text
stringlengths 2
6.14k
| file_name
stringclasses 10
values | node_id
stringlengths 36
36
|
|---|---|---|
Metabolic scaling in small life forms
Mark E. Ritchie & Christopher P. Kempes
March 4, 2024
Abstract
Metabolic scaling is one of the most important patterns in biology.
Theory explaining the 3/4-power size-scaling of biological metabolic rate
does not predict the non-linear scaling observed for smaller life forms.
Here we present a new model for cells <10−8m3that maximizes power
from the reaction-displacement dynamics of enzyme-catalyzed reactions.
Maximum metabolic rate is achieved through an allocation of cell volume
to optimize a ratio of reaction velocity to molecular movement. Small
cells<10−17m3generate power under diffusion by diluting enzyme con-
centration as cell volume increases. Larger cells require bulk flow of cy-
toplasm generated by molecular motors. These outcomes predict curves
with literature-reported parameters that match the observed scaling of
metabolic rates for unicells, and predicts the volume at which Prokary-
otes transition to Eukaryotes. We thus reveal multiple size-dependent
physical constraints for microbes in a model that extends prior work to
provide a parsimonious hypothesis for how metabolism scales across small
life.Introduction
Understanding how and why organisms differ in their demand for and use of
resources is a key objective of biologists [1–3] and critical to understanding the
response of biodiversity and ecosystem function to global changes [4, 5]. A
fundamental and often-debated pattern is how metabolic rate, B, scales with
body mass, M[1, 6, 7], followng the form
B=B0Mβ. (1)
Data for vertebrates and vascular plants [8–12]show an average interspe-
cific scaling exponent of 3/4. This relationship has inspired diverse theories
[10, 11, 13–15], including the network model, which derives 3/4 from the need
for organisms to supply the entire body volume with resources from vascular
resource distribution networks that minimize energy dissipation. More recent
analyses [1, 16–20] that include organisms from the smallest 11 orders of mag-
nitude in size that largely lack vascular distribution networks, show a variable
1arXiv:2403.00001v1 [physics.bio-ph] 16 Dec 2023 | 2403.00001v1.pdf | 50f07c49-d1f4-462b-9b35-d3109cf78c61 |
metabolic scaling exponent that changes across size ranges from scaling expo-
nents as high as 2 – super-linear scaling – for the smallest range and near 3/4
for the largest organisms. Understanding the basis for this variation is impor-
tant, as many different organism features can be derived from this metabolic
scaling, including how maximum growth rates and ribosomal abundances scale
with cell size along with key tradeoffs between features [21, 22]. However, there
is no general, parsimonious theory derived from physical and chemical principles
that addresses this size-dependent scaling for these smallest organisms [16].
Here we propose a combined thermodynamic and reaction description of
metabolic rate for a cell. We assume that natural selection will favor organisms
that can perform more work per time (power) for material synthesis, replica-
tion, repair, locomotion, and other cellular functions [23]. While often co-limited
by materials in the environment, optimizing this power faces cellular trade-offs
concerning both environmental and intra-cellular physical constraints.Conse-
quently, we optimize cellular features to maximize the free energy produced by
the conversion of chemical substrates to products, where heat and products may
inhibit metabolism if not moved away from reaction sites. This thermodynamic
approach provides a framework for considering the two processes, conversion
and displacement, simultaneously.
Thermodynamics of reaction and molecular dis-
placement
We assume that work is proportional to the volume of reactive surface, where
macromolecular catalytic enzymes are freely dispersed in the cytoplasm or at-
tached to the membranes, cytoskeleton, and/or other organelles within the en-
tity [24]. However, a portion of cell volume is needed to allow substrates from
the environment to reach reaction surfaces and to allow displacement of reac-
tion products away from reaction structures. This sets up a conflict between
the volume devoted to metabolic processes and that devoted to transport. Here
we derive a physical and chemical principles-based theory for metabolic scaling
for metabolic rate across the roughly 11 orders of magnitude in volume of or-
ganisms that lack branching vascular systems. | 2403.00001v1.pdf | 30c34430-9648-4d4f-8d96-0e2383bd6fef |
We describe metabolism from
a reaction-displacement thermodynamic system centered on or near reaction
surfaces and its generation of free energy within a spherical space otherwise
obstructed by surfaces at which reactions occur. Our framework applies to uni-
cellular Prokaryotes, Archaea, and Eukaryotes along with eukaryotic organelles.
Our basic model, which we expand on in stages, is to consider metabolism
in a sphere described by the following reaction-diffusion model
∂A
∂t=D1
r2∂
∂r
r2∂A
∂r
−kAZ
∂P
∂t=D1
r2∂
∂r
r2∂P
∂r
+kAZ (2)
2 | 2403.00001v1.pdf | 60c7fa43-505f-49b1-b82f-8ca9e092e0bf |
where the product, P(r), substrate, A(r), and enzyme, Z(r), concentrations
(mol m−3) are all a function of radius, r, inside a sphere, and where the dynamics
are one dimensional in spherical coordinates with spherical symmetry.
Additionally, k=kcat/KMis the reaction constant ((m3mol−1s−1)) and
Dis a displacement coefficient (e.g. molecular diffusivity in some cases) and
has units of (m2s−1) so that each rate is a change in concentration (mol m−3
s−1). We consider the steady-state dynamics, representing a persistent entity
and its metabolism in a fixed environment over time, allowing us to solve for
closed-form solutions of P(r) and A(r) under a given concentration of Z(see
SI).The steady-state free energy production at any location in the cell is given
by
J∗(r) =kRTA∗ZlnA∗Keq
P∗
(3)
where Keqis the equilibrium constant of the energy producing reaction (which
can be interrelated with ∆ G0),Tis temperature, and Ris the ideal gas constant.
The entire metabolism (given as a power in watts) of the entire cell is then
described by
B= 4πkRTZ
1−3Ve(rc)
4πr3cZrc
0x2A(x) lnKeqA(x)
P(x)
dx (4)
where rcis the radius of the cell and Veis the essential volume required for
other for other cellular materials, such as DNA, and is unavailable for energy
generation.
In this system, we can maximize power (free energy/time), B(Watts), as a
function of Z. If there were no tradeoffs, then Bwould be maximized by the
largest feasible Zat a given size (a cell full of enzymes). | 2403.00001v1.pdf | 69527728-14da-4b85-baae-a6afc8e31952 |
However, the need
to move substrate into different regions of the cell and to displace products
away from reaction sites introduces a trade-off between the effective diffusion
coefficient, D, and enzyme concentration, Z. An increase in Zcorresponds to a
decrease in diffusivity [25, 26], and this leads to an optimal enzyme concentration
corresponding to a maximal metabolic rate (see SI).
1 Optimal tradeoffs
Maximizing power over the whole cell, at a given cell size rcsubject to vari-
ation in the concentration of enzymes, Z, we find that the optimal enzyme
concentration Z∗follows
Z=1
e6KeqD0
kr2c1/(1−γ)
(5)
where γis the scaling between diffusivity and enzyme concentration at high
enzyme concentrations, and D0is the normalization constant for that scaling
relationship (see SI). Remarkably, this relationship shows that the optimal en-
zyme concentration in cells can be predicted from a few fundamental thermo-
dynamic, Keq, and kinetic, k, constants (Figure 1). Note that Z∗depends on
3 | 2403.00001v1.pdf | a37d4140-f667-4d9e-a678-1e3982998a11 |
a ratio of reaction rate, k, to a diffusivity normalization for small molecules in
the cytoplasm, D0.
Using the known relationship for diffusivity with enzyme concentration,
where γ≈ −3/2 for sufficiently high concentrations [26], we obtain
Z=Z0V−4/15
c . (6)
Previous studies have shown that total protein count scales with cell size
following a power law with an exponent <1 [22], but these relationships have
no fundamental explanation. Our optimization predicts that protein concentra-
tion should become more dilute as cells become larger following an exponent of
−4/15≈ −0.27, which is indistinguishable from the best fit exponent to data
of−0.30±0.06 (Figure 1).10-2110-2010-1910-1810-1710-160.0050.0100.0500.1000.500
Cell Volume(m3)Enzyme Concentration(mol/m3)PredictionBest Fit
PredictionMedian
-4-20246810020040060080010001200
Log10(k)Frequency
Figure 1: (A) The scaling of the concentration of a single enzyme with cell
volume. The solid line is the prediction from our theory (exponent = 0 .27)
and the dashed line is the best fit to data (exponent = 0 .30±0.06) compiled
across the range of bacteria (calculated from total protein abundance scaling in
[22]). (B) The distribution of kvalues for a few thousand enzymes [27] with the
median value show in red compared to the best fit of our model (dashed black
line).
The prediction for Zwas determined by optimizing metabolic rate and so
we also predict the relationship for maximum metabolic rate:
B=B0V11/15
c
1−ve
Vc
. | 2403.00001v1.pdf | bc737b29-a7a9-42de-8df7-6045ee6debf2 |
(7)
Here B0is a constant that is a complicated function of diffusivity, kinetic and
thermodynamic parameters (see SI), and Veis the volume of other essential
macromolecules which follows Ve=v0Vα
cwith α= 0.83 [22].
Metabolic scaling
Equation 7 provides excellent agreement with observed metabolic rates for
prokaryotes (Fig. 2). Metabolic rate has strong, non-power law curvature for
4 | 2403.00001v1.pdf | 0cef885f-6e74-41b3-be88-fda4844fa6ba |
10-1910-1610-1310-1010-1610-1210-8
Cell Volume(m3)Metabolic Rate(W) = 1.26β = 1/3β = 0.90βProkaryotesEukaryotesFigure 2: The prediction for prokayote (red) and eukaryote (blue) metabolic
rate along with compiled data [16]. The red dashed curve to the right of the
Prokaryote data indicates the scaling of metabolism with an exponent of 1 /3
once diffusivity has saturated to the molecular maximum value. Likewise curva-
ture of the bulk flow model at the largest sizes results from dilution of enzymes
by increasing densities of molecular motors.
small cell sizes driven by the volume V erequired for macromolecules not in-
volved in reactions, such as DNA, and this is what leads to the apparent super-
linear ( β >1) scaling previously observed for prokaryotes [16]. This curvature
quickly relaxes for larger cells due to the sublinear ( β <1) scaling of Ve.For
cell sizes up to about the volume of E. coli (≈10−18m3) the power law ap-
proximation is superlinear (SI Figure 4) with an approximate exponent of 1 .68
which agrees well with the previous estimates of 1 .75 [16].
For larger bacteria, the scaling of Bconverges on the sublinear scaling of
11/15 = 0 .73. For intermediate prokaryote cell sizes, binning the data to re-
move oversampling of certain cell volumes and scatter, shows exceptionally good
agreement between our model and the data (SI Figure 5). These results are not
in conflict with the overall superlinear fit to all of the prokaryotic data [16],
but show that this is a consequence of a complicated set of scale transitions
and shifting constraints. Thus, downstream models that derive results from the
scaling exponent of metabolic rate (e.g. [21, 22]) will gain accuracy and predic-
tive power by starting from our more complex model. For example, models of
growth which assume a power law for metabolism [21] should incorporate the
curvature described in Equation 7.
5 | 2403.00001v1.pdf | bfe88119-0e98-4d75-9856-3e43ca10357d |
Size-dependent failure of diffusion and emergence
of transport mechanisms
The trade-off between decreasing enzyme concentration and increased diffusivity
applies only for high concentrations of the enzyme. Eventually, for large enough
cells, sufficiently low enzyme concentrations are reached such that D=Dmax,
the maximum diffusivity of small molecules in water (see SI). Once diffusivity
saturates, the optimal concentration follows Z∝V−2/3
c (see SI) yielding B∝
V1/3
c(Fig. 2, dashed red curve) and a metabolic rate controlled by diffusion
across the cell membrane. Such scaling also avoids thermodynamic shutdown
(power = 0) that can occur in the center of larger cells. Shutdown can occur
when Dmaxis not fast enough to sufficiently displace reaction products and
avoid reaction reversal.These changes may result in body designs, such as large
central vacuoles, filamented rods, membrane bound organelles, and extreme
polyploidy [22, 28–30] that minimize the potential for interior thermodynamic
shutdown, but yield a relatively slow metabolic rate for prokaryotes larger than
10−17m3[22].
With further increases in cell volume, metabolic rate can only be increased
by increasing the rate of molecular displacement, as Zis well below the concen-
tration that influences diffusivity. A variety of means for increasing transport
in cells exist (e.g. [31–40]) with most attributable to the addition of structures,
such as molecular motors or transport proteins, that each generate active trans-
port of molecules or viscous bulk flow of cytoplasm over at least some local
region within the cell. We model this as an enhanced effective diffusivity of
Dtrans.We model this as an enhanced effective diffusivity of
Dtrans. For example, the movement of molecules along a cytoskeletal network
and randomly arranged regions of bulk flow with a random flow direction can
each be thought of as random walks with long single steps, and so displacement
in the simplest summary model for all of these processes can be described as
an enhanced diffusion (e.g. [31, 34–36, 39–42]). A new trade-off is introduced
between the space occupied by dedicated transport structures versus the space
devoted to catalyzing reactions (enzymes and supporting structures). | 2403.00001v1.pdf | 629eecb3-2cfd-4824-9237-fec5e5e9c9c6 |
This ef-
fective diffusivity and volume tradeoffs are captured by
D′= (1−ρ)Dmax+ρDtrans (8)
and
∂A
∂t=D′1
r2∂
∂r
r2∂A
∂r
−kcatAZ
∂P
∂t=D′1
r2∂
∂r
r2∂P
∂r
+kcatAZ0 (9)
where Dtrans is the enhanced diffusivity in the region affected by the active
transport or bulk flow, ρis the fraction of the average cellular volume that is
associated with active transport or bulk flow, and ϵis the ratio of the effec-
tive volume of active transport to the volume of the structures generating that
6 | 2403.00001v1.pdf | 5bc31627-e19a-4d96-8a65-8dbdee722b5c |
transport (e.g. the ratio of the volume of a region of enhanced transport to a
molecular motor or transport protein’s volume).
We optimize total metabolic rate (see SI) at a given cell volume as a function
ofρand find that ρgrows with cell size following a complicated function (see
SI) that scales like ≈V0.10
cfor small cell sizes and ≈V2/3
cin the limit of large
cell sizes. This density scaling implies that the volume dedicated to transport
structures scales super-linearly with overall cell size ( ≈V1.1
cto≈V5/3
c).Thus,
for cell sizes larger than 10−17m3, the effective diffusivity parameter can increase
with cell volume but at the expense of an ever-increasing proportion of cell
volume devoted to transport. This yields a size-dependent curve for maximum
metabolic power in logarithmic space (blue curve in Fig. 3) that fits the available
data for metabolic rate of Eukaryotes (again with average transport and reaction
kinetic parameters).Over the middle size range of the single cell Eukaryotes,
metabolic rate should approximate a power law with an exponent of 0 .90 which
agrees very well with the power law fit to data of 0 .90±0.17 [16, 21]. However,
the super-linear scaling of the required volume of transport molecules means
that motors and transport proteins begin to substantially reduce the volume for
reaction structures. This transport volume requirement ultimately may limit
metabolic rate, and impose a theoretical upper limit to Eukaryote size of 10−9
m3which compares well to the size of the largest unicellular organisms in the
metabolic database (Fig. 2).
Discussion
These analyses produce a series of metabolic scaling curves (Fig. 2) across a
succession of cell (or organism) volume ranges. Each curve is associated with
a particular mode of reaction volume and molecular displacement mechanisms,
with molecular diffusion at the smallest sizes then with an enhanced diffusion
within devoted transport regions at the largest sizes (Fig. 3).
Across the size spectrum, optimal metabolic rate is set by the tradeoff be-
tween volume for transport and volume for metabolic reactions. | 2403.00001v1.pdf | 86fdaef6-c7ca-48fd-bcdf-0c6f5426a701 |
For prokary-
otes, the smallest cells require space for other macromolecules such as DNA,
which limit the space available for catalytic reaction enzymes. The largest un-
structured prokaryotes face the slow scaling of metabolism set by the maximum
diffusivity of small molecules in water. For unicellular eukaryotes, transport
molecules provide the needed increase in molecular movement and metabolism
scales almost linearly with cell volume. At the largest sizes, unicellular eukary-
otes reach a sharp decrease dictated by the required overpacking of transport
structures to assure the necessary effective diffusivity throughout the cell. Col-
lectively, these curves merge to form an overall nonlinear relationship between
metabolic rate and size that fit observed data very strongly (Fig. 2) and predict
the scaling of protein and other material concentrations and the size at which
major transitions in body plans occur.
These curves clearly depart from the prediction of a single scaling exponent
of 3/4 predicted by the theory for the optimal organization of vascular networks,
7 | 2403.00001v1.pdf | b6ef04a2-9838-450b-8d04-c6ceb599c316 |
EukaryotesSmallest Bacteria - Volume Constrained
(Densely Packed Macromolecules)
-Low DiffusivityMedium Bacteria - Optimal enzyme concentration
set by tradeoff between diffusivity
and metabolismSmall/Medium Eukaryotes - Optimal transport set by tradeoff
between volume for reactions and
movementLargest Eukaryotes - Transport mechanisms
overpack the cellLarge cell without transport - Saturated diffusion
-Enzyme concentration set by
thermodynamic shutdown
Asymptotic Limits
Asymptotic LimitsSuperlinear ScalingSublinear ScalingProkaryotesβ≈0.87—DNA—Ribosomes—Enzymes—Example
Transport—Substrates
β=1/3β=11/15β>2β≈1.26
Superlinear Scaling10-2010-1710-1410-11Cell Volume(m3)rc=100nmrc=1000nmrc=105nm
—Product
Substrate
transport
(e.g. Diffusion)Product
transport
(e.g.Diffusion)Product
transport
(e.g. Diffusion)Figure 3: Representations of cells with increasing cell radius filled by cell com-
ponents DNA, enzymes, ribosomes, and eventually molecular motors [43]. The
scaling regimes and key limitations faced by cells of each category are also in-
dicated.
8 | 2403.00001v1.pdf | 554ce090-e5bd-4030-9f19-b2af300b4105 |
and support the general hypothesis that the smallest forms of life face differ-
ent physical constraints, with different accompanying solutions, for maximizing
metabolic power. However, the derived mechanisms here tend, as a first approx-
imation, to a sublinear scaling not far from 3 /4. For example, the mid range
bacteria are predicted to scale like 11 /15 = 0 .73, and the largest eukaryotes
scale like ≈.80 as curvature sets in. These scalings for both small life and for
organisms with vascular systems occur for very different mechanistic reasons,
but are linked by the fundamental need for life to allocate and organize space
to achieve sufficiently high rates of molecular displacement. Our results thus
support more general arguments for why metabolic scaling should not be either
surface area to volume scaling ( β= 2/3) or strictly proportional to volume
(isometric, β= 1).
Our work suggests that previous analyses of prokaryote metabolic scal-
ing with all of the data [16], which found super-linear scaling, requires size-
dependent shifts in physical constraints that accompany shifts in scaling to
approximate a super-linear power law.Our results demonstrate that the overall
super-linear pattern is driven by strong curvature away from a power law at the
smallest cell sizes.
Downstream extension of the fits using all data have been used to derive
other features such as the increase in growth rate with cell size and the total
abundance of ribosomes [21, 22]. Such models may gain accuracy and improved
resolution in size ranges by incorporating the more complicated metabolic scal-
ing derived here. In addition, other cellular constraint perspectives have been
proposed for explaining scaling in bacteria such as biosynthetic costs of the
membrane [44], ribosome and protein abundances and costs [22, 45], the spatial
location and number of organelles and genomes [16, 29], the increasing number
of genes and their cost [16, 29, 45, 46], and transporter optimizations associated
with the environment (e.g. [23] for a review). In addition, the deployment of
phase separation as another means for enhancing reaction rates [47, 48] and
the spatial architecture of macromolecules [49, 50] may also be important for
cellular scaling. | 2403.00001v1.pdf | 0dbf697e-1e36-4a25-b669-5c99533f892c |
These are all important biophysical, physiological, and evolu-
tionary considerations, and it will be important to integrate these additional
constraints with cellular scaling in a more complicated optimization approach
[23] that combines these costs with the diffusive tradeoffs employed here.
Large prokaryotes also deploy active molecular transport and so the two
models presented here represent bounding cases where large prokaryotes ( E.
coliand bigger) and small unicellular eukaryotes. Multicellular organisms over-
come the limit faced by the largest single cells by evolving mechanisms of bulk
flow among cells. At its extreme, optimizing such bulk flow through tubular
vascular systems to minimize energy dissipation (friction) produces the well-
known B∝M3/4[10, 11, 13, 51–53]. Taken together, our results and vascular
network models show that, organisms across the tree of life may face univer-
sal tradeoffs between biochemical reactions and the space devoted to transport,
where larger organisms must employ increasingly energy-requiring methods to
increase molecular transport velocities.
Our results also predict the scaling of materials within the cell. For small
9 | 2403.00001v1.pdf | 798a60d1-31f8-4959-8522-36176f72d61f |
entities relying on diffusion, we predict the total mass of enzymes (protein)
to scale as V11/15
c , which closely agrees with observed scaling proportional to
V0.70
c[22]. We also predict the minimum size and volume scaling at which cells
should produce molecular motors, microtubules and other cytoskeletal material
and other mechanisms to enhance diffusivity (Fig. 2). These scaling predictions
cannot yet be tested, as they predate the measurement of such volumes or
masses for diverse species and cell sizes, with just recently-developed microscopy
techniques. However, the largest prokaryotes are known for unusual storage
capacity [22, 28], many copies of the genome [29], and surprising amount of
cytoskeleton [54]. All of these characteristics may give hints to the eukaryotic
transition, and indeed Asgard archaea, which are suggested to be the eukaryotic
ancestor, have several atypical cellular morphological characteristics [54–56].Our results suggest that active transport in and/or bulk flow of the cyto-
plasm is likely required to generate free energy even in single cell organisms such
as larger Prokaryotes and single-cell Eukaryotes. Extrapolation of the required
molecular movement in our models further suggests that smaller multicellular
organisms may require mechanisms such as contraction of tissues and/or loco-
motion to generate bulk transport of intercellular fluids without an organized
single source vascular system. Extending further, branching pressurized circu-
latory networks may be required to sustain metabolism for organisms >1 g, a
hypothesis supported by evidence that smallest mechanical pumping circulatory
systems are limited by damping of pulsatile flow [10].
Our results now provide a physical and chemical explanation for the unique
metabolic scaling exhibited by the smallest 11 orders of magnitude in organism
size. We demonstrate that these life forms face significant design constraints, set
by limited space for macromolecules and diffusive molecular movement, and the
continuing need for faster molecular displacement mechanisms for the largest
unicellulars. | 2403.00001v1.pdf | 7381c53c-9d32-4543-a586-dabd94da24b5 |
The predicted scaling relationships as organisms shift across the
different transport limitation domains suggest that natural selection has ad-
dressed this sequence of problems during the evolution of larger cell body sizes.
Such evolution has led to cells with clustered reaction surfaces, channels of cyto-
plasm for transporting molecules, molecular motors and cytoskeletal structures
to enhance diffusion and then finally by mechanical pumping mechanisms that
preview the evolution of powered branched circulatory systems. Our work now
extends prior theory to provide a parsimonious hypothesis for how metabolism
and the materials that support such activity scale across all life.
References
[1] J. H. Brown, J. F. Gillooly, A. P. Allen, V. M. Savage, and G. B. West,
“Toward a metabolic theory of ecology,” Ecology , vol. 85, no. 7, pp. 1771–
1789, 2004.
[2] B. Enquist et al. , “Scaling metabolism from organisms to ecosystems,”
Nature , vol. 423, pp. 639–642, 2003.
10 | 2403.00001v1.pdf | 75ecd593-b11d-47eb-8ba9-08d6067ab221 |
[3] M. Kaspari, “Using the metabolic theory of ecology to predict global pat-
terns of abundance,” Ecology , vol. 85, no. 7, pp. 1800–1802, 2004.
[4] M. Ritchie, Scale, heterogeneity, and the structure and diversity of ecological
communities . Princeton: Princeton University, 2009.
[5] S. Pawar, A. Dell, and V. Savage, “Dimensionality of consumer search space
drives trophic interaction strengths,” Nature , vol. 486, pp. 485–489, 2012.
[6] P. Agutter and J. Tuszynski, “Analytic theories of allometric scaling,” Jour-
nal of Experimental Biology , vol. 214, no. 7, pp. 1055–1062, 2011.
[7] D. Glazier, “A unifying explanation for diverse metabolic scaling in animals
and plants,” Biological Reviews , vol. 85, pp. 111–138, 2010.
[8] K. Schmidt-Nielsen, Scaling: why is animal size so important?[8] K. Schmidt-Nielsen, Scaling: why is animal size so important? Cambridge
UK: Cambridge University Press, 1984.
[9] R. Peters, The ecological implications of body size . Cambridge UK: Cam-
bridge University Press, 1986.
[10] G. B. West, J. H. Brown, and B. J. Enquist, “A general model for the
origin of allometric scaling laws in biology,” Science , vol. 276, no. 5309,
pp. 122–126, 1997.
[11] G. B. West, J. H. Brown, and B. J. Enquist, “A general model for the struc-
ture and allometry of plant vascular systems,” Nature , vol. 400, no. 6745,
pp. 664–667, 1999.400, no. 6745,
pp. 664–667, 1999.
[12] S. Mori, K. Yamaji, A. Ishida, S. G. Prokushkin, O. V. Masyagina, A. Hag-
ihara, A. R. Hoque, R. Suwa, A. Osawa, T. Nishizono, et al. , “Mixed-power
scaling of whole-plant respiration from seedlings to giant trees,” Proceed-
ings of the National Academy of Sciences , vol. 107, no. 4, pp. 1447–1451,
2010.
[13] J. R. Banavar, J. Damuth, A. Maritan, and A. Rinaldo, “Supply–demand
balance and metabolic scaling,” Proceedings of the National Academy of
Sciences , vol. 99, no. 16, pp. 10506–10509, 2002.
[14] J. Banavar et al. , “A general basis for quarter-power scaling in animals,”
Proceedings of the National Academy of Sciences , vol. 107, pp. | 2403.00001v1.pdf | f1bfb8b3-19cc-4eb0-a757-19de85d211b4 |
107, pp. 15816–15820,
2010.
[15] J. Maino, M. Kearney, R. Nisbet, and S. Kooijman, “Reconciling theories
for metabolic scaling,” Journal of Animal Ecology , vol. 83, pp. 20–29, 2014.
[16] J. P. DeLong, J. G. Okie, M. E. Moses, R. M. Sibly, and J. H. Brown,
“Shifts in metabolic scaling, production, and efficiency across major evo-
lutionary transitions of life,” Proceedings of the National Academy of Sci-
ences , vol. 107, no. 29, pp. 12941–12945, 2010.
11 | 2403.00001v1.pdf | 9699da9a-1017-49f6-a42e-920e50983b82 |
[17] D. Glazier, “Metabolic scaling in complex living systems,” Systems , vol. 2,
no. 4, p. 451, 2014.
[18] T. Kolokotrones, V. Savage, E. Deeds, and W. Fontana, “Curvature in
metabolic scaling,” Nature , vol. 464, no. 7289, pp. 753–756, 2010.
[19] G. West and J. Brown, “The origin of allometric scaling laws in biology
from genomes to ecosystems: towards a quantitative unifying theory of
biological structure and organization,” Journal of Experimental Biology ,
vol. 208, pp. 1575–1592, 2005.
[20] A. Makarieva et al. , “Mean mass-specific metabolic rates are strikingly
similar across life’s major domains: Evidence for life’s metabolic optimum,”
Proceedings of the National Academy of Sciences , vol. 105, pp. 16994–16999,
2008.105, pp. 16994–16999,
2008.
[21] C. P. Kempes, S. Dutkiewicz, and M. J. Follows, “Growth, metabolic par-
titioning, and the size of microorganisms,” Proceedings of the National
Academy of Sciences , vol. 109, no. 2, pp. 495–500, 2012.
[22] C. P. Kempes, L. Wang, J. P. Amend, J. Doyle, and T. Hoehler, “Evo-
lutionary tradeoffs in cellular composition across diverse bacteria,” The
ISME journal , vol. 10, no. 9, pp. 2145–2157, 2016.
[23] C. P. Kempes, M. Koehl, and G. B. West, “The scales that limit: the phys-
ical boundaries of evolution,” Frontiers in Ecology and Evolution , vol. 7,
p. 242, 2019.
[24] L. Demetrius, “Quantum statistics and allometric scaling of organisms,”
Physica A: Statistical Mechanics and its Applications , vol. | 2403.00001v1.pdf | 95044ea7-b5ec-4b3e-a2c7-b553a86ff592 |
322, pp. 477–
490, 2003.
[25] D. S. Banks and C. Fradin, “Anomalous diffusion of proteins due to molec-
ular crowding,” Biophysical journal , vol. 89, no. 5, pp. 2960–2971, 2005.
[26] F. Roosen-Runge, M. Hennig, F. Zhang, R. M. Jacobs, M. Sztucki,
H. Schober, T. Seydel, and F. Schreiber, “Protein self-diffusion in crowded
solutions,” Proceedings of the National Academy of Sciences , vol. 108,
no. 29, pp. 11815–11820, 2011.
[27] A. Bar-Even, E. Noor, Y. Savir, W. Liebermeister, D. Davidi, D. S. Tawfik,
and R. Milo, “The moderately efficient enzyme: evolutionary and physico-
chemical trends shaping enzyme parameters,” Biochemistry , vol. 50, no. 21,
pp. 4402–4410, 2011.50, no. 21,
pp. 4402–4410, 2011.
[28] H. Schulz, T. Brinkhoff, T. G. Ferdelman, M. H. Marin´ e, A. Teske, and
B. B. Jørgensen, “Dense populations of a giant sulfur bacterium in namib-
ian shelf sediments,” Science , vol. 284, no. 5413, pp. 493–495, 1999.
12 | 2403.00001v1.pdf | 9186ac76-0d9a-4cce-98ec-5110f3cc4010 |
[29] N. Lane and W. Martin, “The energetics of genome complexity,” Nature ,
vol. 467, no. 7318, pp. 929–934, 2010.
[30] J.-M. Volland, S. Gonzalez-Rizzo, O. Gros, T. Tyml, N. Ivanova, F. Schulz,
D. Goudeau, N. H. Elisabeth, N. Nath, D. Udwary, et al. , “A centimeter-
long bacterium with dna contained in metabolically active, membrane-
bound organelles,” Science , vol. 376, no. 6600, pp. 1453–1458, 2022.
[31] K. Luby-Phelps, “Cytoarchitecture and physical properties of cytoplasm:
volume, viscosity, diffusion, intracellular surface area,” International review
of cytology , vol. 192, pp. 189–221, 1999.
[32] S. M. Hurtley, “Cell biology of the cytoskeleton,” 1998.[33] M. Xu, J. L. Ross, L. Valdez, and A. Sen, “Direct single molecule imaging
of enhanced enzyme diffusion,” Physical review letters , vol. 123, no. 12,
p. 128101, 2019.
[34] N. Fakhri, A. D. Wessel, C. Willms, M. Pasquali, D. R. Klopfenstein, F. C.
MacKintosh, and C. F. Schmidt, “High-resolution mapping of intracellular
fluctuations using carbon nanotubes,” Science , vol. 344, no. 6187, pp. 1031–
1035, 2014.
[35] C. P. Brangwynne, G. H. Koenderink, F. C. MacKintosh, and D. A. Weitz,
“Intracellular transport by active diffusion,” Trends in cell biology , vol. 19,
no. 9, pp. 423–427, 2009.19,
no. 9, pp. 423–427, 2009.
[36] P. C. Bressloff and J. M. Newby, “Stochastic models of intracellular trans-
port,” Reviews of Modern Physics , vol. 85, no. 1, p. 135, 2013.
[37] S. S. Mogre, A. I. Brown, and E. F. Koslover, “Getting around the cell:
physical transport in the intracellular world,” Physical Biology , vol. 17,
no. 6, p. 061003, 2020.
[38] Z. Qu, D. Schildknecht, S. Shadkhoo, E. Amaya, J. Jiang, H. J. Lee, D. Lar-
ios, F. Yang, R. Phillips, and M. Thomson, “Persistent fluid flows defined by
active matter boundaries,” Communications Physics , vol. 4, no. 1, p. 198,
2021. | 2403.00001v1.pdf | 550acf7d-2e3c-494e-ad2a-b799c990d6b7 |
4, no. 1, p. 198,
2021.
[39] M. J. Bowick, N. Fakhri, M. C. Marchetti, and S. Ramaswamy, “Symme-
try, thermodynamics, and topology in active matter,” Physical Review X ,
vol. 12, no. 1, p. 010501, 2022.
[40] M. Guo, A. J. Ehrlicher, M. H. Jensen, M. Renz, J. R. Moore, R. D.
Goldman, J. Lippincott-Schwartz, F. C. Mackintosh, and D. A. Weitz,
“Probing the stochastic, motor-driven properties of the cytoplasm using
force spectrum microscopy,” Cell, vol. 158, no. 4, pp. 822–832, 2014.
[41] R. Lipowsky, S. Klumpp, and T. M. Nieuwenhuizen, “Random walks of
cytoskeletal motors in open and closed compartments,” Physical Review
Letters , vol. 87, no. 10, p.87, no. 10, p. 108101, 2001.
13 | 2403.00001v1.pdf | fad52d6a-2575-4d70-9715-71f39aee5482 |
[42] P. Pierobon, M. Mobilia, R. Kouyos, and E. Frey, “Bottleneck-induced
transitions in a minimal model for intracellular transport,” Physical Review
E, vol. 74, no. 3, p. 031906, 2006.
[43] BioRender.com, “Created with biorender.com.”
[44] B. Trickovic and M. Lynch, “Resource allocation to cell envelopes and the
scaling of bacterial growth rate,” BioRxiv , pp. 2022–01, 2022.
[45] M. Lynch and G. K. Marinov, “The bioenergetic costs of a gene,” Proceed-
ings of the National Academy of Sciences , vol. 112, no. 51, pp. 15690–15695,
2015.
[46] S. Alberti, “Phase separation in biology,” Current Biology , vol. 27, no. 20,
pp. R1097–R1102, 2017.20,
pp. R1097–R1102, 2017.
[47] C. A. Azaldegui, A. G. Vecchiarelli, and J. S. Biteen, “The emergence
of phase separation as an organizing principle in bacteria,” Biophysical
journal , vol. 120, no. 7, pp. 1123–1138, 2021.
[48] D. Zwicker, “The intertwined physics of active chemical reactions and phase
separation,” Current Opinion in Colloid & Interface Science , p. 101606,
2022.
[49] M. Castellana, S. Hsin-Jung Li, and N. S. Wingreen, “Spatial organiza-
tion of bacterial transcription and translation,” Proceedings of the National
Academy of Sciences , vol. 113, no. 33, pp. 9286–9291, 2016.
[50] N. S. Wingreen and K. C. Huang, “Physics of intracellular organization in
bacteria,” Annual Review of Microbiology , vol. 69, pp. 361–379, 2015.69, pp. 361–379, 2015.
[51] G. B. West, J. H. Brown, and B. J. Enquist, “The fourth dimension of life:
fractal geometry and allometric scaling of organisms,” science , vol. 284,
no. 5420, pp. 1677–1679, 1999.
[52] V. M. Savage, L. P. Bentley, B. J. Enquist, J. S. Sperry, D. Smith, P. B.
Reich, and E. Von Allmen, “Hydraulic trade-offs and space filling enable
better predictions of vascular structure and function in plants,” Proceedings
of the National Academy of Sciences , vol. 107, no. 52, pp. 22722–22727,
2010.
[53] A. B. Brummer, V. M. Savage, and B. J. Enquist, “A general model for
metabolic scaling in self-similar asymmetric networks,” PLoS computa-
tional biology , vol. 13, no. 3, p. e1005394, 2017. | 2403.00001v1.pdf | 0e0a640e-1e0b-414b-b2fc-001d6f78a4ae |
13, no. 3, p. e1005394, 2017.
[54] T. Rodrigues-Oliveira, F. Wollweber, R. I. Ponce-Toledo, J. Xu, S. K.-M.
Rittmann, A. Klingl, M. Pilhofer, and C. Schleper, “Actin cytoskeleton
and complex cell architecture in an asgard archaeon,” Nature , vol. 613,
no. 7943, pp. 332–339, 2023.
14 | 2403.00001v1.pdf | 4c77c979-35da-4038-bc9b-79eada745799 |
[55] B. Avcı, J. Brandt, D. Nachmias, N. Elia, M. Albertsen, T. J. Ettema,
A. Schramm, and K. U. Kjeldsen, “Spatial separation of ribosomes and
dna in asgard archaeal cells,” The ISME Journal , vol. 16, no. 2, pp. 606–
610, 2022.
[56] H. Imachi, M. K. Nobu, N. Nakahara, Y. Morono, M. Ogawara, Y. Takaki,
Y. Takano, K. Uematsu, T. Ikuta, M. Ito, et al. , “Isolation of an archaeon
at the prokaryote–eukaryote interface,” Nature , vol. 577, no. 7791, pp. 519–
525, 2020.
2 Supplementary Material
15 | 2403.00001v1.pdf | a31ef284-41ec-453c-a83e-eda856012431 |
3 Diffusions Tradeoffs
The coupled diffusion equations to be solved for steady state are
D(Z)
x2A′′(x) + 2xA′(x)
x2−kZA(x)!
= 0 (10)
D(Z)
x2P′′(x) + 2xP′(x)
x2+kZA(x)!
= 0 (11)
with the boundary conditions
A(rc) = A0 (12)
A′(0) = 0 (13)
P(rc) = P0.(14)
The solution to these equations is
A(r) =A0rccsch
rc√
kZ√
D(Z)
sinh
r√
kZ√
D(Z)
r(15)
Following a similar derivation, the corresponding equation for Pis
P(r) =A0+P0−A0rccsch
rc√
kZ√
D(Z)
sinh
r√
kZ√
D(Z)
r(16)
Returning to the integral which represents the full metabolic rate of the
cell, can can use these solutions for AandPset constraints on Z. To avoid
thermodynamic shutdown we needKeqA(r)
P(r)>1 which, given the solutions above,
leads to the condition
rcsch
r√
kZp
D(Z)!
<A0(1 +Keq)rccsch
rc√
kZ√
D(Z)
A0+P0. | 2403.00001v1.pdf | 78d7ddf6-6bd2-4492-8992-844fe3071d35 |
(17)
Thermodynamic shutdown is a problem for the interior of the cell and so we are
interested in checking this condition for r= 0. Evaluating the above inequality
atr= 0 and expanding in small rcleads to
Z
D(Z)<6(A0Keq−P0)
A0kr2c(1 +Keq). (18)
This equation isn’t fully solved because the diffusivity, D, depends on Z. Previ-
ous work has shown that for large enough Zchanges in diffusivity will approx-
imately follow a power law [26], and so we can consider D(Z) =D0Zγwhich
yields
Z <6D0(A0Keq−P0)
A0kr2c(1 +Keq)1
1−γ
(19)
16 | 2403.00001v1.pdf | e8b6ec9a-8366-4c9a-a69c-1133e5133761 |
which for P≈0 is
Z <6D0Keq
kr2c(1 +Keq)1
1−γ
. (20)
This equation states how the concentration of enzyme should either concentrate
or dilute with cell size to avoid thermodynamic limitations.
Given that γ≈ −3/2 [26] we have that Z∝r−4/5
c or
Z∝V−4/15
c . (21)
It should be noted that given Equation 20, Zis bounded by a concentration
that is diluting out as cells increase in size. This should set a hard upper bound
where the cellular concentrations start to reach discrete numbers of enzymes.
Thus, under a fixed tradeoff for diffusivity an upper bound on cell volume is
predicted by the dilution of enzyme concentrations required to avoid thermody-
namic shutdown.
It should be noted that once Zreaches sufficiently low concentrations diffu-
sivity saturates to a maximum value set by molecular motion in the fluid [26].At this point D(Z) =Dmaxand we have
Z <6DmaxA0Keq
A0kr2c(1 +Keq). (22)
In this limit Zwill scale like V−2/3
c to avoid thermodynamic shutdown.
Thermodynamic Optimization
More generally, we are interested in the concentrations of enzymes that would
optimize cellular metabolic rate. The above arguments give us a bounding
expectation for the scaling of enzyme concentration, but we can replace these
with the complete optimization of cellular metabolic rate, B. Specifically, we
are interested in∂B
∂Z= 0. (23)
where the total metabolism integrated over a cell is
B= 4πRTkZ
1−3Ve(rc)
4πr3cZrc
0r2A(r) logKeqA(r)
P(r)
dr. (24)
Considering cases where the environmental concentration of the product is low,
P0≈0, the integrand is given by
A0rcrZcsch √
kZrcp
D(Z)!
sinh √
kZrp
D(Z)! | 2403.00001v1.pdf | df015ffe-6ce6-468d-852f-89a7a93fdb35 |
sinh √
kZrp
D(Z)!
log
Keqrc
rsinh√
kZrc√
D(Z)
csch√
kZr√
D(Z)
−rc
(25)
17 | 2403.00001v1.pdf | b3440c04-fbfe-45b3-8484-f20b235869fb |
We approximate this integral considering small rcand with, D(Z) =D0Zγ,
the dependence of diffusivity on Z. Given these assumptions integrand is well
approximated by
A0√
krcr2z3/2csch√
kZrc√D0zγ
log
Keqrc
√
D0Zγsinh √
kZrc√
D0Zγ!
√
kZ−rc
√D0Zγ(26)
leading to the solution of the integral as
B=A0RTk3/2rcZ3/2
4πr3
c−3Ve
csch√
kZrc √D0Zγ
log
Keqrc
√
D0Zγsinh √
kZrc√
D0Zγ!
√
kZ−rc
3√D0Zγ.(27)
In the small radius limit the derivative, ∂B/∂Z , is then well approximated by
∂B
∂Z≈A0RTkV e
1−γ−log6KeqD0zγ−1
kr2c
. (28)
Setting this to zero leads to
Zopt=1
e6KeqD0
kr2c1/(1−γ)
. (29)
This optimal solution of Zoptcan be substituted back into the approximate solu-
tion for total metabolic rate, B, which we find does a good job of approximating
the full numerical integral. If we take γ=−3/2 this gives
Zopt=62/5K2/5
eqD2/5
0
ek2/5r4/5
c. (30)
This solution shows that the optimal Zscales like
Z∝r−4/5
c∝V−4/15
c (31)
in agreement with our considerations above for thermodynamic shutdown. | 2403.00001v1.pdf | a89c8894-1494-470c-8e6f-ffcc18f4fed1 |
The
corresponding solution for metabolic rate is given by
B=29/10A0RTk3/5K9/10
eqD2/5
0csch√
6√
Keq
e5/4
4πr3
c−3Ve
log
K3/2
eq
e5/4sinh √
6√
Keq
e5/4!
√
6−√
Keq
10√
3e9/4r4/5
c
(32)
which scales like
B≈r11/5
c∝V11/15
c (33)
18 | 2403.00001v1.pdf | 9a57cf59-55d6-4b02-b6c5-bc39617c6102 |
predicting that metabolic rate will scale slightly sublinearly for sufficiently large
size. It should be note that this is the approximate scaling of Bover a range
of larger cell sizes, but the full form of Bis more complicated a small cells
sizes, especially considering the factor of (1 −Ve(rc)/Vc) multiplying the entire
scaling where Ve=v0Vα
cwith α= 0.83 [22]. Figure 4 shows the full solution of
Bwith more strongly superlinear scaling because of the small-size curvature.
10-2010-1910-1810-1710-1610-15012345
Cell Volume(m3)Effective Scaling Exponent(β)
Figure 4: The approximation of the local scaling exponent across the range of
bacterial volumes. The dashed line is 11 /15 which the complete model converges
on for large cell volumes.
Saturated Diffusivity
It should be noted that at very low Zthe diffusivity saturates to a constant
maximum behavior and stops following a power law.In this case diffusivity
reaches the maximum molecular diffusivity, D=Dmax, and the above equations
lead a optimal enzyme concentration of
Zopt=6DmaxKeq
ekr2c(34)
and corresponding metabolic rate of
B=2√
6A0RTD maxK3/2
eqcschq
6
ep
Keq
4πr3
c−3Ve
log
1√e
6sinh√
6
e√
Keq
K3/2
eq−1
Keq
e3/2r2c.
(35)
Here enzyme concentration will scale like Z∝V−2/3
c andB∝V1/3
c. However,
these scalings are unlikely to be observed because of the shift to employing active
19 | 2403.00001v1.pdf | 1b4db66b-46cb-4d76-a177-dd27710713d7 |
transport, and these results are only shown to illustrate the extreme limitation
faced by large cells if they only used molecular diffusion.
10-2110-2010-1910-1810-1710-1610-1510-1910-1610-1310-10
Cell Volume(m3)Metabolic Rate(W)
Figure 5: Comparison of the model prediction (red curve) with bacterial data
(red points) and the binned averages of those data (black points) same data.
4 Active Transport Tradeoffs
Our model for active transport treats diffusion in the cell as
∂A
∂t=D′1
r2∂
∂r
r2∂A
∂r
−kcatAZ
∂P
∂t=D′1
r2∂
∂r
r2∂P
∂r
+kcatAZ0 (36)
with a modified effective diffusivity, D′, that follows
D′= (1−ρ)Dmax+ρDtrans. (37)
where ρis the fraction of cell volume occupied by transport structures.(37)
where ρis the fraction of cell volume occupied by transport structures. Here
we define an effective volume (region with enhanced transport) of a transport
structure, and ρis the ratio of the total of these effective volumes to the total cell
volume. The optimization of metabolism now depends on tradeoffs between the
adjusted diffusivity and the space occupied by the transport structures, where
total metabolism follows
B= 4πRTkZ (1−ρϵ)Zrc
0r2A(r) logKeqA(r)
P(r)
dr. (38)
where ϵis the ratio of the volume of a transport structure to the effective volume
of enhanced transport that structure creates. The solutions for A(r) and P(r)
are the same as in Equations 15 and 16.
20 | 2403.00001v1.pdf | d0f3167f-bfea-4e31-a161-0b8c2fb15728 |
We are again interested in optimizing cellular metabolic rate, B, where now
we are looking for solutions to
∂B
∂ρ= 0. (39)
The integration of Bis well approximated by
B≈C(1−ρϵ)csch√
kZrc √
ρDtrans +Dmax(1−ρ)
log
Keqrc√
kZ√
ρDtrans +Dmax(1−ρ) sinh√
krc√
Z√
ρDtrans +Dmax (1−ρ)
−rc
3p
ρDtrans +Dmax(1−ρ).
(40)
C≡4πA0RTk3/2r4
cZ3/2(41)
for small values of rc. Again we are considering the regime where transport,
now represented by the adjusted diffusivity D′, is large relative to the scale of
the cell. | 2403.00001v1.pdf | c4c8a8b6-5533-48ca-8598-db508c4355e0 |
Taking the derivative of this this solution and again approximating it
for small rcand also for small Dmax(the maximum molecular diffusivity) we
have
∂B
∂ρ≈r3
cA0RTkZ
1−ρϵ
1 + log
6Dtrans Keqρ
kr2cZ
3ρ= 0. (42)
which gives the optimal fraction of transport structures as
ρ=1
ϵW
6eDtrans Keq
kr2cZϵ (43)
where Wis the Lambert Wfunction (the product logarithm). For small rcthis
is well approximated by
ρ=1
ϵlog
6eDtrans Keq
kr2
cZϵ (44)
Similar to our solutions above for the enzyme concentration tradeoffs in bacteria,
this function again highlights that optimal metabolic values depend onDKeq
kr2c.Taking all of this together we have that the optimal metabolic rate follows
B=1
3A0RTkr3
cZ
1 + 1 /log6DtransKeq
kr2cZϵ
log
6DtransKeq
kr2cZϵ
1 + log
6Dtrans Keq
kr2
cZϵ
.
(45)
This scales like B∝r3
c∝Vcwith a logarithmic correction term. This explains
why the scaling of metabolism in Eukaryotes is close to linear with cell volume
but actually sublinear due to the logarithmic terms. Figure 6 gives the approx-
imate local exponent of Bwith cell size. The full range of Eukaryotes is well
approximated by β= 0.90 which agrees with data as discussed in the maintext.
21 | 2403.00001v1.pdf | 9bfd9766-9f32-404c-9238-d44c1f430b49 |
10-2010-1810-1610-1410-1210-100.50.60.70.80.91.0
Cell Volume(m3)Effective Scaling Exponent(β)Figure 6: The approximation of the local scaling exponent across the range of
bacterial volumes. The dashed line is 11 /15 which the complete model converges
on for large cell volumes.
This function also comes with an asymptotic limit that occurs because trans-
portors begin to overpack the cell. This limit occurs when
W6eDtransKeq
kr2cZϵ
= 1 (46)
which corresponds to an maximum single cell eukaryote of
Vc,max =8√
6πD3/2
transK3/2
eq
k3/2Z3/2ϵ3/2. (47)
This solution mirrors the molecular diffusion limit but with an altered diffusivity.
22 | 2403.00001v1.pdf | 84441c0a-a677-4044-97f7-d6051818429a |
arXiv:2403.00002v1 [math.DG] 24 Dec 2023Gradient Estimate for Fisher-KPP Equation on
Finsler metric measure spaces
Bin Shen and Dingli Xia
Abstract
In this manuscript, we study the positive solutions of the Fi nslerian Fisher-KPP
equation
ut= ∆∇uu+cu(1−u).
The Fisher-KPP equation is widely applied and connected tom any mathematical
branches. We establish the global gradient estimates on com pact Finsler metric
measure manifold with the traditional CD(K,N)condition, which is developed
by S. Ohta and K.-T. Sturm in Finsler geometry. Furthermore, With the assis-
tance of a new comparison theorem developed by the first autho r, we also give
the gradient estimate on forward complete noncompact local ly finite misalign-
ment Finsler metric measure spaces with the mixed weighted c urvature bounded
below and some non-Riemannian curvatures norm-bounded.Keywords: gradient estimate, Fisher-KPP equation, Finsler metric me asure space,
weighted Ricci curvature, mixed weighted Ricci curvature
MSC Classification: 35K55 , 53C60 , 58J35
1 Introduction
.
The Fisher-KPP equation on a complete Riemannian manifold Mis given by
ut= ∆u+cu(1−u), (1.1)
whereuis a real-valued function on M×[0,∞) andcis a positive constant.
The equation was proposed by R. A. Fisher in 1937 to describe the pr opagation
of an evolutionarily advantageous gene in a population, and was also in dependently
describedinaseminalpaperbyA.N.Kolmogorov,I.G.Petrovskii,a ndN.S.Piskunov
in the same year; for this reason, it is often referred to in the litera ture as the name
of Fisher–KPP equation.
1 | 2403.00002v1.pdf | d3c78856-0260-4177-89f7-d0e67eb0af91 |
Since the two papers in 1937, there have been extensive investigat ions on travel-
ing wave solutions and asymptotic behavior in terms of spreading spe eds for various
evolution systems. Traveling waves were adopted to study the non linear PDEs,
such as the nonlinear reaction-diffusion equations modeling physical and biological
phenomena (cf. [ 22][23]), the integral and integrodifferential population models (cf.
[2][7][15][16][30]), the lattice differential systems (cf. [ 6][11][12][14][21][40]), and the
time-delayed reaction-diffusion equations (cf. [ 29][38]).
In 2017, Cao et al. [ 10] derived differential Harnack estimates for positive solutions
to (1.1) on Riemannian manifolds with nonnegative Ricci curvature. The idea comes
from [8][9], in which a systematic method was developed to find a Harnack inequa lity
for geometric evolution equations.Actually, they obtained the follo wing theorem.
Theorem A. [10]Let(M,g)be ann-dimensional complete noncompact Riemannian
manifold with nonnegative Ricci curvature, and let u(x,t) :M×[0,∞)→Rbe a
positive solution to ( 1.1), where uisC2inxandC1int. letf=logu, then we have
∆f+α|∇f|2+βef+φ(t)≥0. | 2403.00002v1.pdf | eb19430e-f9ba-4953-897a-0381fe16b3e2 |
(1.2)
for all xandt, provided that 0< α < 1as well as−cn(2+√
2)
4(1−α)<
β <min{−cn(1+α)
4α2−4α+2n,−cn(2−√
2)
4(1−α)}<0, where φ(t) =u(e2µwt
v−w−1
µ+w)
1−e2µwt,withµ=
βc√
2(1−α)
c(−cn−8β(1−α)), v=(
4β(1−α)
n+c)
·√
2(1−α)
c(−cn−8β(1−α))andw=√
2(1−α)
n.
Utilizing Theorem A, one can integrate along space-time curves to ge t a Harnack
inequality. However, it is different from the classical Li–Yau Harnack inequality [ 20]
in form. Gradient estimates play an important role in studying elliptic an d parabolic
operators.Gradient estimates play an important role in studying elliptic an d parabolic
operators. The method originated first in [ 44] and [13], and was further developed
by Li and Yau [ 20], Li [19], Negrin [ 24], P. Souplet and Q. Zhang [ 37], Y. Yang [ 43],
etc.. Recent gradient estimates under the geometric flow include [ 4] and [3]. For more
results on the nonlinear PDEs, one may refer to [ 1][18].
In 2018,following the line in [ 19], X. Geng and S. Hou [ 17] provedthe positive solu-
tions to the Fisher–KPP equation on complete Riemannian manifolds. T hey derived a
gradient estimate, and got the classic Harnack inequality by using it, which extended
the recent result of Cao, Liu, Pendleton and Ward [ 10].In2009,uponthefoundationalinvestigationsofgradientestimat iononRiemannian
manifolds, S. Ohtainitially put forwardasophisticatedframeworkfo r gradientestima-
tion applicable to the heat equation on compact Finsler manifolds [ 25]. Subsequently,
in the year2014,C.Xia [ 41]undertookacomprehensiveanalysisofharmonicfunctions
within the context of both compact and non-compact (specifically, forward complete)
Finsler structures by utilizing an advanced form of Moser’s iterative technique. Pro-
gressingfurther,Q.Xia[ 42] deliveredintricategradientestimatesforpositivesolutions
to the heat equation on forward complete Finsler spaces, employing methodologies
analogous to those documented in [ 41]. The first author [ 32] proposed a method for
global and local gradient estimates on Finsler metric measure space s, which was also
used to solve the Finslerian Schr¨ odinger equation.Later, the firs t author [ 31] gives the
gradient estimates of bounded solutions to the Finslerian Allen-Cahn equation.
2 | 2403.00002v1.pdf | 7852a098-638a-4ed5-818d-3d7732d96791 |
In this paper, we study the Fisher-KPP equation ( 1.1) on a Finsler metric measure
manifold ( M,F,µ), where cis a positive constant, the solution of ( 1.1) is a function
u(x) onM×[0,∞). Our main theorems in this paper are as follows
Theorem 1.1. Let(M,F,µ)be a compact Finsler metric measure space with dimen-
sionaln≥2, whose weighted Ricci curvature satisfies RicN≥ −K, for some positive
constant K. Assume the reversibility of Mhas upper bound ρ0. Suppose uis a bounded
positive smooth solution of the Fisher-KPP parabolic equat ion (1.1) onM×[0,∞),
then we have
F2(∇u)
u2+2c(1−u)−2ut
u≤2(N+n)ρ2
0K+27√
2(N+n)3
2M1c.(1.3)
where,M1= supu(x,t).(1.3)
where,M1= supu(x,t).
Similarily, on a noncompact forward complete Finsler manifold, the follo wing local
gradient estimates is obtained.
Theorem 1.2. Let(M,F,µ)be a complete noncompact Finsler metric measure space.
Denote by B(p,2R)the forward geodesic ball centered at pwith forward radius 2 R.
Suppose the mixed weighted Ricci curvaturemRicN≥ −K(2R)inB(p,2R)with
K(2R)≥0, and the misalignment αsatisfies α≤A(2R)inB(p,2R). Moreover, the
non-Riemannian tensors satisfy F(U)+F∗(T)+F(divC(V))≤K0.Assume u(x,t)
is a bounded positive smooth solution of the Fisher-KPP para bolic equation ( 1.1) on
M×[0,∞), then we have
F2
∇u(∇u)
u2+sc(1−u)−sut
u≤C1(N,n,ǫ)(s2
t+s2K(2R)
(s−1)+sc
qM1)
+C2(N,n,ǫ,A (2R))s2
R2{
1+R[
1+√
K(2R)]
+s2
s−1}
.(1.4)
onB(p,R)for0< ε <1,s >1,q >0, such that2(1−ε)
N−ε(N−n)s−1
sq≥1
ε−1+(2s−1)2
8,
whereC1,C2are positive constants.
2 Preliminary
.
In this section, we introduce the fundamental principles governing Finsler metric
measure spaces (refer to [ 32] for more details). | 2403.00002v1.pdf | 45fdda5d-6b8e-45c2-8308-be68ab4b53b7 |
A Finsler metric space is a triple
(M,F,µ), which indicatesthatadifferentialmanifold isequipped with aFinslerm etric
Fand a measure µ. Suppose the local coordinate of the tangent bundle is ( x,y), where
xis the point on Mandyis the direction on TxM. A Finsler metric Fis a nonnegative
function F:TM→[0,+∞) satisfies that
(i)Fis smooth and positive on TM\{0};
(ii)Fis a positive homogenousnorm, i.e., F(x,ky) =kF(x,y) for any ( x,y)∈TM
and for any k >0;
3 | 2403.00002v1.pdf | 026e82b5-38a4-434d-a069-00c59bb78a11 |
(iii)Fis strongly pseudo-convex, namely, for any ( x,y)∈TM\ {0}, the
fundamental tensor is a positive definite matrix defined by
gij(x,y) :=1
2∂F2
∂yi∂yj(x,y). (2.1)
Unlike the Riemann metric, the Finsler metric is defined locally as the nor m on
the tangent space at each point, and globally as a metric on the pull-b ack bundle, so
there are a large number of non-Riemannian geometric quantities on a Finsler metric
measure space. The Cartan tensor is defined by
C(X,Y,Z) :=CijkXiYjZk=1
4∂3F2(x,y)
∂yi∂yj∂ykXiYjZk, (2.2)
for any local vector fields X,Y,Z.
There is a unique almost g-compatible and torsion-freeconnection o n the pull-back
tangent bundle π∗TMof the Finsler manifold ( M,F) called the Chern connection.It
is determined by
∇XY−∇YX= [X,Y]; (2.3)
w(gy(X,Y))−gy(∇ZX,Y)−gy(X,∇ZY) = 2Cy(∇Zy,X,Y),(2.4)
for anyX,Y,Z∈TM\ {0}, whereCyis the Cartan tensor. The Chern connection
coefficients are locally denoted by Γi
jk(x,y) in a natural coordinate system, which
could induce the spray coefficients as Gi=1
2Γi
jkyjyk. The spray is given by
G=yiδ
δxi−2Gi∂
∂yi, (2.5)
in whichδ
δxi=∂
δxi−Nj
i∂
∂yjand the nonlinear connection coefficients are locally
induced from the spray coefficients by Ni
j=∂Gi
∂yj. | 2403.00002v1.pdf | 0ac89247-3730-441d-9c35-98f3ef7bd532 |
The Chern connection can define the Chern Riemannian curvature Rand Chern
non-Riemannian connection P. Denote by Ω the curvature form of the Chern
connection, so that
Ω(X,Y)Z=R(X,Y)Z+P(X,∇Yy,Z), (2.6)
for anyX,Y,Z∈TM\{0}, where locally
Ri
j kl=δΓi
jl
δxk+δΓi
jk
δxl+Γi
kmΓm
jl−Γi
lmΓm
jk, Pi
j kl=−∂Γi
jk
∂yl. (2.7)
Customarily, we denote the horizontalChern derivativeby ” |” and the verticalChern
derivative by ”;”. For example,
vi|j=δ
δxjvi−Γk
ijvk, vi;j=∂
∂yjvi, (2.8)
4 | 2403.00002v1.pdf | 4719d2fe-00c0-46d0-87d1-e14f02f0f938 |
for any 1-form v=vidxion the pull-back bundle.
The angular metric form hyis defined by
hy(u,v) =gy(u,v)−1
F2(y)gy(y,u)gy(y,v), (2.9)
for anyy,u,v∈TxMwithy̸= 0 Thus, for any two linearly independent vectors
y,v∈TxM\ {0}, which span a tangent plane Π y=span{y,v}, the flag curvature
with pole yis defined by
K(P,y) =K(y,x) :=Ry(y,u,u,y)
F2(y)hy(u,u), (2.10)
which is locally expressed by
K(y,u) =−Rijklyiujykul
(gikgjl−gilgjk)yiujykul.(2.11)
The Ricci curvature is defined by
Ric(v) :=F2(v)n−1∑
i=1K(y,ei), (2.12)
wheree1,...,e n−1,v
F(v)form an orthonormal basis of TxMwith respect to gy. The
Landsberg curvature of ( M,F) is given by
L:=Li
jk∂i⊗dxj⊗dxk, Li
jk=−yjPi
j kl. (2.13)
By the zero homogeneity, according to the Euler lemma, it easy to se e that
Cijkyi=Lijkyi= 0 (2.14)
H.B. Rademacher introduced the concepts of reversibility and reve rsible manifolds
[28], which are also closely related to the analytical assumptions on Finsle r manifolds. | 2403.00002v1.pdf | 4e02d8ac-5db1-4c14-851b-8b82b49e528f |
A Finsler metric is defined to be reversible if F(x,V) =¯F(x,V) for any point xand
any vector field V, where¯F(x,V) :=F(x,−V) is called the reversed Finsler metric of
F.
ρ:=sup
x∈Msup
V̸=0F(x,V)
F(x,V). (2.15)
Obviously, Fis reversible if and only if ρ≡1. A Finsler manifold ( M,F) is said
to have finite reversibility if ρ <+∞.
Later, K. Ball, E. Carlen, and E. Lieb introduced the concepts of un iform smooth-
ness and uniform convexityin Banachspace theory[ 5], whose geometricinterpretation
in Finsler geometry was provided by S. Ohta [ 26]. We say Fsatisfies uniform convex-
ity and uniform smoothness if there exist uniform positive constant sκ∗andκ, called
5 | 2403.00002v1.pdf | 49f94444-dce0-484f-8d3e-e81dc26ea682 |
the uniform convexity constant and the uniform smoothness cons tant, respectively,
such that for any x∈M,V∈TxM\{0}, andW∈TxM, we have
κ∗F2(x,y)≤gV(y,y)≤κF2(x,y). (2.16)
wheregV= (gij(x,V)) is the Riemannian metric on Minduced from Fwith
respect to the reference vector V. In this situation, the reversibility ρcould be
controlled by κandκ∗as
1≤ρ≤min{√κ.√
1/κ∗}. (2.17)
Fis Riemannian if and only if κ= 1 if and only if κ∗= 1 [26].
The Riemannian structure is inconsistent when the reference vect ors are different.
For example, given three different local non-vanishing vector fields aroundx, namely,
V,W,Y, the norm of YaboutgVandgWmaybe not the same in general case.The
ratiogV(Y,Y) =gW(Y,Y) is a function about V,W,Y. Based on this fact, [ 32]
defined an important constant on a Finsler manifold, called the misalign ment.
Definition 2.1. ([32]). For a Finsler manifold (M,F), the misalignment of a Finsler
metric at point xis defined by
α(x) =sup
V,W,Y∈SxMgV(Y,Y)
gW(Y,Y). (2.18)
Moreover, the global misalignment of the Finsler metric is d efined by
α=sup
X∈MαM(x) =sup
X∈Msup
V,W,Y∈SxMgV(Y,Y)
gW(Y,Y). (2.19)
it also provided in [ 32] some characterizations of the misalignment. Especially, a
Finsler manifold ( M,F) is a Riemannian manifold if and only if αM= 1.Moreover,
a Finsler manifold ( M,F) is uniform convexity and uniform smoothness if and only if
it satisfies finite misalignment.
Since that, we have gaven an important class of Finsler manifold as th e following.
Definition 2.2. We call a Finsler manifold (M,F)has finite misalignment if there
is a positive constant Asuch that α≤A, and has locally finite misalignment if for
any compact subset Ω⊂M, there is a constant A(Ω)depending on Ωsuch that
α(x)≤A(ω)for anyx∈Ω.
So far, we have briefly introduced some Riemannian or non-Riemannia n local
quantities and tensors in Finsler geometry corresponding to Rieman nian geometric
quantities. Next, we will introduce more tensors related to the mea suredµ. | 2403.00002v1.pdf | f481315d-cd10-4e50-ad31-91223da2cb75 |
Next, we will introduce more tensors related to the mea suredµ.
For any smooth function f:M→R, df denotes its differential 1-form and its
gradient ∇fis defined as the dual of the 1-form via the Legendre transformat ion,
namely,∇f(x) :=l−1(df(x))∈TxM. Locally it can be written as
∇f=gij(x.∇f)∂f
∂xi∂
∂xj. (2.20)
onMf:=df̸= 0. The Hessian of fis defined via the Chern connection by
6 | 2403.00002v1.pdf | d882fddf-5f39-48ad-a402-8775093431be |
∇2f(X,Y) =g∇f(∇∇f
X∇f,Y). (2.21)
It can be shown that ∇2f(X,Y) is symmetric [ 27].
For any two points p,qonM, the distance function is defined by
dp(q) :=d(p,q) := inf
γ∫1
0F(γ(t),˙γ(t))dt. (2.22)
where the infimum is taken overall the C1curvesγ: [0,1]→Msuch that γ(0) =p
andγ(1) =q. Fixed a base point ponM, we denote the forward distance function by
r. That is, r(x) =d(p,x), with d denotes the forward distance.
The forward distance function ris a function defined on the Finsler manifold M.
dris a 1-form on M, whose dual is a gradient vector field, noted by ∇r.Precisely,
∇r=gij(x,∇r)∂r
∂xi∂
∂xjTakingtheChernhorizontalderivativeof ∇ryieldstheHessian
of distance function ∇2r. Locally, in a natural coordinate system
∇2r(∂
∂xi,∂
∂xj) =∇j∇ir=∂2r
∂xi∂xj−Γk
ij(x,∇r)∂r
∂xk. (2.23)
In (2.23), the derivative is taken in the direction ∇rnaturally. Generally, we can
take the derivative in any direction. Suppose Vis a local vector field around xonM.
Note that the distance function may not be symmetric about pandqunlessFis
reversible. AC2curveγis called a geodesic if locally
¨γ(t)+2Gi(γ(t),˙γ(t)) = 0. (2.24)
whereGi(x,y) are the spray coefficients. | 2403.00002v1.pdf | 5bf8072a-1ff8-4f7d-9de7-90b5caf83a87 |
A forward geodesic ball centered at p
with radius Rcan be represented by
B+
R(p) :={q∈M:d(p,q)< R} (2.25)
Adopting the exponential map, a Finsler manifold ( M,F) is said to be forward
complete or forward geodesically complete if the exponential map is d efined on the
entireTM. Thus, any two points in a forward complete manifold Mcan be connected
by a minimal forwardgeodesic. Moreover,the forwardclosed balls B+
R(p) are compact.
A Finsler metric measure space ( M,F,dµ) is a Finsler manifold equipped with a
given measure µ. In local coordinates {xi}n
i=1, we can express the volume form as
dµ=σ(x)dx1...dxn. For any y∈TxM\{0}, define
τ(x,y) :=log√
detgij(x,y)
σ(x). (2.26)
which is called the distortion of ( M,F,dµ). The definition of the S-curvature is
given in the following.The definition of the S-curvature is
given in the following.
Definition 2.3. ([35][34]). Suppose (M,F,dµ)is a Finsler metric measure space. For
any point x∈M, letγ=γ(t)be a forward geodesic from x with the initial tangent
vector˙γ(0) =y. The S-curvature of (M,F,dµ)is
S(x,y) :=d
dtτ=d
dt(1
2logdet(gij)−logσ(x))(γ(t),˙γ(t))|t=0.(2.27)
7 | 2403.00002v1.pdf | 29225659-1d05-40cc-937a-c0b0fedbb585 |
Definition 2.3means that the S-curvature is the changing of distortion alo ng the
geodesic in direction y. Modeled on the definition of Tcurvature in [ 36], [32] defined
the difference of ∇τon the tangent sphere, denoted by T,
Definition 2.4. ([32]). The difference of ∇τon the tangent bundle is a tensor denoted
byT, which is given by
T(V,W) :=∇Vτ(V)−∇Wτ(W). (2.28)
for vector fields V,WonM. Locally, it is T(V,W) =Ti(V,w)dxi, with
Ti=δ
δxiτk(V)−δ
δxiτk(W). (2.29)
Obviously, T(V,W) is anti-symmetric about VandW, that is, T(V,W) =
−T(W,V) for any nonvanishing VandW.If in local coordinates {xi}n
i=1, expressing dµ=eφdx1...dxn, the divergence of a
smooth vector field Vcan be written as
divdµV=n∑
i=1(∂Vi
∂xi+Vi∂φ
∂xi). (2.30)
The Finsler Laplacian of a function fonMcould now be given by ∆ dµf:=
divdµ(∇f), noticing that ∆ dµ= ∆∇ff, where ∆∇f
dµf:=divdµ(∇∇ff) is in the view
of weighted Laplacian with
∇∇ff:={
gij(x,∇f)∂f
∂xi∂
∂xjfor x∈Mf;
0 for x /∈Mf.(2.31)
A Finsler Laplacian is better to be viewed in a weak sense due to the lack of
regularity. | 2403.00002v1.pdf | b14e9081-b0a3-4f3a-a35c-e8a416f9ba1d |
Concretely, assuming f∈W1,p(M),
∫
Mφ∆dµfdµ=−∫
Mdφ(∇f)dµ, (2.32)
for any test function φ∈C∞
0(M).
On the other hand, the Laplacian of a function fon a Riemannian manifold is
the trace of the Hessian of fwith respect to the Riemannian metric g. On a Finsler
metric measure space ( M,F,dµ), the weighted Hessian ˜H(f) of a function fonMf=
{x∈M:∇f|x̸= 0}is defined in [ 39] by
˜H(f)(X,Y) =∇2f(X,Y)−S(∇f)
n−1h∇f(X,Y), (2.33)
whereh∇fis the angular metric form in the direction ∇f, given in ( 2.9) It is clear
that˜H(f) is still a symmetric bilinear form with
∆f=tr∇f˜H(f).(2.34)
8 | 2403.00002v1.pdf | 913581dd-02ea-4be7-8f6d-bc86d24fdf90 |
Inspired by this, [ 32] defined the mixed weighted Hessian
˜HV(f)(X,Y) =∇2(f)(X,X)−S(∇f)
n−1hV,∇f(X,X), (2.35)
wherehV,∇fis the mixed angular metric form in the directions Vand∇f, which is
defined by
hV,∇f(X,Y) =gV(X,Y)−1
F2
V(∇f)gV(X,∇f)gV(Y,∇f),(2.36)
for any vector X,Y.
It is necessary to remark that h∇f,∇f=h∇f, so that ˜H∇f(f) =˜H(f) for any
function fonM.
With the assistance of the S-curvature, one can present the defi nition of the
weighted Ricci curvature as the following.
Definition 2.5. ([27][34]).Definition 2.5. ([27][34]). Given a unit vector V∈TxMand an positive integral
numberk, the weighted Ricci curvature is defined by
Rick(V) :=
Ric(x,V)+˙S(x,V) if S(x,V) = 0and k=n or k =∞
−∞ if S(x,V)̸= 0and if k =n
Ric(x,V)+˙S(x,V)−S2(x,V)
k−nif n < k < ∞
(2.37)
where the derivative is taken along the geodesic started fro mxin the direction of V.
AccordingtothedefinitionofweightedRiccicurvature,B.Wudefin edtheweighted
flag curvature when k=N∈(1,n)∪(n,∞) in [39]. We have completely introduced
this concept for any kin [32].
Definition 2.6. ([32]). | 2403.00002v1.pdf | c29d870e-a598-4746-bdfc-87fc5a039804 |
Definition 2.6. ([32]). Let(M,F,dµ)be a Finsler metric measure space, and V,W∈
TxMbe linearly independent vectors. The weighted flag curvatur eKk(V,W)is defined
by
Kk(V,W) :=
K(V;W)+˙S(V)
(n−1)F2(V)if S(x,V) = 0and k=n or k =∞;
−∞ if S(x,V)̸= 0and if k =n;
K(V;W)+˙S(V)
(n−1)F2(V)−S2(V)
(n−1)(k−n)F2(V)if n < k < ∞,
(2.38)
where the derivative is taken along the geodesic started fro mxin the direction of V.
Moreover, it has also been defined the mixed weighted Ricci curvatu re in [32],
denoted bymRick.
9 | 2403.00002v1.pdf | dbd4e6b2-0851-4ea9-897e-c398c65caee4 |
Definition 2.7. ([32])Given two unit vectors V,W∈TxMand an positive integral
numberk, the mixed weighted Ricci curvaturemRick(V,W) =mRick
W(V)is defined by
mRick
W(V) :=
trWRV(V)+˙S(x,V)if S(x,V) = 0and k=n or k =∞;
−∞ if S(x,V)̸= 0and if k =n;
trWRV(V)+˙S(x,V)−S2(V)
k−nif n < k < ∞,
(2.39)
where the derivative is taken along the geodesic started fro mxin the direction of V,
andtrWRV(V) =gij(W)gV(RV(ei,V)V,ej)means taking trace of the flag curvature
with respect to g(x,W).
Remark 2.1.Remark 2.1. The weighted Ricci curvature is a special case of the mixed we ighted
Ricci curvature, i.e., Rick(V) =mRick
V(V). Defining the function ctc(r)as
ctc(r) =
√ccot√cr, c > 0,
1/r, c = 0,√−ccoth√−cr c <0.(2.40)
the following weighted Hessian comparison theorem is cited from Theorem 3.3 in [ 39].
Theorem 2.1. ([32]). | 2403.00002v1.pdf | 7d51f15c-fb9c-44d0-9f54-5275c858bdd9 |
Theorem 2.1. ([32]). Let(M,F,dµ)be a forward complete n-dimensional Finsler
metric measure space with finite misalignment α. Denote the forward distance function
byrand byVa fixed vector field on M. Suppose the mixed weighted Ricci curvature
RicN(V,∇r)ofMis bounded from below by −KwithK >0, for some N > n, as well
as the non-Riemannian curvatures U,TanddivC(V) =Cij
k|i(V)Vkδ
δxjsatisfy the
norm bounds by F(U)+F(T)+F∗(divC(V))≤K0. Then, by setting l=K/C(N,α)
withC(N,α) =N+(α−1)n−α, wherever risC2, the nonlinear Laplacian of rwith
reference vector Vsatisfies
∆Vr≤C(N,α)ctl(r)+√αK0. (2.41)
3 Global gradient estimates on compact Finsler
manifolds
.(2.41)
3 Global gradient estimates on compact Finsler
manifolds
.
Inthissection,wewillfocusontheglobalgradientestimatesofpos itivesolutionsto
the Fisher-KPP equation ( 1.1) on compact Finsler manifolds, with the weighted Ricci
curvature bounded from below. This condition of curvature is exte nsively employed
within the realm of Finsler geometric analysis.
Supposethat uisapositivesolutionon M×[0,∞)totheFisher-KPPequation( 1.1)
. We define the function W(x,t) =u−q, whereqis a positive constant to be fixed
later. We now have the following lemma.
10 | 2403.00002v1.pdf | 3e89f6be-c94e-49d9-b4aa-f48a02518103 |
Lemma 3.1. Letube a positive solution on M×[0,∞)to the Fisher-KPP equation
(1.1) andW(x,t) =u−q, thenWsatisfies that
(∆∇u−∂t)W=q+1
qF2(∇W)
W+cqW−cqWq−1
q. (3.1)
Proof.The gradient of Wwith respect to ∇uis the dual of d Wby the pull-back
metricg∇u. Precisely, ∇W=−qu−q−1∇uonMandWt=−qu−q−1ut, so we have
F2
∇u(∇W) =q2u−2q−2F2
∇u(∇u). (3.2)
tr∇u(∇∇u)2W=q(q+1)u−q−2F2
∇u(∇u)−qu−q−1tr∇u(∇∇u)2u.(3.3)
Wt
W=−qut
u. (3.4)
Then we deduce that in the distributional sense on Mu,
∆∇uW=divµ(∇∇uW) =e−φ∂
∂xi(eφgijWi)
=tr∇u(∇∇u)2W−S(∇W)
=q(q+1)u−q−2F2
∇u(∇u)−qu−q−1∆∇uu.(3.5)
wherever ∇u̸= 0 Combine with ( 1.1) and (3.2), we have
∆∇uW=q+1
qF2
∇u(∇W)
W−qu−q−1[ut−cu(1−u)]
=q+1
qF2
∇u(∇W)
W+Wt+cqW−cqWq−1
q.(3.6)
After the transformation, we obtain the Lemma 3.1.
Now, we follow the line in [ 19]. | 2403.00002v1.pdf | 4ab11559-7232-4b44-982d-a34ea14fb349 |
Now, we follow the line in [ 19]. Define three functions as the follow.
H0(x,t) =F2
∇u(∇W)
W2+ac(1−W−1
q),
H1=Wt
W,
H=H0+βH1,(3.7)
wherea,βare two positive constants to be determined later. Direct calculatio ns
provide that
∇∇uH0=∇∇uF2
∇u(∇∇uW)
W2−2F2
∇u(∇∇uW)∇∇uW
W3+ac
qW−q+1
q∇∇uW,(3.8)
11 | 2403.00002v1.pdf | 31e272d3-c4aa-4c7f-951c-e92f357c3b34 |
and
tr∇u(∇∇u)2H0=2
W2[dW(∇∇u(tr∇u(∇∇u)2W))+Ric(∇∇uW)+(∇∇u)2W2
HS(∇u)]
−8
W3g∇u((∇∇u)2W,∇∇uW⊗∇∇uW)+6F4
∇u(∇∇uW)
W4
−2F2
∇u(∇∇uW)tr∇u(∇∇u)2W
W3−ac(q+1)
q2W−1
q
W2F2
∇u(∇∇uW)
+ac
qW−q+1
qtr∇u(∇∇u)2W,
(3.9)
onMu, where we have employed the Ricci-type identity in [ 33].On the other hand, it
satisfies that
∇∇u(S(∇∇uW)) =˙S(∇∇uW)+g∇u((∇∇u)2W,∇∇uτ⊗∇∇uW),(3.10)
and
∇∇u(tr∇u(∇∇u)2W) =∇∇u(∆∇uW)+˙S(∇∇uW)+g∇u((∇∇u)2W,∇∇uτ⊗∇∇uW).
(3.11)
Plugging ( 3.10) and (3.11) into (3.9) yields
tr∇u(∇∇u)2H0=2
W2[dW(∇∇u(∆∇uW)+g∇u((∇∇u)2W, | 2403.00002v1.pdf | ce967fc3-c09e-4a51-ae5a-41e436107fcd |
∇∇uτ⊗∇∇uW)
+˙S(∇∇uW)+Ric(∇∇uW)+(∇∇u)2W2
HS(∇u)]
+6F4
∇u(∇∇uW)
W4−2F2
∇u(∇∇uW)(∆∇uW+S(∇∇uW))
W3
−8
W3g∇u((∇∇u)2W,∇∇uW⊗∇∇uW)
−ac(q+1)
q2W−1
q
W2F2
∇u(∇∇uW)+ac
qW−q+1
q(∆∇uW+S(∇∇uW)),
(3.12)
wherever ∇u̸= 0.(3.12)
wherever ∇u̸= 0. By a basic computation of Finsler geometry, we could get
∆∇uH0=divµ(∇∇H0) =tr∇u(∇∇u)2H0−dτ(∇∇uH0)+2C∇u
∇2u(∇∇uH0),(3.13)
where we have already employed the fact that C(∇u,·,·) = 0 and C∇u
∇2u(∇∇uH0) =
uk
|iCij
k(∇u)Hj. The notation “ |” denotes the horizontal derivative with respect to the
12 | 2403.00002v1.pdf | 1bff5c80-4c35-4a74-97a4-72c7ccbf2bcd |
Chern connection in the direction ∇u. It follows from
C∇u
∇2u(∇∇uH0) =uk
|iCij
k(∇u)[2WlWl|j−2F2(∇W)W−3Wj+1
qacW−q+1
qWj]
=2uk
|i
W2[(Cij
kWj)|lWl−Cij
k|lWjWl]
=−2uk
|i
W2Lij
kWj
=0,(3.14)
that (3.13) is equal to
∆∇uH0=divµ(∇∇uH0) =tr∇u(∇∇u)2H0−dτ(∇∇uH0). (3.15)
Combining it with ( 3.12), we could find that
∆∇uH0=2
W2[g∇u(∇∇u(∆∇uW),∇∇uW)+Ric∞(∇W)+(∇∇u)2W2
HS(∇u)]
−8
W3g∇u((∇∇u)2W,∇∇uW⊗∇∇uW)−2F2
∇u(∇∇uW)(∆∇uW)
W3
+6F4
∇u(∇∇uW)
W4−ac(q+1)
q2W−1
q
W2F2
∇u(∇∇uW)+ac
qW−q+1
q∆∇uW,
(3.16)
by noticing that
dτ(∇∇uH) = 2dτ[g∇u((∇∇u)2W, | 2403.00002v1.pdf | 041f292f-23f0-4668-9a5c-5fa259d6e9c5 |
∇∇uW)
W2−F2
∇u(∇∇uW)
W3∇∇uW+ac
2qW−q+1
q∇∇uW].
Therefore, H0(x,t) satisfies that
∂H0(x,t)
∂t=∂tF2
∇u(∇W)
W2−2F2
∇u(∇∇uW)Wt
W3+ac
qW−q+1
qWt. (3.17)
Because of ∇Wt= (∇W)t, it follows for ∇∇uH1that
∇∇uH1=∇(Wt
W) =∇Wt
W−Wt∇W
W2. (3.18)
It asserts by direct caiculation that
∂
∂t(∆∇uW) = ∆∇uWt, (3.19)
13 | 2403.00002v1.pdf | 5411f350-f8a5-49f0-951f-146bcf563063 |
arccording to [ 32]. Thus, for H1, one may find that
∆∇uH1=W(∆∇uW)t−2(∇∇uW)t∇∇uW−∆∇uWWt
W2+2WtF2
∇u(∇∇uW)
W3,
(3.20)
so that
(
∆∇u−∂t)
H1=2
qg∇u(∇∇uH1,∇∇ulogW)+cW−1
q−1Wt.(3.21)
Moreover, one could calculate that
2
W2g∇u(∇∇u(∆∇uW),∇∇uW)−1
W2∂tF2(∇W)
=4(q+1)
qg∇u((∇∇u)2W,∇∇uW⊗∇∇uW)
W3
+2cF2
∇u(∇∇uW)
W2[q−(q−1)W−1
q]−2(q+1)
qF4
∇u(∇∇uW)
W4,(3.22)
−2F2
∇u(∇∇uW)
W3(∆∇u−∂t)W=−2(q+1)
qF4
∇u(∇∇uW)
W4−2cqF2
∇u(∇∇uW)
W2
+2cqW−1
qF2
∇u(∇∇uW)
W2,
(3. | 2403.00002v1.pdf | ea2453bf-4ea1-441b-8b9f-5495a47cd327 |
(3.23)
as well as
ac
qW−q+1
q(∆∇u−∂t)W=ac(q+1)
q2F2
∇u(∇∇uW)
W2W−1
q+ac2W−1
q−ac2W−2
q,
(3.24)
Plugging ( 3.22)-(3.24) into (3.16) and (3.17) yields
(
∆∇u−∂t)
H0=4(1−q)
qg∇u((∇∇u)2W,∇∇uW⊗∇∇uW)
W3+2q−2
qF4
∇u(∇∇uW)
W4
+2∥(∇∇u)2W∥2
HS(∇u)
W2+ac2W−1
q(1−W−1
q)+2Ric∞(∇∇uW)
W2
+2cW−1
qF2
∇u(∇∇uW)
W2.
(3.25)
It asserts from the h¨ older inequality that
2g∇u((∇∇u)2W,∇∇uW⊗∇∇uW)
W3≤ε∥(∇∇u)2W∥2
HS(∇u)
W2+1
εF4
∇u(∇∇uW)
W4.(3.26)
14 | 2403.00002v1.pdf | 7545a325-e5e2-487e-a949-50829607c105 |
Thus, when 0 < ε <1, it satisfies that
4(1−q)
qg∇u((∇∇u)2W,∇∇uW⊗∇∇uW)
W3+2(∇∇u)2W2
HS(∇u)
W2+2(q−2)
qF4
∇u(∇W)
W4
≥4
qg∇u((∇∇u)2W,∇∇uW⊗∇∇uW)
W3+2(1−ε)(∇∇u)2W2
HS(∇u)
W2
+2(1−1
ε−2
q)F4
∇u(∇∇uW)
W4.(3.27)
According to the equality(a+b)2
n=a2
N−b2
N−n+N(N−n)
n(a
N+b
N−b)2, for any N > n,
the following inequality holds by substituting a,bandNby ∆∇uW,˙S(∇∇uW) and
N−ε(N−n), respectively. Namely,
(∇∇u)2W2
HS(∇u)≥(tr∇u(∇∇u)2W)2
n=(∆∇uW+S(∇∇uW))2
n
≥(∆∇uW)2
N−ε(N−n)−(S(∇∇uW))2
(1−ε)(N−n),(3. | 2403.00002v1.pdf | 25f64273-3e14-47f0-8d36-6b61bd45d44c |
(3.28)
so that
2(1−ε)(∇∇u)2W2
HS(∇u)
W2+2Ric∞(∇∇uW)
W2
≥2(1−ε)
N−ε(N−n)(∆∇uW)2
W2−2S((∇∇uW))2
(N−n)W2+2Ric∞(∇∇uW)
W2
≥2(1−ε)
N−ε(N−n)(∆∇uW)2
W2+2RicN(∇∇uW)
W2.(3.29)
Moreover, it’s easy to infer that
g∇u(∇∇uH0,∇∇ulogW) =2g∇u((∇∇u)2W,∇∇uW⊗∇∇uW)
W3−2F4
∇u(∇∇uW)
W4
+acW−1
q
qF2
∇u(∇∇uW)
W2.
(3.30)
It is deduced from ( 3.25) by combining ( 3.27)-(3.30) and employing the definition
of weighted Ricci curvature that
15 | 2403.00002v1.pdf | 7d960852-38e9-48fa-9ae5-2f4dcde9f02f |
(
∆∇u−∂t)
H0≥2(1−ε)
N−ε(N−n)(∆∇uW)2
W2+2RicN(∇∇uW)
W2
+2
qg∇u(
∇∇uH0,∇∇ulogW)
+2(
1−1
ε)F4
∇u(
∇∇uW)
W4
+ac2W−1
q(1−W−1
q)+2cF2
∇u(∇∇uW)
W2W−1
q(
1−a
q2)
,(3.31)
onMu. | 2403.00002v1.pdf | 6f895c6b-52a0-4b95-a674-7f73fd5158d4 |
Letβ=a
q, combining with ( 3.21), we have
(
∆∇u−∂t)
H≥2(1−ε)
N−ε(N−n)(∆∇uW)2
W2−2(1
ε−1)F4
∇u(
∇∇uW)
W4
+2
qg∇u(
∇∇uH,∇∇ulogW)
+2c(
1−a
q2)F2
∇u(∇∇uW)
W2W−1
q
+ac2W−1
q(1−W−1
q)+ac
qW1
q−1Wt+2RicN(∇∇uW)
W2.(3.32)
Noticing∆∇uW
W=q
aH+(q+1
q−q
a)F2
∇u(∇∇uW)
W2, and setting a=sq2, we get that
∆∇uW
W=1
sqH+(q+1
q−1
sq)F2
∇u(∇∇uW)
W2=1
sqH+(q+1−1
s
q)F2
∇u(∇∇uW)
W2.
(3.33)
Plugging it into ( 3.32), we arrive at the following lemma.
Lemma 3.2. Let(M,F,µ)be a forward complete Finsler metric measure space, and
denoteMu={x∈M| ∇u(x)̸= 0}. ForH=F2
∇u(∇∇uW)
W2+ac(1−W−1
q)+βWt
Wwith
a=sq2,β=a
q,β=a
q, it satisfies on Muthat
(
∆∇u−∂t)
H≥[2(1−ε)
N−ε(N−n)(sq+s−1)2
s2q2−2(1
ε−1)]F4
∇u(∇∇uW)
W4
+2(1−ε)
N−ε(N−n)1
s2q2H2+4(1−ε)
N−ε(N−n)sq+s−1
s2q2HF2
∇u(
∇∇uW)
W2
+2
qg∇u(
∇∇uH,∇∇ulogW)
+(c−2cs)F2
∇u(∇∇uW)
W2W−1
q
+cW−1
qH+2RicN(∇∇uW)
W2.
(3.34)
16 | 2403.00002v1.pdf | 89673594-1fe8-483a-939b-a35b03918139 |
Employing Lemma 3.2, the following global gradient estimate theorem could be
obtained by utilizing the maximum principle argument, which was first ad opted in
[27] and was applied also in [ 32][31].
Theorem 3.1. Let(M,F,µ)be a compact Finsler metric measure manifold whose
weighted Ricci curvature satisfies RicN≥ −K, for some positive constant K. Assume
the bound of the reversibility on Misρ0. Suppose u(x,t)is a bounded positive smooth
solution of the Fisher-KPP parabolic equation ( 1.1) onM×[0,∞), then we have
F2(∇u)
u2+sc(1−u)−sut
u≤[N−ε(N−n)]
2(1−ε)(s−1)s2K+√
N−ε(N−n)
2(1−ε)s
qM1c.(3.35)
where,0< ε <1,s >1,q >0, such that2(1−ε)
N−ε(N−n)s−1
sq≥1
ε−1 +(2s−1)2
8, and
M1= supu(x,t).
Proof.Lemma3.2implies that ( 3.34) holds in the distributional sense on M. For any
nonnegative test function ϕ, we have
−∫T
0∫
M[dϕ(∇H)−ϕtH]dµdt≥∫T
0∫
Mϕβdµdt. (3.36)
whereβdenotes the RHS of ( 3.34).
Assumey0= (x0,t0)∈B(p,2R)×[0,∞) be the point where Hachieves its maxi-
mum. Without loss of generality, we could assume H(x0,t0)≥0, otherwise the result
will be satisfied trivially. We claim that β(x0,t0)≤0. | 2403.00002v1.pdf | 69552c65-7f97-416c-b0d8-10f9c9cc58c1 |
We claim that β(x0,t0)≤0.
Otherwise, βis strictly positive at ( x0,t0), so that β(x0,t0) is positive in a small
neighborhood of ( x0,t0) onM, which may be denoted by U. Chosen a test function ϕ
whose compact support Vis contained in U, we know from ( 3.36) thatHis a weak,
local subharmonic function in a neighborhood V⊂U.It is a contradiction because
(x0,t0) is a inner point of V.
Because of β(x0,t0)≤0 and∇∇uH(y0) = 0, we arrive at
0≥2(1−ε)
N−ε(N−n)1
s2q2H2+[2(1−ε)
N−ε(N−n)(sq+s−1)2
s2q2−2(1
ε−1)]F4
∇u(∇∇uW)
W4
+4(1−ε)
N−ε(N−n)sq+s−1
s2q2HF2
∇u(∇∇uW)
W2+(c−2cs)F2
∇u(∇∇uW)
W2W−1
q
+cW−1
qH+2RicN(∇∇uW)
W2.
(3.37)
17 | 2403.00002v1.pdf | 34fe1f10-f454-4354-b401-715428b0e792 |
It follows from the h¨ older inequality that
2RicN(∇∇uW)
W2≥ −2Kρ2
0F2
∇u(∇∇uW)
W2
≥ −[2(1−ε)(s−1)2
N−ε(N−n)F4
∇u(∇∇uW)
s2q2W4+(N−ε(N−n))s2q2
2(1−ε)(s−1)2ρ4
0K2]
,
(3.38)
and
(c−2cs)F2
∇u(∇∇uW)
W2W−1
q≥ −[(2s−1)2
4F4
∇u(∇∇uW)
W4+W−2
qc2]
≥ −[(2s−1)2
4F4
∇u(∇∇uW)
W4+M2
1c2]
,(3.39)
whereM1= supu(x). | 2403.00002v1.pdf | 1223a8ac-1473-4aef-819b-0bb9150d05cb |
Now, let q >0 such that2(1−ε)
N−ε(N−n)s−1
sq≥1
ε−1 +(2s−1)2
8.
Combining with ( 3.37), we get
2(1−ε)
N−ε(N−n)1
s2q2H2≤(N−ε(N−n))s2q2
2(1−ε)(s−1)2ρ4
0K2+M2
1c2
−4(1−ε)(sq+s−1)
[N−ε(N−n)]s2q2HF2
∇u(∇∇uW)
W2−cW−1
qH
≤(N−ε(N−n))s2q2
2(1−ε)(s−1)2ρ4
0K2+M2
1c2.(3.40)
Hence
H≤[N−ε(N−n)]
2(1−ε)(s−1)s2q2ρ2
0K+√
N−ε(N−n)
2(1−ε)M1csq. (3.41)
we achieve the global gradient estimates on compact Finsler manifold s.
Remark 3.1. Theorem 1.1follows by taking ε=1
2,s= 2,q=2
27(N+n)in Theorem
3.1.
4 Local gradient estimates on forward complete
Finsler spaces
In this section, we prove the local gradient estimates on forward c omplete Finsler
metric measure spaces with the assistance of Lemma 3.2and the Comparison theorem
(cf. Theorem 2.1).
Theorem 4.1. Let(M,F,µ)be a complete noncompact Finsler metric measure space.
Denote by B(p,2R)the forward geodesic ball centered at pwith forward radius 2R.Suppose the mixed weighted Ricci curvaturemRicN≥ −K(2R)inB(p,2R)with
18 | 2403.00002v1.pdf | 4304fedf-48a2-420a-8f65-d767042f8313 |
K(2R)≥0, and the misalignment αsatisfies α≤A(2R)inB(p,2R). Moreover, the
non-Riemannian tensors satisfy F(U)+F∗(T)+F(divC(V))≤K0.Assume u(x,t)
is a bounded positive smooth solution of the Fisher-KPP para bolic equation ( 1.1) on
M×[0,∞), then we have
F2
∇u(∇u)
u2+sc(1−u)−sut
u
≤N−ε(N−n)
2(1−ε)s2
t+N−ε(N−n)
2(1−ε)(s−1)s2K(2R)+sc
q√
N−ε(N−n)
2(1−ε)M1
+[N−ε(N−n)]s2
2(1−ε)R2{
C3[1+R(1+√
K(2R))]+[N−ε(N−n)]s2
4(1−ε)(s−1)C2
1+2C2
1}
.(4.1)
onB(p,R)for0< ε <1,s >1,q >0, such that2(1−ε)
N−ε(N−n)s−1
sq≥1
ε−1+(2s−1)2
8,
whereC1,C2are positive constants.
Proof.In noncompact situation, we choose the cut-off function ˜ ϕ∈C2[0,+∞), such
that {
˜ϕ(r) = 1r≤1
˜ϕ(r) = 0r≥2.(4.2)
So ˜ϕ(r)∈[0,1]. In addition, we supposed that
−C1≤˜ϕ′(r)
˜ϕ1
2(r)≤0,−C2≤˜ϕ′′(r), (4.3)
whereC1,C2are positive constants.
For a fixed point p, denote by r(x) the forward distance function from pto any
pointx. | 2403.00002v1.pdf | 4097fd8a-8658-453f-9c95-c29100d9a937 |
We define the cut-off function by ϕ(x) = ˜ϕ(r(x)
R). So that
F∇u(∇∇uϕ)≤√αC1
Rϕ1
2. (4.4)
According to the curvature conditions and the Laplacian compariso n theorem on
forward complete Finsler manifolds, it satisfies the Laplacian estimat ion [32] that
∆∇uϕ=˜ϕ′∆∇ur
R+˜ϕ′′F2
∇u(∇∇ur)
R2
≥ −C1
R[
C(N,A)√
K(2R)
C(N,A)coth(
R√
K(2R)
C(N,A))
+C0(K0,A)]
−αC2
R2
≥ −C3
R2[1+R(1+√
K(2R))],
(4.5)
19 | 2403.00002v1.pdf | 78b44975-b0f7-486a-86ec-98c82e3d40b7 |
whereC3=C3(N,A,K 0) is a constant depending on K0,NandA.
Defining Z=tH(x,t), we suppose that the support of function ϕ(x)Z(x,t) is
contained in Bp(2R). For any fixed T >0, let (x0,t0) be the point where ϕ(x)Z(x,t)
achieves its positive maximum, at which it satisfies that
∇(ϕZ) = 0,∂(ϕZ)
∂t≥0,∆(ϕZ)≤0. (4.6)
(4.6) implies that
∇Z=−∇ϕ
ϕZ, (4.7)
and
∆ϕ·Z+2g∇u(∇∇uϕ,∇∇uZ)+ϕ(∆−∂
∂t)Z≤0, (4.8)
at the maximum point ( x0,t0). Letβ=a
q,a=sq2, employing Lemma 3.2,Letβ=a
q,a=sq2, employing Lemma 3.2, we have
(∆−∂
∂t)Z=t(∆−∂
∂t)H−H
≥[2(1−ε)(sq+s−1)2
N−ε(N−n)s2q2−2(1
ε−1)]F4
∇u(∇∇uW)
W4t
+2(1−ε)
N−ε(N−n)1
s2q2Z21
t+4(1−ε)
N−ε(N−n)sq+s−1
ε2q2F2
∇u(∇∇uW)
W2Z
+2
qg∇u(∇∇uZ, | 2403.00002v1.pdf | 2d0a5b1d-0853-436e-9f9e-63bbf9cf984f |
∇∇ulogW)+cW−1
qZ+(c−2cs)F2
∇u(∇∇uW)
W2W−1
qt
+cW−1
qZ−2K(2R)F2
∇u(∇∇uW)
W2t−Z
t.
(4.9)
It follows from the h¨ older inequality that
2K(2R)tF2
∇u(∇∇uW)
W2≤2(1−ε)
N−ε(N−n)(s−1)2
s2q2F4
∇u(
∇∇uW)
W4t
+N−ε(N−n)
2(1−ε)s2q2
(s−1)2K2t,(4.10)
and
(c−2cs)F2
∇u(∇∇uW)
W2W−1
qt≥ −((2s−1)2
4F4
∇u(∇∇uW)
W4t+c2M2
1t)
.(4.11)
20 | 2403.00002v1.pdf | adb3b68f-c689-457c-a3bf-ef4049882630 |
Substituting ( 4.10) and (4.11) into (4.9), and choosing s >1,q >0, such that
2(1−ε)
ns−1
sq≥1
ε−1+(2s−1)2
8, one may find that
(∆−∂
∂t)Z≥2(1−ε)
N−ε(N−n)1
s2q2Z21
t+4(1−ε)
N−ε(N−n)sq+s−1
s2q2F2
∇u(
∇∇uW)
W2Z
+2
qg∇u(
∇∇uZ,∇∇ulogW)
−Z
t−N−ε(N−n)
2(1−ε)s2q2
(s−1)2K2t−c2M2
1t.(4.12)
Substituting ( 4.12) into (4.8) and noticing ( 4.7), we have
2(1−ε)
N−ε(N−n)1
s2q2Z21
tϕ+4(1−ε)
N−ε(N−n)sq+s−1
s2q2F2
∇u(
∇∇uW)
W2Zϕ
−2Z
qg∇u(∇∇uϕ,∇∇uW
W)−Zϕ
t−N−ε(N−n)
2(1−ε)s2q2
(s−1)2K2(2R)ϕt
−C2M2
1tϕ−C3
R2[1+R(1+√
K(2R))]Z−2αC2
1
R2Z≤0. | 2403.00002v1.pdf | 356a7c0a-7b0e-4231-9276-90409fc3faa3 |
(4.13)
It’s easy to know that
2Z
qg∇u(∇∇uϕ,∇∇uW
W)≤4(1−ε)
N−ε(N−n)sq+s−1
s2q2F2
∇u(∇W)
W2Zϕ
+N−ε(N−n)
4(1−ε)s2Z
sq+s−1F2
∇u(∇∇uϕ)
ϕ.(4.14)
Multiplying through by tϕat (4.13) and utilizing ( 4.14), we get that
2(1−ε)
N−ε(N−n)1
s2q2Z2ϕ2−Zϕ−t{
2αC2
1
R2
+N−ε(N−n)
4(1−ε)s2
(sq+s−1)αC2
1
R2+C3
R2[1+R(1+√
K(2R))]}
Zϕ
−t2[N−ε(N−n)
2(1−ε)s2q2
(s−1)2K2+C2M2
1]
≤0.(4.15)
21 | 2403.00002v1.pdf | 3e7f6ccf-bdc2-494e-9a8e-5e7f41b3462b |
which implies that
Zϕ≤[N−ε(N−n)]s2q2
2(1−ε){
1+t(
N−ε(N−n)
4(1−ε)s2
(sq+s−1)C2
1
R2
+C3
R2[1+R(1+√
K(2R))]+2C2
1
R2)}
+t√
[N−ε(N−n)]s2q2
2(1−ε)√
[N−ε(N−n)]s2q2
2(1−ε)(s−1)2AK2+c2M2
1,(4.16)
is satisfied at ( x0,t0).Clearly, corresponding to the assumption of ( x0,t0), one can deduce that
sup
x∈BpZ(x,T)≤Z(x0,t0)ϕ(x0)
≤[N−ε(N−n)]s2q2
2(1−ε){
1+T(
N−ε(N−n)
4(1−ε)s2
(sq+s−1)C2
1
R2+2C2
1
R2
+C3
R2[1+R(1+√
K(2R))])}
+T√
[N−ε(N−n)]s2q2
2(1−ε)√
[N−ε(N−n)]s2q2
2(1−ε)(s−1)2AK(2R)2+c2M2
1.(4.17)
Recall a result that, if a1,a2,a3≥0 then√a1√a2+a3≤√a1a2+√a1a3. | 2403.00002v1.pdf | 461531e3-2610-4d79-8669-ca9870c4fe45 |
Thus,
F2
∇u(∇u)
u2+sc(1−u)−sut
u
≤N−ε(N−n)
2(1−ε)s2
t+N−ε(N−n)
2(1−ε)(s−1)s2K(2R)+sc
q√
N−ε(N−n)
2(1−ε)M1
+[N−ε(N−n)]s2
2(1−ε)R2{
C3[1+R(1+√
K(2R))]+[N−ε(N−n)]s2
4(1−ε)(s−1)C2
1+2C2
1}
,
(4.18)
onB(p,2R), which is the desired inequality.
Remark 4.1. Theorem 1.2follows from Theorem 4.1directly.
References
[1] G. Albanese and M. Rigoli. Lichnerowicz-type equations on complet e manifolds.
Adv. Nonlinear Anal, 5(3), 223–250, 2016.
22 | 2403.00002v1.pdf | ed20d3bc-eead-45b5-b130-5ce5002ffb54 |
[2] C. Atkinson and G. Reuter. Deterministic epidemic waves. Math. P roc. Cam-
bridge Philos. Soc., 80(2), 315–330, 1976.
[3] M. Bailesteanu. Gradient estimates for the Heat equation under the Ricci-
Harmonic map flow. Adv. Geom., 15(4), 445–454, 2015.
[4] M. Bailesteanu, X. Cao, and A. Pulemotov. Gradient estimates fo r the Heat
equation under the Ricci flow. J. Funct. Anal., 258(10), 3517–354 2, 2010.
[5] K. Ball, E. A. Carlen, and E. H. Lieb. Sharp uniform convexity and s moothness
inequalities for trace norms. Invent. Math., 115, 463–482, 1994.
[6] P. W. Bates, X. Chen, and A. J. Chmaj.Traveling waves of bistab le dynamics on
a lattice. Sima. J. Math. Anal., 35(2), 520–546, 2003.
[7] K. Brown and J. Carr. Deterministic epidemic waves of critical velo city. Math.
Proc. Cambridge Philos. Soc., 81(3), 431–433, 1977.
[8] X. Cao. Differential Harnack estimates for backward Heat equa tions with
potentials under the Ricci flow. J. Funct. Anal., 255(4), 1024–103 8, 2008.
[9] X. Cao and R. S. Hamilton. Differential Harnack estimates for time -dependent
Heat equations with potentials. Geom. Funct. Anal., 19, 989–1000, 2009.
[10] X. Cao, B. Liu, I. Pendleton, and A. Ward. Differential Harnack estimates for
Fisher’s equation.Differential Harnack estimates for
Fisher’s equation. Pac. J. Math., 290(2), 273–300, 2017.
[11] X. Chen and J. Guo. Existence and asymptotic stability of trave ling waves of
discrete quasilinear monostable equations. J. Differ. Equations, 18 4(2), 549–569,
2002.
[12] X. Chen and J. Guo. Uniqueness and existence of traveling wave s for discrete
quasilinear monostable dynamics. Math. Ann., 326(1), 123–146, 20 03.
[13] S. Cheng and S. -T. Yau. Differential equations on Riemannian ma nifolds and
their geometric applications. Commun. Pure Appl. Math., 28(3), 333 –354, 1975.
[14] S.-N. Chow, J. Mallet-Paret, and W. Shen. | 2403.00002v1.pdf | 42345d05-d7d7-4b4d-9a5b-cfcbb259bb13 |
Chow, J. Mallet-Paret, and W. Shen. Traveling waves in latt ice dynamical
systems. J. Differ. Equations, 149(2), 248–291, 1998.
[15] O. Diekmann. Thresholds and traveling waves for the geograph ical spread of
infection. J. Math. Biol., 6(2), 109–130, 1978.
[16] O. Diekmann and H. G. Kaper. On the bounded solutions of a nonlin ear
convolution equation. Nonlinear Anal., 2(6), 721–737, 1978.
[17] X. Geng and S. Hou. Gradient estimates for the Fisher–KPP equ ation on
Riemannian manifolds. Boundary Value Probl., 2018(1), 1–12, 2018.
23 | 2403.00002v1.pdf | d223289e-fe18-47db-bfd5-1402f20c1074 |
[18] M. Ghergu and V. Radulescu. Nonlinear PDEs: Mathematical mod els in biol-
ogy, chemistry and population genetics. Springer Monographs in Ma thematics.
Springer, Heidelberg, 2011.
[19] J. Li. Gradient estimates and Harnack inequalities for nonlinear p arabolic and
nonlinear elliptic equations on Riemannian manifolds. J. Funct. Anal., 10 0(2),
233–256, 1991.
[20] P. Li and S.-T. Yau. On the parabolic kernel of the Schr¨ odinge r operator. Acta
Math. 156, 153–201, 1986.
[21] J. Mallet-Paret. The global structure of traveling waves in spa tially discrete
dynamical systems. J. Dyn. Differ. Equations , 11, 49–127, 1999.
[22] J. D. Murray. Mathematical biology: I: An introduction. interd isciplinary applied
mathematics.interd isciplinary applied
mathematics. J. Math. Biol., Springer, New York, 2002.
[23] J. D. Murray. Mathematical biology: II: Spatial models and biom edical applica-
tions. J. Math. Biol., Springer, New York, 2003.
[24] E. Negrin. Gradient estimates and a Liouville type theorem for th e Schr¨ odinger
operator. J. Funct. Anal., 127(1), 198–203, 1995.
[25] S. Ohta. Finsler interpolation inequalities. Calc. Var. Partial Diffe r. Equations,
36(2), 211–249, 2009.
[26] S. Ohta. Nonlinear geometric analysis on Finsler manifolds. Eur. J . Math. 3(4),
916–952, 2017.
[27] S. Ohta and K.-T. Sturm.[27] S. Ohta and K.-T. Sturm. Bochner–Weitzenb¨ ock formula and L i–Yau estimates
on Finsler manifolds. Adv. Math., 252, 429–448, 2014.
[28] H.-B. Rademacher. A sphere theorem for non-reversible Finsle r metrics. Math.
Ann., 328(3), 373–387, 2004.
[29] K. W. Schaaf. Asymptotic behavior and traveling wave solutions for parabolic
functional differential equations. Trans. Am. Math. Soc., 302(2) , 587–615, 1987.
[30] K. Schumacher. Traveling-front solutions for integro-differe ntial equations. I.,
Walter de Gruyter, New York, 1980.
[31] B. Shen. Gradient estimates of the Finslerian Allen-Cahn equatio n. preprint.
[32] B. Shen. | 2403.00002v1.pdf | bbdf434e-cbd6-4dd7-a6fa-1b6b64bd94be |
[32] B. Shen. Operators on nonlinear metric measure spaces I: A ne w Laplacian
comparison theorem on Finsler manifolds and a generalized Li-Yau app roach to
gradient estimates of Finslerian Schr¨ odinger equation. arXiv:2312 .06617v1,2023.
24 | 2403.00002v1.pdf | 7ed88134-fe55-4bfd-9d80-de7d794e12fb |
[33] B. Shen. Vanishing of killing vector fields on compact Finsler manifo lds. Kodai
Math. J, 41(1), 1–15, 2018.
[34] Y. Shen and Z. Shen. Introduction to modern Finsler geometry . World Scientific
Publishing Company, 2016.
[35] Z. Shen. Volume comparison and its applications in Riemann–finsler geometry.
Sib. Adv. Math. Siberian, 128(2), 306–328, 1997.
[36] Z. Shen. Lectures on Finsler geometry. World Scientific, Singap ore, 2001.
[37] P. Souplet and Q.-S. Zhang. Sharp gradient estimate and Yau’s L iouville theorem
for the Heat equation on noncompact manifolds. Bull. Lond. Math. S oc., 38(6),
1045–1053, 2006.S oc., 38(6),
1045–1053, 2006.
[38] H. R. Thieme and X.-Q. Zhao. Asymptotic speeds of spread and t raveling waves
for integral equations and delayed reaction–diffusion models. J. Diff er. Equations,
195(2), 430–470, 2003.
[39] B.Wu.ComparisontheoremsinFinslergeometrywithweightedcu rvaturebounds
and related results. J. Korean Math. Soc., 52(3), 603–624, 2015 .
[40] J. Wu and X. Zou. Asymptotic and periodic boundary value proble ms of mixed
fdes and wave solutions of lattice differential equations. J. Differ. E quations ,
135(2), 315–357, 1997.
[41] C. Xia. Local gradient estimate for Harmonic functions on Finsle r manifolds.
Calc. Var. Partial Differ. | 2403.00002v1.pdf | 844dc9bc-83a5-4076-9032-c4c754c0da88 |
Calc. Var. Partial Differ. Equations , 51(3-4), 849–865, 2014.
[42] Q. Xia. Geometric and functional inequalities on Finsler manifolds. J. Geom.
Anal., 30(3), 3099–3148, 2020.
[43] Y. Yang. Gradient estimates for a nonlinear parabolic equation o n Riemannian
manifolds. Proc. Am. Math. Soc., 136(11), 4095–4102, 2008.
[44] S.-T. Yau. Harmonic functions on complete Riemannian manifolds. Commun.
Pure Appl. Math. , 28(2), 201–228, 1975.
25 | 2403.00002v1.pdf | 69131eb6-2a3b-4d43-8f5d-96e04fc98b45 |
arXiv:2402.00001v1 [math.GM] 20 Sep 2023Proof the Collatz Conjecture by a new view of natural numbers
FENG Jishe
(School of Mathematics and Information Engineering, Longd ong University, Qingyang, Gansu, 745000, China.)
Abstract
The two main strategies used in this study are the binary represent ation and the decomposition of a natural
number into many compound functions of odd function and even fun ction. The Collatz conjecture regarding
odd even numbers in number theory can be examined and discussed u sing them in this way: The sequence
createdby the finite iterationsofthe Collatz function becomesthe ultimately periodic sequenceifanynatural
number is the beginning value, proving the conjecture that has bee n held for 85 years.
Keywords: binary representation; composite function; Collatz conjecture; ultimately periodic sequence;
chaos
1. Introduction
In the study of number theory, odd and even numbers are a funda mental pair of ideas. Natural number
sets come in two different varieties.Natural number
sets come in two different varieties. Numerological theory frequen tly examines the connections between
various numbers. There are numerous conjectures that attemp t to generalize the law of different sorts of
natural numbers discovered in a restricted range to the entire infi nite set of natural numbers. This article
will examine the famous Collatz conjecture, which states that for e ach natural number n, if it is even, divide
by 2, if it is odd, multiply by 3, add 1, and so on, the result must finally re ach 1. It is also referred to
as the 3n+ 1 conjecture and was put forth in 1937 by Lothar Collatz, also kno wn as the 3 n+1 problem.
The mathematician Paul Erdos once said of this conjecture: ”Math ematics may not be ready for such
problems”[1,2].
The inconsistencies between the finite and the infinite, as well as the relation between various kinds,
present difficulties in the study of number theory problems. | 2402.00001v1.pdf | 926c5003-4a43-42c4-b22c-67e38da9ef09 |
We are t alking about the connection between
two different mathematical ideas: the iterated sequence is a ultimat ely periodic sequence whether the initial
value is odd or even.
The finite and the infinite can be connected by the useful mathemat ical construct known as a function,
and the resulting outcomes will also be finite and infinite. A special fun ction that has particular significance
in discrete mathematics is the piecewise function. Compound functio ns and piecewise functions combined is
Email address: E-mail: [email protected] (FENG Jishe)
0Received date: 2023-9-11.
0Foundation item: Educational technology innovation proje ct of Gansu Province (No. 2022A-133)
Preprint submitted to Elsevier February 2, 2024 | 2402.00001v1.pdf | 3dd27cd3-908e-40e2-bd6d-b4f586e6abf9 |
a highly clever mathematical trick. Particularly in number theory, wh ich is the most fundamental idea and
mathematical expression, the sequence of numbers is a close conn ection between functions and finite and
infinite. Numerous conjectures are obtained through restricted iteration of an iterative algorithm, which is
a widely used method in number theory, but people typically lack the me ans to demonstrate the accuracy
and reasoned nature of conjectures. A fresh technique or new k nowledge is frequently used to support a
hypothesis. For the Collatz conjecture, we can describe it as a fun ction:
T(n) =
3n+1,ifnis odd number,
n
2ifnis even number.(1.1)
The following sequence is obtained via the composite function (iterat ion):
a={n,T(n),T(T(n)),T(T(T(n))),···}={n,T(n),T2(n),T3(n),···}.Consequently, the Collatz conjecture can be stated as follows: Th e sequence aalways leads to the integer
1, regardless of where you start with the natural number n, name lyTm(n) = 1. The series ais an infinite
sequence of ultimately period[9,10]: the preperiod η(n) varies with the initial value n, but the ultimately
period is always {4,2,1}.
2. Composition of odd and even functions and binary represen tation of natural numbers
If a natural number can be divided by 2, it is said to be an even number ; otherwise, it is said to be
odd number. The Peano’s Axiom states that 1 is the smallest natural number. The set of natural numbers
N={1,2,3,···}may be separated into odd and even sets, we will utilize the standard definition of natural
numbers in this work.
{natural number }={odd number }⋃{even number }.{natural number }={odd number }⋃{even number }.
In the set of natural numbers where 1 is the smallest odd number an d 2 is the smallest even number, we
can use the expression n= 2k−1 to indicate that it is an odd, and the expression n= 2kto indicate that
it is an even, where kis any natural number.
We introduce two functions O(x) = 2x+ 1 to express odd numbers greater than 1, and E(x) = 2xto
express even numbers.
Definition 1 A natural number nis obtained by composition of the odd function O(x) = 2x+1 and
the even function E(x) = 2xseveral times, namely n=f(f(···f(1))) =fk(1), the function fis either odd
function O(x) or even function E(x). | 2402.00001v1.pdf | 99816c42-54d5-429d-a8d9-7d9ec8a220ba |
For example, 7 = 2 ∗3+1= 2 ∗(2∗1+1+1)= O(O(1)),189 = 2∗2∗(2∗4+1)= 2 ∗(2∗(2∗(2∗1))+1)=
E(O(E(E(1)))). In order to more clearly express the odd-even compositio n process of a natural number, we
usebinary representation of a natural number n, for example: 7 = (111) 2, 18 = (10010) 2.
Similarly, we can have a piecewise function
f−1(n) =
n+1
2,ifnis odd number,
n
2,ifnis even number.(2.1)
Ifn=f(f(···f(1))) =fk(1), then the inverse function is f−k(n) = 1.
For generalization, we give the definitions of three natural number s:
2 | 2402.00001v1.pdf | 24c0587e-d46e-47db-9c76-77d636c92eb8 |
Definition 2(i) By applying the odd function O(x)mcompositions, a natural number, namely Om(1) =
2m−1 = 2m−1+2m−2+···+2+1= (11 ···1)2, is obtained. | 2402.00001v1.pdf | 5d46421e-9982-4bad-863e-b9ad5e658284 |
such as 3 = (11) 2,7 = (111) 2,15 = (1111) 2,31 =
(11111) 2,63 = (111111) 2,···, which we call it as pure odd number ;
(ii)By applying the even function E(x)mcompositions, a natural number, namely Em(1) = 2m=
(10···0)2, is obtained, such as, 2 = (10) 2,4 = (100) 2,8 = (1000) 2,16 = (10000) 2,32 = (100000) 2,64 =
(1000000) 2,···, which we call it as pure even number ;
(iii)The natural number obtained by the composition of odd function O(x) and even function E(x), we
call itmixed number . Such as, 18 = (10010) 2,28 = (11100) 2,67 = (1000011) 2,309 = (100110101) 2.In particular, the naturalnumbersobtained bythe finite alternat elycomposition ofthe odd function O(x)
and the even function E(x), namely, [ E(O(1))]m= (101···101)2. Such as 5 = (101) 2,21 = (10101) 2,85 =
(1010101) 2,341 = (101010101) 2,1365 = (10101010101) 2,5461 = (1010101010101) 2,···, which we call hard
number .
Definition 3 The binary string of a natural number is a representation of its odd -even composite
function, where the 1 in the i(i >0)-bit from right to left is the i(i >0) sub-odd function, and 0 is the
corresponding sub-even function.
For a natural number n, if its binary string has kbits, then the degree of composite function is k−1.The binary string of a pure odd number is made of all 1; and the binary string of a pure even number is all
0 aside from one 1 in left; the binary string of a mixed number is made of many 0 and 1. Fig.1 shows the
decomposition of the composite function (inverse of the composite function) of 60.
60=2* 2* 2* 2* 2*1+1）+1）+1））=E(E(O(O(O(1))))=(111100) 2
Figure 2.1: Natural number 60 is obtained from 1 through the c omposition of five even and odd functions, its binary string i s
(111100) 2.In this way, we can classify the set of natural numbers in another w ay,
Property 4 The set of natural numbers can be divided into three different sets :
{natural number }={pure odd number }⋃
{pure even number }⋃
{mixed number }
The pure odd number is an odd number, the pure even number is an ev en number, and mixed numbers
can be either an odd number or an even number when compared to th e conventional classification. The last
bit of the binary string, which is either 0 or 1, indicates whether it is ev en or odd. The entire binary string
implicitly indicates one of the following three types: all bits are 0s with t he exception of one 1 in the left
3 | 2402.00001v1.pdf | b7831281-0846-4916-8505-f78b024a9a93 |
implicitly pure even, all bits are 1s implicitly pure odd, and there are 0 an d 1 in any number of bits of the
binary string implicitly mixed.
Example 1 (1)60, 97 are mixed numbers.
(2)64,1180591620717411303424 are pure even numbers.
(3)63,1180591620717411303423 are pure odd numbers.(3)63,1180591620717411303423 are pure odd numbers.
60 = 2∗30 = 2∗(2∗15) = 2∗(2∗(2∗7+1))) = 2 ∗(2∗(2∗(2∗(2∗3)+1)+1))) = 2 ∗(2∗(2∗(2∗(2∗
(2∗1+1)+1)+1))) = E(E(O(O(O(1)))) = (111100) 2;
97 = 2∗48+1 = 2 ∗(2∗24)+1 = 2 ∗(2∗(2∗12))+1 = 2 ∗(2∗(2∗(2∗6)))+1 = 2 ∗(2∗(2∗(2∗(2∗3)))+1 =
2∗(2∗(2∗(2∗(2∗(2∗1+1))))+1 = O(E(E(E(E(O(1)))))) = (1100001) 2.64 = 26= (1000000) 2,1180591620717411303424= 270= (10000 ···0)2are pure even numbers;
63 = 26−1 = (111111) 2,1180591620717411303423= 270−1 = (11···1)2are pure odd numbers.
This composite function has special significance, for example,
Example 2 Let [x] bexthe integer part, that is, [ x] =max{y∈Z:y≤x},{x}be fractional part,
namely{x}=x−[x]. Assuming g(x) =1
[x]−{x}+1, is rational set Q+and rational set Qcan be expressed
as respectively
Q+={g(0),g2(0),g3(0),···},
Q={0,g(0),−g(0),g2(0),−g2(0),g3(0),−g3(0),···}.
See the references[5]for proof.
3. | 2402.00001v1.pdf | d0e8fd55-7fa3-4e56-a970-e9a57b3f7d9e |
See the references[5]for proof.
3. Tabular and algebraic expressions
For the Collatz conjecture, if expressed by the function T(n), we find that Tk(n) = 1.
For the sake of discussion, we combine the compound function (iter ative relation) to get the reduced
Collatz function
RT(n) =
3n+1
2m,ifnis odd number,
n
2rifnis even number.(3.1)
The result is an odd number. We introduce tabular form and algebraic expression to express the reduced
Collatz function RT(n).
For example, for n= 67 we use the formula (1.1) and iteration, get the following table. Wh ere the last
column is its algebraic expression:.Wh ere the last
column is its algebraic expression:.
67→202→10138
219·6737
219
101→304→152→76→38→1936
218
19→58→2935
214
29→88→44→22→1134
213
11→34→1733
210
17→52→26→1332
29
13→40→20→10→53
27
5→16→8→4→2→11
24
1
4 | 2402.00001v1.pdf | 026e93da-94dc-442f-823c-d1e89020c727 |
Representing the numbers in the table by its binary string to get
1000011→11001010 →110010138
219·6737
219
1100101→100110000 →10011000 →1001100→100110→1001136
218
10011→111010→1110135
214
11101→1011000→101100→10110→101134
213
1011→100010→1000133
210
10001→110100→11010→110132
29
1101→101000→10100→1010→1013
27
101→ 10000→ 1000→ 100→ 10→ 11
24
1
The above table is simplified with the help of the reduced Collatz functio n, and the simplified binary
string table obtained is as follows, and we will use this form as the defa ult table in the rest of this article.1000011→11001010 →110010138
219·6737
219
1100101→100110000 →1001136
218
10011→111010→1110135
214
11101→1011000→101134
213
1011→100010→1000133
210
10001→110100→110132
29
1101→101000→1013
27
101→ 10000→ 11
24
1
In the future, we will use this table as a research tool to write the a lgebraic expression of the last column
in the table:
T27(67) =T(8,19,67) =1
24+3
27+32
29+33
210+34
213+35
214+36
218+37
219+38
219·67 = 1
In general, the algebraic expressions are obtained: Starting from the last row of the table and going up
to the binary corresponding to the initial value n of the first row, th e numerator is 3k,k= 0,1,2,3,···, | 2402.00001v1.pdf | 5e905a15-f590-4ba1-a34e-982607bd360e |
k= 0,1,2,3,···, The
denominator is 2rk,r0=m0,rk=rk−1+mk,k= 1,2,3,···,mkhere for the before k lines at the end of
the second column of binary string number 0. The details are expres sed in the last line of the corresponding
table, and we write out the algebraic expression of them as follows:
5 | 2402.00001v1.pdf | 8193844e-5eef-4514-bb9b-f41a650ebe68 |
T106(31) =1
24+3
29+32
210+33
211+34
214+35
218+36
220+37
222+38
226+39
227
+310
228+311
229+312
232+313
233+314
234+315
235+316
236+317
237+318
239
+319
240+320
242+321
243+322
244+323
247+324
249+325
250+326
251+327
252
+328
254+329
255+330
256+331
258+332
259+333
261+334
263+335
264+336
265
+337
266+338
267+339
267·31
= 1
T107(63)1
T107(63) =1
24+3
29+32
210+33
211+34
214+35
218+36
220+37
222+38
226+39
227
+310
228+311
229+312
232+313
233+314
234+315
235+316
236+317
237+318
239
+319
240+320
242+321
243+322
244+323
247+324
249+325
250+326
251+327
252
+328
254+329
255+330
256+331
258+332
259+333
263+334
264+335
265+336
266
+337
267+338
268+339
268·63
= 1
T118(97)1
T118(97) =1
24+3
29+32
210+33
211+34
214+35
218+36
220+37
222+38
226+39
227
+310
228+311
229+312
232+313
233+314
234+315
235+316
236+317
237+318
239
+319
240+320
242+321
243+322
244+323
247+324
249+325
250+326
251+327
252
+328
254+329
255+330
256+331
258+332
259+333
261+334
263+335
264+336
265
+337
266+338
269+339
270+340
271+341
273+342
275+343
275·97
= 1
T91(10027) | 2402.00001v1.pdf | 58e19e34-3f24-4b50-8a10-e0ebe9bcff38 |
1
T91(10027) =1
24+3
29+32
210+33
211+34
214+35
218+36
220+37
222+38
226+39
227
+310
228+311
229+312
232+313
233+314
234+315
235+316
236+317
237+318
239
+319
242+320
243+321
244+322
245+323
248+324
249+325
253+326
256+327
258
+328
260+329
261+330
261·10027
= 1
6 | 2402.00001v1.pdf | 94a06d0a-4f6f-4651-ac27-9cd00da7da12 |
4. The characteristics of sequence of binary string of the Co llatz function iteration
The Collatz function is expressed in binary form as
T(n) =
(11)2·(1×···×1)2+1 = (1 ×××10···0)2,ifnis odd number,
(1×···×10···00)2
(10)2= (1×···×10···0)2,ifnis even number.(4.1)
The characteristics of the first and last parts are represented b y the figure 2.
1
0
Grows to the left
by two bits 10.The last bit must be 0,and
the penultimate bit does not
change.11y x1
11y x1
0z x 1
1
0
Grows to the left by
one bit 1.100 x1
100 x1
11 x
The last bit must be 0,and
the penultimate bit does not
change.Figure4.1: Binaryrepresentation ofthe Collatzfunction, add up atmosttwo digitsfromthe leftsideofthe binary strin g,subtract
at least one digit from the right side of the binary string.
Byobservingthe binarystringsofiteratedCollatzfunctionswith init ialvaluesof31,63(seethe tabulation
procedure in reference [7]) and 97,10027 as the follows, we get some laws of pure even, pure odd and mixed
numbers with initial values. In the following, we will analyze and discuss the table of the iterative process
of the Collatz functions from the two dimensions of row and column of the tables.
7 | 2402.00001v1.pdf | 248e2d7f-5bd8-4cf3-a3e8-b09546be3392 |
343
275·97
97 = (1100001) 2→ (100100100) 2→ (1001001) 2342
275
73 = (1001001) 2→ (11011100) 2→ (110111) 2341
273
55 = (110111) 2→ (10100110) 2→ (1010011) 2340
271
83 = (1010011) 2→ (11111010) 2→ (1111101) 2339
270
125 = (1111101) 2→ (101111000) 2→ (101111) 2338
269
47 = (101111) 2→ (10001110) 2→ (1000111) 2337
266
71 = (1000111) 2→ (11010110) 2→ (1101011) 2336
265
107 = (1101011) 2→ (101000010) 2→ (10100001) 2335
264
161 = (10100001) 2→ (111100100) 2→= (10100001) 2→ (111100100) 2→ (1111001) 2334
263
121 = (1111001) 2→ (101101100) 2→ (1011011) 2333
261
91 = (1011011) 2→ (100010010) 2→ (10001001) 2332
259
137 = (10001001) 2→ (110011100) 2→ (1100111) 2331
258
103 = (1100111) 2→ (100110110) 2→ (10011011) 2330
256
155 = (10011011) 2→ (111010010) 2→ (11101001) 2329
255
233 = (11101001) 2→ (1010111100) 2→ (10101111) 2328
254
175 = (10101111) 2→ (1000001110) 2→ (100000111) 2327
252
263 = (100000111) 2→ (1100010110) 2→ | 2402.00001v1.pdf | edf17af0-5b9e-443f-aa43-d926729ffaeb |