Q The Query Complexity Zoo
\(\mathrm{ApproxIndex}_k\)

Approximate index

The outer function that refutes the randomized composition conjecture: a biased guess at the index is good enough, which is cheap noisily and expensive classically.

partial separating arity: $n = k + 2^k$ bits (a $k$-bit address and a $2^k$-cell array over $\{0,1,2\}$, binary-encoded)

Description

The input is an address $a \in \{0,1\}^k$ together with an array $x \in \{0,1,2\}^{2^k}$ indexed by $\{0,1\}^k$. The promise: every cell $b$ within Hamming distance $k/2 - 2\sqrt{k\log k}$ of $a$ holds one and the same bit, and every other cell holds $2$. Output that bit.

The point is the slack. You never have to find $a$ -- any $b$ in a ball of radius just barely under $k/2$ around it reads off the answer -- and a ball of that radius catches a string built by guessing each bit of $a$ with only a small bias. Bias is cheap in the noisy model and expensive in the query model, and that difference is the whole counterexample. The alphabet $\{0,1,2\}$ costs only a constant factor after binary encoding, so this is treated as a partial Boolean function.

Formal definition

$\mathrm{ApproxIndex}(a,x) = x_a$ if $x_b = x_a \in \{0,1\}$ for every $b$ with $|b-a| \le k/2 - 2\sqrt{k\log k}$ and $x_b = 2$ for every other $b$; undefined otherwise.

Measure values

MeasureValueNotes
\(\mathrm{R}_{\varepsilon}(f)\) Bounded-error randomized query complexity \(\Theta(\sqrt{k\log k}) = \Theta(\sqrt{\log n\,\log\log n})\) Lemma 27. Upper: read the first $8\sqrt{k\log k}$ bits of $a$, fill the rest of $b$ with fair coins, query $x_b$. Lower: an algorithm making $\sqrt{k\log k}$ queries can be assumed never to query the array (pretend the answer is $2$), and then it must still name a small set of candidate addresses one of which lands in the ball -- i.e. guess $k/2 + \sqrt{k\log k}$ of the unread bits of a uniformly random string correctly, which Chernoff forbids. BenDavidBlais20composition
\(\mathrm{D}(f)\) Deterministic query complexity \(\Theta(\sqrt{k\log k})\) putative So this function is not a $\mathrm{D}$-vs-$\mathrm{R}$ separation, which is worth knowing given how hard its $\mathrm{R}$ lower bound works. Lower: $\mathrm{D} \ge \mathrm{R}$. Upper: read the first $q = 4\sqrt{k\log k}$ bits of $a$ and let $b$ agree with them and be $0$ elsewhere, $b'$ agree with them and be $1$ elsewhere. Then $|b-a| + |b'-a| = k-q$, so one of the two is within $(k-q)/2 = k/2 - 2\sqrt{k\log k}$ of $a$ -- inside the ball. Query $x_b$; if it is $2$, query $x_{b'}$, which cannot be. Total $4\sqrt{k\log k}+O(1)$ queries, deterministically. Randomness buys nothing here: it only replaces "read $4\sqrt{k\log k}$ bits and use antipodal guesses" with "read $8\sqrt{k\log k}$ bits and guess at random".
\(\mathrm{R}_{0}(f)\) Zero-error randomized query complexity \(\Theta(\sqrt{k\log k})\) Sandwiched: $\mathrm{R} \le \mathrm{R}_0 \le \mathrm{D}$, and the two ends agree.
\(\mathrm{noisyR}(f)\) Noisy randomized query complexity \(O(\log k) = O(\log\log n)\) Not stated as a value in the source, but it is exactly what its two theorems say. Their composed algorithm gives $\mathrm{R}(\mathrm{ApproxIndex}_k \circ \mathrm{GapMaj}_k) = O(k\log k)$, and Theorem 3 says $\mathrm{R}(f \circ g) = \Omega(\mathrm{noisyR}(f)\,\mathrm{R}(g))$ with $\mathrm{R}(\mathrm{GapMaj}_k) = \Theta(k)$; divide. Directly, in the noisy model: buy each of the $k$ address bits at bias $\gamma = c\sqrt{\log k/k}$ for $\gamma^2$ each, total $c^2\log k$, which puts the guess inside the ball with high probability, then read one array cell at bias $1$. BenDavidBlais20composition
\(\bar{\mathrm{R}}_{\varepsilon}(f)\) Expected-cost randomized query complexity \(\Theta(\sqrt{k\log k}) = \Theta(\sqrt{\log n\,\log\log n})\) $\bar{\mathrm{R}}_{\varepsilon}(f) = \Theta(\mathrm{R}_{\varepsilon}(f))$ for every $f$ (\(\bar{\mathrm{R}}_{\varepsilon}(f)\) vs \(\mathrm{R}_{\varepsilon}(f)\)), and $\mathrm{R}_{\varepsilon}(f) = \Theta(\sqrt{k\log k}) = \Theta(\sqrt{\log n\,\log\log n})$ here.
\(\mathrm{C}(f)\) Certificate complexity \(\Theta(1)\) putative One array cell. By the promise, a cell holding a value in $\{0,1\}$ must lie in the ball around $a$, so reading any non-$2$ cell both locates you inside the ball and reads off the answer -- two bits after the binary encoding. So $\mathrm{C} = \Theta(1)$ against $\mathrm{D} = \Theta(\sqrt{k\log k})$: the difficulty is entirely in finding a good cell, never in recognising one.
\(\mathrm{C}^0(f)\) 0-certificate complexity \(\Theta(1)\) putative See $\mathrm{C}$; the two sides are symmetric.
\(\mathrm{C}^1(f)\) 1-certificate complexity \(\Theta(1)\) putative See $\mathrm{C}$; the two sides are symmetric.
\(\mathrm{C}_{\min}(f)\) Minimum certificate complexity \(\Theta(1)\) putative See $\mathrm{C}$: the same certificate works at every input, so $\mathrm{C}_{\min} = \mathrm{C} = \Theta(1)$.
\(\mathrm{s}(f)\) Sensitivity \(0\) putative No single flip stays inside the promise. Flipping a bit of the address $a$ moves the ball, leaving cells inside the new ball holding $2$; flipping a bit of an array cell either breaks the "$v$ inside the ball" condition or the "$2$ outside" one. So $\mathrm{s}^0 = \mathrm{s}^1 = 0$.
\(\mathrm{s}^0(f)\) 0-sensitivity \(0\) putative See $\mathrm{s}$.
\(\mathrm{s}^1(f)\) 1-sensitivity \(0\) putative See $\mathrm{s}$.
\(\mathrm{bs}(f)\) Block sensitivity \(1\) putative Exactly one, so no two sensitive blocks are disjoint. After any answer-flipping change every non-$2$ cell holds $1-v$, whereas at present every one of the $|B(a,r)|$ cells of the ball holds $v$; so every sensitive block contains the whole ball, and any two of them meet. There is at least one (rewrite everything), so $\mathrm{bs} = 1$.
\(\mathrm{bs}^0(f)\) 0-block sensitivity \(1\) putative See $\mathrm{bs}$.
\(\mathrm{bs}^1(f)\) 1-block sensitivity \(1\) putative See $\mathrm{bs}$.
\(\mathrm{fbs}(f)\) Fractional block sensitivity \(1\) putative The load constraint at any single cell of the ball already caps $\sum_B w_B$ at $1$, since every sensitive block contains that cell.
\(\mathrm{FC}(f)\) Fractional certificate complexity \(1\) Equal to $\mathrm{fbs}$ by LP duality.
\(\mathrm{RC}(f)\) Randomized certificate complexity \(\Theta(1)\) $\mathrm{RC} = \Theta(\mathrm{fbs}) = \Theta(1)$; also $\mathrm{RC} \le \mathrm{C} = \Theta(1)$.
\(\lambda(f)\) Spectral sensitivity \(0\) putative The sensitivity graph on the promise is empty (see $\mathrm{s} = 0$).
\(\mathrm{D}^{\mathrm{na}}(f)\) Non-adaptive deterministic query complexity \(\Theta(\sqrt{k\log k})\) putative Adaptivity is worth nothing: read the first $4\sqrt{k\log k}$ bits of $a$ and query both antipodal completions $x_b$ and $x_{b'}$ at once -- one of them is inside the ball. And $\mathrm{D}^{\mathrm{na}} \ge \mathrm{D}$. Note the lemma "$\mathrm{D}^{\mathrm{na}} = n$ for functions depending on all variables" does not apply here: with $\mathrm{s} = 0$, no single coordinate is relevant within the promise.
\(\mathrm{R}^{\mathrm{na}}(f)\) Non-adaptive randomized query complexity \(\Theta(\sqrt{k\log k})\) putative Same remark: the randomized algorithm reads a fixed prefix of $a$ and one array cell chosen from the coins alone, so it is non-adaptive; and $\mathrm{R}^{\mathrm{na}} \ge \mathrm{R}$.
\(\mathrm{D}^{\mu}_{\varepsilon}(f)\) Distributional query complexity \(\Theta(\sqrt{k\log k})\) By Yao, $\max_\mu \mathrm{D}^\mu = \mathrm{R}$.
\(\mathrm{D}^{\mathrm{prod}}(f)\) Product-distribution distributional complexity \(0\) putative A rectangle inside the promise is a single input. First the address is pinned: if inputs $(a,x)$ and $(a',x')$ with $a \ne a'$ both lay in the rectangle, so would $(a, x')$, and $x'$ marks the ball around $a'$, a different set of cells from the ball around $a$. With $a$ fixed, the array is pinned too: every one of the $\binom{k}{\le k/2 - 2\sqrt{k\log k}}$ cells in the ball must hold the same bit, so a rectangle with even one free array coordinate contains an input where two ball cells disagree. Hence $\mathrm{D}^\mu = 0$ for every product $\mu$, against $\mathrm{R} = \Theta(\sqrt{k\log k})$. See \(\mathrm{forr}\).
\(\mathrm{Q}_{\varepsilon}(f)\) Bounded-error quantum query complexity \(O(\sqrt{k\log k})\) At most $\mathrm{R}$. Nothing better is known and nothing better is needed: the point of the family is the $\mathrm{R}$-vs-$\mathrm{noisyR}$ gap, and quantum plays no role in it.
\(\mathrm{Adv}^{\pm}(f)\) General (negative-weights) adversary bound \(O(\sqrt{k\log k})\) $\mathrm{Adv}^{\pm} = \Theta(\mathrm{Q})$, hence at most $O(\sqrt{k\log k})$.
\(\gamma_2^{\alpha}(f)\) Approximate gamma_2 norm \(O(\sqrt{k\log k})\) $\gamma_2 = \Theta(\mathrm{Q})$.
\(\mathrm{MAdv}(f)\) Multiplicative adversary \(O(\sqrt{k\log k})\) Sandwiched between $\mathrm{Adv}^{\pm}$ and $\mathrm{Q}$.
\(\widetilde{\deg}_{\varepsilon}(f)\) Approximate degree \(O(\sqrt{k\log k})\) $\widetilde{\deg} \le 2\mathrm{Q} \le 2\mathrm{R} = O(\sqrt{k\log k})$. From below, only $\Omega(1)$, and closing that is a real question rather than an oversight -- note that $\deg = 2$ below is no help, because approximate degree is the one polynomial measure that insists on boundedness off the promise. What a bounded polynomial can do here has a clean reading. Suppose $c_b(a)$, $b \in \{0,1\}^k$, are polynomials in the address of degree $\le d$ with $c_b \ge 0$ and $\sum_b c_b(a) \le 1$ on the whole cube, and $\sum_{b \in B(a,r)} c_b(a) \ge 2/3$ for every $a$ -- a guessing scheme, a distribution over cells of low degree in $a$ that lands in the ball. Then $p(a,x) = \sum_b c_b(a)\big(\mathbb{1}[x_b=1] - \mathbb{1}[x_b=0]\big)$ has degree $d+2$, satisfies $|p| \le 1$ everywhere because the $c_b$ are nonnegative and sum to at most $1$, and equals $(2v-1)\sum_{b \in B(a,r)} c_b(a)$ on the promise; so $\widetilde{\deg} \le d+2$. Two schemes are available and neither beats the algorithm: reading a prefix of $8\sqrt{k\log k}$ address bits and guessing the rest uniformly has degree $8\sqrt{k\log k}$, and biasing every bit by $\gamma = c\sqrt{\log k/k}$ -- the \(\mathrm{noisyR}(f)\) strategy -- gives $c_b(a) = \prod_i (\tfrac12 + \gamma(1-2b_i)(1-2a_i))$, degree $k$, which is worse. Whether some cleverer nonnegative scheme, or a polynomial not of this shape at all, does better is open.
\(\mathrm{N}_{p}(f)\) Noisy decision tree complexity \(\Theta(\sqrt{k\log k})\) putative At least $\mathrm{R}$; and reading each of the $O(\sqrt{k\log k})$ address bits $O(1)$ times and majority-voting still lands inside the ball, so $O(\mathrm{R})$. Note the sharp contrast with \(\mathrm{noisyR}(f)\) $= O(\log k)$: $\mathrm{N}_p$ charges a fixed constant noise rate per read, while the noisy-query model lets the algorithm buy an arbitrarily weak bias at cost $\gamma^2$, and weak bias is exactly what this problem needs.
\(\mathrm{prt}_{\varepsilon}(f)\) Partition bound \(O(\sqrt{k\log k})\) From $\mathrm{prt}_\varepsilon \le 2\mathrm{R}_\varepsilon$.
\(\overline{\mathrm{prt}}_{\varepsilon}(f)\) Relaxed partition bound \(O(\sqrt{k\log k})\) From $\overline{\mathrm{prt}}_\varepsilon \le \mathrm{prt}_\varepsilon$.
\(\mathrm{CA}(f)\) Classical adversary bound \(O(\sqrt{k\log k})\) At most $\mathrm{R}$. Since $\mathrm{fbs} = 1$ here, the classical adversary is the only combinatorial measure left with any room, but no lower bound better than $\Omega(1)$ is available.
\(\mathrm{Adv}(f)\) Positive-weights adversary bound \(\Theta(1)\) The certificate barrier is decisive: $\mathrm{Adv} \le \sqrt{\mathrm{C}^0\mathrm{C}^1} = \Theta(1)$, an inequality that does hold for partial $f$ [SpalekSzegedy06, Zhang05]. So the positive adversary method proves nothing about this function, whatever $\mathrm{Q}$ turns out to be. Consistent with $\mathrm{bs} \le \mathrm{Adv}^2$, since $\mathrm{bs} = 1$. SpalekSzegedy06 Zhang05
\(\mathrm{QC}(f)\) Quantum certificate complexity \(\Theta(1)\) $\mathrm{QC} = \Theta(\sqrt{\mathrm{fbs}}) = \Theta(1)$: a constant-size quantum certificate, matching the constant-size classical one. Aaronson08
\(\mathrm{ndeg}(f)\) Nondeterministic degree \(\Theta(1)\) $\mathrm{ndeg} \le \mathrm{C}^1 = \Theta(1)$ -- take the sum over $1$-certificates of the product of their literal indicators, which is nonzero exactly on the $1$-inputs of the domain (a $0$-input containing a $1$-certificate would be a domain point of both values). And $\mathrm{ndeg} \ge 1$.
\(\mathrm{NQ}(f)\) Nondeterministic quantum query complexity \(\Theta(1)\) $\mathrm{NQ} = \lceil \mathrm{ndeg}/2\rceil = \Theta(1)$.
\(\mathrm{rdeg}(f)\) Rational degree \(\Theta(1)\) $\mathrm{rdeg} = \max\{\mathrm{ndeg}(f),\mathrm{ndeg}(\bar f)\} = \Theta(1)$, both sides having constant certificates.
\(\mathrm{rdeg}_{\varepsilon}(f)\) Approximate rational degree \(\Theta(1)\) $\mathrm{rdeg}_\varepsilon \le \mathrm{rdeg} = \Theta(1)$.
\(\mathrm{PostQ}_{\varepsilon}(f)\) Post-selected quantum query complexity \(\Theta(1)\) $\mathrm{PostQ}_\varepsilon = \Theta(\mathrm{rdeg}_\varepsilon) = \Theta(1)$. A postselecting algorithm guesses a cell, reads it, and postselects on not seeing $2$ -- against $\mathrm{R} = \Theta(\sqrt{k\log k})$ this is an unbounded gap, and it is the mirror image of the $\mathrm{noisyR}$ story: both models can afford to guess the address. MahadevdeWolf15
\(\mathrm{CG}(f)\) Certificate game complexity (private coin) \(O(k^2) = O(\log^2 n)\) putative A private-coin strategy, since the total-function sandwich $\mathrm{CG} \le \mathrm{C}^0\mathrm{C}^1$ is unavailable on a promise problem. Each player, with probability $1/2$, outputs a uniformly random one of the $k$ address bits, and otherwise outputs a random bit of the cell indexed by its own address $a$. If the two addresses differ, they agree on a differing address bit with probability $\ge 1/(4k^2)$; if they are equal, they both name the same cell, which must differ, with probability $\Omega(1)$. So $\mathrm{CG} = O(k^2)$, while $\mathrm{CG}^{\mathrm{pub}} = O(\sqrt{k\log k})$. CGLMSG23
\(\mathrm{CG}^{\mathrm{pub}}(f)\) Certificate game complexity (public coin) \(O(\sqrt{k\log k})\) $\mathrm{CG}^{\mathrm{pub}} = \Theta(\mathrm{CA}) = O(\mathrm{R})$; the shared-randomness version of the strategy in $\mathrm{CG}$ also gives $O(k)$. From below only $\Omega(\mathrm{fbs}) = \Omega(1)$. CGLMSG23
\(\mathrm{CG}^{*}(f)\) Quantum certificate game complexity \(O(\sqrt{k\log k})\) At most $\mathrm{CG}^{\mathrm{pub}}$; all three non-private variants are $\Theta(\mathrm{CA})$. CGLMSG23
\(\mathrm{CG}^{\mathrm{ns}}(f)\) Non-signaling certificate game complexity \(O(\sqrt{k\log k})\) At most $\mathrm{CG}^{*}$. CGLMSG23
\(\mathrm{EC}(f)\) Expectational certificate complexity \(O(k) = O(\log n)\) putative Weight $1$ on each of the $k$ address bits and on the two bits of the cell $a$ pointed at by the input's own address. Two inputs of different value with different addresses differ in some address bit, where both put weight $1$; with the same address they differ at cell $a$, which both weight. Cost $k + 2$. This does not go through $\mathrm{EC} \le \mathrm{C} = \Theta(1)$, which is a total-function statement -- its proof needs a $0$-certificate and a $1$-certificate to conflict, and here two non-conflicting certificates stitch together into a string outside the promise. Whether $\mathrm{EC}$ is really $\omega(1)$ here is open; from below only $\mathrm{EC} \ge \mathrm{fbs} = 1$.
\(\mathrm{R}^{1}(f)\) One-sided-error randomized query complexity \(\Theta(\sqrt{k\log k})\) Squeezed between $\mathrm{R}$ and $\mathrm{R}_0$, which agree.
\(\mathrm{Q}_{0}(f)\) Zero-error (Las Vegas) quantum query complexity \(O(\sqrt{k\log k})\) At most $\mathrm{R}_0$; at least $\mathrm{Q}$.
\(\mathrm{Q}_{\mathrm{E}}(f)\) Exact quantum query complexity \(O(\sqrt{k\log k})\) At most $\mathrm{D}$. From below the polynomial method gives $\mathrm{Q}_E \ge \deg/2 = \Omega(1)$, i.e. nothing at all, since $\deg = \Theta(1)$ here.
\(\deg(f)\) Exact degree \(\Theta(1)\) putative At most $2$, and at least $1$ since $f$ is not constant. Count the marked cells instead of finding one: the ball $B(a,r)$ has $V = \sum_{j \le r}\binom kj$ cells whatever the address $a$ is, and on the promise every one of them holds $v$ and no other cell does, so $\tfrac1V\sum_b \mathbb{1}[x_b = 1]$ equals $v$ exactly on the whole domain. Each $\mathbb{1}[x_b=1]$ is a degree-$\le 2$ function of that cell's two encoding bits, so $\deg \le 2$ -- and $\deg = 1$ under the encoding that spends one bit on $\mathbb{1}[x_b=1]$. The address is never looked at. Nothing survives this off the promise, where the sum runs up to $2^k \gg V$: it is unbounded, which is legal for $\deg$ (agree on $\mathrm{Dom}(f)$, unconstrained elsewhere) and is exactly what $\widetilde{\deg}$ forbids. So $\deg = \Theta(1)$ against $\mathrm{D} = \Theta(\sqrt{k\log k})$, and the unbounded polynomial method is worthless here; cf. \(\mathrm{SEARCH}^{\le 1}_n\), where the same collapse happens for the same reason.
\(\mathrm{sdeg}(f)\) Sign degree \(\Theta(1)\) $\mathrm{sdeg} \le \deg = \Theta(1)$: the exact polynomial above, shifted to $\tfrac2V\sum_b \mathbb{1}[x_b=1] - 1$, sign-represents $f$ on the domain.
\(\widetilde{\deg}_{+}(f)\) One-sided approximate degree \(\Theta(1)\) Between $\mathrm{sdeg}$ and $\deg$, both $\Theta(1)$. Like $\deg$ and $\mathrm{sdeg}$, and unlike $\widetilde{\deg}$, one-sided approximate degree never asks the polynomial to stay bounded off the promise.
\(\mathrm{UC}(f)\) Unambiguous certificate complexity \(O(\sqrt{k\log k})\) Between $\mathrm{C} = \Theta(1)$ and $\mathrm{D}$. Unambiguity is not free here: the constant-size certificates overlap heavily, since many cells of the ball certify the same value.
\(\mathrm{UC}_{\min}(f)\) Minimum unambiguous certificate complexity \(O(\sqrt{k\log k})\) Between $\mathrm{C}_{\min} = \Theta(1)$ and $\mathrm{UC}$.
\(\mathrm{SP}(f)\) Subcube partition complexity \(O(\sqrt{k\log k})\) Between $\mathrm{C}$ and $\mathrm{D}$: the leaves of the deterministic tree form a subcube partition of the promise.
\(\mathrm{rank}(f)\) Decision tree rank \(O(\sqrt{k\log k})\) At most $\mathrm{D}$; at least $1$.
\(\chi(f)\) Conflict complexity \(O(\sqrt{k\log k})\) $\chi \le \bar\chi = O(\mathrm{LR}) \le \mathrm{R}$.
\(\bar{\chi}(f)\) Max-conflict complexity \(\Omega\big((k\log k)^{1/4}\big)\) From $\mathrm{R} = O(\bar\chi^2)$, which holds for partial $g$; and $\bar\chi = O(\mathrm{LR}) \le \mathrm{R} = O(\sqrt{k\log k})$.
\(\mathrm{LR}(f)\) Linearised randomized complexity \(\Omega\big((k\log k)^{1/4}\big)\) $\mathrm{LR} = \Omega(\sqrt{\mathrm{R}})$, and $\mathrm{LR} \le \mathrm{R} = O(\sqrt{k\log k})$. Do not confuse this with \(\mathrm{noisyR}(f)\) $= O(\log k)$: $\mathrm{LR}$ charges $1/\gamma$ for bias $\gamma$ and $\mathrm{noisyR}$ charges $\gamma^2$, so only the latter makes cheap guessing cheap.
\(\mathrm{RS}(f)\) Sabotage complexity \(O(\sqrt{k\log k})\) $\mathrm{RS} = O(\bar\chi) = O(\mathrm{R})$.
\(\mathrm{R}^{*}(f)\) Asymptotic composition rate \(O(\sqrt{k\log k})\) $\mathrm{R}^{*} \le \mathrm{R}$; the lower bound $\mathrm{R}^{*} = \Omega(\mathrm{LR})$ is recorded only for total $f$, and $\mathrm{R}^{*} = \Omega(\mathrm{Q})$ gives nothing without a $\mathrm{Q}$ lower bound.
\(\mathrm{cjdeg}_{\varepsilon}(f)\) Conical junta degree \(O(\sqrt{k\log k})\) $\mathrm{cjdeg}_\varepsilon \le \mathrm{R}_\varepsilon$.
\(\mathrm{Q}^{\mathrm{rob}}(f)\) Robust quantum query complexity \(\widetilde O(\sqrt{k\log k})\) $\mathrm{Q}^{\mathrm{rob}} = O(\mathrm{Q}\log \mathrm{Q})$ with $\mathrm{Q} = O(\sqrt{k\log k})$.
\(\mathrm{psD}(S)\) Pseudo-deterministic query complexity \(\Theta(\sqrt{k\log k})\) Degenerate on a partial Boolean function: $\mathrm{psD} = \Theta(\mathrm{R}) = \Theta(\sqrt{k\log k})$.
\(\mathrm{cbs}(S)\) Critical block sensitivity \(1\) Degenerate on a Boolean function: $\mathrm{cbs} = \mathrm{bs} = 1$. GoosPitassi14

Separations its values prove — computed, not recorded

Read straight off the table above by comparing two values: no relation record states these, and where one does and says more, the record wins and the pair is not listed here. 16 in total, strongest first.

… and 2 more, all of them visible in the matrix (widen it to all measures) and on the pair pages.

Separations it witnesses

Cited as a witness by