Where the flop count stopped predicting the time
The same arithmetic at a different price
A blocked and an unblocked elimination perform 72,568 operations each — the same operations, associated differently — choose the same pivots, and return a factorisation identical to the last bit: ‖PA − LU‖/‖A‖ = 4.487946226420872·10⁻¹⁶ in both. One of them moves 41,332 words between fast and slow memory and the other moves 19,476.
A block size is a property of the machine
Three lines of counting say the best block size is √(M/3). Scanned over every integer at five fast memories, the measured optimum is √M − 2 — exactly, at all five. The count has the right scaling and the wrong constant, low by a factor of 1.56, and the wrong form: the answer is affine in √M rather than proportional to it.
A reduction that changes the order
A tall-skinny QR computed as a tree of independent block factorisations touches a 512×12 matrix once instead of twelve times, computes a completely different sequence of roundings from the sweep it replaces, and returns ‖AᵀA − RᵀR‖/‖AᵀA‖ = 1.65·10⁻¹⁵ against the sweep's 9.95·10⁻¹⁵. On the same matrix classical Gram–Schmidt returns 4.6·10⁻¹⁰.
The message and the word
Three factorisations of one matrix on sixteen processors: 48 communication rounds, 4, and 4. The words sent are 1,170, 1,170 and 2,160 — so the method with the fewest rounds sends the most words, and the count that separates the three is the one no operation count can see.
Doing it twice
Cholesky QR squares the condition number — a fitted slope of 1.95 in κ against the Householder sweep's 1.00. Run the identical routine a second time on the Q it returned and the slope is 0.93, the orthogonality is at or below the sweep's at every κ, and the price is one more all-reduce.
Memory bought with messages
Holding four copies of the data instead of one is supposed to cut a matrix multiplication's communication by √4. Measured on a machine of 64 processors it costs 14% more traffic; at 576 it saves 44%, which is 72% of what the law promises. The memory is exactly four times, and that part is not asymptotic.
Where the format starts paying
A hierarchical solve costs 1.48 times a dense factorisation at 64 unknowns and 0.16 times it at 512. The crossover is between 64 and 128, it walks right when the accuracy is tightened, and the exponent between consecutive sizes is 2.13, 1.93, 1.74 — falling towards one and never arriving.
The order the products are taken in
The sparsity field's first essay says the elimination order decides the memory. This is the same sentence about arithmetic: a contraction of several tensors over shared indices has one value and many evaluation orders, and on the inner product of two trains they differ by a factor of two million.
The last digit is the cheapest
Every cost curve measured here has the same shape: the first digits are cheap and the last ones are not. One method inverts it. Doubling the work buys twice as many digits as the previous doubling did, so the price of a digit halves every time it is paid.
The recursion that was never told the memory
A blocked elimination has to be tuned to its fast memory, and tuned to one memory it costs up to 2.9 times the best at another. A recursive elimination splits the columns in half down to one and reads no memory size at all. On eight fast memories from 36 to 576 words it moves between 0.94 and 1.28 times the words of the best tuned block, with the same 585,200 operations and the same pivots — and on a machine with two caches it beats the block tuned to either cache on six machines of seven.
The block size a recursion still has
A recursive elimination is sold as having no block size, and every real one switches to plain loops below some width. Swept over that width, the traffic is a staircase with its steps at the halvings of n, and its cliff sits where the blocked elimination's does — the first panel wider than √M − 2 moves 1.53 to 4.47 times the words, on six memories of six. On three caches the innermost decides, and a third cache costs every tuned block up to 14 per cent and the recursion nothing.
A ceiling is not a target
Asked to hold the smallest possible intermediate, an evaluation order costs a median of 1.55 times the cheapest order's arithmetic. Asked to hold no more than that same amount, it costs 1.15. The two answers hold exactly the same number of numbers, and on one network they are 5.17 times apart in work.
The plan that was right at rank four
An evaluation order is chosen once and paid for thousands of times, and the dimensions it was chosen at are not the dimensions it runs at. Compiled at rank four and run at rank 256 it costs 8.01 times the order that rank deserves; compiled at 256 and run at 2 it costs 301 times. A one-line rule recomputed on arrival costs 2.92 and 1.11.
The search that got worse as it widened
Keeping two candidate orders instead of one removes a third of the cheapest-product rule's excess, and by eight it has removed all of it — the two greedy rules this field separated become the same rule. On twenty-four of sixty networks a wider search returns a dearer order than a narrower one, and it is the better rule that it more often makes worse.
A beam ranked on what remains
A beam over contraction orders ranked on cost so far returns a dearer order when it is widened on 24 of 60 networks. Rank it on cost so far plus the largest group still to be paired — a lower bound on what remains, and free — and that falls to 12, eleven of them from the original 24. Rank it on a stronger estimate that is not a bound and the count stays at 23, but only 10 are the same networks: the anomaly has moved, not gone. At nine tensors, where the unranked beam's median at width 8 was worse than at width 1, both rankings make widening pay again.
Widen the beam where the ranking is right
A beam that is wide at its first pairings and narrow at its last looks like the right shape, since the early commitments are the damaging ones. Measured, it is worse than no beam: at nine tensors a width of 16, 8, 4, 2 and then 1 prices 742 pairings and has a median of 1.263 against width 1's 1.062 for 120. The reverse shape — width 1 early, doubling to 16 at the end — prices 78 pairings on seven tensors and finds the exhaustive order on 37 of 60 networks, against 30 for a constant width of 4 at 161. A beam should be wide where cost so far is nearly the whole cost.
The leaf that sits on the edge
A recursive elimination's base case was found to be a block size in disguise, with a cliff where the blocked elimination's is, and the choice read as a trade: the processor wants wide leaves, the cache wants narrow ones. Measured at every width rather than at the halvings of 96, there is no trade inside the edge. Leaves exactly √M − 2 wide are the cheapest the recursion can have in words as well as calls — 0.81, 0.80 and 0.84 of the pure recursion's traffic at 64, 144 and 256 words — and one column wider moves 1.77 to 3.15 times it. And a matrix of 100 columns, which halves unevenly, meets the cliff in two steps rather than one.
A near-tie is a factor of four
A beam over contraction orders spends its width by level — wide late was the best schedule, finding the exhaustive order on 37 of 60 seven-tensor networks for 78 pairings priced. The alternative was to widen only where the ranking is a near-tie. Measured, near-ties buy nothing: keeping every candidate within a quarter of the best changes nothing at all, because the pairing the exhaustive order wants is almost never tied with the ranking's first choice. It sits a factor of two to four further down. Widen to keep everything within a factor of three and the beam finds the exhaustive order on 46 networks for 150 pairings, and at nine tensors has a median of 1.007 where every schedule tried before had 1.056 or worse below 1,380 pairings.
Leaves cut to the edge on purpose
A recursive elimination's cheapest leaf is exactly the square root of M, less 2, columns wide, and the rule drawn from it was to set the base case there and let the halvings put the leaves at or below it. On thirty-two sizes from 96 to 127 columns, halving to that base case moves more words than the pure recursion on sixteen of them at 144 words of fast memory, because the halvings stop at six and seven columns, not ten. Cut every dimension at a multiple of the edge instead and every size keeps the saving: 0.79 to 0.86 of the pure recursion's words, against halving's 0.91 to 1.07, with the same arithmetic and the same pivots. What it cannot make full is the one leftover leaf, and a leftover of one column is where it loses.
The circle between two eigenvalues
A contour count's error is not approximately governed by the nearest eigenvalue; it is exactly one closed-form term per eigenvalue, and summing those terms reproduces the quadrature to a millionth at 1,720 radius and point pairs. Three things follow. The rate is the ratio of the two moduli the circle sits between, not a distance, so two circles 0.2 from their nearest eigenvalue converge three times apart. Ten digits cost about 21 points divided by log₁₀ of that ratio, 71 points with twelve eigenvalues inside and 3,476 with six. And the best circle is not halfway: at the geometric mean of two eigenvalues on one ray their two terms are equal and opposite, and ten digits cost 27 points where the midpoint needs 84.