  • {1,8}\{1,8\} and {7,2}\{7,2\} are forbidden because they are complement-to-10 pairs,
  • {4,5}\{4,5\} is forbidden because it is a sum-to-9 pair.

Therefore the support of DD must be one of the following:

  • one-value supports: {1}, {2}, {4}, {5}, {6}, {7}, {8}, \{1\},\ \{2\},\ \{4\},\ \{5\},\ \{6\},\ \{7\},\ \{8\},
  • two-value supports: {1,2}, {1,5}, {1,6}, {1,7}, {2,4}, {2,6}, {4,8}, {5,6}, {5,7}, {6,7}, {6,8}, {7,8}, \{1,2\},\ \{1,5\},\ \{1,6\},\ \{1,7\},\ \{2,4\},\ \{2,6\},\ \{4,8\},\ \{5,6\},\ \{5,7\},\ \{6,7\},\ \{6,8\},\ \{7,8\},
  • multi-value supports: {1,2,6}, {1,5,6}, {1,5,7}, {1,6,7}, {5,6,7}, {6,7,8}, {1,5,6,7}. \{1,2,6\},\ \{1,5,6\},\ \{1,5,7\},\ \{1,6,7\},\ \{5,6,7\},\ \{6,7,8\},\ \{1,5,6,7\}.

If DD is supported on at most two values, then Lemma 11 applies and we are done. So it remains only to handle the seven explicit multi-value supports.

Support {1,2,6}\{1,2,6\} Let (a,b,c)(a,b,c) denote multiplicities of (1,2,6)(1,2,6), so a+b+c=9a+b+c=9.

  • If a4a \ge 4, use 10=(1+1+1)2+1. 10 = (1+1+1)^2 + 1.
  • Else if b4b \ge 4, use 10=222+2. 10 = 2\cdot2\cdot2+2.
  • Else if b2b \ge 2 and c1c \ge 1, use 10=6+22. 10 = 6+2^2.
  • Else if b=1b=1 and a2a \ge 2, use 10=(2+1)2+1. 10 = (2+1)^2+1.
  • Else if a1a \ge 1 and c4c \ge 4, use 10=1+(6+6+66)2. 10 = 1+\left(\frac{6+6+6}{6}\right)^2.
  • Otherwise c6c \ge 6, so use 10=(6+66)2+6. 10 = \left(\frac{6+6}{6}\right)^2+6.

Support {1,5,6}\{1,5,6\}

  • If c52c_5 \ge 2, then 10=5+510=5+5.
  • Otherwise c5=1c_5=1, so c1+c6=8c_1+c_6=8.
  • If c14c_1 \ge 4, use 10=(1+1+1)2+110=(1+1+1)^2+1.
  • If c13c_1 \le 3, then c65c_6 \ge 5; form 3=(6+6+6)/63=(6+6+6)/6 and then 10=1+32. 10 = 1 + 3^2.

Support {1,5,7}\{1,5,7\}

  • If c52c_5 \ge 2, then 10=5+510=5+5.
  • Otherwise c5=1c_5=1, so c1+c7=8c_1+c_7=8.
  • If c14c_1 \ge 4, use 10=(1+1+1)2+110=(1+1+1)^2+1.
  • If c13c_1 \le 3, then c75c_7 \ge 5; use 10=(7+7+77)2+1. 10 = \left(\frac{7+7+7}{7}\right)^2 + 1.

Support {1,6,7}\{1,6,7\}

  • If c14c_1 \ge 4, use 10=(1+1+1)2+110=(1+1+1)^2+1.
  • If c63c_6 \ge 3, use 10=7+6+6+66. 10 = 7 + \frac{6+6+6}{6}.
  • Otherwise c62c_6 \le 2, so c74c_7 \ge 4; use 10=(7+7+77)2+1. 10 = \left(\frac{7+7+7}{7}\right)^2 + 1.

Support {5,6,7}\{5,6,7\}

  • If c52c_5 \ge 2, then 10=5+510=5+5.
  • Otherwise c5=1c_5=1, and since the support is exactly {5,6,7}\{5,6,7\} we may use 10=6+(75)2. 10 = 6 + (7-5)^2.

Support {6,7,8}\{6,7,8\}

  • If c72c_7 \ge 2, use 10=8+7+77. 10 = 8 + \frac{7+7}{7}.
  • Otherwise c7=1c_7=1.
  • If c63c_6 \ge 3, use 10=7+6+6+66. 10 = 7 + \frac{6+6+6}{6}.
  • If c62c_6 \le 2, then c86c_8 \ge 6; use 10=(8+8+88)2+88. 10 = \left(\frac{8+8+8}{8}\right)^2 + \frac 88.

Support {1,5,6,7}\{1,5,6,7\}

  • If c52c_5 \ge 2, then 10=5+510=5+5.
  • Otherwise c5=1c_5=1, and since 6 and 7 are present we may use 10=6+(75)2. 10 = 6 + (7-5)^2.

Thus in all cases, DD admits a 1010-kernel.

Now the proof of our original claim is straightforward.

Proof of Claim 9.

Let k11k \ge 11 and D[9]kD \subset [9]^k. By pigeonhole, some digit ee appears at least twice. Reserve two copies and form Zee=0. Z \doteq e - e = 0. Remove these two digits; the remaining multiset DD' has size D=k29\left| D' \right| = k-2 \ge 9. By Lemma 12, DD' contains a 1010-kernel TT.

Let RR be any expression using all remaining unused digits of DD exactly once (for example, their product in any parenthesization). Then T+ZR=10 T + Z \cdot R = 10 uses every digit exactly once and evaluates to 1010.

k9k \le 9 (Mar 07)

Claim 13 (Achievability for k9k \ge 9). For every k9k \ge 9 and every multiset D[9]kD \subset [9]^k, there exists a valid expression using each digit exactly once that evaluates to 1010.

The k11k \le 11 argument leaves two digits to manufacture Z=ee=0Z=e-e=0 and then searches for a 9-digit 1010-kernel. We can push that architecture one step further by proving a 7-digit kernel.

Lemma 14 (Unit pair from 5 digits when 3,93,9 are absent). If E[9]5E \subset [9]^5 and 3,9E3,9 \notin E, then EE contains a unit-producing pair.

Proof of Lemma 14.

If some digit repeats, we have an equal pair (i,i)(i,i) and therefore a unit pair.

So assume all five digits are distinct. If two are adjacent, we are done. Hence assume no two are adjacent. Then the support is an independent 5-set in the path 129. 1-2-\cdots-9. But the unique independent 5-set in this path is {1,3,5,7,9}. \{1,3,5,7,9\}. Since 3,9E3,9 \notin E, this is impossible. Therefore EE must contain a unit pair.

Lemma 15 (A 10-kernel from 7 digits). For every multiset D[9]7D \subset [9]^7, there exists a valid expression using some of the digits of DD (each at most once) that evaluates to 1010.

For the case analysis below, it is convenient to fix a small kernel catalog: K1=(1+1+1)2+1,K2=222+2,K4=4+4+4+44,K6=(6+66)2+6,K7=(7+7+77)2+77,K8=8+8+88. \begin{aligned} K_1&=(1+1+1)^2+1,\\ K_2&=2\cdot2\cdot2+2,\\ K_4&=4+4+\frac{4+4}{4},\\ K_6&=\left(\frac{6+6}{6}\right)^2+6,\\ K_7&=\left(\frac{7+7+7}{7}\right)^2+\frac77,\\ K_8&=8+\frac{8+8}{8}. \end{aligned} and the mixed kernels: 4+4+2=10,8+84=10,6+22=10,(2+1)2+1=10,6+(51)=10,6+(75)2=10,7+1+1+1=10,7+5(1+1)=10,1+(6+6+66)2=10,1+(7+7+77)2=10,7+6+6+66=10,8+7+77=10,(86)2+6=10,6+67+77=10,7+(76)2+(1+1)=10. \begin{aligned} 4+4+2&=10,\qquad 8+\frac84=10,\qquad 6+2^2=10,\qquad (2+1)^2+1=10,\\ 6+(5-1)&=10,\qquad 6+(7-5)^2=10,\qquad 7+1+1+1=10,\qquad 7+5-(1+1)=10,\\ 1+\left(\frac{6+6+6}{6}\right)^2&=10,\qquad 1+\left(\frac{7+7+7}{7}\right)^2=10,\qquad 7+\frac{6+6+6}{6}=10,\\ 8+\frac{7+7}{7}&=10,\qquad (8-6)^2+6=10,\qquad 6+6-\frac{7+7}{7}=10,\qquad 7+(7-6)^2+(1+1)=10. \end{aligned} We will explicitly track multiplicity thresholds against the number of digit occurrences each displayed kernel consumes.

Lemma 16 (Family kernel lemma 1\star_1: support inside {2,4}\{2,4\}). If D[9]7D \subset [9]^7 and supp(D){2,4}\operatorname{supp}(D)\subseteq \{2,4\}, then DD contains a 1010-kernel.

Proof of Lemma 16.

Let (c2,c4)(c_2,c_4) be multiplicities with c2+c4=7c_2+c_4=7.

  • If c24c_2\ge 4, use K2K_2.
  • Else c23c_2\le 3, so c44c_4\ge 4.
  • If c2=0c_2=0, then c4=7c_4=7; use K4K_4.
  • If c21c_2\ge 1, then c42c_4\ge 2; use 4+4+2=104+4+2=10.

Lemma 17 (Family kernel lemma 2\star_2: support inside {4,8}\{4,8\}). If D[9]7D \subset [9]^7 and supp(D){4,8}\operatorname{supp}(D)\subseteq \{4,8\}, then DD contains a 1010-kernel.

Proof of Lemma 17.

Let (c4,c8)(c_4,c_8) be multiplicities with c4+c8=7c_4+c_8=7.

  • If c82c_8\ge 2 and c41c_4\ge 1, use 8+84=108+\frac84=10.
  • Else if c81c_8\le 1, then c46c_4\ge 6; use K4K_4.
  • Else c82c_8\ge 2 and c4=0c_4=0, hence c8=7c_8=7; use K8K_8.

Lemma 18 (Family kernel lemma 3\star_3: support inside {1,2,6}\{1,2,6\}). If D[9]7D \subset [9]^7 and supp(D){1,2,6}\operatorname{supp}(D)\subseteq \{1,2,6\}, then DD contains a 1010-kernel.

Proof of Lemma 18.

Let (a,b,c)(a,b,c) be multiplicities of (1,2,6)(1,2,6) with a+b+c=7a+b+c=7.

  • If a4a\ge 4, use K1K_1.
  • Else if b4b\ge 4, use K2K_2.
  • Else if b2b\ge 2 and c1c\ge 1, use 6+22=106+2^2=10.
  • Else if b=1b=1 and a2a\ge 2, use (2+1)2+1=10(2+1)^2+1=10.
  • Else if a1a\ge 1 and c4c\ge 4, use 1+(6+6+66)2=101+\left(\frac{6+6+6}{6}\right)^2=10.
  • Otherwise c6c\ge 6, so use K6K_6.

Lemma 19 (Family kernel lemma 4\star_4: support inside {6,7,8}\{6,7,8\}). If D[9]7D \subset [9]^7 and supp(D){6,7,8}\operatorname{supp}(D)\subseteq \{6,7,8\}, then DD contains a 1010-kernel.

Proof of Lemma 19.

Let (c6,c7,c8)(c_6,c_7,c_8) be multiplicities with c6+c7+c8=7c_6+c_7+c_8=7.

If c81c_8\ge 1:

  • If c84c_8\ge 4, use K8K_8.
  • Else if c73c_7\ge 3, use 8+7+77=108+\frac{7+7}{7}=10.
  • Else c72c_7\le 2, so c62c_6\ge 2; use (86)2+6=10(8-6)^2+6=10.

If c8=0c_8=0, then c6+c7=7c_6+c_7=7:

  • If c76c_7\ge 6, use K7K_7.
  • Else if c73c_7\ge 3, then c62c_6\ge 2; use 6+67+77=106+6-\frac{7+7}{7}=10.
  • Else if c71c_7\ge 1, then c64c_6\ge 4; use 7+6+6+66=107+\frac{6+6+6}{6}=10.
  • Else c6=7c_6=7; use K6K_6.

Lemma 20 (Family kernel lemma 5\star_5: support inside {1,5,6,7}\{1,5,6,7\}). If D[9]7D \subset [9]^7 and supp(D){1,5,6,7}\operatorname{supp}(D)\subseteq \{1,5,6,7\}, then DD contains a 1010-kernel.

Proof of Lemma 20.

Let (a,b,c,d)(a,b,c,d) be multiplicities of (1,5,6,7)(1,5,6,7) with a+b+c+d=7. a+b+c+d=7.

If b2b\ge 2, use 5+5=105+5=10. So assume b1b\le 1.

Case 1: b=1b=1.

  • If d5d\ge 5, use K7K_7.
  • Else if a1a\ge 1 and c1c\ge 1, use 6+(51)=106+(5-1)=10.
  • Else if c1c\ge 1 and d1d\ge 1, use 6+(75)2=106+(7-5)^2=10.
  • Else if a2a\ge 2 and d1d\ge 1, use 7+5(1+1)=107+5-(1+1)=10.
  • The only remaining possibilities are pure supports {1,5}\{1,5\} or {5,6}\{5,6\}, so use K1K_1 or K6K_6.

Case 2: b=0b=0. Then a+c+d=7a+c+d=7.

  • If a4a\ge 4, use K1K_1.
  • Else if d6d\ge 6, use K7K_7.
  • Else if a3a\ge 3 and d1d\ge 1, use 7+1+1+1=107+1+1+1=10.
  • Else if a1a\ge 1 and c4c\ge 4, use 1+(6+6+66)2=101+\left(\frac{6+6+6}{6}\right)^2=10.
  • Else if a1a\ge 1 and d4d\ge 4, use 1+(7+7+77)2=101+\left(\frac{7+7+7}{7}\right)^2=10.
  • Else if c4c\ge 4 and d1d\ge 1, use 7+6+6+66=107+\frac{6+6+6}{6}=10.
  • Else if c2c\ge 2 and d3d\ge 3, use 6+67+77=106+6-\frac{7+7}{7}=10.
  • Else if a2a\ge 2, c1c\ge 1, and d2d\ge 2, use 7+(76)2+(1+1)=107+(7-6)^2+(1+1)=10.
  • Otherwise c=7c=7, so use K6K_6.

Proof of Lemma 15.

As in the k11k \le 11 section, we use the identity 10=9+110=9+1 and seek a disjoint nine-gadget and unit pair.

First remove immediate 2-leaf kernels:

  • no complement-to-10 pair in (1,9),(2,8),(3,7),(4,6)(1,9),(2,8),(3,7),(4,6) occurs,
  • and c51c_5 \le 1 (so no 5+5=105+5=10 kernel).

If 3D3 \in D or 9D9 \in D, we get a 1-leaf nine-gadget. Removing that one digit leaves 6 digits, and Lemma 10 yields a disjoint unit pair. So in the reduced case: 3Dand9D. 3 \notin D \quad \text{and} \quad 9 \notin D.

Now consider 2-leaf nine-gadgets. If DD contains any of (1,8),(2,7),(4,5),(1,4),(4,7),(2,5),(5,8), (1,8),(2,7),(4,5),(1,4),(4,7),(2,5),(5,8), then removing that gadget leaves 5 digits. By 3,9D3,9 \notin D and Lemma 14, those 5 digits contain a disjoint unit pair. Hence 10=9+110=9+1.

Therefore, in the hard case we may assume all those pairs are absent. Equivalently, the support is an independent set in the forbidden graph HH on vertices {1,2,4,5,6,7,8} \{1,2,4,5,6,7,8\} with forbidden edges 14,47,25,58,18,27,45,28,46. 14,47,25,58,18,27,45,28,46. The maximal independent sets of HH are: {2,4}, {4,8}, {1,2,6}, {6,7,8}, {1,5,6,7}. \{2,4\},\ \{4,8\},\ \{1,2,6\},\ \{6,7,8\},\ \{1,5,6,7\}. Define M1={2,4},M2={4,8},M3={1,2,6},M4={6,7,8},M5={1,5,6,7}. M_1=\{2,4\},\quad M_2=\{4,8\},\quad M_3=\{1,2,6\},\quad M_4=\{6,7,8\},\quad M_5=\{1,5,6,7\}. Every reduced support SS is an independent set of HH, hence SMiS\subseteq M_i for some i{1,,5}i\in\{1,\dots,5\}. Therefore it is enough to prove, for each ii, the stronger family statement: (i):every 7-digit multiset D with supp(D)Mi contains a 10-kernel. (\star_i):\quad \text{every 7-digit multiset } D \text{ with } \operatorname{supp}(D)\subseteq M_i \text{ contains a 10-kernel.} These are exactly lemmas Lemma 16 through Lemma 20. Hence the hard case is solved, and every 7-digit multiset admits a 1010-kernel.

We may now prove our original claim.

Proof of Claim 13.

Let k9k \ge 9 and D[9]kD \subset [9]^k.

If some digit repeats, reserve two copies (e,e)(e,e) and set Zee=0. Z \doteq e-e=0. Removing those two digits leaves k27k-2 \ge 7 digits; pick any 7 of them and apply Lemma 15 to obtain a 1010-kernel TT. Let RR be any expression using all remaining unused digits exactly once. Then T+ZR=10 T + Z\cdot R = 10 uses every digit exactly once.

If no digit repeats and D=9\left| D \right|=9, then necessarily D={1,2,3,4,5,6,7,8,9}, D=\{1,2,3,4,5,6,7,8,9\}, and we have an explicit full-use expression: 1+2+3+4+5(98(76)2)=10. 1+2+3+4+5\bigl(9-8-(7-6)^2\bigr)=10.

Hence every 9-digit multiset is solvable, so k9k \le 9.

k5k \le 5 (Mar 07)

Theorem 21 (Exact threshold). Every multiset D[9]5D \in [9]^5 admits a valid expression (using each digit exactly once and operations in MM) that evaluates to 1010.

We will want the below lemma.

Lemma 22 (Triple lemma). For every multiset T[9]3T \in [9]^3, there is a valid expression using exactly the three digits of TT that evaluates to either 55 or 88.

Proof of Lemma 22.

We split by support size.

Case 1: one-value support

For T={d,d,d}T=\{d,d,d\}, the following witnesses work: {1,1,1}: (1+1)2+1=5,{2,2,2}: 22+22=5,{3,3,3}: 3233=8,{4,4,4}: 4+44=5,{5,5,5}: 555=5,{6,6,6}: 666=5,{7,7,7}: 7+77=8,{8,8,8}: 888=8,{9,9,9}: 999=8. \begin{aligned} \{1,1,1\}&:\ (1+1)^2+1=5,\\ \{2,2,2\}&:\ 2^2+\frac22=5,\\ \{3,3,3\}&:\ 3^2-\frac33=8,\\ \{4,4,4\}&:\ 4+\frac44=5,\\ \{5,5,5\}&:\ 5\cdot\frac55=5,\\ \{6,6,6\}&:\ 6-\frac66=5,\\ \{7,7,7\}&:\ 7+\frac77=8,\\ \{8,8,8\}&:\ 8\cdot\frac88=8,\\ \{9,9,9\}&:\ 9-\frac99=8. \end{aligned}

Case 2: two-value support

Write T={x,x,y}T=\{x,x,y\} with xyx\ne y. Since xx=1\frac xx=1, we can form y1y-1, yy, and y+1y+1. Hence if y{4,5,6,7,8,9}y\in\{4,5,6,7,8,9\}, one of these is 55 or 88.

So only y{1,2,3}y\in\{1,2,3\} remains. Exhaustive witnesses: {1,1,1}: (1+1)2+1=5,{2,2,1}: 2+2+1=5,{3,3,1}: 321=8,{4,4,1}: 1+44=5,{5,5,1}: 551=5,{6,6,1}: 661=5,{7,7,1}: 1+77=8,{8,8,1}: 1+8+8=5,{9,9,1}: 991=5,{1,1,2}: 1+(21)2=5,{2,2,2}: 22+22=5,{3,3,2}: 2+33=5,{4,4,2}: 22+44=5,{5,5,2}: 5+52=5,{6,6,2}: 22+66=5,{7,7,2}: 772=5,{8,8,2}: 22+88=5,{9,9,2}: 22+99=5,{1,1,3}: 1+1+3=5,{2,2,3}: 3222=5,{3,3,3}: 3233=8,{4,4,3}: 4+43=5,{5,5,3}: 3+55=8,{6,6,3}: 3266=8,{7,7,3}: 7+732=5,{8,8,3}: 883=5,{9,9,3}: 3299=8. \begin{aligned} \{1,1,1\}&:\ (1+1)^2+1=5, & \{2,2,1\}&:\ 2+2+1=5, & \{3,3,1\}&:\ 3^2-1=8,\\ \{4,4,1\}&:\ 1+\sqrt{4\cdot4}=5, & \{5,5,1\}&:\ \sqrt{\frac{5\cdot5}{1}}=5, & \{6,6,1\}&:\ \sqrt{6\cdot6}-1=5,\\ \{7,7,1\}&:\ 1+\sqrt{7\cdot7}=8, & \{8,8,1\}&:\ 1+\sqrt{8+8}=5, & \{9,9,1\}&:\ 9-\sqrt9-1=5,\\ \{1,1,2\}&:\ 1+\left(\frac21\right)^2=5, & \{2,2,2\}&:\ 2^2+\frac22=5, & \{3,3,2\}&:\ 2+\sqrt{3\cdot3}=5,\\ \{4,4,2\}&:\ 2^2+\frac44=5, & \{5,5,2\}&:\ \frac{5+5}{2}=5, & \{6,6,2\}&:\ 2^2+\frac66=5,\\ \{7,7,2\}&:\ \sqrt{7\cdot7}-2=5, & \{8,8,2\}&:\ 2^2+\frac88=5, & \{9,9,2\}&:\ 2^2+\frac99=5,\\ \{1,1,3\}&:\ 1+1+3=5, & \{2,2,3\}&:\ 3^2-2-2=5, & \{3,3,3\}&:\ 3^2-\frac33=8,\\ \{4,4,3\}&:\ 4+4-3=5, & \{5,5,3\}&:\ 3+\sqrt{5\cdot5}=8, & \{6,6,3\}&:\ 3^2-\frac66=8,\\ \{7,7,3\}&:\ 7+7-3^2=5, & \{8,8,3\}&:\ \sqrt{8\cdot8}-3=5, & \{9,9,3\}&:\ 3^2-\frac99=8. \end{aligned}

Case 3: three-value support with a consecutive pair

Let a<b<ca\lt b\lt c be distinct. If b=a+1b=a+1, then ba=1b-a=1, so we can make c1,c,c+1c-1,c,c+1; if c=b+1c=b+1, then cb=1c-b=1, so we can make a1,a,a+1a-1,a,a+1. Thus only finitely many edge cases remain; witnesses: {1,2,3}: 2+31=5,{1,3,4}: 3+41=5,{1,4,5}: (1+4)5=5,{1,5,6}: (61)5=5,{1,6,7}: 6217=5,{1,7,8}: (1+7)8=8,{1,8,9}: 891=5,{2,4,5}: 2+4+5=5,{2,5,6}: 22(56)=5,{2,6,7}: 22(67)=5,{2,7,8}: 22(78)=5,{2,8,9}: 22(89)=5,{3,5,6}: 5+63=8,{3,6,7}: 36+7=5,{3,7,8}: 7+83=5,{3,8,9}: 893=5. \begin{aligned} \{1,2,3\}&:\ \frac{2+3}{1}=5, & \{1,3,4\}&:\ \frac{3+\sqrt4}{1}=5, & \{1,4,5\}&:\ \sqrt{(1+4)\cdot5}=5,\\ \{1,5,6\}&:\ \sqrt{(6-1)\cdot5}=5, & \{1,6,7\}&:\ \frac{6^2-1}{7}=5, & \{1,7,8\}&:\ \sqrt{(1+7)\cdot8}=8,\\ \{1,8,9\}&:\ \frac{8-\sqrt9}{1}=5, & \{2,4,5\}&:\ 2+\sqrt{4+5}=5, & \{2,5,6\}&:\ 2^2-(5-6)=5,\\ \{2,6,7\}&:\ 2^2-(6-7)=5, & \{2,7,8\}&:\ 2^2-(7-8)=5, & \{2,8,9\}&:\ 2^2-(8-9)=5,\\ \{3,5,6\}&:\ 5+6-3=8, & \{3,6,7\}&:\ \sqrt{3\cdot6+7}=5, & \{3,7,8\}&:\ \frac{7+8}{3}=5,\\ \{3,8,9\}&:\ 8-\frac93=5. \end{aligned}

Case 4: three-value support with no consecutive pair

The sparse triples (3535 total) are exhausted by the following witnesses: {1,3,5}: 1+(53)2=5,{1,3,6}: 1+(63)2=5,{1,3,7}: 1(37)=5,{1,3,8}: 831=5,{1,3,9}: 931=5,{1,4,6}: 1+46=5,{1,4,7}: 741=5,{1,4,8}: 1(48)=5,{1,4,9}: 941=5,{1,5,7}: 1+(57)2=5,{1,5,8}: (58)21=8,{1,5,9}: 1(59)=5,{1,6,8}: 1+(68)2=5,{1,6,9}: 1+(69)2=5,{1,7,9}: 1+7+9=5,{2,4,6}: 4+62=5,{2,4,7}: 2(47)=5,{2,4,8}: 4+82=5,{2,4,9}: 4+92=5,{2,5,7}: (72)5=5,{2,5,8}: 2(58)=5,{2,5,9}: (2+9)5=5,{2,6,8}: 62+822=5,{2,6,9}: 2(69)=5,{2,7,9}: 7+92=5,{3,5,7}: 3(57)=5,{3,5,8}: (83)5=5,{3,5,9}: 3+95=5,{3,6,8}: 3(68)=5,{3,6,9}: 6+93=5,{3,7,9}: 3(79)=5,{4,6,8}: 62+824=5,{4,6,9}: (69)24=5,{4,7,9}: 4+79=5,{5,7,9}: 7+95=5. \begin{aligned} \{1,3,5\}&:\ 1+(5-3)^2=5, & \{1,3,6\}&:\ 1+\left(\frac63\right)^2=5, & \{1,3,7\}&:\ 1-(3-7)=5,\\ \{1,3,8\}&:\ \frac{8-3}{1}=5, & \{1,3,9\}&:\ 9-3-1=5, & \{1,4,6\}&:\ \sqrt{1+4\cdot6}=5,\\ \{1,4,7\}&:\ \frac{7-\sqrt4}{1}=5, & \{1,4,8\}&:\ 1-(4-8)=5, & \{1,4,9\}&:\ \frac{9-4}{1}=5,\\ \{1,5,7\}&:\ 1+(5-7)^2=5, & \{1,5,8\}&:\ (5-8)^2-1=8, & \{1,5,9\}&:\ 1-(5-9)=5,\\ \{1,6,8\}&:\ 1+(6-8)^2=5, & \{1,6,9\}&:\ 1+\left(\frac6{\sqrt9}\right)^2=5, & \{1,7,9\}&:\ 1+\sqrt{7+9}=5,\\ \{2,4,6\}&:\ \frac{4+6}{2}=5, & \{2,4,7\}&:\ 2-(4-7)=5, & \{2,4,8\}&:\ \frac{\sqrt4+8}{2}=5,\\ \{2,4,9\}&:\ 4+\sqrt9-2=5, & \{2,5,7\}&:\ \sqrt{(7-2)\cdot5}=5, & \{2,5,8\}&:\ 2-(5-8)=5,\\ \{2,5,9\}&:\ \sqrt{(2+\sqrt9)\cdot5}=5, & \{2,6,8\}&:\ \frac{\sqrt{6^2+8^2}}{2}=5, & \{2,6,9\}&:\ 2-(6-9)=5,\\ \{2,7,9\}&:\ \frac{7+\sqrt9}{2}=5, & \{3,5,7\}&:\ 3-(5-7)=5, & \{3,5,8\}&:\ \sqrt{(8-3)\cdot5}=5,\\ \{3,5,9\}&:\ 3+\sqrt{9-5}=5, & \{3,6,8\}&:\ 3-(6-8)=5, & \{3,6,9\}&:\ \frac{6+9}{3}=5,\\ \{3,7,9\}&:\ 3-(7-9)=5, & \{4,6,8\}&:\ \sqrt{\frac{6^2+8^2}{4}}=5, & \{4,6,9\}&:\ (6-9)^2-4=5,\\ \{4,7,9\}&:\ \sqrt{4+7\sqrt9}=5, & \{5,7,9\}&:\ 7+\sqrt9-5=5. \end{aligned}

All 3-digit multisets are covered, proving the lemma.

Lemma 23 (Pair graph for value 22). Let G2G_2 be the graph on vertices {1,,9}\{1,\dots,9\} where abab is an edge iff pair {a,b}\{a,b\} can make 22, and where loop aaaa exists iff {a,a}\{a,a\} can make 22. Then the available loops and edges are exactly: loops: 11,22,44,88; \text{loops: }11,22,44,88; edges: 12,13,14,15,19,24,26,28,35,36,37,46,48,57,59,68,69,79. \text{edges: }12,13,14,15,19,24,26,28,35,36,37,46,48,57,59,68,69,79.

Proof of Lemma 23.

Witnesses: 11: 1+1,22: 2+2,44: 44,88: 8+8,12: 12,13: 31,14: 4,15: 51,19: 91,24: 4/2,26: 622,28: 8/22,35: 53,36: 6/3,37: 327,46: 64,48: 8/4,57: 75,59: 59,68: 86,69: 6/9,79: 97. \begin{aligned} 11&:\ 1+1,\quad 22:\ \sqrt{2+2},\quad 44:\ \frac4{\sqrt4},\quad 88:\ \sqrt{\sqrt{8+8}},\\ 12&:\ 1\cdot2,\quad 13:\ 3-1,\quad 14:\ \sqrt4,\quad 15:\ \sqrt{5-1},\quad 19:\ \sqrt9-1,\\ 24&:\ 4/2,\quad 26:\ 6-2^2,\quad 28:\ 8/2^2,\quad 35:\ 5-3,\quad 36:\ 6/3,\quad 37:\ 3^2-7,\\ 46&:\ 6-4,\quad 48:\ 8/4,\quad 57:\ 7-5,\quad 59:\ 5-\sqrt9,\quad 68:\ 8-6,\quad 69:\ 6/\sqrt9,\quad 79:\ 9-7. \end{aligned}

Proof of Theorem 21.

Let D[9]5D\in[9]^5. If DD contains a G2G_2-pair, use that pair to build 22. The complementary triple builds 55 or 88 by Lemma 22, so we get 25=10or2+8=10. 2\cdot 5=10 \quad \text{or} \quad 2+8=10. Therefore any counterexample must contain no G2G_2-pair at all.

So we only need to solve 5-multisets independent in G2G_2. The maximal independent supports are: {2,5}, {2,7}, {4,5}, {4,7}, {5,6}, {5,8}, \{2,5\},\ \{2,7\},\ \{4,5\},\ \{4,7\},\ \{5,6\},\ \{5,8\}, {1,6,7}, {1,7,8}, {2,3,9}, {3,4,9}, {3,8,9}. \{1,6,7\},\ \{1,7,8\},\ \{2,3,9\},\ \{3,4,9\},\ \{3,8,9\}. These are obtained by exhaustive verification over all supports in {1,,9}\{1,\dots,9\}. Because 1,2,4,81,2,4,8 have loops, each can appear at most once inside these candidates.

It remains to list one witness per admissible 5-multiset family:

Family {2,5}\{2,5\}

25555: 5+5+5+52=10. 25555:\ \frac{5+5+5+5}{2}=10.

Family {2,7}\{2,7\}

27777: 2(77+77)=10. 27777:\ 2\left(7-\frac{7+7}{7}\right)=10.

Family {4,5}\{4,5\}

45555: 4555+5=10. 45555:\ 4\cdot\frac{5\cdot5}{5+5}=10.

Family {4,7}\{4,7\}

47777: 7+7+774=10. 47777:\ 7+7+7-7-4=10.

Family {5,6}\{5,6\}

55555: 5(5+5+5+5)=10, 55555:\ \sqrt{5(5+5+5+5)}=10, 55556: 555+(5+6)=10, 55556:\ \sqrt{5\cdot5\cdot\sqrt{5+(5+6)}}=10, 55566: 5((6+6)55)=10, 55566:\ 5\bigl((6+6)-5-5\bigr)=10, 55666: 525(6/(6+6))=10, 55666:\ \frac{5^2}{5\cdot\bigl(6/(6+6)\bigr)}=10, 56666: 566+66=10, 56666:\ 5\sqrt{6-\frac{6+6}{6}}=10, 66666: 6+(66+66)=10. 66666:\ 6+\left(6-\frac{6+6}{6}\right)=10.

Family {5,8}\{5,8\}

55558: 55285+5=10. 55558:\ \sqrt{\frac{5\cdot5^2\cdot8}{5+5}}=10.

Family {1,6,7}\{1,6,7\}

16666: 1+(6+6+66)2=10, 16666:\ 1+\left(\frac{6+6+6}{6}\right)^2=10, 16667: 6+6+671=10, 16667:\ 6+6+6-7-1=10, 16677: 6+(677)1=10, 16677:\ 6+\left(6-\frac77\right)-1=10, 16777: 1+(677+7)2=10, 16777:\ 1+\left(\frac{6\cdot7}{7+7}\right)^2=10, 17777: 1+(7+7+77)2=10. 17777:\ 1+\left(\frac{7+7+7}{7}\right)^2=10. 66667: 6+6+66+7=10, 66667:\ \frac{6+6+6}{6}+7=10, 66677: 7(6+66)2+7=10, 66677:\ 7-\left(\frac{6+6}{6}\right)^2+7=10, 66777: (6+6)7+77=10, 66777:\ (6+6)-\frac{7+7}{7}=10, 67777: 7(7+7)/6+7=10, 67777:\ \frac{7}{(7+7)/6}+7=10, 77777: 7+7+77+7=10. 77777:\ \frac{7+7+7}{7}+7=10.

Family {1,7,8}\{1,7,8\}

17778: 1+7+787=10. 17778:\ 1+\frac{7+7\cdot8}{7}=10. 77778: 77+77+8=10. 77778:\ \frac77+\frac77+8=10.

Family {2,3,9}\{2,3,9\}

33339: 3+333+9=10, 33339:\ \frac{3+3-3}{3}+9=10, 33399: 3+3+39+9=10, 33399:\ \frac{3+3+3}{9}+9=10, 33999: 9(33)99=10, 33999:\ 9-\frac{(3-3)-9}{9}=10, 39999: 39+99+9=10, 39999:\ 3-\frac{9+9}{9}+9=10, 23333: 3+3+3+32=10, 23333:\ 3+3+3+3-2=10, 23339: 3+3+932=10, 23339:\ 3+3+9-3-2=10, 23399: 9+9332=10, 23399:\ 9+9-3-3-2=10, 23999: 3(9(9+9))2=10, 23999:\ 3-\bigl(9-(9+9)\bigr)-2=10, 29999: 9+9+992=10. 29999:\ 9+\sqrt{9+9-9}-2=10.

Family {3,4,9}\{3,4,9\}

33333: 3+3+(3+33)2=10, 33333:\ 3+3+\left(\frac{3+3}{3}\right)^2=10, 33334: 3+3+3+43=10, 33334:\ 3+3+3+4-3=10, 33349: 3+4+933=10, 33349:\ 3+4+9-3-3=10, 33499: 3+3+4+99=10, 33499:\ 3+3+4+9-9=10, 34999: 4(9(9+9))3=10, 34999:\ 4-\bigl(9-(9+9)\bigr)-3=10, 49999: 4+9+9+9+9=10. 49999:\ 4+\sqrt{9+9+9+9}=10.

Family {3,8,9}\{3,8,9\}

33338: 3(3+8)33=10, 33338:\ \frac{3(3+8)-3}{3}=10, 33389: 3+3+(3+89)2=10, 33389:\ 3+3+(3+8-9)^2=10, 33899: 3(3+8(9+9))=10, 33899:\ 3-\bigl(3+8-(9+9)\bigr)=10, 38999: 9+9+9832=10, 38999:\ 9+9+9-8-3^2=10, 89999: 9+9+998=10, 89999:\ 9+9+9-9-8=10, 99999: 999(9+9)=10. 99999:\ 9-\frac{9}{9-(9+9)}=10.

Thus every G2G_2-independent 5-multiset is solvable, and every non-independent one is solvable by the pair-plus-triple reduction above. Hence every 5-digit multiset is solvable. Together with Claim 3, this gives k=5k=5.

  1. We exclude 0 by convention, it makes the problem unsolvable; given all 0s we can't ever reach 10.
  2. Square root is only defined for perfect integer squares, i.e., 1, 4, 9.
  3. I ended up using a Python program for enumerating some of the casework (especially for k5k \leq 5), but most of this was done without computer assistance.
  4. This turned out to not be nearly as useful as I had hoped, most of the progress came from casting it as a graph problem.