1: \documentclass[12pt]{article}
2: \usepackage{graphicx}
3: \input{amssym}
4: \title{Very elementary interpretations of the Euler-Mascheroni constant
5: from counting divisors in intervals}
6: \author{David Feldman}
7:
8: \begin{document}
9:
10: \maketitle
11:
12: \section{Introduction}
13:
14: After $\pi$, and then $e$, or perhaps the golden ratio $\phi$,
15: the Euler-Mascheroni number $\gamma$ stands among the most famous
16: mathematical constants. We aim here for a formulation of $\gamma$
17: that makes it accessible to the widest possible public.
18:
19: Now the public knows $\pi$ best by dint of its connection to
20: computing circular perimeters and areas. Though a mathematician
21: would analyze these computations by employing the apparatus of
22: limits \`a la Cauchy, merely {\em communicating} the meaning of
23: $\pi$ should not depend on Cauchy's sophisticated, abstract,
24: universal, rigorous formulation of limits. Cauchy's approach
25: synthesized diverse mathematical discourses, but it did not
26: abolish them. The public face of a mathematical constant should
27: preferably not depend on familiarity with Cauchy style limits.
28:
29: While a mathematical constant will possess a single, definite
30: value, it may admit many interpretations according to the diverse
31: contexts where it arises. For example, formulated properly, we
32: may say that, with probability $6/\pi^2$, two natural numbers
33: chosen at random share no factor greater than $1 $. That $\pi$
34: occurs here despite the lack of any apparent connection to circles
35: beautifully exemplifies the sort of excitement associated with
36: pure mathematics!
37:
38: The public does not know $e$ as well as it knows $\pi$, but $e$
39: too admits accessible narrative interpretations. If the public
40: knew hyperbolas as well as it knows circles, one could effectively
41: characterize $e$ as that number (greater than 1) such that that
42: the area under $y=1/x$ over the interval $[1,e]$ equals $1$.
43: Closer to practical concerns, one can observe that $\$1$ left in
44: the bank for a year at $100\%$ interest, compounded continuously,
45: grows to $\$ e$. A seemingly very different take on $e$ involves
46: {\em derangements}. Supposing that $n$ people participate in a
47: Christmas party grab bag, we can ask for the probability that no
48: one gets their own gift back. All the probabilities with $n$ even
49: exceed all the probabilities with $n$ and only the number $1/e$
50: lies in between.
51:
52: The Euler-Mascheroni constant $\gamma$ cries out for a canonical
53: narrative interpretation suitable for public consumption. Steven
54: R. Finch's encyclopedic {\em Mathematical Constants} lists several
55: candidates, where the most compelling takes the form
56: $$
57: \lim_{n\to\infty}\frac{1}{n}\sum_{k=1}^{n}\left\{\frac{n}{k}\right\}=
58: 1-\gamma$$ a result of de la Vall\'ee Poussin. As Finch
59: paraphrases de la Vall\'ee Poussin's result:
60: \begin{quote}
61: \ldots if a large integer $n$ is divided by each integer $1\leq
62: k\leq n$, then the average fraction by which the quotient $n/k$
63: falls short of the next integer is not 1/2, but $\gamma$!
64: \end{quote}
65: As an elementary interpretation of $\gamma$, de la Vall\'ee
66: Poussin result has two nice features not shared by Finch's other
67: examples. First, $\gamma$ occurs more or less directly, rather
68: than embedded in a formula such as $e^{-\gamma}$. Second, de la
69: Vall\'ee Poussin's formula for $\gamma$ refers only to basic
70: arithmetic and in particular avoids mention of natural logarithms.
71:
72: We offer here a novel elementary interpretation (indeed a vast
73: family of such interpretations) of $\gamma$ sharing the stated
74: advantages of de la Vall\'ee Poussin's and the additional
75: advantage, perhaps, that it arises very naturally if one considers
76: a very modest variation on a very familiar mathematical situation.
77:
78: We mean to address two sorts of readers at once, namely those who
79: have had (or remember) only high school mathematics and would like
80: to learn about the Euler-Mascheroni constant from scratch, and
81: those who know enough calculus to digest the usual definition and
82: wish to understand its equivalence with our reformulation. The
83: former may just skip without loss some remarks obviously directed
84: at the latter, who should exercise patience with details spelled
85: out for the former.
86:
87: We begin by recalling the usual formula for the Euler-Mascheroni
88: and then offer an alternative formula in the same spirit which
89: nevertheless eliminates the explicit appearance of natural
90: logarithms. Our first attempt at attaching a very simple,
91: compelling narrative interpretation to our formula for $\gamma$
92: produced only a fallacy, albeit an instructive one. Rather than
93: suppress this initial failure, we start there, so that the reader
94: will appreciate the mildly technical but unavoidable modification
95: required for a valid interpretation.
96:
97: \section{The standard definition of the Euler-Mascheroni constant}
98:
99: We begin by explaining in elementary terms the usual definition
100: $$ \gamma = \lim_{n\to\infty}\left( \sum_{k=1}^n
101: \frac{1}{k}-\ln(n)\right)\ .$$
102: We wish to interpret $\gamma$ geometrically. For
103: this purpose it does no harm to make the modification
104: $$ \gamma = \lim_{n\to\infty}\left( \sum_{k=1}^{n-1}
105: \frac{1}{k}-\ln(n)\right),$$
106: or (after reindexing)
107: $$\gamma =\lim_{n\to\infty}\left( \sum_{k=1}^{n}
108: \frac{1}{k}-\ln(n+1)\right).$$
109: Note that the general term of the original sequence and of the
110: (first formulation of the) new
111: sequence differ by $1/n$, and the difference approaches $0$ as $n$ grows,
112: justifying the modification.
113:
114: The (reindexed) new sequence leads to the area of the shaded region in the
115: following diagram:
116:
117: \includegraphics{Eulerconstantdefinition.ps}
118:
119: Here the curve in the diagram represents the graph of $y=1/x$.
120: Indeed the first term of the new sequence gives the area of the
121: leftmost black wedge, the second term the area of the two leftmost wedges,
122: and generally, the $n^{\rm th}$ term the area of the $n$ leftmost wedges.
123: Explicitly, the sum $\sum_{k=1}^{n}
124: \frac{1}{k}$ gives the area of the $n$ leftmost rectangles
125: and $\ln(n+1)$ {\em means} the area under the curve and within these
126: rectangles. Taking the limit gives the area of all the wedges.
127:
128: Now imagine all the wedges sliding horizontally to the left
129:
130: \includegraphics{sliding.ps}
131:
132: \noindent
133: until we have the stacked vertically within the square, our original left
134: most rectangle:
135:
136: \includegraphics{slided.ps}
137:
138: From this picture we see (or at least glean the tools needed to prove)
139: the finiteness of $\gamma$. As a subset of the unit square it must
140: have an area between $0$ and $1$, and the picture even makes clear
141: that the area of all the wedges must exceed $.5$. Moreover the
142: area of the first $n$ wedges falls short of $\gamma$ by no more than
143: $1/n$ (since the remaining wedges fit in 1 by 1/n rectangle), but also
144: by at least $1/2n$ (since they fill more than half of that rectangle).
145:
146: Observe that, as with $\pi$, we can interpret $\gamma$ as the area
147: of a region in the plane that can construct explicitly. Of course
148: this region seems highly artificial compared with the unit circle.
149: To a student of integral calculus the region should seem less unnatural.
150: In that context, $\gamma$ bounds the error that occurs when approximating
151: the areas defining natural logarithms of natural numbers by means of
152: {\em upper sums}. Of course one can approximate a give area by many
153: different upper sums, but {\em these} upper sums often arise in their
154: own right, as {\em harmonic sums}
155: $\sum_{k=1}^{n} 1/k$. One often has occasion to turn
156: the story around, and using (sophisticated but easily manageable
157: natural logarithms to approximate (elementary but awkward)
158: harmonic sums. As an {\em a priori} estimate of the error involved,
159: $\gamma$ can help us improve such approximations, and in this role
160: it enters many formulas.
161:
162: \section{Getting rid of the logarithms}
163:
164: The following pictures suggest some calculations to approximate
165: $\gamma$ which don't involve logarithms, and thus lead to a way
166: of framing $\gamma$ for an audience that doesn't know about logarithms
167: (and doesn't want to hear about them):
168:
169: \includegraphics{nolog1.ps}
170:
171: \vspace{1em}
172: \includegraphics{nolog2.ps}
173:
174: \vspace{1em}
175: \includegraphics{nolog3.ps}
176:
177: By way of explanation, we would like to estimate the total area of
178: all the wedges without computing exactly the area {\em under} the
179: curve. We do this by now also approximating the region {\em under}
180: the graph by a union of rectangles, but we let these approximations
181: get more refined as we go.
182:
183: As far as concerns estimating $\gamma$,
184: we now have two sources of error. First, the $n^{\rm th}$ picture
185: only takes account of the first $n$ wedges. Second, we have unwanted
186: area now below the various wedges.
187:
188: We have already bounded the magnitude of the first type of error by $1/n$.
189: We can also approximate the second type of error by sliding wedges,
190: this time the new wedges we have created under the graph. In the
191: $n^{\rm th}$ pictures, these will all slide horizontally to fit
192: inside a $1/n$ by $1$ rectangle. So $1/n$ also bounds
193: the second type of error. These two types of error, moreover,
194: carry opposite signs, so certainly $1/n$ bounds the total error.\footnote{
195: In actually the two types of error tend to cancel. It turns out
196: that $n^2$ times the error approaches $2/3$. Our reformulation
197: converges to $\gamma$ rather must faster than the original definition.}
198:
199: Numerically, the area of the regions in the three diagrams equal,
200: as the reader may easily check,
201: $$ 1 - \frac{1}{3} -\frac{1}{4},$$
202: $$ 1 +\frac{1}{2}- \frac{1}{4}
203: -\frac{1}{5}- \frac{1}{6} -\frac{1}{7}- \frac{1}{8} -\frac{1}{9},$$
204: $$ 1 +\frac{1}{2}+ \frac{1}{3}-\frac{1}{4}
205: -\frac{1}{5}- \frac{1}{6} -\frac{1}{7}- \frac{1}{8} -\frac{1}{9}-\frac{1}{10}
206: -\frac{1}{11}- \frac{1}{12} -\frac{1}{13}-
207: \frac{1}{14} -\frac{1}{15}-\frac{1}{16},$$
208: or in general
209: $$ \sum_{k=1}^{n-1} \frac{1}{k} - \sum_{k=n+1}^{n^2} \frac{1}{k}$$
210: for the ${n-1}^{\rm st}$ picture.
211:
212: In words, to approximate $\gamma$, for $q=n^2$, we sum the reciprocals of
213: numbers less than $\sqrt{q}$ and subtract off the reciprocals of all
214: numbers greater than $\sqrt{q}$ up to $q$. Indeed we need not require
215: making $q$ a perfect square. We see this easily by comparing the recipe
216: applied to a general $q$ with the recipe applied to the largest square
217: below it.
218:
219: So, as a slogan, for large numbers $q$, $\gamma$ approximates
220: the sum of the reciprocals of the numbers below the square root of $q$
221: minus
222: the sum of the reciprocals of the numbers above the square root of $q$, up to
223: $q$.
224:
225: \section{A fallacy, first}
226:
227: Roughly speaking, for random $q$, the probability that $d$ divides $q$
228: equals $1/d$ (since dividing $q$ by $d$ can leave $d$ possible
229: remainders, all equally likely, with 0 just one among them).\footnote{We
230: must say ``roughly speaking'' because we cannot make
231: literal sense of ``for random $q$'' since the set of all
232: natural numbers does not carry any uniform probability distribution.
233: We may of course speak of a random $q$ between $1$ and $B$, but
234: depending upon the $B$, the probability may not equal
235: exactly $1/d$; the larger the $B$, though, the smaller the error.}
236: In probability theory one typically introduces a quantity that equals
237: $1$ when an event occurs and $0$ when it doesn't. The {\em expectation}
238: of this sort of quantity (intuitively, its value on the average)
239: coincides with its probability. The virtue of working with expectations
240: rather than directly with probabilities lies in the linearity of expectation:
241: the expectation of a sum equals the sum of the expectations.
242:
243: So suppose we have a set $D=\{d_1,\ldots,d_j\}$. Again, roughly speaking,
244: the expected number of elements of $D$ that divide a random $q$ should
245: equal $$\frac{1}{d_1}+\cdots+\frac{1}{d_j}.$$
246:
247: Notice that when $D$ consists of many consecutive natural numbers,
248: the expected number of elements of $D$ that divide a random $q$ has the form
249: of the sort of quantities that come into our approximations for $\gamma$.
250:
251: This perhaps suggests asking if $\gamma$ approximates the expectation
252: of $Z$, defined as the number of divisors of $q$ below $\sqrt{q}$ \hspace{1em}
253: minus \hspace{1em} the number of divisor of $q$ above $\sqrt{q}$.\footnote{We do not
254: have to end with ``up to $q$'' since no number larger than $q$ divides $q$.}
255:
256: $Z$ does indeed have an expectation, but its expectation turns out
257: equal to $0$, not $\gamma$!
258:
259: Indeed, if $d$ divides $q$, so does $q/d$, and if one lies below
260: $\sqrt{q}$ the other lies above, and vice versa. For example, if
261: $d<\sqrt{q}$ and also $q/d < \sqrt{q}$, we have
262: $$q=d\cdot(q/d)<(\sqrt{q})^2=q,$$
263: a contradiction, and similarly for $d,q/d>\sqrt{q}$.
264: Thus {\em every} number $q$ has exactly the name
265: number of divisors below $\sqrt{q}$ as above.
266:
267: Of course, the reader already trained to refuse even to hear
268: all but the most rigorous analysis will find no fallacy here.
269: However, in mathematics, our type of heuristic reasoning does often
270: lead, after careful formulation, to true statements, albeit often
271: these statements turn out much harder to prove than the heuristics
272: suggest. So even though we made clear when we left the realm of
273: rigorous reasoning, perhaps it still comes as a surprise that we
274: have failed so badly, that the gaps we left do not admit any repair.
275:
276: The reader may well wish to think upon the question of what sort of
277: burden a failed heuristic imposes. We have proved that it lead us
278: to a wrong conclusion. Generally speaking we don't feel we need to
279: explain why erroneous proofs lead to false conclusions! Nevertheless,
280: when an erroneous proof depends on the unproved assumption that
281: certain quantities vary independently when in fact they don't, we
282: ought enquire into the nature of their interdependence. Alternatively,
283: and we take this approach here, we can see if can rescue the heuristic
284: by some slight change of the situation.
285:
286: We surely can make perfect sense of
287: ``the expected number of elements of $D=\{d_1,\ldots,d_j\}$ that divide a random $q$
288: equals $\frac{1}{d_1}+\cdots+\frac{1}{d_j}$'' provided that we {\em keep
289: $D$ fixed}, bound $q$, and accept some small error that tends to vanish
290: as the size of the bound on $q$ grows. But our purported interpretation
291: of $\gamma$ had the ``$D$'' varying along with $q$.
292:
293: This suggests a first, but admittedly ugly, fix. First fix $q$. Now
294: given another
295: quantity $Q$, consider, $Z_q$, the number of divisors of $Q$ minus than
296: $\sqrt{q}$ minus the number of divisors of $Q$ between $\sqrt{q}$ and $q$.
297: The expectation of $Z_q$ takes the form of one of our approximates to
298: $\gamma$, but we must let $q$ grow and take a bald limit to get $\gamma$
299: itself, so not the stuff of a popular interpretation.
300:
301: \section{A surprisingly satisfactory fix}
302:
303: We shall now formulate a family of valid probabilistic interpretations of
304: $\gamma$, all very much in the spirit of the fallacious one, albeit just
305: slightly more complicated.
306:
307: {\bf Theorem 1} {\em Let $F:{\Bbb N}\rightarrow{\Bbb R}$ stand for any
308: function which\\ a) $F$ monotonically weakly increases;\\ b) $F$ tends
309: to infinity; and\\ c) such that $q/F(q)$
310: tends to infinity.
311:
312: Let $Z_F(q)$ equal the number of divisors of $q$
313: less than $\sqrt{F(q)}$ minus the number of divisors of $q$ between
314: $\sqrt{F(q)}$ and $F(q)$.
315:
316: Then, on the average\footnote{Of course by ``on the average'' we mean taking the limit of
317: averages that arise with $q$ bounded by $B$ as $B$ increases.}, $Z_F(q)$ equals $\gamma$.}
318:
319: \vspace{1em}
320:
321:
322: Considering our original goal, a popular interpretation of $\gamma$,
323: we could perhaps just set $F(x)=\sqrt{x}$. We then get
324: \begin{quote}
325: $\gamma$ means the average by which the count of divisors of a number
326: that sit below
327: its fourth root exceeds the count of divisors that lie
328: between the fourth
329: root and the square root.
330: \end{quote}
331:
332: The gist of the previous section consists in telling
333: us that we cannot entirely
334: dispense with the condition that $n/F(n)$ tends to infinity, since
335: the conclusion fails when taking $F(n)=n$.
336:
337: {\bf Proof}
338:
339: In the following diagram,
340:
341: \includegraphics{diagram1.ps}
342:
343: \noindent
344: circles in row $r$ (counting up)
345: have area $1/r$. We have colored green those
346: circles in column $q$ having row
347: number less than $\sqrt{F(n)}$, and those with row number
348: in the half-open interval
349: $[\sqrt{F(n)},F(n))$ red. (While we have in mind a general
350: $F$, satisfying the conditions of Theorem 1, the diagram
351: shows the situation specifically for $F(x)=\sqrt{x}$.)
352:
353: Consider a particular column. By our previous work\footnote{
354: Just for the sake of simplicity now, here we choose to approximate
355: $\gamma$
356: by the sum of the reciprocals of the numbers below the square root
357: of $q$
358: minus
359: the sum of the reciprocals of the numbers {\em equal to or}
360: above the square root of $q$
361: }
362: the excess of
363: the green area over the red area takes the form of an approximation
364: to $\gamma$ with the approximations approaching perfection as we
365: move to the right, on account of the assumption that $F(q)$ grows without
366: bound. So certainly if we consider together
367: all the columns up to column $B$, the total green area less the total
368: red area divided by $B$ approaches $\gamma$ as $B$ tends toward infinity.
369:
370: Now compare the following diagram with the previous:
371:
372: \includegraphics{diagram2.ps}
373:
374: Here all circles now have area 1, but this time
375: we only color circles if the row
376: number divides the column number.
377:
378: For the second diagram, for a given column,
379: the excess of green area over red area constitutes just the sort of
380: quantity we have claimed averages to $\gamma$ in the long run.
381:
382: It suffices to show that
383: if we consider together
384: all the columns up to column $B$, the total green area less the total
385: red area divided by $B$ approaches $\gamma$ as $B$ tends toward infinity.
386:
387: While the two diagrams appear quite different
388: column-by-column, a row-by-row comparison works out quite simply, as follows.
389:
390: Fix a row number, say $r$, and consider the corresponding $r$-rows
391: in the two diagrams, with the aim of estimating the discrepancy
392: between, first, the total red areas they hold,
393: and second, their total green
394: areas.
395:
396: In the $r$-row of the second diagram, consider any colored circle
397: if one occurs. Call it $C_1$; $C_1$ has area 1. Write $C_2$
398: for the next colored circle to its right (in the infinite version of
399: the second diagram).
400: Next, consider the circle $c_1$
401: in the first diagram corresponding position-wise to $C_1$
402: together with
403: the $r$ circles in the first diagram
404: in positions corresponding to those circles
405: strictly between $C_1$ and $C_2$ (all these diagram 1 circles
406: together have total area 1).
407:
408: The previous paragraph shows that
409: if the total red areas in the $r$-rows of the diagrams differ, they
410: differ on account of what happens when, moving left to right say,
411: as we enter and leave the first diagram's
412: ``red island''.
413:
414: Thus the
415: red area discrepancy in row $r$ cannot exceed magnitude $1$, and likewise
416: for the green area discrepancy.
417:
418: As for the green area minus the red
419: area in the two $r$-rows, the discrepancy between the diagram one
420: difference and the diagram two difference
421: cannot exceed magnitude $2$.
422:
423: For rows with no colored circles in either diagram we obviously have
424: no discrepancy at all, and at most
425: $F(B)$ rows have colored circles.
426:
427: We have now bounded the total green area minus red area discrepancy
428: (for all rows) between the two diagrams by $2F(B)$. By assumption,
429: $2F(B)/B$ approaches $0$ as B grows. Thus, as $B$ increases,
430: the values for the average green area minus red area per column for
431: two types of diagrams converge.
432:
433: Since this average approaches $\gamma$
434: for diagrams of the first type, it also does for diagrams of the second
435: type, as desired.
436:
437: \section{The case of $F(x)=\alpha x$}
438:
439: Theorem~1 does not
440: speak to the case of $F(x)=\alpha x$ for any $\alpha\in(0,1)$;
441: such an $F$ could produce as many as $\alpha x$
442: rows that exhibit a discrepancy. Nevertheless we can make
443: the proof technique yield up a complete analysis.
444:
445: {\bf Theorem 2} {\em Fix $\alpha\in(0,1)$. Write $A$ for the average
446: number of divisors of $n$ that lie in $(0,\sqrt{\alpha n})$
447: minus the number of that lie in $(\sqrt{\alpha n},\alpha n)$.
448: Then
449: $$A=\sum_{i=1}^{\lceil \frac{1-\alpha}{\alpha} \rceil } \frac{1}{i}
450: -\ln(\frac{1}{\alpha}).
451: $$}
452:
453: Before turning to the proof, we offer a few remarks.
454:
455: First, except
456: when $1/\alpha$ has integral value,
457: $\lceil \frac{1-\alpha}{\alpha} \rceil=
458: \lfloor \frac{1}{\alpha} \rfloor$, which looks a bit simpler.
459:
460: The formula correctly predicts a balance between divisors above
461: and below the square root of $n$, the $\alpha=1$ case. Moreover,
462: as $\alpha$ approaches $0$, the values of the formula converge
463: to $\gamma$, just as one might hope based on Theorem 1.
464:
465: The discontinuities in the graph below
466:
467: \centerline{\includegraphics{osc4.ps}}
468:
469: \centerline{Graph of $A$ as a function of $\alpha$}
470:
471: \noindent
472: come as no surprise. As $\alpha$ shrinks past $1/k$, we lose,
473: from the second diagram, divisors of $n$ of the form $n/k$ when
474: they occur, which they do for one $n$ out of $k$. For those $n$ large
475: compared to $k$ we will have these divisors colored red
476: (since n/k will exceed $\sqrt(\alpha n)$),
477: so we expect the graph to jump up (as we move to the left) by $1/k$.
478:
479: For all $\alpha < 1$ we have $A>0$, so we expect, on
480: the average, more divisors in
481: $(0,\sqrt{\alpha n})$ than in $(\sqrt{\alpha n},\alpha n)$.
482: This leads us to guess that numbers $n$ with
483: more divisors in
484: $(0,\sqrt{\alpha n})$ than in $(\sqrt{\alpha n},\alpha n)$
485: should occur with a positive density. But this does not follow
486: immediately. Logically speaking, relatively rare numbers with
487: {\em many} more divisors in
488: $(0,\sqrt{\alpha n})$ than in $(\sqrt{\alpha n},\alpha n)$
489: might possibly make all the necessary contribution to
490: the average behavior.
491: Nevertheless, such number cannot occur {\em too} rarely, since,
492: overall, relatively few numbers $n$ possess even a {\em total}
493: number of divisors large compared with $\ln n$.
494:
495: Because the graph oscillates about the value $\gamma$, for
496: infinitely many special values of $\alpha$ (namely those of the
497: form $e^{\gamma-(1+\cdots+1/k)}$), $A$ takes the value $\gamma$,
498: the right answer for the wrong reason, if you will. Note that
499: this characterizes $\gamma$:
500: the {\em only} average realized for infinitely many values of $\alpha$.
501:
502: One might wonder about the average value of the average if we
503: choose $\alpha$ from a uniform distribution on $(0,1)$.
504: Curiously, integrating $A$
505: as $\alpha$ varies over $(0,1)$ gives
506: $\zeta(2)-1=\pi^2/6-1=.644934068\ldots$.
507:
508: \vspace{1em}
509: \noindent
510: {\bf Proof of Theorem 2} We refers here
511: to the same two sorts of diagrams as the last proof,
512: but now we assume them {\em square}, just so that the average per row
513: excess of green area over red area equals the average per column excess.
514:
515: We wish to compare, asymptotically, the the average per row
516: excess of green area over red area in the two types of square
517: diagrams.
518:
519: Since we have a uniform bound on the excess that occurs in any
520: {\em single} row, we can safely ignore the green circles entirely! The
521: green circles occur in only $\sqrt{\alpha n}$ rows, so the
522: variation in green areas between the two diagrams
523: will tend to vanish when we divide by $n$ and let $n$ grow.
524: (Compare with the previous proof, where the condition on $F$
525: meant that switching to a row-by-row analysis ultimately
526: allowed us to ignore {\em everything}. Even with $F=\alpha x$,
527: the old reasoning still applies to $\sqrt{F}$.)
528:
529: As for the variation in the {\em red} area between the two diagrams,
530: we employ a straight-forward integral approximation, getting
531: $$\int_0^\alpha
532: \left(\lfloor 1/y \rfloor - \lfloor 1/\alpha \rfloor\right ) -
533: \frac{1-\frac{y}{\alpha}}{y} \,dy .$$
534: The first term, in parentheses, captures the contribution for the second
535: diagram, and from this we subtract off the contribution from the first
536: diagram. Specifically, we have estimated the average per row excess
537: of the red area in the second diagram over red area in
538: the first.
539:
540: Please note, for clarity, that since red dots count {\em
541: negatively}, and by the remark above concerning the possibility of
542: ignoring the green area, this expression also estimates the amount
543: by which the average per row excess of green area over red area in
544: the first diagram exceeds the average per row excess of green area
545: over red area in the second diagram.
546:
547: After some routine calculation, the integral in question evaluates
548: to\footnote{In the case of $\alpha=1$, mechanical evaluation of this
549: integral constitutes the essence of a proof of the theorem of
550: de la Vall\'ee Poussin mentioned in the introduction -- $\gamma$ emerges
551: directly from the definition in the form of its usual definition.
552: But from the pairing of divisors
553: of $n$ above and below $\sqrt{n}$ we actually
554: know the value of the integral
555: in advance, albeit just in this case. That means
556: we have actually have in hand two
557: independent proofs of de la Vall\'ee
558: Poussin's theorem.
559: }$$\gamma
560: +\ln(\frac{1}{\alpha})
561: -\sum_{i=1}^{\lceil \frac{1-\alpha}{\alpha} \rceil } \frac{1}{i}.
562: $$
563: Since we know that the first diagram has an average per row
564: excess of green area over red area equal to $\gamma$, while
565: we seek the corresponding information for the second diagram,
566: the result follows
567: when we subtract this quantity from $\gamma$.
568:
569: \section{Final Remark} Even though we set as our original goal
570: the crafting of novel interpretations for $\gamma$, a great variety of
571: curious
572: statements arise when we force $\gamma$ to leave the story. Here we
573: give just
574: one example. By Theorem~1, a number $n$ tends to have
575: $\gamma$ more divisors in $(0,n^{1/4})$ than in $(n^{1/4},n^{1/2})$,
576: and likewise $\gamma$ more divisors in $(0,n^{1/8})$ than in
577: $(n^{1/8},n^{1/4})$. Subtract these two differences, we see that:
578: \begin{quote}
579: on the average, $n$ has exactly twice as many divisors in
580: $(n^{1/4},n^{1/2})$ as it does in $(n^{1/8},n^{1/4})$.
581: \end{quote}
582: Since $\gamma$ no longer appears in the statement, one should naturally
583: enquire about the possibility of a $\gamma$-free proof.
584:
585:
586:
587:
588:
589:
590:
591:
592: \end{document}
593: